GRL Thesis

Red Blood Cell mechanics:
from membrane elasticity

to blood rheology
Guillermo Rodríguez Lázaro
Aquesta tesi doctoral està subjecta a la llicència Reconeixement- NoComercial –

CompartirIgual 3.0. Espanya de Creative Commons.
Esta tesis doctoral está sujeta a la licencia Reconocimiento - NoComercial – CompartirIgual

3.0. España de Creative Commons.
This doctoral thesis is licensed under the Creative Commons Attribution-NonCommercial-

ShareAlike 3.0. Spain License.
Red Blood Cell mechanics:
from membrane elasticity to
blood rheology
Ph. D. Thesis
Guillermo Rodrı́guez Lázaro
Advisors:
Aurora Hernández-Machado
Ignacio Pagonabarraga
Programa de Doctorat en Fı́sica

Departament d’Estructura i Constituents de la Matèria
Acknowledgments
The results presented in this Thesis involve the work carried out during several
years of research. During this time in Barcelona I have received the help of many
people that have contributed in different ways, both scientifically and on a personal
level, to make this Thesis possible. I would like to thank you all for your support
and help.
En primer lugar, me gustarı́a agradecer a mis dos directores de tesis, Igna-
cio Pagonabarraga y Aurora Hernández-Machado, el ofrecerme la posibilidad de
realizar el doctorado en un campo tan fascinante como la biofı́sica. Las muchas
horas de reuniones, a veces en horarios intempestivos, los consejos y discusiones,
sin duda han sido determinantes para mejorar mi formación.
I would like to acknowlegde Kathryn A. Melzak the time dedicated to discuss
and show me a more biological and experimental perspective about the world of
red blood cells.
Algunos de los compañeros de la Facultad han contribuı́do decisivamente con
interesantes discusiones o aportaciones que muchas veces me han ayudado a re-
solver diferentes problemas. Destacar especialmente a Xavier Clotet, Clara B.
Picallo, Claudia Trejo, Francisco Alarcón, Carles Blanc y David Palau.
También querrı́a agradecer las discusiones, y especialmente los consejos, de
Rodrigo Ledesma y Félix Campelo, quienes precedieron en parte mi trabajo y
cuyas aportaciones han resultado en más de una ocasión decisivas para poder
avanzar.
En el plano personal, quiero agradecer en primer lugar el apoyo incondicional
prestado por mis padres, Belén y Miguel, especialmente cuando las cosas no han
ido bien y más lo he necesitado. También a toda la familia en general, con mención
especial para mi hermana Paula y mi tı́a Begoña por su cariño y atención.
Sin duda estar en un ambiente de trabajo en el que me he sentido tan a gusto
ha contribuı́do a ir con alegrı́a dı́a tras dı́a a la Facultad. Ahi han estado siempre
Xavifú, Laura, Pau, Claudia, Paco, Xavier, Carles, Jordi, Marta, y otros muchos
junto a los que he sufrido de innumerables tuppers, menús y exquisitos cafés.
i
ii
Por supuesto, también agradecer al resto de amigos que me han rodeado estos
años, y con los que tan buenos ratos he pasado. En especial a Manolo, Melchor y
Pesquera, por aguantarme a todas horas y estar siempre ahı́ cuando lo he necesi-
tado. Lara siempre me apoyó en los buenos momentos, pero especialmente en los
malos. Martı́, Alberto, Marce, Aida, Albert, Marcel, y todos los demás con los
cuales he compartido alegrı́as, bien delante de unas cervezas o en Aiguamolls o
Tarragona. A la gente de Santander, por acogerme las pocas veces que vuelvo, en
especial a David, Jose y Coso. Finalmente, no quiero dejar de agradecer a todos
con los que he compartido viajes durante estos años, con sus aventuras y desven-
turas, y que me han ayudado a mantener la dosis de adrenalina que tanto necesito.
En especial a Dani y Manolo, que siempre se han ofrecido a acompañarme por
estrambótico que fuese el plan.
Barcelona, 7 de Mayo de 2014.

Contents
I Physical description of cell membranes 7
1 Biological introduction 9
1.1 Cell membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 The lipid bilayer . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.2 The cytoskeleton . . . . . . . . . . . . . . . . . . . . . . . 14
1.1.3 Active processes in the membrane . . . . . . . . . . . . . . 15
1.2 The human Red Blood Cell . . . . . . . . . . . . . . . . . . . . . 16
1.2.1 RBC properties . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.2 RBC disorders . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3 Circulatory system and blood . . . . . . . . . . . . . . . . . . . . 20
1.3.1 Functionality . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.2 Blood properties . . . . . . . . . . . . . . . . . . . . . . . 22
2 Physical approach to membranes 25

2.1 Theory of elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.1 The bending mode . . . . . . . . . . . . . . . . . . . . . . 28
2.2 The Helfrich bending energy . . . . . . . . . . . . . . . . . . . . . 31
2.2.1 Microscopic realization . . . . . . . . . . . . . . . . . . . . 32
2.2.2 Equilibrium equation and membrane fluctuations . . . . . 34
2.2.3 Shapes of vesicles . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Other membrane models . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.1 Area-difference elasticity . . . . . . . . . . . . . . . . . . . 37
2.3.2 Cytoskeleton elasticity . . . . . . . . . . . . . . . . . . . . 39
2.4 Elastic parameters values . . . . . . . . . . . . . . . . . . . . . . . 41
iii
iv CONTENTS
II Membrane elasticity and RBC morphological transi-

tions 43
3 The disco-echinocyte transition 45

3.1 Morphological sequence . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Agents inducing echinocytosis . . . . . . . . . . . . . . . . . . . . 48
3.3 AFM-induced RBC shape change . . . . . . . . . . . . . . . . . . 50
3.3.1 Experimental results . . . . . . . . . . . . . . . . . . . . . 51
3.3.2 Interpretations and discussion . . . . . . . . . . . . . . . . 53
4 Membrane elasticity and RBC morphology 55

4.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.1.1 Membrane elastic energy . . . . . . . . . . . . . . . . . . . 56
4.1.2 Shape description . . . . . . . . . . . . . . . . . . . . . . . 57
4.1.3 Minimal shapes of the model . . . . . . . . . . . . . . . . . 60
4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.1 Energy contributions . . . . . . . . . . . . . . . . . . . . . 63
4.2.2 Shape diagram . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2.3 Morphological characterization . . . . . . . . . . . . . . . 67
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Discussion of the AFM experiments . . . . . . . . . . . . . . . . . 75
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
III A phase-field method for membrane modeling 79
5 Phase-field modeling of biological membranes 81

5.1 Phase-field method . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.1.1 Thermodynamics of phase-field models . . . . . . . . . . . 83
5.2 Elastic properties of the phase-field interface . . . . . . . . . . . . 86
5.2.1 Cell membrane model . . . . . . . . . . . . . . . . . . . . 89
5.2.2 Generalized membrane model . . . . . . . . . . . . . . . . 94
5.3 Membrane dynamics and hydrodynamic coupling . . . . . . . . . 96
CONTENTS v
5.4 Macroscopic equations . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4.1 Sharp-interface limit . . . . . . . . . . . . . . . . . . . . . 98
5.5 Linear stability analysis . . . . . . . . . . . . . . . . . . . . . . . 103
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6 Numerical scheme 107
6.1 Lattice-Boltzmann method . . . . . . . . . . . . . . . . . . . . . . 107
6.1.1 Single-relaxation time lattice-Boltzmann . . . . . . . . . . 108
6.1.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . 109
6.2 Phase-field integration . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2.1 Advection . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2.2 Derivative calculations . . . . . . . . . . . . . . . . . . . . 110
6.2.3 Area conservation . . . . . . . . . . . . . . . . . . . . . . . 111
6.3 Numerical realization of tensionless interfaces . . . . . . . . . . . 114
6.4 Several cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.5 Parameter steering . . . . . . . . . . . . . . . . . . . . . . . . . . 117
IV Red Blood Cells flow in confined systems 119
7 Blood circulation 121
7.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.1.1 Blood properties . . . . . . . . . . . . . . . . . . . . . . . 121
7.1.2 Cell manipulation . . . . . . . . . . . . . . . . . . . . . . . 123
7.2 RBC behaviour in shear and parabolic flows . . . . . . . . . . . . 126
7.2.1 Shear flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.2.2 Parabolic flow . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.3 Scope of Part IV . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

vi CONTENTS
8 Rheology and deformability of RBCs 133

8.1 Simulation of RBC and vesicle flow . . . . . . . . . . . . . . . . . 134
8.2 RBC behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.3 Effective viscosity of RBC suspensions . . . . . . . . . . . . . . . 142
8.4 Flow description . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
8.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9 On the mechanisms of RBC deformation and migration 151

9.1 Effective viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
9.2 Elastic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.3 Energy dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.3.1 External power . . . . . . . . . . . . . . . . . . . . . . . . 156
9.3.2 Elastic membrane power . . . . . . . . . . . . . . . . . . . 156
9.3.3 Viscous dissipation . . . . . . . . . . . . . . . . . . . . . . 158
9.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10 RBC focusing 161

10.1 Effective viscosity and focusing . . . . . . . . . . . . . . . . . . . 162
10.2 RBC focusing and alignment . . . . . . . . . . . . . . . . . . . . . 164
10.2.1 Focusing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
10.2.2 Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
10.2.3 Effect of walls confinement . . . . . . . . . . . . . . . . . . 166
10.3 Vesicle shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
10.4 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . 170
11 Collective flow of RBCs 173

11.1 Hydrodynamic interactions between RBCs . . . . . . . . . . . . . 174
11.1.1 Regular arrays . . . . . . . . . . . . . . . . . . . . . . . . 174
11.1.2 Screening between cells . . . . . . . . . . . . . . . . . . . . 176
11.2 Several RBCs in flow . . . . . . . . . . . . . . . . . . . . . . . . . 179
11.2.1 Order vs disordered configurations . . . . . . . . . . . . . 180
CONTENTS vii
11.2.2 Migration and orientation in ordered configurations . . . . 182

11.2.3 Spatial ordering . . . . . . . . . . . . . . . . . . . . . . . . 184
11.2.4 Focusing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
11.2.5 Rheology . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
11.3 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . 190
V Conclusions 193
12 Conclusions and future perspectives 195

12.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
12.2 Future perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . 201
13 Resumen en castellano 203

13.1 Introducción . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
13.1.1 Introducción biológica . . . . . . . . . . . . . . . . . . . . 204
13.1.2 Modelos fı́sicos de membranas . . . . . . . . . . . . . . . . 204
13.2 Resultados . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
13.2.1 Elasticidad de la membrana y transición del
discocito al equinocito . . . . . . . . . . . . . . . . . . . . 205
13.2.2 Métodos de interfase difusa para modelado de membranas 207
13.2.3 Flujo de glóbulos rojos en canales confinados . . . . . . . . 209
13.3 Conclusiones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
A Notes on differential geometry 213

A.1 Mean curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
A.2 Extension ratios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
A.3 Differential operators in curvilinear coordinates . . . . . . . . . . 215
B The sharp-interface limit 217

B.1 The inner and the outer region . . . . . . . . . . . . . . . . . . . 217
B.2 Differential operators . . . . . . . . . . . . . . . . . . . . . . . . . 219
C List of publications 223
References 225
General introduction
In recent years, the application of physical techniques and rapid technological

development have driven biological research towards a more quantitative descrip-
tion. A wealth of new experimental data at the molecular and cellular scale has
been obtained, allowing a more direct connection with physical models. Thereby,
the closer connection between biologists and physicists has facilitated the emer-
gence of biophysics as a new field, and research has benefitted from the profitable
interplay between both disciplines.
Cell membranes are complex objects which involve biological processes span-
ning from the flip of the lipids they contain, on the length scale of tens of nanome-
ters, to the overall response of the cell, at typically ten microns. Membranes have
fascinated physicists since the earliest stages of development of the field, due to
their very specific properties, which are not shown but any other material (such as
non-classical elastic behaviour). Additionally, in spite of the membrane intrinsic
complexity, physical models have explained an astounding number of phenomena
observed experimentally, demonstrating that membranes invite to an extensive
theoretical exploration. Furthermore, the recent access to the cellular and molec-
ular scale in experiments has led to the discovery of new phenomena that must
be both understood and explained.
At the scale in which the cell responds to mechanical pressures in the tissue,
the study of the system generally requires a coarse-grained description. Macro-
scopic variables such as membrane shape geometry and elasticity become useful
tools for the purposes of studying the system. This Thesis is framed in this bio-
physical context. We will study, from a theoretical perspective, the mechanics of
membranes and cells at different situations of biological interest that cells may
face during their life.
Elasticity of cell membranes

The membrane is a fundamental structure in all living organisms, as it defines the
cell as an entity, the basic unit of life. Membranes separate the external environ-
1
2 CONTENTS
ment from the cell inner region which contains all the organelles and molecular
machinery that constitute the basic ingredients with which the whole organism
is constructed. The formation of closed membranes from lipid aggregates repre-
sented a major step that allowed the development of life. In a parallel way to the
evolution towards more complex organisms, such as vertebrates and specifically
mammals, cell membranes have also increased their complexity, incorporating a
huge quantity of larger molecules and microstructures that enable a complex func-
tionality.
For all these reasons, cell membranes have attracted interest from biologists for
a long time, and from the pioneer model of the fluid mosaic by Singer and Nicolson
(1972), our knowledge about membrane molecular composition and functioning
has continuously increased. In the last 40 years, membranes have also been stud-
ied by physicists, providing a complementary picture about membrane behaviour
and properties. The subject was first approached by Canham (1970) and Helfrich
(1973), and based on their seminal models an outstanding number of membrane
phenomena have been understood and explained from a physical perspective. In
particular, the characterization of the elastic properties of the membrane is the
scope of many studies. Whereas we have achieved a considerable understanding
about the mechanics of simple biomimetic membranes such as bilayers of homoge-
neous lipid composition, the elastic behaviour of more complex membranes, such
as those present in mammalian cells, is still under lively debate in the literature.
In this context, most research has focused on the study of the human red blood
cell as a model system, due to its structural simplicity. Red blood cells present
a remarkable capability to deform and pass through very thin capillaries, and in
microcirculation they acquire strange shapes whose benefits are still unknown.
They also develop a number of different morphologies if their membrane is altered
or damaged, as known from a number of anemias, malaria, or during blood stor-
age. The delicate membrane equilibrium at the molecular scale ultimately affects
mechanisms taking place in a much larger scale, such as cell shape and blood
properties.
Blood rheology
Newtonian fluid dynamics has proven to be an accurate theory for describing the
behaviour of gases and liquids containing small molecules (with a molecular weight
of less than roughly 1000 uma). If, however, the liquid contains larger molecules,
a number of mechanisms (such as molecule orientation, spatial organization or
aggregation) become dominant, introducing a more complex internal physics in the
liquid microstructure that affects to the macroscopic behaviour, which differs from
CONTENTS 3
the classic Newtonian dynamics. These phenomena are usually shear- or time-
dependent, and they also depend on other factors such as molecule concentration
or system geometry. The variable response of these liquids under the action of
different external stresses has led to the widely extended denomination of complex
fluids. Classic examples of this kind of fluids are polymer melts, glass-forming
liquids and micellar solutions.
It is known for a long time that the rheological behaviour of blood is largely
determined by the deformability of red blood cells, as suggested by the abnormal
properties of blood notized in experiments of macroscopic rheology with altered
samples (eg spherocytes). However, the problem remains poorly understood, due
to the difficulty of a controlled experimental approach at the single-cell microscale.
In the theoretical frame, most studies are based on a simplified description of the
cells, ignoring important details of their shape and elasticity, and hence neglecting
the important aspect of cell elasticity. In this Thesis, we deal in depth with
the deformability of red blood cells flowing in confined channels, paying special
attention to the importance of membrane elasticity and its consequences in blood
properties.
Thesis aims and structure

This Thesis deals with the physical properties of cell membranes, partially based
on our background on simpler interfaces. Our starting point is the microstructure
of the plasma cell membrane, and we devote special effort to the conciliation
between the biological picture and the physical modeling. The next step is to
connect the membrane physical properties with the cell morphology, and identify
the role of each membrane ingredient in the overall response of the cell. Finally, we
study how the elasticity and morphology of the cell affect the physical properties
of the whole tissue. Our research is focused in the particular case of red blood
cells and blood, a prime element of our organism.
The main aim of this Thesis is to understand and explain red blood cell de-
formability and dynamic behaviour from the elastic properties of its membrane,
and connect this behaviour with the rheological properties of blood. For this pur-
pose, we need to set the physical properties of the membrane that our model should
account for, after applying this model to specific problems at the cell scale. Two
different problems are studied: the development of stationary altered shapes when
the physiologic membrane equilibrium is broken, and the deformations undergone
by healthy cells when exposed to confined flows. The first case corresponds to a
mechanical equilibrium problem, whereas the second is stricly a non-equilibrium
problem in which the hydrodynamics of the external fluid plays a central role.
4 CONTENTS
Attending to these objectives, the Thesis is structured in four separated parts.

Part I contains two introductory chapters. In Chapter 1, the main aspects of
the biological systems of interest are described, including an overview covering
membrane composition and structure, red blood cell characteristics, and a brief
description of the circulatory system and blood. In Chapter 2 we present the most
relevant physical theories describing cell membranes. Based on the Helfrich theory
framework, new and more depured models have been proposed, achieving a more
realistic description that accounts for more complex phenomena than the initial
Helfrich theory. We outline the main models in order to provide the physical
framework necessary to understand the subsequent studies.
Once the biophysical frame has been stablished, we then present our original
research. Part II is devoted to the study of the membrane elastic response during
morphological transformations in the so-called disco-echinocyte transition of the
red blood cell. It is known that alterations in the membrane molecular struc-
ture, such as lipid rearrangements or conformational changes in the proteins of
the cytoskeleton, lead to the formation of spicules and bumps, in a sequence of
increasingly crenated shapes. The purpose of this Part is to elucidate the role of
both the lipid bilayer and the cytoskeleton during these morphological changes
and generically understand the response of each membrane component during
any kind of deformation. In a series of experiments performed by Kathryn A.
Melzak and José Luis Toca-Herrera, perturbation of altered cells by an AFM tip
is shown to induce a reverse morphological transformation towards the healthy
discocyte. We develop a theoretical model, based on a Cassini ovals parametriza-
tion of the cell membrane, that allows us to study and understand the first stages
of this transition. Thereby, Chapter 3 first presents the fundamental aspects of
the disco-echinocyte transition and some of the molecular mechanisms running
in the cell membrane that have been discovered in previous experiments. We
then describe the AFM experiments and discuss the molecular mechanisms in-
volved. These conclusions are based on the theoretical study, which is presented
in Chapter 4. In this chapter, we analyze the main morphologies observed during
the transition, identifying the key mechanisms that explain each cell shape. Our
results show that the cytoskeleton is relevant for stabilizing the discocyte and op-
poses resistance to the formation of bumps, but it shows weak disturbance when
cell shape is close to the initial discocyte. One of the main assumptions of our
model is that the cytoskeleton is relaxed at the discocyte stage, based on numer-
ous experimental observations. This point is, however, controversial and remains
as one of the less understood aspects of cell elasticity, and we discuss the subject
in detail throughtout the Thesis.
Afterwards, we drive our attention to the development of a phase-field method
for membrane modeling, corresponding to Part III. The phase-field approach is
CONTENTS 5
based on the Helfrich theoretical framework. In Chapter 5, we dedicate special

effort to the formalization and rationalization of the model, showing its conver-
gence to the Helfrich theory. We obtain the stress tensor of the membrane from
which the membrane local force can be derived. This is used to incorporate the
membrane phase-field model to a more general model in which membrane dy-
namics is coupled with an external fluid, whose hydrodynamics is dictated by the
Navier-Stokes equation. The consideration of an external fluid is important as in
many membrane-related problems the hydrodynamics plays a central role. From
the complete phase-field Navier-Stokes model, we derive the macroscopic equilib-
rium equation of the membrane. Finally, we perform a linear stability analysis,
studying the relaxation of flat membranes when subjected to a sinusoidal pertur-
bation, in order to validate the model and test if it captures the correct membrane
dynamics. In Chapter 6 we describe the numerical implementation of the model,
based on a lattice-Boltzmann scheme.
The last piece of work, Part IV, focuses on the behaviour of red blood cells
while flowing in confined channels. The rheological behaviour of blood in narrow
conduits, in which the cell concentration is considerably lower than in the thickest
arteries, is dominated by the elasticity and deformability of the red blood cells.
However, the problem of non-linear rheology caused by deformable objects is still
poorly understood. Red blood cells flowing in small channels are known to exhibit
a complex morphological behaviour, often assuming asymmetric shapes termed
slippers, whose advantages with respect to symmetric shapes are still unclear.
The main aims of this Part are to link the rheological behaviour of red blood
cell suspensions with the elasticity of the cells they contain, as well as to explore
the cell morphological response. In Chapter 7, we present the main applications
and interests of this subject, especially in the context of microfluidics and lab-
on-a-chip devices, whose development is being conducted towards the field of
pathology diagnosis and other medical applications. In this Chapter, we also
outline some previous results about the behaviour of red blood cells in different
flow conditions. In Chapter 8, we identify the morphological regimes of red blood
cells flowing isolated in confined channels, characterizing the observed cell shapes
and analyzing the elastic contributions to the membrane in each case. We also
describe the effective viscosity of the cell suspension, discussing the shear-thinning
behaviour that we obtain. In Chapter 9, we focus on the understanding of why red
blood cells develop asymmetric (slipper) shapes, comparing these configurations
with the symmetric shapes in order to identify the benefits of a lateral position in
the channel, and to adress its implications in the suspension fluidity. In Chapter
10, we drive our attention to the control of red blood cell migration and lateral
position in the channel by tuning the channel and flow properties, a subject of
relevance for the improvement of chip devices to manipulate and separate single
6 CONTENTS
cells.
We end with a separated part containing the conclusions of the Thesis, where
we conciliate the three previous parts and provide an all-encompassing picture
of the elasticity and deformability of red blood cells. Finally, we discuss future
perspectives of the work presented in the Thesis.
Part I
Physical description of cell

membranes
7
Chapter 1
Biological introduction
In order to set the proper biophysical context that allows us to identify the main
features to be captured in our physical models, we start with a biological intro-
duction of the systems of study. Thereby, this Chapter is organized in a sequence
of increasingly complex structures. First, we cover the cell membrane composi-
tion and assembly, driving then our attention to the particular properties of the
red blood cells. Besides, we describe the circulatory system which represents the
biological frame in which red blood cells live.
1.1 Cell membrane

In this Section we present the most important biological characteristics of cell
membranes necessary to understand their elastic and mechanic response. We
will briefly outline the molecular composition of the membrane before identifying
the main components that form its microstructure. Then, the importance of
membrane properties will be analyzed in the case of human erythrocytes, a unique
case among human cells because it lacks any organelle or internal structure, so
that its overall shape and mechanical properties can be directly related to that of
its membrane.
Cell membranes represent an essential element in the development of living
organisms. They constitute the cells frontiers, separating the interior of the cell
from the external environment. Membranes enclose the organelles and components
that all together form the basic unit of life. However, membrane functionality is
not limited to its simple structural role, but membranes are also responsible of the
interactions of the cell with neighbour cells. These interactions are mediated by a
certain type of transmembrane proteins that coordinate the cell signaling, enabling
the cell response to environmental pressures. Additionally, membranes maintain
9
10 Chapter 1. Biological introduction
ion gradients which allow the synthesis of ATP, the basic energetic molecule. The
plasma membrane is the most important membrane of the cell, but other types
of membrane are present in organelles like the nucleus, the Golgi aparatus, the
endoplasmic reticulum and the mitochondria. Considering all the membranes of
the cell, they concentrate around 30% of the total proteinic activity (Alberts et al.,
1994).
All biological membranes share a common structure and composition in spite
of being part of different entities, and regardless of their function. Membranes are
composed by different lipid molecules that assemble forming bilayers. Lipid bilay-
ers are selectively permeable to the exchange of polar molecules and host a high
density of transmembrane proteins, which essentially define the specific membrane
functionality. Lipids are bound by relatively weak, non-covalent interactions that
allow a rapid lateral interchange of positions, leading to a huge surface diffusion
over the membrane plane. Typically, a lipid flips with neighbour molecules 107
times per second, exploring the whole cell surface in just a few seconds. The lipids
practically behave as a fluid in the bilayer plane, a property with important im-
plications for the cell activity. Transmembrane proteins waft in the fluid bilayer
(Singer and Nicolson, 1972), and they are also able to diffuse laterally. The mem-
brane is connected with the inner (cytosolic) cytoskeleton, a three-dimensional
mesh formed by actin filaments which provides compactness and structural or-
dering, and determines the cell shape, which in turn depends on the type of cell
and its function. In some cells, an exterior (cortical) cytoskeleton also exists, and
it connects with neighbouring cells in order to facilitate a coordinate response of
the tissue. The fragile membrane equilibrium is controlled by a number of active
processes, including flip-flop rearrangement of the different lipid species of the
bilayer, remodeling of the cytoskeleton, or the balance of lipid densities during
vesiculation processes (eg. during endo and exocytosis), which is achieved by the
existence of lipid reservoirs in the interior of the cell.
1.1.1 The lipid bilayer
Lipid composition
Lipids represent up to 50% of the total mass of the membrane in mammalian

cells. They are amphiphilic molecules with a polar head (which prefers to contact
and interact with other polar molecules, such as water) and a tail formed by two
hydrocarbon chains which present a strong hydrophobicity, and therefore tails
avoid the interaction with water. The term amphiphile, which means in greek both
lover, derives from the ancient observation that amphiphilic molecules organize in
the presence of oil and water due to their polar nature, as they assemble in the
1.1. Cell membrane 11
surface separating these fluids, orienting their heads to the water region and the
tails to the oil region. If lipids are immersed in water, they tend to self-assemble
to avoid the hydrophobic interactions with the surrounding water. Two basic
structures can be formed by these aggregates. Sometimes they assemble forming
micelles, a closed structure, with all the tails in the inner, free-water region, and
the lipid heads oriented to the exterior, in contact with water. Another possibility
is the formation of bilayers, when two lipid monolayers fold in opposite directions,
so that the heads form two parallel sheets whereas the tails are trapped in the
intermediate region, without contact with the aqueous environment. Still, at
the edges of the bilayer the tails of the boundary lipids do interact with water
molecules. Lipids rearrange to avoid the presence of edges, forming closed surfaces
in which the water is both at the inner and outer regions, but there is no direct
interaction with the tails. The strong hydrophobicity causes that these closed
structures are much energetically favourable, thus ensuring large stability under
thermal fluctuations and other mechanical disruptions. The shape of the lipid
favours the formation of one structure in detritment of the other. Conical shaped
lipids (thus with a large head section compared to the tail), such as lysolipids,
pack forming micelles, whereas cylindrical lipids, such as phospholipids, are more
symmetric and form bilayers.
Despite their structural simplicity, bilayers are formed by an extensive variety
of lipid species. An eukaryotic cell is typically composed by 500-1000 different
species, although many occur just incidentally and the major components reduce
to the phospholipids, with just a few species present in mammalian cells. All
phospholipids have two fatty chains of variable length (ranging from 14 to 24
carbons), with the particularity that one of the chains is unsaturated (thus it
contains a cis bond), and hence it is not completely straight but presents a kink,
resembling a broken rod (see Figure 1.1), whereas the second chain is saturated.
The presence of an unsaturated tail is crucial, since it breaks the symmetry of
the molecule and reduces the packing capacity of the lipids, leading to an in-
crease in the fluidity of the bilayer. In mammalian cells, the bilayer is basically
composed by sphingomyelin, phosphatidycholine, phosphatidylethanolamine, and
phosphatidylserine. They are asymmetrically distributed in the monoloyers: the
first two are always present in the outer monoloyer whilst the later are restricted
to the inner one (van Meer et al., 2008). This asymmetry responds to the presence
of certain chemical groups (such as an amino group in the particular case of phos-
phatidylserine), that allows the cell to maintain a strict control of the membrane
asymmetry.
In addition to the phospholipids, animal cell membranes also contain choles-
terol and glycolipids (Yeagle, 1993). Cholesterol is a small molecule with a polar
hydroxyl group and a short hydrocarbon chain. Cholesterol occupies the space
between phospholipid tails in the inner region of the bilayer, with its head oriented
close to the phospholipid head. Mammalian cell membranes are rich in cholesterol
and this molecule plays an important role in the control of bilayer fluidity, and it
also affects the membrane rigidity when present at abnormal high densities. The
glycolipids are found exclusively in the outer leaflet of the plasma membrane and,
in spite of their usual low concentrations (roughly 2%), they are important in the
control of bilayer asymmetry.
Figure 1.1: (A) Comparison between a phospholipid and the much smaller cholesterol
molecule. Phospholipids are characterized by a typical structure with a polar head
which interacts with the molecules of water and a non-polar tail of high hydrophobicity.
The tail is composed by two carbon tails, one of which is unsaturated. This symmetry
breaking reduces the packing capability of the lipids enabling a higher fluidity of the
bilayer. Credits: (left) Biology Department at College of the Siskiyous, California;
(right) Biochemistry Dictionary, Jeff D Cronk. (B) Different aggregates formed by lipids:
micelles, bilayers, and closed bilayers forming vesicles or liposomes. The preference of
the lipids to aggregate in one or other structure is determined by the shape of the lipid;
phospholipids form bilayers. Credits: Mariana Ruiz Villareal.
Transmembrane proteins
Membrane proteins are responsible of the main processes that take place in the
membrane and therefore they define the membrane functionality (Alberts et al.,
1994). Depending on the membrane, proteins represent 25-75% of the total mass
of the membrane. Membranes of bacteria, such as E. coli, characterized by a low
protein activity, present protein concentrations below 20%. In the inner membrane
of the mitochondria of eukariotic cells, however, proteins may represent as much
as 80% of the total mass, reflecting the high concentration of ion gradients and
catalytic activity of these structures (Cooper and Hausman, 2009). In the case
of plasma membranes, a typical value is 50%. Since proteins are much larger
than lipids, this concentration corresponds to a protein per ∼ 50 − 100 lipids.
Transmembrane proteins are also amphiphilic and orient their polar groups to the
aqueous environment (cytosol and exterior of the cell) whereas the hydrophobic
groups interact with the lipid tails. Other molecules are located at either side of
the membrane and interact externally with the bilayer; an example is the spectrin
cytoskeleton that we will study in the subsequent section. Although the action
of transmembrane proteins presents many fascinating aspects, for the purposes of
this Thesis they are not especially important and we will not extend further on
this subject. Still, it is remarkable that these molecules also suppose an important
contribution to the membrane elasticity, but this contribution will be directly
incorporated to the models by the effective parameters that capture the averaged
membrane properties.
Bilayer properties
The bilayers of mammalian cells, as stated before, are complex structures with
a bewildering number of proteins working on and through them. They have a
typical thickness of 4nm, while most eukaryotic cells are ∼ 5μm length. Thus,
the membrane thickness is much smaller ∼ 10−3 than the overall cell length.
Lipids can diffuse freely throughout their own monolayer, with diffusion coeffi-
cient D ∼ 10−12 m2 /s, but the exchange of molecules between monolayers (the
so-called flip-flop) is inhibited by the presence of a strong hydrophobic potential
(Alberts et al., 1994). The time scale of spontaneous flip-flop is, roughly, several
days. There are, however, explicit mechanisms to regulate this lipid rearrangement
which control the appropriate bilayer asymmetry. Among others, the translocases
are transmembrane proteins specialized in the flip-flop of certain lipids. Cells reg-
ulate their bilayer asymmetry to control shape and mechanical properties of the
membrane, or to mediate in the interactions with the environment. For instance,
since translocases action is ATP-driven, when this molecule is depleted or the cell
undergoes apoptosis, lipids start to diffuse freely and the lipid composition of the
external monolayer changes. This process is used for cell signaling, alerting espe-
cialized cells which detect the abnormality and phagocyte the dead cell (Devaux
et al., 2008).
Although the bilayer is usually fluid, this property presents a strong depen-
dence on the temperature and lipid composition. The high diffusivity of each lipid
along the leaflets would suggest that they might be randomly distributed along
the cell surface. However, in certain molecules, van der Waals interactions be-
tween the hydrocarbon tails are strong enough to form aggregates and inhibit the
lateral diffusion. These aggregates are known as lipid rafts and are composed of
sphingomyeline, phosphatidylcholine and especially cholesterol (Jacobson et al.,

2007). Experiments suggest that cholesterol are present in much higher densities
in these rafts than in the surrounding areas of the bilayer. It seems that the or-
ganization in rafts has specific functions, since some signaling molecules are only
assembled in these microdomains, and therefore rafts are important for regulation
of neurotransmission, and receptor traffiking (Allen et al., 2007). The fluidity of
the bilayer, presence and formation of rafts and generically the microstructure of
the lipid bilayer are subjects of intense debate, and recent studies suggest that the
current description represents just a rough picture of the actual bilayer structure
(Engelman, 2005; Leslie, 2011).
Figure 1.2: (A) Plasma membrane scheme. The plasma membrane is composed by a
fluid lipid bilayer and a high concentration of transmembrane proteins that account
for the enzymatic activity. The bilayer contains a rich composition of diverse lipid
species, including cholesterol, a small molecule inserted in contact with the phospholipid
tails. The underneath spectrin cytoskeleton is a two dimensional fold attached to the
bilayer. Source: www.shmoop.com/biology-cells/plasma-membrane.html (B) Electron
microscopy imaging of the cytoskeleton network. Note the 6-vertices links. Source:
www.lbl.gov/Science-Articles/Archive/LSD-single-gene.html
1.1.2 The cytoskeleton

Most cells have a complex mesh formed by actin filaments which occupies most of
the inner cytosolic volume, and connects the different organelles and microstruc-
tures of the cell. This structural element provides mechanical strength to the cell
and it often participates in cell shape and cell mobility. This structure, known as
cortical cytoskeleton, is connected with the membrane in order to coordinate the
response to external perturbations. Despite its importance in several aspects of the
cell mechanics, we concentrate here in other and much simpler cytoskeletal struc-
ture, the so-called membrane cytoskeleton which lies underneath the lipid bilayer.
The membrane cytoskeleton has a structural functionality, providing strenghten

and preventing from certain shape deformations, such as vesiculation or pinch-off
of the bilayer. Its presence also limits the transmembrane diffusion on the bilayer,
defining small patches that effectively act as domains where the protein diffusion
is restricted. The membrane cytoskeleton is a two-dimensional spectrin network
anchored to the inner (cytosolic) monolayer of the plasma bilayer of certain cells
(Bennett, 1989), such as the human erythrocyte. Spectrin is a long rod of around
75nm length. The spectrin molecules are crosslinked by vertices formed by actin,
which can form different type of junctions, depending on the number of molecules
that they connect (from 3 to 7). In healthy cells, the 6-vertices junction is dom-
inant, with as low as 3% of 5-vertices and 8% of 7-vertices (Liu et al., 1990).
The presence of these defect junctions is, however, important as it allows local
relaxation of the network tension.
The spectrin cytoskeleton is an active tissue. The presence of ATP is crucial
for maintaining the cytoskeleton properties, and when this molecule is depleted
the cell experiences drastic shape changes. Although this phenomenon is not com-
pletely understood, the fluid gel hypothesis assumes that the network is subjected
to continuous remodeling, which allows relaxation of the cytoskeleton tensions (Li
et al., 2005). Hence, when active processes cease, the cytoskeleton loses its fluidic
behaviour and stiffens.
1.1.3 Active processes in the membrane

The membrane description presented so far corresponds to a static picture. How-
ever, both the bilayer and the cytoskeleton are controlled by a sophisticated engi-
neering maintained by active processes. Therefore, the membrane is not in ther-
modynamic equilibrium, but it is subjected to a continuous renewal of molecules
that requires of an expensive energy consumption, in order to guarantee the deli-
cate steady state of the membrane.
As previously mentioned, the lipid bilayer asymmetry is modulated by three
main proteins that control the composition of each leaflet (Devaux et al., 2008).
Flippases transport aminophospholipids from the outer to the inner leaflet, al-
thouth it seems that incidentally they are capable of flipping other lipid molecules.
Floppases transport amphiphilic drugs and phospholipids with little selectivity
from the inner to the outer leaflet. Scramblases flip phosphatidylserine from the
inner to the outer leaflet of the membrane. The presence of this lipid in the cell
surface triggers platalet aggregation, so that the very specific function of scram-
blases relates with the response to certain abnormal environmental pressures. The
fragile balance of membrane asymmetry is achieved by the coordinate action of
these (among others) transmembrane proteins.
The presence of active processes affecting the cytoskeleton chemical compo-

sition is known for a long time ago (Pinder et al., 1977), but its molecular ba-
sis has been adressed in detail only recently. Several studies have identified the
cytoskeletal proteins (mainly β-spectrin and protein 4.1R) that under phospho-
rylation can modify the network connectivity, affecting the mechanical stability
of the membrane (Manno et al., 2005). However, it is still unclear whether these
processes occur spontaneously and serve as a regulation system of the cytoskeletal
mechanics. A different mechanism of control of the cytoskeleton structure has
been proposed (Wong, 1999). The flux of anions mediated by the transmembrane
protein Band 3 is used to fold and unfold the spectrin filaments, giving rise to
a contraction or relaxation of the cytoskeleton. This may be used by the cell to
control its shape in certain conditions. These and other active mechanisms will
be discussed in detail in Chapter III.
1.2 The human Red Blood Cell

Eukaryotic cells present an extensive variety of shapes, as an adaptation to their
specific function and location within the different tissues. The cortical cytoskele-
ton and the plasma membrane are the two main elements responsible of the cell
shape and mechanic response. Still, the different organelles occupy an impor-
tant portion of the cell volume, and their presence implies that the cell must
accomodate them. Hence, while studying the mechanical properties of the cell it
is difficult to discern between the different effects, obscuring the understanding
of the specific properties of the membrane. Taking into account this problem,
the erythrocyte or red blood cell (RBC hereafter) represents an interesting case.
Mammalian RBCs are anucleated and lack any internal structure, so that their
unique components are the plasma membrane with its underlaying cytoskeleton.
Accordingly, the shape of the RBC can be directly understood as the result of its
membrane properties. The RBC is therefore studied as a model system in order
to understand plasma membrane properties and, indeed, many of the studies that
have elucidated key insights in membrane biology focused on RBCs. Nevertheless,
RBCs are interesting not only as a model system but also due to their crucial role
in our lifes, as they are the main component of blood and the unique carriers of
oxygen.
1.2.1 RBC properties
Mammalian RBCs have different shapes and sizes, depending on the animal phys-
iological requirements (eg oxygen consumption in animals inhabiting high altitude
1.2. The human Red Blood Cell 17
Figure 1.3: Healthy red blood cells present a distinct shape, the so-called discocyte,
characterized by a discoidal shape with a concave region in the center. The disc has a
typical diameter of 8μm and thickness 2μm. Credits: CDC / Janice Carr.
mountains). Very similar species have different RBC shapes, such as sheeps and
goats. The human red blood cell has a disk shape of typical diameter 8μm, with
a concave region in the center where the cell achieves its minimum thickness of
1μm, and a convex outer rim where it reaches a maximum thickness of 2μm. This
particular shape is usually known as the biconcave discocyte, and it corresponds
to the healthy state of the cell. Typical cell area and volume on a healthy individ-
ual are 140μm2 and 90μm3 (Zarda et al., 1977), respectively. Cells present specific
regulatory systems to maintain their area and volume constant, and thus ensuring
that their resting shape is fixed. In humans, RBCs exhibit a huge intraindividual
variability, with strong correlation with sex and age. Cells of men are up to 20%
larger than in women, and men also present higher hematocrit (ie the volume
fraction of RBCs in blood) in the circulatory system. Ageing affects to the RBC
membrane rigidity, so that old individuals present more rigid cells.
The biconcave discocyte, however, represents just one of the many morpholo-
gies exhibited by RBCs, and it responds to a very specific conditions of area to
volume ratio, bilayer and cytoskeleton elastic properties, membrane internal asym-
metry and pH of the surrounding aqueous environment, among others. Other
well-known morphologies are the stomatocyte, when the cell acquires a cup-like
shape, and the echynocyte, when the cell becomes spherical and it develops many
spikes around its contour, resembling a sea urchin. The entire shape deformation
comprises a sequence of different morphologies usually known as stomato-disco-
echynocyte (Besis, 1973), and it is triggered by disruption of the membrane mi-
crostructure which changes the membrane asymmetry. These phenomena will be
studied in detail in the Part II of this Thesis.
RBCs are produced in the bone marrow, developed from stem cells. RBCs
mature during the subsequent 7 days in a process known as erythropoiesis, before

they develop the healthy discocytic state and are released in the circulatory sys-
tem. Approximately, 2.5 million cells per second are produced in adult humans.
They have a 120 functional life span in which they complete a cycle each 20 sec-
onds. During each cycle, RBCs pass through capillaries of 5μm that force the cell
to undergo extreme deformations. Ageing reduces cell deformability, and even-
tually RBCs enter in eryptosis, in which they are removed from the circulatory
system.
The origins of the peculiar discocyte shape have been subject of debate for
decades. It seems reasonable that the large cell area compared to volume (com-
paring withthat of a sphere, the so-called reduced volume vred = V /(4πR2 /3)=0.6,
where R = (A/4π) is the radius of a sphere with equal area to the cell; thus, for
a sphere vred = 1), responds to the necessity of optimizing the diffusion of oxygen
across the membrane. Alternatively, it has been proposed that the disk has a low
inertial momentum, so that it does not rotate when flowing in the main arteries,
minimizing the formation of turbulent flows (Uzoigwe, 2006). Other hypothesis
postulates that the discocyte is an appropiate shape to undergo strong deforma-
tions and pass through the smallest capillaries, after recovering the normal relaxed
shape (Reinhart and Chien, 1986).
1.2.2 RBC disorders
The healthy running of RBC circulation and oxygen transport can be affected by
different disorders. For instance, iron cell anemia, in which low concentrations
of iron imply a reduced storage capacity of oxygen, is a regular and well-known
disorder even in healthy individuals. However, most interesting disorders concern
inherited pathologies which affect the RBC membrane, producing abnormalities in
the RBC shape or deformability, which potentially reduce the healthy functioning
of blood circulation. In addition, malaria (which does not have a genetic origin)
is also known to impair the membrane microstructure leading to cell stiffening.
These membrane alterations provide important information about the membrane
structural balance, and their main causes and consequences are outlined below.
• Sickle cell anemia (drepanocytosis). This disease is characterized by

the formation of sickle cells that lose their capability to deform and recover
the discocyte shape, altering the oxygen delivery. The molecular basis is
found in an abnormal phosphorilation of hemogoblin that promotes a mas-
sive aggregation of this molecule under low concentrations. The formation of
these aggregates affects the concentration of the protein Band 3, and the cell
1.2. The human Red Blood Cell 19
membrane is damaged in a process similar to ageing, becoming rigid. Pa-

tients affected by this pathology present a reduced life expectancy, although
modern medical treatments allow a normal life.
• Hereditary spherocytosis. Patients present a high concentration of sphe-

roidal shaped RBCs, as a consequence of defects in several proteins of
the membrane (mainly from the bilayer-cytoskeleton links) which produce
fragility of the membrane. The alteration of membrane properties allows
vesiculation and loss of membrane surface, triggering the cell shape defor-
mation. Spherocytes are rapidly retired from circulation by the spleenic
system, leading to hemolysis. Patients must be treated with blood transfu-
sions for critical levels of anemia (Gallagher, 2005).
• Hereditary elliptocytosis. This pathology is characterized by abnormal-

ities in the spectrin dimers, producing weakness of the cytoskeleton which
impairs the membrane stability. RBCs deform into ellipsoidal (or cigar-
shaped) cells. The RBC functionality might not be severely affected, as
most patients are asymptomatic and only 10% present anemia. Interest-
ingly, elliptocytes are more resistant to malaria than normal RBCs, and
presumably this fact explains why hereditary elliptocytosis presents a much
higher incidence in Africa and Mediterranean Europe than any other regions
(Gallagher, 2005).
• Malaria. Malaria is caused by the infection of a parasit of the genus Plas-

modium. Malaria is currently the disease causing the highest number of
deaths in the world, roughly 1-2 million per year (most of them in Africa),
with up to 200 million documented cases each year. The parasit infects dif-
ferent cell species. Infected RBCs develop an advanced proteinic machinery,
including the formation of organelles similar to the Golgi apparatus which
serve for nutrient transport and storage, and allow enzimatic activity. The
parasit is hosted in a vacuole, and during its maturation it reaches the size
of a nucleus in a typical eukaryotic cell. Appart from this new internal struc-
ture, the parasit produces changes in the membrane proteins that affect the
deformability of the cell. At least 50 protein abnormalities have been iden-
tified (Cooke et al., 2004). The RBC also adopts a more spherical shape,
and proteins allocated in the external face of the membrane promote aggre-
gation with other infected cells, avoiding the hemolysis in the spleen. All
these conformational changes strongly affect the cell mechanical properties,
modifying the rheological properties of blood (Dondorp et al., 2000).
Figure 1.4: Red blood cell membrane disorders. Alterations in the membrane mi-
crostructure produce and imbalance in the membrane that lead to abnormal cell
shapes. (Left) Elliptocyte. (Right) Drepanocyte (sickle cell anemia). Source:
http://www.mclno.org/WebResources/kbase/cellatlas/Ovalocyte.html
Cross-sectional area (cm2 ) Velocity(cm/s) Re

Aorta 2.5 33 3000
Small arteries 20 4.12 80
Arterioles 40 2.05 0.5
Capillaries 2500 0.033 0.05
Small veins 80 1.03 0.2
Table 1.1: Values of the total cross-sectional area, flow velocity and Reynolds number
for different conduits of the circulatory system. Adapted from Hall (2011).
1.3 Circulatory system and blood
The circulatory system is an organ system that circulates the blood along all
the cells and tissues, facilitating the transport of oxygen and nutrients which
allow the nourishment of the cells. It also serves as carrier of other molecules or
matter, such as transport of waste products towards the excretory system, or a
fast transport of hormones from one part of the body to another in response to a
certain environmental condition. Generally, the main function of the circulatory
system is to provide the molecules that the body tissues need at each moment.
The circulatory system is divided into two main parts, the pulmonary circulation
(which represents a closed circuit from the heart towards the lungs in order to
oxigenate the blood, and back to the heart) and the systemic circulation (which
corresponds to the network that distributes the blood all along the body). The
circulatory system is coupled to the cardiovascular system, formed by the heart
and adjacent venule circuits, which serves as the engine that pumps blood.
1.3. Circulatory system and blood 21
1.3.1 Functionality
The systemic circulation comprises a structured network in which the conduits

are increasingly thinner, as they distance from the heart. The arteries transport
oxygen-rich blood under high pressure to the different tissues. The aorta is the
thickest conduit of the body as it collects all the blood from the heart, which is
then distributed between the different arteries. These conduits have strong vas-
cular walls, and blood flows at high velocities. The arterioles are small branches
which release blood into the capillaries. Arterioles have strong muscular walls
that control the conduit width, adapting to the specific tissue needs. The cap-
illaries are the thinnest conduits of the system, with a cross section that can be
considerably smaller than the surface of the RBC disk. When passing through-
out the capillaries, the RBC membrane is largely in contact with the capillary
wall. It is here where the oxygen delivery occurs, as well as diffusion of other
molecules transported in the blood. Molecules pass through the capillary wall to
the intersticial fluid, and from here they freely diffuse, reaching all the cells of
the tissue. Capillary walls have numerous pores that facilitate the exchange of
substances. Oxygen-depleted blood is returned to the heart by the vessel system.
Veins transport blood under very low pressures. For this reason, they serve as
blood reservoirs, controlling the volume of extra blood by muscular contraction.
The amount of blood hosted by veins and arterioles is inhomogeneously dis-

tributed. Of the 84% from the total body blood contained by the systemic cir-
culation, 64% corresponds to veins, 13% to arteries, and 7% to capillaries (Hall,
2011). The total cross sectional area for each type of conduit is shown in Table
1.1. In spite of the reduced thickness of the capillaries, they form a very dense
mesh and therefore the total area is enormous. Since the total blood flow Q is
constant, an estimate of each conduit velocity can be obtained from v = Q/A,
and it is shown in Table 1.1, in addition to the correspondent Reynolds number.
The results show that highly turbulent flows are present in the aorta, but laminar
flow is expected in most arterioles and venules and indeed Re is very low in the
capillaries.
The heart pumps blood by periodic muscular contractions. Accordingly, the

blood pressure is also periodic as the pulse wave travels through the circulatory
system. In the large arteries, where the systolic pressure is usually measured, the
periodic pulses are easily detected, and as it is widely known that this represents
one of the main medical indicators of life. However, the pressure damping along
the smaller arterioles and capillaries attenuates the amplitude of the pulse, and
actually blood pressure is almost constant in capillaries, as shown in Figure 1.5.
Figure 1.5: (Left) Blood pressure pulse for different conduits of the circulatory system.
The periodic heart beats produce a periodic pulse of the blood pressure in arteries.
However, the pressure damping in thinner conduits attenuates this pulse and pressure
is almost constant in the thinnest capillaries. Adapted from Hall (2011). (Right) In
vivo red blood cell flow in capillaries. To pass through the thinnest capillaries, RBCs
are forced to undergo severe deformations, squeezing and adopting asymetric shapes.
Source: http://quizlet.com/4383143/circulatorylymph-quiz-flash-cards/
1.3.2 Blood properties
Blood comprises a fluid known as plasma and a high concentration of immersed

cells. Among them, three main species deserve mention: red blood cells, leuko-
cytes, and platalets. RBCs are the major component of blood, representing up
to 95% of the total number of cells. Leukocytes are large cells found in very low
concentrations (around 0.1%), but with crucial functions in the immunity system.
Platalets represent roughly 5% of cells and are responsible of blood coagulation to
control blood loss when the circulatory system is damaged. In healthy humans,
RBC concentration in blood is typically 5.2 · 106 cells per cubic milimeter in men
and 4.7 · 106 in women. This represents a volume concentration (known as hemat-
ocrit) of approximately 40-45%. Nevertheless, RBC concentration is not uniform
in the circulatory system, and it decays in the thinnest conduits. For instance,
concentrations of 10-20% are found in the capillaries (Fung, 1997).
Blood plasma is a solution of diverse molecules (including proteins, min-
eral ions, hormones, and glucose) and 92% of water. It is characterized by
a straw-yellow colour, an adult individuals contain roughly 3 litres of plasma.
Blood plasma density is approximately 1025 kg/m3 , and cells density is around
1125kg/m3 , leading to a value for the whole blood of typically 1060 kg/m3 . Blood
viscosity is, at normal conditions, 3 − 4 · 10−3 Pa s, but it strongly depends on the
hematocrit and RBC mechanical properties, including RBC deformability and
RBC aggregation, as we will study in Part IV of this Thesis. Alterations of blood
1.3. Circulatory system and blood 23
viscosity can be relevant in many disease processes, and hemorheology is an im-

portant area in a medical frame (Baskurt et al., 2007).
Chapter 2
Physical approach to membranes
The complexity of biological membranes constitutes a challenge for any theoret-

ical description. In order to capture their rich behaviour, one needs to carefully
consider the scales of interest of the system. If one is interested in the study of
the overall membrane response at the cell scale, the atomic description is clearly
unaffordable: the complexity of atomic interactions, even in simple molecules such
as lipids, discards any treatment at this scale. Coarse-grained descriptions, which
represent each lipid by a number of grains (typically 3-10) that encompass a re-
gion of the molecule with similar properties (Marrink et al., 2007; Shillcock and
Lipowsky, 2006), offer a path for the study of small sized patches of membranes.
The state-of-the-art numerical methods are able to describe the kinetics of typi-
cally 106 molecules (Marrink and Tieleman, 2013), involving membrane domains
of roughly 100 x 100nm, but still far from the macroscopic cell scale, 10μm x 10
μm. It is clear that this scale is not feasible if one pretends to study the overall
cell response. For this purpose, it is convenient to invoke mesoscopic theories
(Deserno, 2009). By considering the membrane as locally homogeneous and intro-
ducing a continuum description, each small part of the membrane is characterized
by some certain local properties. These properties must be consistent with the lo-
cal molecular structure of the membrane, so that a connection between the micro
and meso scales should be derived.
The molecular complexity of biological membranes only affects to a few essen-
tial aspects of relevance in a physical description of membranes. Namely, these
main features can be summarize as follows
1. Length scale separation. The membrane thickness is of the molecular scale,

∼ 4nm, whereas the radii of curvature of typical objects of interest (such as
cells, vesicles and vacuoles) are typically of microns. This difference ensures
that in the cell scale the membrane can be considered a two-dimensional
sheet.
25
26 Chapter 2. Physical approach to membranes
2. Fluidity. At normal conditions (including lipid composition and physiologi-

cal temperature), the lipid bilayer is fluid in the membrane plane. Thus, it
does not present resistance to shear stresses.
3. Hydrophobicity of the lipid tails. The strong hydrophobicity of the nonpolar

tails of the phospholipids implies two different considerations. First, that
lipid molecules are insolube in water, so that the total amount of lipids in
the membrane is constant (though this fact does not affect the distribution
of lipids between leaflets). Second, that even when assembled in a bilayer,
lipid tails are sensitive to the interaction with water molecules (eg if the
bilayer is stretched and the distance between lipid increases, tails can be
exposed to water). The hydrophobic repulsion forces that the bilayer is
almost incompressible, with a very high resistance to stretch.
4. Bilayer architecture. Although by assertion 1 the membrane can be consid-

ered as a fold, the internal microstructure of the membrae must be taken
into account. The bilayer is formed by two leaflets separated by a constant
distance. However, the leaflets slide each other, accomodating the local lipid
densities to the local shape and curvature of the membrane.
5. RBCs present a spectrin network beneath the bilayer, similar to a string

mesh, whose mechanical properties can be important – or dominant – under
certain conditions. In the typical size of a cytoskeleton patch, ∼ 50 nm2 , the
overall composition of the lipid bilayer is relatively homogeneous throught
the cell membrane, and hence this invites to a continuous description.
6. Active processes. At resting conditions (ie in the absence of external hy-

drodynamic fluxes), RBCs are in a stable state given by the mechanical
equilibrium of their membrane. This equilibrium is determined by the in-
ternal properties of the membrane, which actually are maintained by active
processes. Thus, strictly speaking, cells are out of thermodynamic equi-
librium, since energy is required to maintain the steady properties of the
membrane, but these properties determine the mechanical dynamics and
equilibrium of the cell. Unless otherwise specified, in this Thesis equilibrium
refers to mechanical equilibrium.
In this context, the Helfrich bending energy represents the fundamental theory
of membrane elasticity. Helfrich adapted the general theory of elasticity to the
particular characteristics of membranes, accounting for the structural membrane
properties. The theory has proven to be largely succesful, explaining an extensive
phenomenology of membrane related problems. In this Chapter, we first outline
some important results of the general theory of elasticity, focusing on the bend-
ing deformations which are of special interest within the membrane framework.
2.1. Theory of elasticity 27
We then formally present the Helfrich theory, and briefly describe some of its
main results regarding cell and vesicle morphology. Afterwards, we consider other
alternative or complementary models that haven been proposed to explain new
and more complex phenomena, such as the cytoskeleton elasticity or the bilayer-
couple model. We end with an overview of the characterization of the material
parameters of the model as measured by different experimental techniques.
2.1 Theory of elasticity

Membranes are elastic structures whose mechanical properties are usually de-
scribed by the general theory of elasticity. The deformation of any object is
described by the deformation vector uα = xα − xα , where x are the coordinates
in the relaxed state and x represent the coordinates under deformation. For sim-
plicity we shall assume an isotropic, homogeneous material. The strain tensor is
defined as

1 ∂uα ∂uβ
uαβ = + . (2.1)
2 ∂xβ ∂xα
The energy of an elastic object subjected to a certain deformation, given by

the tensor uαβ , reads

λ 2
Fel = (uγγ ) + μ (uαβ uαβ ) dV, (2.2)
2
where λ and μ are material parameters known as Lamé coefficients. Fel actually
represents an excess energy with respect to the undeformed state. The interpre-
tation of the diagonal terms of the deformation tensor uαβ is simple. Considering
a stretching deformation x = ψx (see Fig. 2.1A), a simple calculation shows
that the non-diagonal terms of the strain tensor are zero, leading to an energy
of deformation Fel = (λ + 2/3μ)ψ 2V , where V is the volume of the underformed
object. Conversely, in a shear deformation x = αx, y = βy (see Fig. 2.1B), which
maintains constant surface, the trace of the tensor vanishes. Given that any pla-
nar deformation can be decomposed into a pure stretching and a pure shear, we
rewrite the elastic energy,
2
1
Fel = μ uαβ − δαβ uγγ + K(uγγ )2 dV, (2.3)
3
where μ is renamed as shear modulus and K = λ + (2/3)μ is the bulk (or com-
pression) modulus. Within the frame of biological membranes, often the elasticity
is described in terms of two new modulii, the Young modulus E and the Poisson
ratio ν, and the expression for the elastic energy now reads

E ν
Fel = u2αβ + u 2
dV, (2.4)
2(1 + ν) 1 − 2ν γγ
where E = 9Kμ/(3K + μ) and ν = (3K − 2μ)/2(3K + μ).

Figure 2.1: Scheme of the main deformations. Bold lines represent the underformed
state and dashed lines the deformation. (A) Stretching, when the object increases its
area; (B) shear, when the object deforms without a change on its area; and (C) bending,
which represents a normal displacement ξ to the surface.
2.1.1 The bending mode
In order to comprehensively understand the elasticity of membranes, we explore

here the elastic properties of plane objects of small thickness compared to their
surface. Thus, we represent the membrane as a generic object, such as a plate,
subjected to a pure bending deformation. Cell membranes have a typical thickness
of 4nm, whereas the overall cell length is usually 5 − 10μm. The separation
in the length scales suggests that the approximation of the membrane as a two
dimensional fold, as previously commented, is largely appropriate. After averaging
the elastic parameters over the plate section and adopting the two-dimensional
description, the Helfrich free energy is recovered.
Let us consider a flat plate of thickness h. Consider now the bending defor-
mation shown in Figure 2.1C, with no in-plane deformation ux = uy = 0 but
out-of-plane displacement uz = ξ(x, y) in the Monge representation (do Carmo,
1976). The stress tensor is given by the variations of the elastic energy with
respect to the deformation,
∂Fel
σαβ (r) = . (2.5)
∂uαβ (r)
2.1. Theory of elasticity 29
The specific dependence on the strain tensor reads
1
σαβ = K(uγγ δαβ ) + 2μ(uαβ − uγγ δαβ ). (2.6)
3
And the force exerted by the object can be derived from
∂σαβ
Fαplate = . (2.7)
∂xβ
The deformation of the object may respond to the action of an external forcing,
Fαext ,
on the object surface. This force is balanced by the internal stresses of the
plate, which are given by the force −σαβ nβ dA. The equation of force balance on
the object surface, where the external forces apply, reads
σαβ nβ = Fαext, (2.8)
where n is the normal vector to the object surface. We suppose that, given that
the plate is very thin, the external forces required to bend it are small compared
to the internal tensions across the membrane (Landau and Lifshitz, 1999), so that
F ext are neglected in (2.8). Note that this hypothesis is a priori not obvious, but
still let us assume its validity under certain conditions. For a strictly flat plate,
nz = 1 and therefore σxz = σyz = σzz = 0. If the plate is only slightly bent,
these components are strictly non-zero but remain small compared to the rest of
components of σαβ , so we can equate them to zero and use the resulting equations
to solve the strain tensor components. For this deformation, all the terms of the
strain tensor can be calculated explicitly:
uxx = −z∂x2 ξ,
uyy = −z∂y2 ξ,
uzz = −(ν/(1 − ν))z(∂x2 ξ + ∂yy
2
ξ),
2
(2.9)
uxy = uyx = −z∂xy ξ,
uxz = uzx = 0,
uyz = uzy = 0.
By introducing these expressions in (2.2), the elastic energy of the bent plate
is

2 2
E 1 ∂2ξ ∂2ξ ∂2ξ ∂2ξ ∂2ξ
Fel,b = z2 + + 2(1 − ν) − 2 2 dV.
1+ν 2(1 − ν) ∂x2 ∂y 2 ∂x∂y ∂x ∂y
(2.10)
Integrating across the plate thickness, from −h/2 to h/2, in the z-direction,

2 2
h3 E ∂2ξ ∂2ξ ∂2ξ ∂2ξ ∂2ξ
Fel,b = + + 2(1 − ν) − 2 2 dS.
24(1 − ν 2 ) ∂x2 ∂y 2 ∂x∂y ∂x ∂y
(2.11)
Some considerations are required to readily identify the ξ-dependent terms in
(2.11). In the Monge parametrization, the normal vector to the surface can be
obtained from n̂ = ∇ξ/|∇ξ|. The total and Gaussian curvatures are defined in
terms of the normal vector as C = −∇ · n̂ and G = 1/2[(∇s · n̂)2 − ∇s · n̂ : ∇s · n̂],
respectively. Hence, straightforward calculations lead to (do Carmo, 1976)
C = −[(1 + (∂x ξ)2 )∂yy

2
ξ + (1 + (∂y ξ)2 )∂xx
2 2
ξ] − 2(∂x ξ)(∂y ξ)(∂xy 2
ξ) ≈ −[∂xx 2
ξ + ∂yy ξ]
(2.12)
,
2 2 2
(∂xx ξ)(∂yy ξ) − (∂xy ξ)2 2 2 2
G= ≈ (∂xx ξ)(∂yy ξ) − (∂xy ξ)2 , (2.13)
(1 + ((∂x ξ)2 )2 + ((∂y ξ)2 )2 )2
for the total (2.12) and Gaussian (2.13) curvatures. We have assumed that the
out-of-plane displacement ξ is small relative to the lengthscale of the plate, so that
the gradients are small ∂ξ 1. By direct comparison of (2.11) with expressions
(2.12) and (2.13), the elastic energy for the bending deformation can be rewriten
as

κ 2
Fel,b = C + κG G dS, (2.14)
2
where we have introduced the bending rigidity κ = Eh3 /12(1−ν 2 ) and the saddle-
splay modulus κG = −κ(1 − ν). This derivation applies generically for objects
such as plates or, in the present context, monolayers. However, a bilayer consists
of two monolayers which glide each other, so cannot transmit shear stresses and
the bending rigidity of the bilayer, κbil arises as the sum of the rigidities of the
monolayers, κm . Considering monolayers of thickness h = d/2, the bilayer bending
modulus is given by
d3 E KA
κbil = κm m
1 + κ2 = = , (2.15)
48(1 − ν 2 ) 24(1 + ν)
where we have introduced the area-compression modulus KA = dE/2(1+ν), which

represents the energetic cost of expand/compress the area of the plate, and it is
2.2. The Helfrich bending energy 31
related with the volumetric bulk modulus K. Hence, assuming a homogeneous

plate and considering a pure bending deformation, the general elastic energy (2.2)
reduces to the bending contribution (2.14). The particular elastic modulii used to
describe cell mechanics depend on the cell species. In the frame of RBC elasticity,
it is usual to characterize the membrane mechanics with the bending and shear
modulii. On the contrary, in cells with an inner structure (such as leucocytes or
epithelial cells), especially in the framework of atomic force microscope experi-
ments, most studies focus on the Young modulus (Kuznetsova et al., 2007). We
hereafter concentrate on the elastic description based on the bending modulus,
which is widely extended in the membrane mechanics field.
2.2 The Helfrich bending energy
The theoretical study of membranes at the cell scale was first performed by Can-
ham (1970), Helfrich (1973), and Evans (1974). They concentrated on the iden-
tification of the relevant elastic properties of the RBC membrane by trying to
reproduce its distinctive discocyte shape. The main assumption of their approach
is that the cell membrane can be described as a two-dimensional fold, based on
its minute thickness compared to the cell length. They initially focused on lipidic
membranes, neglecting any contribution from the cytoskeleton. Helfrich proposed
that from the three main type of deformations that a layer can undergo, shear, tilt
and bending, only the last does play a relevant role in the membrane elasticity.
Accordingly, he generalized the curvature energy (2.14) to describe the elasticity
of lipid membranes, proposing the free energy

κ 2
Fb = (C − c0 ) dA + κG GdA + γdA + ΔpdV, (2.16)
2
where c0 is the so-called spontaneous curvature that accounts for any asymmetry
in the membrane internal structure, whereas γ and Δp generically represent a
surface tension and a pressure difference across the membrane. In the Helfrich
initial description, these two components are Lagrange multipliers to ensure that
cell area and volume, respectively, are conserved. It is remarkable that in part of
the literature there is a misleading use of the mean curvature H = 1/2(c1 + c2 ),
instead of the total curvature C = c1 + c2 , in (2.16). Although the factor 2 does
not obviously affect to the geometrical description of the shape, it is relevant for
the value of the associated material parameters, κ and c0 , which depending on the
specific notation will deviate in a factor of 4 and 2, respectively. In this Thesis
we use the total curvature C (sometimes also represented as K in the literature).
The Gaussian curvature is a topologic invariant, as known from the Gauss-Bonnet
theorem,

GdA = 4π(1 − g), (2.17)
where g is the topological genus, eg number of holes or handles. Consequently,

the Gaussian curvature only plays a role in the membrane elasticity in processes
involving topological transformations. For the case of closed membranes, such as
cells, the Gaussian term remains constant and for simplicity it can be ignored.
The minimization of (2.16) for an ellipsoidic shape under the appropriate values
of area and volume leads to the biconcave discocyte of the RBC as the equilibrium
shape. Ensuing studies investigated the properties and minimal shapes of this en-
ergy and the theory has been refined in order to incorporate other mechanisms, as
explained later on. In the last years, the model has been incorporated to different
dynamic theories, offering the possibility of studying new and more complicate
phenomena. Many of the results of this theory have proven good agreement
with experiments; nice examples include the theoretical prediction of shapes of
the stomatocyte-echynocyte transition (Lim et al., 2002), the study of tubulation
when polymers are attached to a lipidic vesicle and effectively induce a sponta-
neous curvature (Campelo and Hernández-Machado, 2008), or the experiments of
stretching of red blood cells with optical tweezers (Li et al., 2005).
2.2.1 Microscopic realization

The mesoscopic description assumes that there exist domains of small size com-
pared to the length scale of the system, considered homogeneous, so that the vari-
ables at the mesoscale effectively capture the relevant properties of the microscale.
With the objective of explaining this fundamental assumption, and validate the
mesoscopic membrane description of Helfrich, we present here the simple model
proposed by Petrov and Bivas (1984) which, in spite of being highly non realistic,
is useful to naively illustrate the connection between both scales. The model as-
sumes a rough description of the interactions between lipids, and from this simple
basis the Helfrich free energy (2.16) for a bilayer is derived.
The model assumes a harmonic approximation of the free energy per molecule,
2 2
1 AH 1 AT
fm = kH −1 + kT −1 (2.18)
2 A0H 2 A0T
where AH/T are the areas per molecule of the head/tail, respectively, and kH/T are
the harmonic constants related to the respective interactions between each group.
A0 are the preferent areas, related with the equilibrium intermolecular distance
in the relaxed monolayer. Still, the effective constants kH/T should be related
with the specific bonds between molecules, but this is difficult to address at this
simple level of description. If one defines the neutral surface as the point of the
lipid where the forces are balanced, and we call A the lipid section at this point,
the head and tail areas can be expanded in terms of the curvature at this neutral
point as
2
AH/T = A[1 + δH/T C + δH/T G]. (2.19)
where δH/T are the respective distances to the neutral surface. In the case of a
flat interface (C = G = 0), when the areas per lipid at the head, tail and neutral
points are the same, the energy density reduces to
2
1 A
fm = Km −1 (2.20)
2 A0
where Km = kH /A0H + kT /A0T is the stretching modulus per molecule. On the

contrary, for a positively curved membrane, the head and tail areas are expanded
and contracted, respectively, with respect to the neutral one. Thus, introducing
(2.18) in (2.19), the density energy of the deformation leads to the Helfrich energy
(2.16), with the elastic parameters
ψH ψT
κm = δ 2 Km = δH δT Km ,
(ψH + ψT )2
0 0 ψH ψT
G = (AH − AT )(ψH − ψT )
κm , (2.21)
(ψH + ψT )2
κm
G 1
cm
0 = .
κ δH − δT
where ψH/T = kH/T /A0H/T , and the superscript m indicates that it refers to one
of the monolayers. Some properties of the elastic parameters can be infered from
these expressions. First, the bending parameter is always positive, as opposed to
the Gaussian modulus which depends on the specific properties of the lipid. Anal-
ogously, the sign of the spontaneous curvature depends on the relative position
of the neutral surface with respect to the head and chain. Thus far we have just
considered a monoloyer. For a bilayer, the elastic parameters depend on whether
the monolayers are connected or not. For simplicity, we suppose that there is no
interaction and then
κ = 2κm , G − 2κ c0 ΔN ),
κG = 2(κm m m
c0 = 0. (2.22)
where ΔN is the distance between neutral surfaces. In spite of the enormous

simplifications considered in this model, it shows how the Helfrich energy can be
derived from a simple harmonic description of the lipid-lipid interactions.
2.2.2 Equilibrium equation and membrane fluctuations

The membrane equilibrium equation was first derived by Ou-Yang and Helfrich
(1987) for a spherical shape. They considered the free energy (2.16) and use a
variational argument to solve the equation δF /δA = 0. The resulting equation
reads
1
Δp = γC − κ(C − c0 )[C(C − c0 ) − 2C 2 + 4G] − κΔs C. (2.23)
2
This expression can be understood as a generalization of the Young-Laplace
equation for contributions of higher orders in the curvature O(C 2 ). This general
equation has a major limitation, as it is only valid in the limit of low curvatures.
The general equations have also been derived (Capovilla and Guven, 2002; Kozlov,
2006).
Out-of-equilibrium, expression (2.23) relates with the force exerted by the
membrane, fmem = δF /δA, which corresponds to the right hand side of equation
(2.23). This expression can be used to obtain the undulation modes of the mem-
brane. We show here a simplified derivation, though more strict and complicate
procedures have been developed (Kuriabova and Levine, 2008). Suppose an al-
most flat membrane whose height with respect to the equilibrium plane is given
by y = h(x, t). We shall assume that the surrounding fluid is important in the
membrane dynamics, so that the hydrodynamic effects must be considered. The
Langevin equation describing the membrane dynamics is

∂h(x, t)
=− dxΛ(|x − x |)fmem + ξ(x, t), (2.24)
∂t
where ξ is a white noise and Λ is the Oseen tensor accounting for the hydrody-
namics,
1
Λ(x) = . (2.25)
8πηx
Assuming that the membrane profile can be decomposed in a Fourier series,

h = k hk,0 e
iqx
, the membrane force in Fourier space to linear order O(h20 ) is
given by fmem = Δp + γq 2 + κq 4 . Therefore, equation (2.24) reads
∂hq (t)
= −ω(q)hq (t) + ξq (t) (2.26)
∂t
where the relaxation frequency is given by
Δpq −1 + γq + κq 3
ω(q) = . (2.27)
4η
The terms associated to q 1 and q 3 are the surface tension and bending, respec-
tively, whereas q −1 is a chemical term related with the volume conservation. Using
the equipartition theorem, the equilibrium spectrum of undulations is obtained,
kB T
< hq h−q >= a. (2.28)
Δp + γq 2 + κq 4
where a is the area of the membrane patch considered. This expression provides
an important pathway to measure membrane elastic parameters by analyzing the
fluctuation spectrum of the membrane in locally flat regions of the cell, such as
the main disk of the RBC (Brochard and Lennon, 1975; Park et al., 2010b).
2.2.3 Shapes of vesicles
Once the free energy of the membrane has been determined, the first step is
to obtain the minimal shapes for this energy. This minimization is, however,
generally complex and in many cases requires of advanced numerical methods.
Still, if one limits the search to axisymmetric shapes, a few cases are analytically
affordable or, at most, require of simple numerical integrations. We overview here
some of the first results obtained based on the Helfrich theory, as they provide
intuition and key insights for the understanding of more complex morphologies in
the subsequent studies.
Unduloids
It is obvious that for a bilayer of spontaneous curvature c0 , minimal shapes include

a sphere of radius 1/c0 or a cylinder of radius 2/c0 , as both have zero curvature
energy. Interestingly, there is a family of axisymmetric surfaces with curvature
C = 1/c0 , called unduloids. This family comprises from a cylinder to an undulated
tube and eventually a necklace of spheres, as shown in Figure 2.2. These shapes
naively remind phenomena of tubulation and pearling (Tsafrir et al., 2001, 2003)
that are observed when a spontaneous curvature is artificially induced in vesicle
membranes. Likewise, the formation of pearls connected by fine necks can be
associated with the budding and vesiculation phenomena that is known to occur
during the formation of spherocytes.
Figure 2.2: Different unduloids: cylinder, undulated tube and necklace of sphere. These
family of shapes, of strictly zero curvature energy, reminds the shapes observed in vesi-
cles when subjected to prompt changes in their membrane asymmetry, as observed when
vesicles are exposed amphiphilic proteins that anchor to the bilayer and introduce and
spontaneous curvature. Figure from Seifert and Lipowsky (1995).
Shape diagrams for axisymmetric vesicles
The determination of the minimal shapes for closed vesicles is more complicate,
even considering axisymetric shapes. The procedure is as follows. From the com-
plete free energy for the membrane (2.16), the Euler-Lagrange equation is obtained
in the appropriate parametrization of the vesicle contour. Stationary of this equa-
tion is imposed, and (usually numerically) solving this relatively simple equation
provides the equilibrium shape. For symmetric bilayers (c0 = 0), three different
type of shapes are found (Seifert et al., 1991), depending on the specific area and
volume of the vesicle: stomatocytes, oblates and prolates. The relation between
vesicle area A and volume V is captured by the reduced volume, as defined in
Chapter 1. Stomatocytes are found for low vred < 0.592. The vesicle accomo-
dates its area bending and acquiring a cup-like shape. For 0.592 ≤ vred ≤ 0.651,
oblates become minimal shapes. It is remarkable that the discocyte belongs to
this subfamily of shapes. For vred > 0.591, vesicles present prolate shapes.
If asymmetric bilayers are considered, a complex shape diagram in the param-

eter space (vred , c0 ) is obtained. However, the three family of shapes commented
before are invariably present at low values of c0 . Negative spontaneous curva-
ture favours the development of stomatocytes, and in general the transitions are
swifted towards this shape. Interestingly, at high c0 a new branch of shapes is
found: pears and buds. Budding phenomena is therefore necessarily associated to
high asymmetries of the membrane.
2.3. Other membrane models 37
Figure 2.3: Axisymmetric vesicle shapes of lowest bending energy with c0 = 0 as a

function of the reduced volume ν. Stomatocytes, oblates and prolates are found for
increasing ν. The transition between each subfamily of minimal shapes is discontinuous.
Figure from Seifert et al. (1991).
2.3 Other membrane models

The Helfrich model has been extended to incorporate other structural properties
of membranes. Specifically, the Helfrich theory assumes that the membrane can be
considered as a two-dimensional layer, neglecting the internal architecture of the
bilayer. The area-difference elasticity accounts for the coupling between leaflets.
In addition, different descriptions have been proposed to model the elasticity of
the spectrin cytoskeleton, which is not included in the Helfrich theory.
2.3.1 Area-difference elasticity
The Helfrich theory does not consider any internal structure of the bilayer, such
as the coupling between leaflets when both try to accomodate their lipid densities
to the local shape, gliding each other. This effect is incorporated in the so-called
area-difference model, closely related to the bilayer-coupled model. Although the
theoretical basis of these models was adressed by Sheetz and Singer (1974) and
Evans (1974), it was formally proposed by different independent groups in 1991
(Seifert et al., 1991; Waugh et al., 1992) .
The area-difference model can be derived by introducing two densities φ± for
the two leaflets at their own neutral surface. In order to rationalize this description
with the single-fold scheme, these densities are expressed as projected densities
onto the neutral surface of the bilayer,
φ± = φ±
proj (1 ∓ dC), (2.29)
where we have neglected higher order contributions (C 2 , G, etc). Assuming a

harmonic free energy for each monolayer, in the spirit of the microscopic model
presented in Section 2.2.1,
f (m) = (k (m) /2)(φ± /φ0 − 1)2 ≈ (k (m) /2)(ρ± ∓ dC)2 , (2.30)
where k (m) is the compression modulus of the monolayer and ρ± (φ±

proj /φ0 − 1) is
the deviation of the projected density from the equilibrium value for a relaxed
membrane. The total membrane for the bilayer reads
f (b) = (k (m) /2)[(ρ+ − dC)2 + (ρ− + dC)2 ]. (2.31)
For simplicity, we have assumed a symmetric bilayer, d+ = d− = d and k + =

k − = k (m) . Introducing the average density ρ̄ = (ρ+ + ρ− )/2 and the density
difference ρ = (ρ+ − ρ− )/2, the local free energy of the bilayer is rewritten as
f (b) = k (m) [ρ̄2 + (ρ − dC)2 ]. (2.32)
From this relation it can be seen that ρ̄ does not couple to the local shape,
and thus ρ̄ = 0 minimizes the total free energy of the membrane. ρ, however, is
inhomogeneous as it adapts to the local surface. The minimization of the total
free energy of the membrane leads to
ρ(s) = d[C(s) − M/A] + (N + − N − )/(2Aφ0 ), (2.33)
where A is the total membrane area, M is the averaged total curvature, and we
have made use of the relation

dAρ = (N + − N − )/φ0, (2.34)
which fixes the total number of lipids and relates it with the surface membrane
area. By introducing expression (2.33) into the free energy of the membrane
(2.32), and integrating over the membrane surface,
k (m)
F (b) = (ΔA − ΔA0 )2 , (2.35)
2A0
where A0 = A is introduced to avoid confusion with the area-difference ΔA,
calculated from

ΔA = 2d CdA. (2.36)
2.3. Other membrane models 39
ΔA and ΔA0 represent the differences in area between the two leaflets, for the
actual and preferent (relaxed) shapes, respectively. Adding this non-local bending
contribution to the Helfrich free energy, we obtain the complete area-difference
model (ADE),

κ κN L π
F = (C − c0 )2 dA + (ΔA − ΔA0 )2 , (2.37)
2 A0 d2
where we have introduced the non-local bending rigidity, κN L d2 /π. This param-
eter can be related with the bending rigidity, κN L = ακ. The theoretical predic-
tion provides the value α = 3/π, but this parameter is best left free (Seifert and
Lipowsky, 1995). For large α, the non-local term is dominant and this approxima-
tion is usually known as bilayer-couple model. The interpretation of c0 and ΔA0
responds to the framework in which each model is derived. The area-difference
specifically refers to the change in area of each leaflet, typically associated to a
translocation of lipids from leaflet to the other, or a chemical change in the lipids
that lead to a reduction of the relaxed area per lipid. The spontaneous curvature
is however more general, and it can account for different imbalances such as in-
sertion of proteins in the bilayer, among others. Both parameters are, however,
undoubtedly connected and indeed it is easy to prove that any c0 = 0 can be
mapped on a renormalization of the effective area difference.
2.3.2 Cytoskeleton elasticity
The cytoskeleton is a triangulated mesh of spectrin rods. The elasticity of the

cytoskeleton has been studied by many different techniques, but much insight
is still required to completely understand its basic mechanics. The cytoskeleton
presents a low resistance to bend, with a bending modulus at least two orders of
magnitude lower than that of the bilayer (Mukhopadhyay et al., 2002). It does
present, however, resistance to shear and compression in the membrane plane,
and it is known to play a fundamental role in inhibiting budding and vesiculation
processes.
Different models have been formulated to model the cytoskeleton elasticity. A
simple way is to represent it as a spring mesh, relating the spring constant with
the elastic modulii. In the same spirit, a more refined model was propposed by Li
et al. (2005), based on previous studies (Discher et al., 1997; Lee et al., 1999). The
membrane surface is organized in a triangle tessellation, with 6-vertices junctions
that can be flipped, mimicking an adaptative connectivity. The cytoskeleton free
energy is given by a worm-like chain model (Marko and Siggia, 1995), which
provides the elastic force for each spectrin rod

kB T 1 1 L
fW LC (L) = − − +x , x≡ , (2.38)
p 4(1 − x)2 4 Lmax
where p, the persistence length of the chain, and Lmax , the maximum length of
each chain, are parameters of the model related to the spectrin properties. The
total free energy of the cytoskeleton is given by
Li
Fcyt = − dξfW LC (ξ). (2.39)
i 0
where i runs over all the spectrin links of the network. The results of this model
fit well with experimental results of optical tweezers elasticity, among others.
However, it requires of a fine tuning of the model parameters and a considerable
computational cost.
A different approach is to recover the continuum mechanics description and
consider the finite strain theory . It can be shown that the contributions of shear
and stretching for a 2D fold are given by (Evans and Skalak, 1980)

K 2 2 2
Fin−plane = dS0 (1 − λ1 λ2 ) + μ(λ1 − λ2 )/4λ1 λ2 . (2.40)
2
√
where λi are the principal extension ratios, defined as λi = 2i + 1, where i are
the eigenvalues of the Green’s strain tensor,

1 ∂xγ ∂xγ
αβ = − δαβ . (2.41)
2 ∂x0α ∂x0β
In these expressions, x are the coordinates of the actual membrane surface, S,

whereas x0 are the coordinates of the reference surface S0 , in which the cytoskele-
ton is relaxed. The relaxed state of the cytoskeleton is one of the remaining fun-
damental questions about the cytoskeleton elasticity (Hoffman, 2001). Different
experiments and theoretical predictions have pointed towards a relaxed cytoskele-
ton with an elliptic shape (Svodoba et al., 1992; Lim et al., 2002), a spherical
shape (Fischer, 2004), or a discocyte shape (Yu et al., 1973; Stokke et al., 1986;
Li et al., 2005), and to date there is no agreement about the subject.
One important aspect considering the cytoskeleton elasticity and relaxed shape
is the presence of active processes affecting to the spectrin scaffold properties. Gov
and Safran (2005) were the first to theoretically approach this problem, based on
the experimental observation that after ATP-depletion, RBCs exhibit membrane
undulations of lower amplitude than at physiological conditions (Zeman et al.,
2.4. Elastic parameters values 41
1990). This surprising observation was explained postulating that the cytoskeleton
is subjected to continuous phosphorylation of the spectrin-actin links, so that the
bilayer undergoes local detachments of the cytoskeleton that allow larger thermal
fluctuations. The active remodelling of the network represents a relaxation of
the structural and elastic resistance of the cytoskeleton. Gov and Safran (2005)
proposed a simple model to explain the abnormal membrane undulations observed
in RBCs with normal ATP concentrations, which we briefly explain here. They
define the occupation probability of an actin link, from standard thermodynamics
nAT P e
nd = . (2.42)
1 + nAT P e
In this expression, nAT P is the ATP concentration and = (ΔEAT P − ΔEsa −

kB Sf − ΔEmec )/kB T . ΔEAT P is the energy gain per ATP molecule, ΔEsa is the
cost of detaching a spectrin-actin link, Sf is the entropy of the spectrin filaments
and ΔEmec = μ(R − R0 ) is the mechanical energy released by the molecule disso-
ciation, where μ is the shear modulus and R and R0 are the typical and relaxed
length of the spectrin filaments. Considering the regimes in which the bending
modes are dominant, the fluctuation spectrum when active processes are present
is given by,
ΔEmec μa nd
< h2 >AT P 2
. (2.43)
κ 1 + (τ ωb )−1
where ωb = κ/4ηa3/2 is the static fluctuation frequency and τ is the characteristic

time of filament diffusion that enable the actin-spectrin re-attachment. The model
by Gov and Safran (2005) quantitatively explains the large ATP-driven membrane
fluctuations, and highlight the importance of active processes in the cytoskeleton
elasticity.
2.4 Elastic parameters values

The elastic properties of the RBC membrane are highly dependent on the spe-
cific bilayer lipid composition, ATP concentration, age of the cell, and temper-
ature. They are also known to vary with the morphogolical state of the cell,
and echinocytes or spherocytes are considerably more rigid than discocytes (Park
et al., 2010b). Bearing in mind this huge variability, we provide below typical
values for the elastic parameters as measured in experiments.
The bending rigidity has been measured by different experimental techniques,
for instance flicker spectroscopy and phase contrast microscopy, analyzing the
membrane amplitude oscillations due to thermal fluctuations, or disturbing tech-

niques such as micropipette aspiration and by optical tweezers (see Seifert and
Lipowsky (1995) and references therein). Typical values fall between 10-50kB T
with slight deviations depending on the specific technique. The shear modulus of
the bilayer is negligible due to its fluidic nature in the membrane plane, given that
that any shear stress is instantaneously relaxed by the rapid lateral rearrangement
of lipids. The area-compression modulus of flat bilayers has been measured for
different lipid species, with typical values of 100-250 mJ/m2 (Evans and Need-
ham, 1987; Rawicz et al., 2000). Note that, for the given values κ = 2.0 · 10−19 J,
KA = 10−1 J/m2 and cell radius R = 8 · 10−6m, the energy scale of stretching
is much larger than that of the bending KA R2 /κ = 3 · 108 . This huge energetic
penalty imposed to the expansion of area implies that any membrane deformation
driven by bending is required to effectively maintain a constant area. For practical
purposes, it is usual to remove the elastic contribution of the area-compression
and strictly impose the constraint of constant area, for instance by introducing a
Lagrange multiplier, as proposed by Canham and Helfrich and shown in (2.16).
The molecular basis of this phenomenon is found in the strong hydrophobicity
of the lipid tails. An expansion of the membrane area implies that the area per
molecule increases from its relaxed state and the lipid tails are therefore exposed
to the water molecules. To avoid this situation, the attractive interaction be-
tween molecules enforces the bilayer incompressibility. Finally, it is important
to remark that whereas membranes are permeable to transfer of nutrients and
other molecules, they do not allow water exchange. This imply that for closed
membranes (ie cells, vesicles), the amount of enclosed water remains constant.
Assuming incompressibility, this condition can be translated into a constant vol-
ume, and accordingly another Lagrange multiplier, accounting for this volume
conservation, is incorporated to the elastic membrane energy. Hence, the addition
of the bilayer incompressibility and constant volume to the bending contributions
of (2.14) allow to recover the complete Helfrich free energy (2.16).
The cell spectrin cytoskeleton, which plays a fundamental role in the cell re-
sponse under certain deformations (such as morphological changes during crena-
tion (Melzak et al., 2013) or squeezing during optical tweezers experiments (Li
et al., 2005)) adds a shear contribution to the membrane elasticity. Although
the value of the cytoskeleton shear modulus has been under debate until recently,
typically measurements agree around μ=50mJ/m2 (Dimitrakopoulos, 2012).
Part II
Membrane elasticity and RBC

morphological transitions
43
Chapter 3
The disco-echinocyte transition
It has long been realized that healthy RBCs deform from the discocyte shape when
exposed to the action of a number of agents, adopting different morphologies that
can be divided into stomatocytes (cup-like shapes), and echinocytes (crenated
shapes) (Besis, 1973). The first agents to be identified included ATP-depletion
(Nakao et al., 1960), Ca2+ loading (White, 1974), and incubation in the presence
of certain amphiphilics (Sheetz and Singer, 1974), but the list was extended in sub-
sequent research. In particular, the pioneer studies by Sheetz and Singer (1974)
demonstrated that the conversion into stomatocytes or echinocytes is actually dic-
tated by the chemical nature of the amphipatic compound that triggers the shape
deformation: crenator molecules are those that equilibrate in the outer leaflet,
whereas cup formers are incorporated to the outer leaflet. The morphologies ob-
served are virtually independent of the agent, suggesting that the alteration affects
to a very specific and basic elements that determine membrane stability. These
observations provide an important insight about the bilayer leaflets coupling that
gave rise to the bilayer-couple and area-difference models that were presented in
Chapter 2.
The complete morphological sequence is usually known as stomato - disco
- echinocyte, and has attracted interest from biologists, and subsequently from
physicists, for many years as it provides a wealth of information about the mem-
brane molecular balance and functioning. Among many other reasons, echinocytes
can be easily obtained in any laboratory, and actually RBCs spontaneously trans-
form into echinocytes if not properly treated during blood storage. Additionally,
the fact that the shape deformation responds to the action of several agents offers
different ways to understand the problem.
In this Chapter, we present the main characteristics of the disco-echinocyte
transition and some of the hypotheses and explanations that have been proposed
to understand this phenomenon. We then concentrate on the experiments realized
45
46 Chapter 3. The disco-echinocyte transition
by Kathryn A. Melzak and José Luis Toca-Herrera of shape transformations in-

duced by AFM perturbation, which allow to induce shape transitions between the
different morphologies in a controlled manner, offering new information about the
deformations of the cells. A proper characterization of the morphological tran-
sition is required to understand the theoretical study of Chapter 4, in which we
focus on the identification of the physical mechanisms controlling the transition.
3.1 Morphological sequence

The RBC morphologies observed can be divided into three main groups: stoma-
tocytes, discocytes and echinocytes. The later group, however, is subdivided into
different stages (Besis, 1973). The shapes are developed sequentially for increas-
ing intensity of the agent acting, forming an ordered sequence. Typically, the
morphologies are classified following a criteria that accounts for the degree of cre-
nation and sphericity of the cell. The morphological index (MI) characterizes the
deformation state of each cell, associating a number to each morphology. Given
that the discocyte is the healthy morphology, its MI is asigned to 0 (ground state);
crenated shapes are associated to positive numbers, of increasing value for larger
number of spicules and bumps; and cup-shapes are associated to negative values
of the MI. Hence, a high MI implies a very deformed cell from the discocyte,
with a considerable degree of membrane damage due to the strong and prolonged
action of the agent. These shapes are usually referred as advanced stages of the
morphological transition, whereas slightly deformed shapes, closer to the healthy
discocyte, are referred as lower stages. The transition can be understood in two
different but relatively equivalent ways: (i) as a sequence of stable shapes for
increasing intensity in the action of the agent; (ii) as the temporal evolution of
the cell shape, if the time scale of shape deformation is much shorter than that of
the agent effects. For instance, the absorption of lipid molecules occurs in a very
fast time scale, and discocytes deform directly towards the corresponding mor-
phological stage (specified by the amount of lipids incorporated to the membrane)
in seconds or minutes. ATP-depletion, however, induces changes in the mem-
brane composition in typically hours or days, and the cells sequentially develop
the different morphologies, covering the entire transition during their temporal
evolution. The MI of a cell is, by definition, an integer number (see below specific
values for each morphology). In experiments, however, the MI is usually aver-
aged over the entire RBC population of the blood sample, leading to a decimal
value. Given that RBCs are relatively inhomogeneous both in their geometrical
and elastic properties, the morphological response of some cells is faster than in
others, and in a blood sample it is usual to simultaneously observe discocytes and
different types of echinocytes.
3.1. Morphological sequence 47
The complete transition is thereby divided into (see examples of each mor-
phology in Figure 3.1)
• Stomatocyte (MI=-1). The cell bends towards a cup-like shape, with a

variable degree of curvature.
• Discocyte (MI=0). As extensively commented before, the discocyte is a disc

with a biconcave profile, with a concave central region and a convex outer
rim.
• Discoechinocyte I (MI=+1). The first stage of the echinocyte region is

characterized by the presence of some undulations in the contour of the cell
disk. These undulations are relatively smooth, and the cell maintains a flat
shape though they are appreciably less biconcave than discocytic cells.
• Discoechinocyte II (MI=+2). This stage is characterized by the development

of bumps normal to the disc plane, often known as out-of-plane bumps, main-
taining the overall shape of the Discoechinocyte I. The number of bumps is
typically low, ∼1-5.
• Echinocyte III (MI=+3). The cell adopts a more compact shape, close to
an ellipsoid, with numerous bumps (∼ 20) that are increasingly sharp.
• Echinocyte IV (MI=+4). The cell is nearly spherical and its surface presents
a high density of small, spiky bumps. Budding and vesiculation lead to a
loss of membrane area.
• Spherocyte. The cell equilibrates in a small spherical shape when it achieves

the appropriate area to volume ratio.
Once the action of the external agent ceases, the membrane eventually reaches
an internal balance and the cell stabilizes in one of the morphologies. The specific
morphology depends on the final properties of the membrane. At these conditions,
the final morphology is usually very stable and no shape changes are observed for
days. The transition is, however, reversible for those agents whose action can
be driven in the opposite direction; for instance, echinocytes formed by ATP-
depletion return to discocytes if the physiological ATP concentration is restored.
The reversibility is limited to typically MI≤3, since more advanced stages are
usually accompanied by severe membrane damage and surface loss.
Figure 3.1: Scanning electron micrographs of different stages of the disco-echinocyte

transition: (a) discocyte; (b) discoechinocyte I; (c) discoechinocyte II; (d) echinocyte
III; (e) echinocyte IV; and (f) spherocyte. Figure reproduced from Ferrell et al. (1985).
3.2 Agents inducing echinocytosis

The biological processes associated to echinocytosis affect both the bilayer and the
cytoskeleton, and the different experimental techniques can selectively alter one
or the other, offering an interesting tool to understand the interplay and balance
between both structures.
The first subset of studies group those agents that affect the bilayer structure.
The first agent studied in detail was the effect of incubation with amphiphilic
molecules. Lipid molecules exposed close to the RBC surface are incorporated
to the cell membrane and equilibrate at their natural position (inner or outer
leaflet). Direct observations of the process by fluorescent markers (Matayoshi,
1980) confirm the initial hypothesis of Sheetz and Singer (1974). The imbalance
induced by the different amount of lipids between leaflets force the cell to expand
(echinocytosis) or contract (stomatocytosis) the outer leaflet with respect to the
inner one, depending on the preference of the lipid to occupy the outer or inner
leaflet, respectively. A similar effect is observed when RBCs are exposed to high
concentrations of Ca2+ , which activate the phospholipase C that is able to degra-
date some specific phospholipid species of the inner leaflet into simpler molecules
of smaller effective area (Ferrell et al., 1985). Hence, although the amount of lipids
remains constant, the mean area per molecule decreases in the inner leaflet and
the total area of the outer leaflet expands with respect to the inner one. ATP-
3.2. Agents inducing echinocytosis 49
depletion and the addition of vanadate are also known to trigger echinocytosis by
means of the inhibition of the lipid translocases action. The lipids spontaneously
flip towards a non-physiological membrane asymmetry, and this imbalance leads
to crenation. This process is generally much slower and more reversible than
others described here, but it is of great relevance in blood handling and storage,
since echinocytosis occurs if blood is not suitably preserved. Finally, lipid balance
can be also modified by mechanical disruption of the membrane, as demonstrated
by micropipette aspiration (Artmann et al., 1997). Micropipette aspiration is a
widely extended technique for measuring the RBC elastic modulii, but in these ex-
periments aspiration pressures are typical low, and after aspiration the cell rapidly
recovers the discocytic shape. If, however, one applies a high pressure, inducing
a severe aspiration of typically ∼ 1.5μm, the cell adopts a echinocytic shape after
being released. Apparently, the extremely high curvatures that the membrane
adopts inside the pipette force lipid translocation. The degree of crenation of the
cell correlates with the duration of the aspiration experiment, supporting this hy-
pothesis. Moreover, shape recovery was studied both for normal cells and for cells
in presence of vanadate. The later shows much larger time scales for recovering
the discocyte shape, as expected since in normal cells lipid translocases rapidly
regulate the bilayer asymmetry, restoring the initial conditions.
The second group of techniques involves those agents that alter the stability
of the cytoskeleton. For instance, if urea is added to crenated cells, spicules
disappear and the cells deform towards the discocyte (Khairy et al., 2010). On
the contrary, if urea is added to discocytic cells, they assume a elliptocytic shape
(Khairy et al., 2010). These observations have been rationalized by the hypothesis
that urea weakens the cytoskeleton scaffold as it is known to denature spectrin.
However, this hypothesis should be confirmed by other techniques since urea is
an aggresive agent that could potentially affect other molecules of the membrane.
Wong (1999) proposed a mechanism for echinocytosis in which the lipids play
a secondary role, and cell shape change is entirely controlled by the proteins of
the cytoskeleton that attach to the bilayer. The cytoskeleton is attached to the
bilayer by ankyrin molecules, which in turn are bound to Band 3, the protein
that mediates influx and efflux of anions and cations. Depending on the ion
concentration, Band 3 assumes an inward or outward position in the membrane,
and thus it modulates the cytoskeleton tension, inducing relaxation or contraction
of the network. Thereby, alterations in the pH or addition of Ca2+ affect to the
normal anion equilibrium, thus modifying the cytoskeleton state and leading to
echinocytosis. Likewise, amphiphilic drugs and detergents inhibit the Band 3
anion transport, inducing a similar response of the cell. This hypothesis, however,
fails to explain other phenomena such as the mechanical induction of echinocytosis.
The experiments explained so far focused on the formation of echinocytes by
manipulation of discocytes, but the reverse process has also been investigated in a
few studies. Additional experiments have reversed echinocytosis by affecting the
bilayer balance, though their interpretation in terms of the bilayer-couple model
is less direct. For instance, it has been shown that hemolyzed echinocytes produce
discocytic ghosts (Lange and Slayton, 1982), and that electroporation is able to
avoid shape changes triggered by one of the different crenator agents (Schwarz
et al., 1999). The interpretation of these effects is based on the perforation of
the membrane and formation of pores, which allow an easier flip of lipids between
lipids.
Although many different agents capable of inducing echinocytosis have been
identified, allowing the description of several new mechanisms of shape control,
our knowledge on quantitative aspects of the transition is very limited. Experi-
mentally, only Ferrell et al. (1985) were able to provide accurate measurements of
the area expansion required to induce shape transformations. Theoretical stud-
ies based on physical membrane models (Mukhopadhyay et al., 2002; Khairy and
Howard, 2011) have described the basic physics behind these phenomena, recover-
ing a number of shapes with nice agreement with those observed in experiments,
but a quantitative characterization of the transition is still lacked.
3.3 AFM-induced RBC shape change
Atomic force microscope (AFM) has been used in different ways to characterize
mechanical properties of cells and membranes (Alcaraz et al., 2003; Kuznetsova
et al., 2007). Although many of the techniques are conceived for determining the
properties of the cell (or material, depending on the context), AFM has numerous
biomedical applications as an active tool to manipulate or disrupt very specific
regions of the cells, eg in plasmid delivery to viable cells, which then express the
encoded proteins, through a pore formed by the AFM tip (Lamontagne et al.,
2008). Other examples include the extraction of membrane lipids and proteins
(Ikai and Afrin, 2003), and localized disruptions of the cytoskeleton (Lamontagne
et al., 2008). Therefore, AFM offers a pathway for RBC shape manipulation
by mechanical alteration of the membrane properties, and the experiments by
Kathryn A. Melzak and José-Luis Toca-Herrera exploit this technique to explore
RBC membrane response to an external mechanical disruption. The study is based
on the observation that echinocytes pushed by the AFM tip deform into other mor-
phologies, implying that conformational changes in the membrane structure are
involved. The link between the disruption of the AFM and the membrane balance
is not direct. In order to clarify the principles driving these shape transforma-
tions which could help us to understand the effect of the AFM on the membrane
3.3. AFM-induced RBC shape change 51
structure, we perform a theoretical study (presented in Chapter 4), providing a

quantitative explanation and description of the these transitions. The theoretical
results allow us to interpret and understand the experiments described below.
Figure 3.2: Shape changes for different initial RBC morphologies after puncturing with
the AFM tip. Cells are shown (a) before the interaction with the tip, (c) after the
measurements, and (c) after a period of 2.1, 2.5 and 2 hours after the measurement, for
the cells 2, 4, and 5 respectively, in order to demonstrate the stability of the obtained
shape. For cells 1-3 the puncture produced the disappereance of the out-of-plane bumps
placed in the central region, characterizing a transition from the discoechinocyte II to
the discoechinocyte I (1,3), or to the discocyte (2).
3.3.1 Experimental results
The study focuses on the morphological response of crenated RBCs distorted by

the tip of an AFM. The series of experiments described here presents two main
novelties with respect to the methods explained in the previous section: (i) only a
mechanical disruption is induced, so that chemical processes such as anion fluxes
are not involved; and (ii) experiments focus on the recovery of discocytes from
damaged cells, instead of inducing echinocytosis.
The details of the experimental setup are not described here, but we outline
briefly the main features of the experiments (for further information see Melzak
et al. (2012)). RBCs from a single individual were washed and preserved in buffer.
RBCs are known to undergo echinocytosis during storage due to partial depletion
of ATP. After one day, most cells were seen to transform into echinocytes, and
vanadate was added at a concentration of 200μM to ensure that lipid transloca-
tors were inhibited. Discoechinocytes II (thus, disregarding echinocytes IV and
spherocytes which had potentially undergone vesiculation) were selected visually,
Figure 3.3: (A) Percentage of shape change after the AFM measurements, as a func-
tion of the maximum force exerted by the cantilever. The percentage of shape change
increases with the force up to nearly 100% for the maximum load 20nN. This curve is
expected to depend on the particular properties of the AFM tip. (B) Force-distance
curve of the AFM. The discontinuity marked with an arrow coincides with the typical
membrane width, suggesting that the membrane was punctured at that point.
focusing on cells with flat main body but with 1-5 spicules in the central region of
the cell. The loading rate at which the tip approaches the membrane was main-
tained constant at 1 μm/s, before maintaining a constant force for 10 s. AFM
measurements of the force exerted by the membrane during the interaction with
the AFM cantilever were performed. Force-distance curves were recorded at max-
imum loads of 2 to 20 nN. After the measurement, a considerable proportion of
cells were observed to transform into discocytes or lower stages of the transition
than the initial one. Shape response was systematically recorded for 390 cells.
Echinocytes with a range of different initial shapes were shown to recover
less advanced shapes of the transition, as shown in Figure 3.2. In the case of
echinocytes IV it is difficult to identify the final morphology, although they clearly
transform into a smoother shape. Dischoechinocytes II and I, however, clearly re-
cover a previous stage, either discoechinocyte I or discocytes. The shape obtained
after the interaction with the AFM tip was largely stable, and no perceptible
shape changes were observed after 2 hours. The proportion of cells that changed
shape increased with the applied force, as shown in Figure 3.3 A, in a set of exper-
iments with the same AFM tip. The results show that at weak loads ∼ 5nN, the
percentage of cells exhibiting shape changes remains low, but it increases linearly
reaching ∼ 100% for a 20nN load. This curve is expected to depend on the AFM
tip geometry and size, but it reflects the general trend regardless of these factors.
Many of the force-distance curves showed a discontinuity of 5nm lenght, as
shown in Figure 3.3 B. This length scale corresponds to the typical membrane
width, suggesting that the AFM tip was puncturing the membrane at this point.
There is no apparent correlation between punctures and shape change, since mem-
3.3. AFM-induced RBC shape change 53
brane puncture was sometimes not accompanied by shape change, whereas some
cells deformed even if the curve-distance curve did not indicate any puncture.
However, it is remarkable that this discontinuity cannot be recorded during the
10 s period that the force was held constant, so that the percentage of punctures
recorded from the force-distance curves does not necessarily correspond to the
total of punctures performed, and likely capture just a low proportion of the total
punctures. During the constant force loading, direct observation showed that the
cell maintained its echinocytic shape, suggesting that the force was not strong
enough to disrupt the cytoskeleton.
Other rare events deserve mention. When the AFM tip was dried in air after
a series of experiments, subsequent experiments showed a low percentage of shape
changes. If, however, the tip was cleaned, it recover its ability to induce shape
changes. Finally, ghosts were rarely observed after the AFM measurement and
they were likely produced due to the formation of large pores.
3.3.2 Interpretations and discussion

The experiments described in the previous Section evidence that mechanical dis-
ruption of the membrane drives shape changes in RBCs, in this case from cre-
nated to discocytic shapes. However, there is no direct observation of which kind
of disruption that the AFM tip induces. Possible mechanisms that explain this
phenomenon include distortion of the cytoskeleton, the bilayer, or both.
Given that chemical conditions are held constant (ie there is no anion increase
and ATP is depleted in the entire series of experiments), the only explanation
concerning the cytoskeleton is direct mechanical disruption of the network. In
this hypothesis, weakness of the cytoskeleton would lead to a solely lipid mem-
brane, whose minimal shape corresponds to the discocyte, explaining the observed
echinocyte - discocyte transition. This picture requires, however, that the for-
mation of echinocytes after ATP-depletion is driven by conformational changes
in the cytoskeleton, a process that has not been described and seems unlikely.
Large disruptions in the cytoskeleton are expected to produce cell deformation
into elliptocytes or pear shapes, as occurs in lipid vesicles under high membrane
asymmetries (Seifert et al., 1991). Very local perturbations of the network should
be accompanied by budding or the formation of very thin necks at the bump
base. In addition, many echinocytes were seen to be stable after the AFM tip
disruption if puncture did not occur, suggesting that the force exerted by the tip
was not sufficient to disrupt the network. This phenomenology indicates that the
cytoskeleton remained largely unperturbed during the experiments.
The most plausible explanation is the alteration of the bilayer microstructure
when the AFM tip punctures the membrane. Within this scenario, echinocytes are
Figure 3.4: Sketch of the mechanism proposed for shape change when the AFM tip
interacts with the echinocyte membrane. The temporal breaking of the hydrophobicity
barrier by the formation of a pore allows flip flop of lipids, and the natural balance
between lipid densities of the leaflets can be restored.
formed after ATP-depletion by an increase in area of the outer leaflet, which can be
due to either an imbalance in the lipid densities between leaflets, due to inhibition
of lipid translocases, or to degradation of some lipid species. ATP depletion also
avoids remodelling of the cytoskeleton. The AFM disruption would affect this
imbalance between the monolayer areas, allowing the recovery of the initial state.
We propose that the puncturing of the membrane forms a pore, which is open
during the maximum load of the AFM (pores at normal conditions are rapidly
self-healing). The pore induces a temporary breaking of the hydrophobic barrier
and hence it permits flip flop of lipids between leaflets during roughly 10 s, allowing
a relaxation of the lipid density imbalance and recovery of the healthy discocyte
shape, as shown in Figure 3.4. The hypothesis is consistent with the experimental
observation of membrane puncturing and the fact that cells return to previous
stages of the transition; as commented before, disruptions of the cytoskeleton are
associated to a different class of shapes. Other possible explanations, such as
mechanically induced degradation of ceramids that would balance the membrane
asymmetry (López-Montero et al., 2005) are unlikely as the time scale of this
process (minutes) is much larger than the prompt shape changes observed in the
experiments.
Chapter 4
Membrane elasticity and RBC

morphology
The altered shapes of the disco-echinocyte transition, unless until budding oc-
curs, can be explained from conformational changes in the microscopic structure
of the cell membrane. This has attracted the interest of physicists, as the predic-
tion of crenated shapes serves to test physical membrane models. Many of the
key insights achieved recently about membrane elasticity relate with the interpre-
tation of echinocytes. However, the shape at which the cytoskeleton is relaxed
remains under debate. As commented in different sections of this Thesis, recent
approaches to the microstructure of the cell membrane have shown strong ev-
idence of ATP-dependent active remodelling of the network, both theoretically
(Gov, 2007; Li et al., 2007) and experimentally (Pinder et al., 1977; Park et al.,
2010b). Accordingly, as postulated by Stokke et al. (1986) and Li et al. (2005),
the cytoskeleton could be able to relax its stress energy over long timescales by
remodelling of the actin vertices of the network, but acting as an elastic mesh if
subjected to prompt perturbations. With this assumption, on healthy cells the
resting shape of the network would be identified as the discocyte. Conversely,
macroscopic models focusing on the cell morphologies have assumed a static sce-
nario, with spherical or ellipsoidal shapes fixed as resting configurations of the
cytoskeleton, and indeed they succesfully predicted some of the morphologies ob-
served in the discoechinocyte transition (Mukhopadhyay et al., 2002; Peng et al.,
2010; Khairy and Howard, 2011). Nevertheless, a global picture, encompassing all
the experimental evidence, is still lacked.
In this Chapter, we perform a theoretical study of the membrane elasticity
during the disco-echinocyte transition, paying attention to the minimal shapes
observed and characterizing the transition in terms of the energetic contributions
of the membrane components. The study has two main objectives. First, to
55
56 Chapter 4. Membrane elasticity and RBC morphology
provide a quantitative description of the morphological sequence, facilitating the

understanding of the elastic and structural role of each membrane component.
Afterwards, the identification of the physics behind the shape transitions will allow
to interpret the AFM experiments presented in Chapter 3 in terms of membrane
internal balance between the cytoskeleton and the bilayer.
We focus on the first three stages of the disco-echinocyte transition, the dis-
cocyte, the discoechinocyte I and the discoechinocyte II. The characteristics of
the spicules and the elastic properties of the crenated shapes have been studied in
the later stages of the transition (Ferrell et al., 1985; Lim et al., 2002), within the
regime of the spheroechinocyte, but little is known about its first stages. We will
study the shape transformations driven by the increase in the preferred asymmetry
and how it affects to the energy landscapes.
Figure 4.1: Comparison between the minimal shape obtained in the Cassini ovals
parametrization, at zero asymmetry c0 = ΔA0 = 0, and a RBC profile experimentally
estimated from a picture takken by high-resolution camera C2, and assuming Beer’s
law. Adapted from Lewalle and Parker (2011).
4.1 The model
4.1.1 Membrane elastic energy

We consider the area-difference elasticity (ADE) model for describing the bilayer
and the finite strain theory description for modeling the cytoskeleton, both pre-
sented in Chapter 2, but outlined here again for the sake of clarity. Therefore, the
RBC membrane elastic energy, Emem , arises from the interplay between the lipid
bilayer, Ebil , and the spectrin cytoskeleton, Ecyt ,
Emem = Ebil + Ecyt . (4.1)

4.1. The model 57
where the ADE model includes the Helfrich bending energy, and a non-local term
that penalizes any deviation from a preferred area difference between the two
leaflets ΔA0 ,

κ κN L π
Ebil = (C − c0 )2 dS + 2
(ΔA − ΔA0 )2 , (4.2)
2 S ARBC d
The integration of expression (4.2) is performed over the current surface of the
cell S. Both contributions to the membrane asymmetry are grouped in a unique
dimensionless parameter,
Δa0 = ΔA0 /ARBC + κN L dc0 /κ. (4.3)
The cytoskeleton elastic contribution is represented by two terms correspond-

ing to shear and stretching:

K 2 (λ1 − λ2 )2
Ecyt = (λ1 λ2 − 1) + μ dS0 , (4.4)
S0 2 2λ1 λ2
The integration is performed over S0 , the resting shape of the cytoskeleton at

which its stress energy vanishes. We assume a fluid cytoskeleton with an initially
relaxed network for a discocyte shape (the minimal shape of the membrane free
energy (4.2)); the discocyte is therefore a ground state of the morphological tran-
sition, in the sense that it does not present a constant storage of membrane stress
energy. When the asymmetry is increased, the timescales of shape change are
shorter than those of network remodelling, and hence the cytoskeleton displays
memory, acting as an elastic layer. Since we will focus on the early stages of the
transition, we have not included non-linear terms (Lim et al., 2002) as they are
only relevant when the protrusions are very spiky, in the latest echinocyte regimes.
4.1.2 Shape description
We perform a parametrization of the cell surface based on the Cassini ovals curves
for the cell profile, but specific treatment is required to describe the spicules and
bumps developed by the cell during echinocytosis. We therefore introduce two
specific parametrizations for the in-plane and out-of-plane bumps.
The modified Cassini ovals.
The minimization of the Helfrich free energy (corresponding to the first term in
equation (4.2)) requires numerical methods, since to date no analytical solution
Figure 4.2: Different RBC morphologies: (0) discocyte, (I) discoechinocyte I, and (II)
discoechinocyte II. (e) In vitro cells as observed by optical microscopy. The white bar
corresponds to 2μm. Images were acquired with a transmission optical microscope (Axio
Observer D1 Zeiss, Germany), using a 100x oil immersion objective lens (NA 1.3). (t)
Theoretical shapes obtained by the model under constraints of total area 140 μm2 and
volume 90μm2 , at Δa0 = 0.001, 0.017 and 0.024 (defined in equation (4.3)), respectively.
The increase in the asymmetry forces the discocyte to develop in plane undulations.
After the undulations have acquired a considerable height, bumps appear in the central
flat region of the disk. Even at this state, the cell remains as a nearly flat disk. The
following stages, characterized by a spherical central body with numerous conical spikes,
are known as echinocytes.
has been found. In an attempt to obtain an analytical description as close as pos-

sible to the exact solution, different expressions have been proposed (Evans and
Skalak, 1980). Among them, the Cassini ovals represent an elegant parametriza-
tion of the cell profile, ranging from the critical lemniscate of Bernoulli to a circle,
in a shape transformation controlled by just one parameter. Even if these curves
do not represent an exact solution of the minimal shape of the Helfrich free en-
ergy, they offer a consistent basis for an analytical study of the properties of the
curvature energy. They have been previously used as a parametrization of the cell
profile, e.g. to study spherocytosis of the cell (Martino and Zampirollo, 1978), or
for fitting data from neutron interferometry of the RBC resting shape (Hellmers
et al., 2006). In our model, we make use of the Cassini oval family to parametrize
the cell surface, in the spirit of Canham’s original paper. A comparison between
the minimal shape obtained from a Cassini oval parametrization at zero asymme-
try and an experimental profile of a RBC is shown in Figure 4.1.
The geometrical properties of the Cassini ovals have been studied in detail
(Angelov and Mladenov, 2000). This plane curve obeys the expression:
4.1. The model 59
(x2 + z 2 + a2 )2 − 4a2 x2 = c4 . (4.5)
To obtain a discocyte shape (see Figure 4.2 0) the curve must be rotated
around the polar axis, leading to an axisymmetric surface. Thus, expressed in
cylindrical coordinates, the cell surface is characterized by

√
z(r) = 4a2 r 2 + c4 − a2 − r 2 , (4.6)
where the ratio ≡ a/c, usually known as biconcavity, describes the shape of the
profile. In the limit → 0 it reduces to a circle, but in the range 0.7 < < 1
the curve presents a concave region in the center, increasing the ratio between
the maximum and minimum heights with . The radius of the disk is defined by
√
R = c 1 + 2 . In the case of the RBC, both volume and area are conserved, and
then another parameter must be introduced (Canham, 1970), reading

√
z(r) = B 4a2 r 2 + c4 − a2 − r 2 , (4.7)
where B fixes the height of the profile and the parameter set (, R, B) entirely
determines the shape and the size of the manifold subject to the constraints of
constant area and volume (ARBC , VRBC ).
In-plane bumps.
To study the in-plane undulations that appear in the discoechinocyte I (Figure

4.2 I), the radial contour is modified introducing an angular dependence,
R(θ) = Rd (1 + Y(θ)), (4.8)

where Rd represents the radius of the central disk. The function Y(θ) = m am (1+
sin(mθ)), expressed in terms of sinusoidal functions, will determine the shape of
the undulation. It is also necessary to allow for an angular dependence of the
parameter B = B(θ), to smooth the undulated surface in the central region. We
found that over extensive ranges of Δa0 , there is a specific mode l which assumes

the main contribution to the sequence, typically ( m=l am )/al < 0.1. Thus, the
undulation is basically specified by just one mode and the rest of modes define de-
tails of the shape. We use this fact to introduce a simplification of the model: each
virtual configuration is generated with a fixed l, separating the different modes as
different configurations. This method also allows for an interesting direct compar-
ison between modes, as well as an easier identification of the competition between
modes in spite of the obvious loss of degree of freedoms. The undulation height
can be identified as 2δRd , where we have defined δ ≡ al for the specific mode l.
Out-of-plane bumps.
The out-of-plane bumps that characterize the discoechinocyte II grow on the flat
central region of the disc, presenting an apparent random distribution over the
surface, as opposed to the in-plane bumps which present a more regular spacing
(Besis, 1973) along the contour (see Figure 4.2 II). Following Waugh (1996) and
Lim et al. (2002), we parametrize these bumps as modified gaussians modulated
by a smooth truncation function. The algebraic expression for the cell height at
the bump region takes the form:
α
zbump (r) = z(r) + s(rb ) he−((rb −Rb )/c) , (4.9)
where z(r) is the height (4.7), Rb is the central position from where the bump
grows (thus Rb < Rd ), h is the bump height, α fixes the shape of the bump and c
is related with the bump diameter at its base, c(Db ). We introduce the function
s(r) = exp(−A2 /(Db − rb )2 ) to ensure that the bump contribution vanishes at
rb = Db . Note that in this expression, rb is the radial variable with respect to the
center of the bump, r = R b + rb . Thus, we explore the size (h, Db , c), shape (α, A)
and location in the cell (Rb ) of the bump.
4.1.3 Minimal shapes of the model

The expansion of the outer area with respect to the inner one forces the cell to
deform, leading to morphological changes as previously described. To reproduce
this shape transition, we increase the preferred area-difference Δa0 , obtaining the
shapes which minimize the membrane energy, (4.1), for each value of the asym-
metry. Given that the preferred area-difference is time-dependent as it increases
gradually with the action of the damaging agent, the sequence of minimal shapes
obtained mimics the temporal evolution of the cell. The rate of change of Δa0 is,
however, not properly controlled and it is defined by the timescale in which the
agent impairs the membrane structure, typically minutes (López-Montero et al.,
2005) to hours (Ferrell et al., 1985), but we will not enter into a dynamic charac-
terization of the transition.
Elastic parameters
We choose typical values for the elastic modulii κ̄N L = κN L /κ = 1/π, K̄ =

KARBC /κ = 3.3 · 103 , and μ̄ = 2K̄. This corresponds to K = 2.0 · 10−6 J/m2 , a
4.1. The model 61
Figure 4.3: Energy for a sequence of RBC shapes with increasing degree of crenation,
corresponding to the shapes shown in Fig. 1. The competition between the bilayer
and the cytoskeleton, which opposes resistance to the undulation growth, defines the
position and depth of the energetic well. The heights of the protrusions are rewritten
in a dimensionless form ū = 2δRd /R0 and h̄ = h/R0 , where R0 is the radius of the
initial discocyte. In (a), the low asymmetry Δa0 = 0.0024 implies a relaxed bilayer
for a discocyte morphology. (b) When the asymmetry is increased, the presence of
membrane protrusions involves an important bilayer relaxation that favours crenation;
at Δa0 = 0.017, a minimal shape is found at ū = 0.095, corresponding to an undulation
height of 0.39 μm (in (a) and (b), h̄ is found to be zero). (c) At higher Δa0 = 0.024, the
out-of-plane bumps appear and the energy minimum moves to h̄ = 0.1, corresponding
to 0.41μm; in this plot the minimal ū ranges from 0.096 to 0.13.
typical value found in experimental measurements (Dimitrakopoulos, 2012). The

energetic scale of the system is therefore defined by the bending modulus κ, with
a typical value of ∼ 50kB T . The energy is expressed in a dimensionless form,
Ē = E/κ.
We fix the RBC area and volume at ARBC =140 μm2 and VRBC =90μm3
respectively, and they are maintained constant during the transition. This is
imposed by exploring a wide range of values of {Rd , B} for a specific set of the
rest of parameters, and selecting those which fulfill the constraints (Canham,
1970).
Geometrical parameters are rewritten in units of the radius of the discocyte
at physiological conditions, R0 . We define the height of the undulations with the
new variable ū = 2δRd /R0 , and analogously the dimensionless form of the bumps
height reads h̄ = h/R0 .
Energy minimizations
Since the integrals of expressions (4.2) and (4.4) over the resulting surface is not
generally analytical, we perform numerical integrations by means of the following
procedure. We discretize the space, and at each point the local energy densities
of the bilayer and the cytoskeleton can be analytically calculated for the elastic
parameters given previously (ie the problem reduces to compute the surface ele-
ment, mean curvature, and extension ratio at each grid node; for further details
see Appendix A). The integration is then straightforward, obtaining the desired
total energy for a specific set of the geometrical parameters (, δ, h, B, α). We
perform integrations over domains of 500x500 nodes, achieving a high precision in
the energy calculation (eg errors in the case of a sphere, which can be calculated
analytically, of order ΔE/E ≈ 10−4 ).
The energy minimization is performed as follows. The parameter space is dis-

cretized in a grid of roughly 50 points for each parameter (with some differences
between parameters). For each Δa0 , the energy is evaluated for all the config-
urations of the parameter set. To obtain the absolute energy minimum at the
relevant asymmetry, the parameter grid is explored sequentially. The parameters
are ordered, and the whole parameter space is explored by successive sweeps of the
one-dimensional subspace corresponding to each parameter. For example, fixing
all the parameters except the last one α, the minimization problem is reduced
to the one-dimensional case. After sweeping the entire range for the rest of pa-
rameters, the new energy landscape involves only the parameters (, δ, h, B) and
each new configuration is minimal for α. The absolute minimum for the complete
parameter space is reached once the minimization process has been carried out
for the whole sequence. The elastic energy has two contributions (bilayer and
cytoskeleton), each one displaying a simple monotonic behavior over the whole
parameter range. The global minimum strongly dominates the whole space. This
has been tested by permutations in the order of the sequence, in order to check
if the minimization procedure is trapped in a metastable state related to local
minima. We find that, regardless of the order of the sequence, the minima ob-
tained fall in a range with variations below 0.5%. These results point out that the
minimization method is robust and the minima obtained can be safely identified
as the absolute minima of the system. Hence, given a Δa0 , an energy minimum
Emin (Δa0 ) is found for some specific values of the geometrical parameters. The
morphology state of the cell can then be identified: the discocyte is obtained at
biconcavities > 0.7 and no protrusions {δ = 0, h = 0}; the discoechinocyte I is
characterized by the in plane undulations, but it still lacks out-of plane bumps
{h = 0}; and finally, the discoechinocyte II, in which both bumps and undulations
are simultaneously present {δ = 0, h = 0}.
4.2. Results 63
Parameter Interpretation Value

κ Bending modulus 1.0
κ̄N L Non-local bending modulus 1/π
K̄ Shear modulus 3.3 103
μ̄ Stretching modulus 2K̄
R0 Discocyte radius 4.08μm
d Bilayer width 3 · 10−3 μm
Rd Disk radius A,V constraints (*)
B Disk height A,V constraints (*)
l Number of undulations 2-15
δ Undulation height 0.0-2.0 μm
h Bump height 0.0-2.0 μm
α Bump shape 1.0-8.0
Db Bump diameter 0.4-4.0 μm
Rb Bump position 0.0-0.8RRBC
Table 4.1: Parameters of the model an their numerical values. The elastic modulii are
expressed relative to the bending modulus κ. The geometrical parameters are expressed
in dimensions, for typical values for RBCs (μm) to facilitate an intuitive understanding.
In the text, morphological parameters are expressed in units of the discocyte radius
R0 . For the parameters that are varied, the table indicates the range explored in the
different minimizations. Parameters marked with (*) are directly fixed by applying the
constraints of constant RBC area and volume.
4.2 Results
4.2.1 Energy contributions
When the membrane asymmetry is increased, cell shape transformations can be

understood from the interplay between its two main components, the bilayer and
the cytoskeleton. The bilayer, with an increasing excess of area in the outer layer,
forces the cell to deform and curves the membrane. The growth of spicules and
protrusions represents a natural way to favour the expansion of the outer leaflet.
In contrast, the cytoskeleton offers resistance to spiculated shapes, preserving
the discocyte morphology of its resting shape. The competition between the ex-
panding bilayer and the resistance of the cytoskeleton defines the morphological
response of the cell to Δa0 , as shown in Figure 4.3. Let us consider the case (a) of
the Figure in the context of the previous section. The energy is represented here
as a function of ū, which depends on δ, and the membrane energy of each config-
uration has been obtained as a minimum of the subspace (, h, B, α). From the
plot, the energy minimum is attained at ū = 0, representing the minimal shape at
the given asymmetry that is identified as the shape observed in the experiments.
In Figure 4.3, the energy of both the bilayer and the cytoskeleton is plot-
ted for a sequence of shapes with increasing degree of crenation (given by ū and
h̄). The cytoskeleton (which does not depend on the value of Δa0 ) is distorted
when deformed from its reference state, and hence its elastic energy rapidly grows
with the undulation height. At low asymmetries, Figure 4.3 (a), the bilayer is
relaxed and the minimal shape corresponds to the discocyte. When the preferred
area-difference is increased, Figure 4.3 (b), the bilayer is stressed by its intrinsic
asymmetry, and therefore prefers undulations with large mean curvatures. An
energy minimum of the total membrane arises as a compromise between an undu-
lation high enough to considerably relax the bilayer, but implying a limited elastic
penalization by the cytoskeleton. If we consider the dependence on h̄, we see that
the presence of out-of-plane bumps, Figure 4.3 (c), require of further increase in
Δa0 , as the cytoskeleton distortion of these structures is considerable larger than
in the previous case. This can be tested comparing insets in Figure 4.3 (a) and
Figure 4.3 (b), where the energy barrier of the out-of-plane bump is several times
larger than that of the undulations. The energy wells of these protrusions, of
several hundreds of kB T , ensure a large stability of the shapes, in agreement with
RBCs that do not exhibit visible fluctuations when observed through an optical
microscope.
4.2.2 Shape diagram
From the energy minimizations, for each preferred area-difference a unique mini-
mal shape is found, allowing to build up the sequence of real shapes and identifying
the stability range for each morphology. In Figure 4.4, the energy difference be-
tween the three configurations of interest is presented for each membrane compo-
nent (bilayer, cytoskeleton and total membrane). In our formalism, the discocyte
represents a subfamily of shapes of the discoechinocyte I. Similarly, both stages
are particular cases of the discoechinocyte II. Thus, within the range of stability
of the discocyte, minimal discoechinocytes I and II degenerate into the discocyte.
Dashed lines represent the minimal energies for the bilayer contribution (4.2),
whilst bold lines represent the evolution of the shapes minimizing the total en-
ergy, (4.1). The transition to positive energy differences defines the shape change
from one configuration into the other. To attain an undulated shape, the bilayer
needs to store enough energy to beat the cytoskeleton resistance to gain curvature.
In the case of the first in-plane undulations, the bilayer is favoured by crenation at
low asymmetries Δa0 ≈ 0.003. However, the negative effect in the cytoskeleton,
presented in Figure 4.4 (a), delays the developement of undulations until a consid-
erable increase in the preferred area-difference, Δa0 = 0.0063. Discoechinocytes
I are then energetically favourable for the total membrane in detritment of dis-
4.2. Results 65
cocytes, which are no longer the stable shapes. At this asymmetry, the energy
difference for the bilayer is ∼ 12κ ∼ 500kB T , consistent with the elastic energy of
deformation of the cytoskeleton. A similar process is observed in the transition be-
tween discoechinocytes, from I to II, in Figure 4.4 (b). In this case, the growth of
the out-of-plane bumps characteristic of the later is again triggered by the bilayer
at low asymmetries. The action of the cytoskeleton shifts the shape transition up
to Δa0 ≈ 0.0175. Bumps therefore require of larger asymmetries to grow up than
in-plane undulations, in agreement with the experimentally observed sequence.
(a) 20
(E 0 -E I )/κ
10
0
bilayer (eqn 2)
bilayer (eqn 1)
−10
cytoskeleton
membrane
−20
(b) 0 0.005 0.01 0.015 0.02 0.025
Δa0
(E I -E II )/κ
10
0
bilayer (eqn 2)
bilayer (eqn 1)
−10 cytoskeleton
membrane
0 0.005 0.01 0.015 0.02 0.025

Δa0
Figure 4.4: Energy differences between RBC stages as a function of Δa0 . Each config-
uration is labeled as discocyte (0), discoechinocyte I (I) and discoechinocyte II (II). (a)
The discoechinocyte is favourable with respect to the discocyte at 0.0063, though the
bilayer would prefer to undulate at lower asymmetries. (b) The energy barrier of the
cytoskeleton resistance to the growth of bumps of the discoechinocyte II is considerably
larger than the discoechinocyte I. The transition occurs at 0.0175.
In RBCs, out-of-plane bumps are rarely present before the discocyte has de-
veloped undulations. The order of appearence of the configurations is very strict,
as a consequence of the enormous differences in energy between configurations.
From the minimal shapes for the bilayer, dashed lines in Figure 4.4, it can be in-
ferred that out-of-plane bumps and in-plane undulations achieve a similar benefit
in terms of bilayer relaxation. However, the larger energy gap that the bumps in-
duce due to a higher distortion of the cytoskeleton derives in a delayed appearence
of these structures in the total membrane energy.
The discocyte is found to be the minimal shape of the curvature energy for a
symmetric membrane, but the cell shape is highly sensitive to the variation of the
asymmetry in the absence of a cytoskeleton. In this case, the discocyte destabi-
lizes into a prolate ellipsoid already for Δa0 ∼ 0.0001 (Khairy and Howard, 2011).
In lipid vesicles –in which the cytoskeleton is absent, thus only the bilayer contri-
bution is considered –, systematic studies on the dependence of the vesicle shape
with the asymmetry lead to a different sequence of shapes (Svetina et al., 2004):
discocytes, elliptocytes, dumbbells and eventually pearling are found for increasing
asymmetries, but crenated shapes are not recovered. Therefore, the cytoskeleton
arises as an essential ingredient to understand the appearence of protrusions. Al-
though we have not directly analyzed this transition, different indirect evidence
also supports the importance of the cytoskeleton. The presence of the cytoskeleton
implies a high energy storage, with energy barriers of typically > 103 kB T , con-
sistent with experimental observations which show that undulations and bumps
are long-term structures with a well-defined shape. In contrast, in the absence
of a network, the discocyte becomes unstable at Δa0 = 0.003 –when undulated
shapes become minimal over a short range of preferred area-difference– but the
energy barriers between configurations are typically of ∼ 5kB T , and hence cre-
nated shapes would be highly unstable under thermal fluctuations. A structural
component is required to explain the large stability of these structures. Accord-
ingly, the cytoskeleton plays a fundamental role in the stabilization of the different
RBC morphologies.
The energy landscapes of Figure 4.3 allow to identify the different stages that
are sequencially found as energy minima when the asymmetry is increased. To
make a clear connection with the observed transition, a purely morphological
description is obviously more appropiate. The shape diagram defining the presence
of undulations and bumps, and their height, is presented in Figure 4.5. Each
point of the diagram represents a minimal shape, and the energy corresponding
to each configuration can be directly obtained from Figure 4.4. Only when the
energy difference in Figure 4.4 (a) is positive, and discoechinocytes I are favourable
with respect to discocytes, the undulation height ū is non-zero and increases with
the membrane asymmetry. Analogously, out-of-plane bumps appear when the
energy difference between discoechinocyte I and II is non-zero. A comprehensive
description of the morphologies found is presented in the subsequent section.
4.2. Results 67
Figure 4.5: Shape diagram of the first stages of the RBC shape transition: (0) discocyte,
(I) discoechinocyte I, and (II) discoechinocyte II. The number of undulations is fixed
at l = 5 (the first to appear after the desestabilization of the discocyte), although it
does not correspond with the minimal configuration for the entire range of Δa0 (see Fig.
4.7). The growth of in-plane undulations saturates when out-of-plane bumps appear.
The later grow much faster and rapidly acquire similar heights.
4.2.3 Morphological characterization

The discocyte.
Assuming a discocytic-relaxed cytoskeleton (where Ecyt = 0, and the shape is

determined only by the bending contribution of the bilayer), the minimal shape
obtained at zero asymmetry corresponds to = 0.98 and R0 = 4.08μm, shown in
Figure 4.2 (0.t). The thickness of the cell is 2.11μm at its maximum and 1.03μm at
the center, giving a ratio of 2.05, consistent with results for the discocyte found in
previous Cassini-based calculations (Canham, 1970; Angelov and Mladenov, 2000;
Eremina et al., 2006). This shape remains minimal over a small range of positive
asymmetry, but its energy decreases until reaching an absolute minimum at Δa0 =
0.00089. This is consistent with previous theoretical works (Liu et al., 1999) that
also found global energy minima for discocytes with small positive asymmetries.
This discocyte is fixed as S0 , mimicking the relaxation of the cytoskeleton over
the bilayer, and the cytoskeleton will hereafter present resistance to deformations
from this reference shape.
Transition to discoechinocyte I.
When Δa0 is increased from the initial discocyte, the biconcavity shows a slow
decay to less biconcave profiles, forced to reduce the negative curvature of the
(a) x 10−3
9.075
9.050
Δa
9.025
9.000
8.975
0.5 0.6 0.7 0.8 0.9 1

(b)0.96
0.955
0.95

0.945
0.94
0.935
0 0.005 0.01 0.015 0.02 0.025
Δa0
Figure 4.6: (a) Total area-difference of RBC shapes as a function of the biconcavity
in the discocyte regime. The area-difference is lower at small biconcavities because the
large area-to-volume ratio implies very flat disks. (b) Rate of thickening of the cell when
Δa0 is increased. Even at high asymmetries the shape remains slightly biconcave.
center of the disk. But the area difference does not increase monotonously when
the biconcavity is reduced, and it actually achieves a maximum at = 0.88,
as shown in Figure 4.6 (a). Convex profiles do not imply a net benefit in the
outer area, because shapes fulfilling the high area to volume ratio of the RBC
demand large radius and low thickness, close to a very flat disk. Instead of losing
biconcavity as a mechanism to increase its curvature, the cell can also break its
axisymmetry and it develops undulations along its in-plane contour. This shape
transformation leads not only to an explicit gain of curvature, but also to an
increase of the surface area of the contour, allowing for an overall thickening of
the central body of the cell whilst the total area remains constant. Therefore, these
undulations allow the cell to deform into more spherical shapes. The rate at which
the cell becomes thicker with Δa0 seems to be slightly low compared with previous
works (Lim et al., 2002), in which the cell is markedly ellipsoidal for Δa0 ∼ 0.02,
whereas in our case at this point the shape remains considerably flat (typically
≈ 0.9, see Figure 4.6 (b)). Discocytes become unstable earlier (Δa0 = 0.0063)
than predicted with resting cytoskeletons in an oblate ellipsoid (Δa0 = 0.0082)
(Khairy and Howard, 2011). This delay in the transition might be related to the
more compact resting shape of the cytoskeleton, as the ellipsoid would provide
4.2. Results 69
more resistance to the undulations than a discocytic relaxed network.
Figure 4.7: Dependence of the number of undulations l and their height with the pre-
ferred area-difference in a RBC within the discoechinocyte regime. Each l represents a
minimal shape only over the range in which the line is bold; vertical dashed lines limit
the range in which each mode is dominant. Contours with more undulations present
lower heights and appear later in the transition.
The discoechinocyte I.
An example of a typical discoechinocyte I obtained is shown in Fig. 4.2 (I.t). The

diagram of stable shapes within the discoechinocyte I regime predicted by the
model is presented in Figure 4.7. A sequence of minimal shapes with increasing
number of undulations is found, from l = 5 at low asymmetries up to l = 10 at
the end of the discoechinocyte I regime, when the cell first develops out-of-plane
bumps. l = 1 − 4 are sequentially found as minimal shapes over a very short range
right before l = 5 becomes dominant,and for this reason are not shown here.
The height of the undulation is very sensitive to the current overall number of
undulations; for low numbers, the protrusions are considerably larger than those
found for more undulated cells, giving a more spiky appearence to the cell, in
accordance with experimental observations (Khairy et al., 2010). For instance,
cells with 5 undulations show protrusions up to ū ∼ 0.15, reduced to ∼ 0.1 for
8 undulations and ∼ 0.75 for 10; these values correspond to undulation heights
of 0.6, 0.4 and 0.3 μm, respectively. This dependence can be explained on the
basis of the stretching that the undulation undergoes. For high l, the base of the
undulation is small and reaching large heights involves large stretchings, which
disappear for sufficiently large bases associated to low-l. As previously explained,
each mode has been analyzed here separately (as a single shape), which allows
an easier comparison between modes and identification of the dominant one at

each Δa0 , but the curve in Figure 4.7 would be actually continuous if the entire
combination of modes is considered.
Transition to discoechinocyte II.
The cell develops in-plane undulations in an earlier stage than out-of-plane bumps
in spite of being, a priori, similar ways to increase the outer area. In-plane undu-
lations grow on a curved surface, and the network disruption from the reference
state is limited, involving only a smooth deformation of the contour. In contrast,
out-of-plane bumps are drastic events which grow on an almost flat surface, and
the local perturbation of the network is considerably larger. In Figure 4.8 (a), the
dimensionless energy density map fcyt (x) = (Rd2 /κ)δEcyt /δA of the cytoskeleton
for a discoechinocyte II is presented. The stress energy is largely localized in the
out-of-plane bump, whilst the in-plane undulations bear a much less severe penal-
ization. Figure 4.8 (b) shows the local area-difference for the same configuration.
Both kind of protrusions achieve an area-difference benefit of the same order of
magnitude, albeit slightly larger for the out-of-plane bump. Accordingly, the more
stressed cytoskeleton inhibits the appearence of bumps in favour of the in-plane
undulations, which imply a similar bilayer relaxation to the former.
Figure 4.8: (a) Stress-energy density of the cytoskeleton fcyt (x) = (Rd2 /κ)δEcyt /δA
in the discoechinocyte II stage. (b) Local area-difference Δa between layers for a
discoechinocyte II. The out-of-plane bumps are highly penalized for a limited area-
difference benefit, and consequently they appear when the bilayer has stored more en-
ergy than the in-plane undulations.
In-plane undulations induce a limited cytoskeleton distortion as far as their

height is ū 0.125; from this point onwards, the cytoskeleton perturbation affects
4.2. Results 71
to significantly large regions of the entire network, and the rate of increase with
Δa0 of the undulations height decays. The cell develops new protusions in the
central area of the disk, the so-called out-of-plane bumps, which characterize a
new stage of the transition. As shown in Figure 4.5, the rate of growth of the
undulations saturates coinciding with the development of the bumps, and the
undulation stabilizes at ū ∼ 0.13 (around 0.5 μm). The energy to relax the
bilayer is now concentrated in the bump rather than continue with the growth of
undulations.
Figure 4.9: (a) Energetic cost associated with the growth of a bump as a function of the
position of the bump in the cell (the resting shape of the cytoskeleton is shown in (b)
for comparison). Bumps avoid the center of the cell, where the distortion of the concave
cytoskeleton is around 15% larger (see main text), prefering the outer convex ring.
The discoechinocyte II.
Figure 4.2 (II.t) shows a typical example of discoechinocyte II. At this stage, the
cell develops out-of-plane bumps which grow faster than the undulations, as shown
in Figure 4.5. The model predicts bumps with a fairly rounded profile and typical
base diameter Db ≈ 0.9μm. This is in good agreement with experimental observa-
tions, Figure 4.2 (II). The bumps maintain a nearly constant base, increasing their
height and becoming thicker with Δa0 . Bump growth is limited by the constraints
of area and volume, since a large bump size implies a considerable deformation of
the overall cell shape, leading to typical heights of h̄ ∼ 0.2. The bumps, even if
they imply a locally large disruption of the cytoskeleton, are isolated events with
a limited global impact over the network. The area and volume occupied by the
bump also allows the cell to reduce its radius and become thicker. Interestingly,
bumps growing over convex profiles require less energy storage of the bilayer to
obtain a similar shape than those appearing at concave surfaces. For this reason,
it is found that out-of-plane bumps preferably grow over the convex ring, close to
the contour, and avoiding the central concave region, as shown in Figure 4.9. The
energy necessary to grow a bump in the center is higher than close to the rim,
(Ecenter − Ering )/Ering ≈ 0.15. This implies that the out-of-plane bump might
show a distribution pattern over the disk. Studies on the membrane fluctuations
of the RBC showed that the amplitude of the fluctuations are also larger at this
convex region (Park et al., 2010a). It may be feasible that the larger fluctuation
amplitudes and the tendency of bumps to grow on the convex ring are related
and due to geometrical restrictions, but further study is required in this direction.
However, it has been suggested (Lim et al., 2002; Gov and Safran, 2005) that the
development of the bumps is highly sensitive to isolated defects of the network
implying local weakening of the cytoskeleton, which could facilitate the growth in
the surrounding region. Such a dependence on the network defects could affect the
predictions on the asymmetries required for the bump development and it would
obscure any pattern in the bump distribution.
4.3 Discussion
The results presented show that a simple description of the cell shape via a Cassini
oval parametrization provides a good account of the shape transformation in re-
sponse to an increase of the membrane asymmetry. The analytical description
allows a clearer interpretation and identification of the main ingredients deter-
mining the RBC morphological evolution, in spite of the limitations of the model
in comparison with numerical free minimizations. Eremina et al. (2006) com-
pared the minimal profile based on a Cassini ovals parametrization with the curve
proposed by Evans and Skalak (1980), showing similar results. In spite of the vari-
ability shown by a population of RBCs, including size, cell profile, physiological
membrane asymmetry and elastic modulii, most cells exhibit all the morpholo-
gies of the sequence in the correct order, suggesting that the transition presents
a certain lack of sensitivity to variations in these parameters. The restriction of
the cell profile being subject to a predetermined curve such as the Cassini oval
is therefore expected to introduce small quantitative deviations with respect to
the exact solution, but the overall process should be reasonable robust to profile
differences.
The increase in the bilayer asymmetry triggers the formation of crenated
shapes, driven by a stressed bilayer that tries to expand its outer leaflet to acco-
4.3. Discussion 73
Figure 4.10: Comparison between the experimental results from Ferrell et al. (1985)
(black symbols; the different types of symbols correspond to different lipid species), and
theoretical results from Mukhopadhyay et al. (2002), Khairy and Howard (2011) (from
which only the range of stability of the discocyte can be extracted because they do not
distinguish between different types of echinocytes) and the results presented in this The-
sis. The morphological index refers to the shape stage; 0-5 correspond to the discocyte,
discoechinotes I and II and echinocytes I, II and III, respectively. The morphological
index is a continuous quantity only when it is averaged over a population of cells; hence
the theoretical results are expressed as a stepped curve. The measured magnitude in ex-
periments is the preferred area difference, Δa, whereas the parameter from simulations
is the preferred area-difference Δa0 , which tends to be slightly larger. All models show
a qualitative agreement with experimental results, but the model of Lim et al. develops
the most spiculated morphologies over a short range around Δa0 ∼ 0.017 − 0.020; in
contrast, our results predict a more gradual transition. These divergences are already
present in the discocyte regime, as our model predicts a shorter range of stability for
this shape (the lines at morphological index equal to zero have been intentionally shifted
in order to discern the exact range predicted by each model). Adapted with permission
from Ferrell et al. (1985). Copyright 1985 American Chemical Society.
modate the excess area. It has been shown that the presence of the cytoskeleton is
required to explain the whole transition, as a stabilizing component which favours
compact axisymmetric shapes against vesiculation. The growth of protrusions
has two direct implications in the shape transformation: (i) the gain of positive
curvature allows a relaxation of the bilayer; (ii) the presence of undulations and
bumps entails a relaxation of the large area to volume ratio, as they locally involve
large surfaces enclosing small volumes. This allows the central body of the cell
to become more spherical, and eventually protrusions will undergo vesiculation
and the cell derives in a small spherocyte, with a considerably reduced area. The
energy differences between configurations are large, ensuring a huge stability and
explaining the marked hierarchy of the transition. Discoechinocytes I are therefore
necessarily found earlier along the transition than discoechinocytes II.
The reference state of the cytoskeleton was fixed as a discocyte shape. The
present results support this assumption as a valid option. As with previous the-
oretical studies in which the resting shape was fixed as a prolate ellipsoid, we
sequentially recover shapes in nice agreement with the RBC shapes observed in
experiments. The quantitative analysis highlights the differences in the rate of de-
velopment of the crenated morphologies. The shape diagram in Figure 4.5 shows a
gradual shape transformation when the preferred area-difference is increased. The
discocyte remains as a minimal shape up to Δa0 = 0.0063, when the cell develops
in-plane undulations. These structures grow up smoothly, increasing sequentially
their number, and its stability range expands up to Δa0 ∼ 0.02, when out-of-plane
bumps appear in the central region, defining the transition to discoechinocyte II.
The bumps grow up over a short range of Δa0 compared with the undulations,
thus requiring a small amount of excess area in the outer leaflet. A typical dis-
coechinocyte II such as that shown in Figure 4.2 (II.t) is attained at Δa0 = 0.0024.
In Figure 4.10, these results are compared with the experimental results of Ferrell
et al. (1985), as well as with the theoretical results found by Lim et al. (2002) and
Khairy and Howard (2011). All the models predict a similar order of magnitude in
the preferred area difference to induce the transition. Nevertheless, studies which
assumed an ellipsoidic relaxed cytoskeleton found a markedly steep transition; for
instance, the discocyte, discoechinocyte I and echinocyte I and II are found at
Δa0 = 0.00143, 0.01717 , 0.01788 and 0.02003 respectively. Whilst a consider-
ably increase in Δa0 is required to convert a discocyte into a discoechinocyte, the
subsequent stages appear for a proportionally very reduced increase of Δa0 , and
actually the transition saturates around a preferred area-difference of 0.02. These
discrepancies might concern the different resting shape of the cytoskeleton. In-
plane undulations represent a lower distortion for a discocyte than for an ellipsoid
relaxed cytoskeleton, and thus lower Δa0 are required to destabilize the discocyte
in the former case. Conversely, when the cell is more spherical, the cytoskele-
ton is more stressed for a discocytic-reference state than for the ellipsoidal case,
explaining why discocytic-relaxed shapes present more gradual transitions. Exper-
imental evidence supports a regular transition; discoechinocytes I and II require
an increase in Δa of ∼ 0.008 and 0.014 from the discocyte, whilst an echinocyte
II is not found up to ∼0.03. Thus, they found a similar increase in Δa0 to induce
both transitions, in accordance with our results, but at lower values than what
our findings suggest. However, the magnitude of the outer monolayer expansion
associated with these shapes has not been properly clarified, and significant diver-
gences have been found; in comparison with the results in Figure 4.10, Lange and
Slayton (1982) reported an increase of ∼ 0.015 necessary to obtain a echinocyte
II, whilst Matayoshi (1980) suggested that this quantity must be exceeded several
4.4. Discussion of the AFM experiments 75
times. Therefore, more accurate results are necessary to determine the actual
shape (the discocyte or the ellipsoid) at which the cytoskeleton is relaxed.
It has been shown that both the rigidity of the bilayer and the tension of the
cytoskeleton are modified when ATP is depleted (Betz et al., 2009). Typically,
the elastic modulii increase during morphological transformations in the absence
of ATP (Park et al., 2010b), producing stiffening of the cell membrane. The
rate of shape change with the bilayer asymmetry depends on the specific values
of the elastic parameters. If K̄ and μ̄ decrease from their normal values (which
would correspond to a relative increase in the bilayer rigidity with respect to the
cytoskeleton stiffness), the ranges of stability of the discocyte and discoechinocytes
are reduced and the asymmetries required to induce the shape transformation are
lower. The order in the shape sequence is maintained, but undulations and bumps
develop earlier. If the cytoskeleton is weakened enough, we found that eventually
out-of-plane bumps can appear at very early stages, even before the in-plane
undulations have been developed. Future studies may explicitly incorporate the
ATP-dependent mechanisms of cytoskeleton relaxation and how it affects to the
cell morphologies, though a comprehensive understanding and description of this
change in the elastic response is still required.
4.4 Discussion of the AFM experiments
The theoretical results show a considerable separation in the energy scale between
each morphology. Accordingly, weak reductions of the membrane asymmetry may
be accompanied by a strong relaxation of the deformation energy of the cell mem-
brane, and therefore cells will change their shape even if minute changes in their
membrane conformation are induced. The fact that the discocyte represents the
ground state of the entire transition also explains the strong directionality of the
shape recovery, as nearly in all cases cells were observed to deform into lower
stages of the transition (towards the discocyte).
The stable shapes show an exteme sensitivity to the internal balance between
the cytoskeleton and the bilayer. This fact, as well as the observation that cells
invariably return to discocytic shapes, indicate that the AFM disruption does
not affect the cytoskeleton equilibrium, or at least the potential damage is not
important for shape stability. The effect induced by the AFM disruption must
be then associated to the bilayer. The internal processes taking place in the
bilayer cannot be identified in this model, altough we provide a measure of the
area-expansion necessary to induce each shape transformation. The hypothesis
presented in Section 3.3.2, in which we propose that the puncture of the membrane
by the AFM tip allows the flip of lipids due to a breaking of the hydrophobic
potential, is plausible and consistent with the present results.
Finally, in the experiments it is shown that when crenated shapes (regardless
of their morphological index) are punctured, they return to lower stages of the
transition but showing a certain dispersion in the final stage induced. For instance,
echinocytes IV are observed to deform into either echinocytes III, II or I. This
observation supports a gradual shape transition with respect to an increase in the
asymmetry. During each puncture, the amount of lipids transferred to the inner
leaflet is likely different, giving rise to a dispersion in the induced change of the
asymmetry. For a sharp transition, the final shape will be strongly dominated
by the stable configuration for a wider range of Δa0 , and hence the transition
towards discoechinocytes III (which, in this scenario, are stable only for a very
reduced range of Δa0 ) would be very unlikely.
4.5 Conclusions
The mechanisms driving cell deformations and the role of each structural compo-
nent of the RBC membrane have been identified. The reported results highlight
the relevance of understanding the process of membrane asymmetry control, and
how it interplays with the cytoskeleton remodelling, in order to stablish the time-
dependent behaviour of the shape transformation. The theoretical results support
the hypothesis of lipid rearrangement induced by AFM puncturing proposed in
the previous section.
The spectrin cytoskeleton has been shown to play an important role in the
stabilization of the discocyte shape. The results presented here agree with other
theoretical and experimental studies that indicate an (at least partially) relaxed
cytoskeleton in the discocytic state. The cytoskeleton is highly stressed when the
membrane deforms and bumps and spicules develop, though its energetic storage
is very low when the cell remains weakly deformed from the initial discocyte. This
may imply that even if the cytoskeleton is relevant in some kind of deformations,
opposing resistance to crenation and confiring stability under budding and vesic-
ulation, it could have a secondary contribution to RBC elasticity when cell shape
remains close to the discocyte, such as during blood flow.
Active processes seem fundamental to clarify the kinetic behaviour of the cell
morphology and elasticity; whereas recent theories have incorporated the effect
of ATP in several mechanisms affecting membrane elasticity, the approach to
study cell morphology still assumes a static picture, with active processes been
considered negligible or of little importance. In that regard, more studies are
required to elucidate the resting shape of the cytoskeleton, ideally including local
4.5. Conclusions 77
relaxation, as it seems realistic that the network is able to conveniently change

its connectivity at certain regions depending on the enviromental conditions (Li
et al., 2007). All these questions are crucial to throw additional light into the
issue of how active processes modulate membrane elasticity and achieve a deeper
understanding of RBCs mechanics.
Part III
A phase-field method for

membrane modeling
79
Chapter 5
Phase-field modeling of biological

membranes
In Chapter 2, we have presented the main physical models describing membrane

elastic behaviour. One of these theories, the ADE model, has been used in Chap-
ter 4 to gain a deeper understanding of the role of each membrane component in
the membrane elasticity, and we have predicted some equilibrium morphologies of
the cell on the basis of an analytic parametrization of the membrane. However,
when more complex problems are faced, such as any dynamic instability or out-
of-equilibrium process, a more flexible framework is usually required. Given that
the representation of the membrane as a two-dimensional layer is reasonably accu-
rate, the simplest and most direct formulation consists in defining a mesh of points
which represents the membrane neutral surface, and from here extract the local
mean curvature or deformation tensor necessary to compute the elastic energy.
This must be combined with a minimization procedure (such as a Monte Carlo
free minimization (Lim et al., 2002)) to obtain equilibrium shapes, or introducing
the elastic mesh in a dynamic theory if one is interested in membrane kinet-
ics. Most important examples include the immersed boundary methods (Peskin,
2002; Kaoui et al., 2012), integral boundary methods (Pozrikidis, 1992, 1995) or
multiparticle collision dynamics (Malevanets and Kapral, 1999; McWhirter et al.,
2008). Methods in this direction have been succesfully applied to the study of
many membrane related topics (Li et al., 2005; Peng et al., 2013). All these meth-
ods require of a explicit tracking of the membrane position and the calculation
of the deformation variables, ie the curvature. A different approach, based on an
Eulerian rather than a Lagrangian description, are the phase-field models. The
membrane is identified from an auxiliary scalar field defined in the entire space,
and the method details the dynamics of the field, instead of specifically deal with
the evolution of the interface. This formulation also avoids the problematic of
defining the boundary conditions at the membrane surface.
81
82 Chapter 5. Phase-field modeling of biological membranes
This Part of the Thesis is devoted to the characterization, rationalization and

study of a phase-field method for membrane modeling. We connect the phase-
field model with the theory of elasticity, explaining how the elastic properties
of the membrane can be controlled. The macroscopic equations that define the
membrane dynamics, including the general membrane equilibrium equation, are
derived. Finally, we numerically perform a linear stability analysis in order to test
the reliability of the model. The phase-field method presented here corresponds
to a bending membrane, based on the Helfrich theory, and thus it represents a
purely lipidic description of the membrane. The method does not incorporate any
shear contribution characteristic of the cytoskeleton. Accordingly, this method
can be applied to study phenomena associated with vesicles and cells (Campelo
and Hernández-Machado, 2007a, 2008; Campelo et al., 2012; Wang and Du, 2008)
in which the cytoskeleton plays a subdominant role, as we will see in the Part VI
of this Thesis.
5.1 Phase-field method
Phase field models have been widely applied in the study of different phenomena of
phase transitions, such as superconductivity and solidification (Steinbach, 2009),
due to their universality as well as their utility as a mathematical tool in the
numerical and analytical fields. Although the application of phase-field methods
to amphiphilic systems was extensively investigated in the past (Gompper and
Schick, 1994), it has not been until recently that these models have been used in
the study of cell morphology and dynamic response. One of the main advantages
of the phase-field modeling is that the evolution and shape of the membrane does
not need to be tracked, as in the explicit methods, but it spontaneously evolves
with the phase field dynamics. Phase-field methods also invite to a deep analytical
exploration as they have a robust physical basis.
Ginzburg-Landau free energy
The origins of phase-field models are found in the mean-field approach to phase
transitions. The Landau theory consists in a free energy which is expanded in
powers of a scalar field, called order parameter φ, which receives different inter-
pretations depending on the system, as we will see below, but, for instance, in a
ferromagnetic system it is readily identified as the magnetization. The symmetries
of the system specify the value of the coefficients of the expansion. However, this
model does not account for the presence of interfaces which could have an ener-
getic cost associated with the interaction between the components of each phase.
5.1. Phase-field method 83
Ginzburg and Landau generalized this expression in their studies about supercon-
ductors and they incorporate to the free energy of the system, F [φ], powers of the
gradients of the order parameter,

2

F [φ] = L(φ, ∇φ, ∇ φ)dV = f (φ) + g(φ)2(∇φ)2 + c4 (∇2 φ)2 dV. (5.1)
where L is the energy density, f (φ) is the bulk potential, so that in equilibrium
f (φeq ) = 0 and φeq are the stable bulk phases. The coefficients associated to the
order parameter gradients, g(φ) and c, represent the energetic cost of having an
interface. Typically, f (φ) is chosen to form a symmetric double-well potential,
and hence two phases are present, φeq = ±1. The interface is characterized by
a smooth profile characterized by a width , and it is usual to fix the interface
position at the isosurface φ = 0. The method does not necessarily need that
operates on the same scale of the width of the real interface, which in some
systems can be the atomic scale, but it is sufficiently to require that (i) is much
smaller than any other length of the system; and (ii) the interface incorporates
the relevant information from the microscale trought the effective constants of the
model.
The construction of a Ginzburg-Landau free energy often responds to a purely
phenomenological basis and it generally attends to the symmetries of the system.
The order parameter is related to a characteristic physical property depending
on the specific system; in the particular case of a membrane, it can be described
as a function of the concentration of lipids, φ(x) = 1 ∓ ρlip (x)/ρ0 , where ρ0 is
the reference concentration of maximum package and the ∓ corresponds to the
equilibrium phase, -1 and +1, in the inner or outer regions of the membrane,
respectively. It is noteworthy that φ is defined in a spatial scale sufficiently large
to have a statistically good average of the molecular densities, but yet small enough
to capture the mesoscopic spatial variations of the density.
5.1.1 Thermodynamics of phase-field models
In thermodynamics, the chemical potential of a species in a mixture is defined

as the free energy change for deviations in the concentration from the reference
value. Analogously, within the phase-field framework the chemical potential is
obtained from the free energy (5.1),
δF ∂L ∂L ∂L
μ[φ] = = − ∇β + ∇2 . (5.2)
δφ ∂φ ∂(∇β φ) ∂(∇2 φ)
In equilibrium, the chemical potential is constant and in addition to the bound-

ary conditions in the bulk phase φ(x → ±∞) = ±1, this equation can be solved
obtaining the equilibrium profile of the interface. Typically, the relaxed profile is
a smooth tanh-like function with interface width . As we will see later on, the
chemical potential is a relevant variable as it governs the diffusion of the order
parameter (Section 5.3). In the presence of a non-uniform phase field, there is a
thermodynamic local force at each point (eg this force can be naively understood
in the case in which φ is a concentration, and then gradients of concentration
force diffusive fluxes towards a uniform concentration). Given that in the bulk
μ = const., the force density is localized at the interface.
The stress tensor is an important magnitude to characterize the deformation
of membranes, though it is not obvious how to compute this tensor from a phase-
field free energy of the form (5.1). The work required to deform an object by a
small displacement δxα is given by (Landau and Lifshitz, 1999)

δF = − ∇β σαβ δxα dV = σαβ ∇β δxα dV. (5.3)
In the phase-field framework, we need to specify the variations in the order

parameter due to the deformation δx. Assuming that these small variations only
correspond to convective fluxes, ie diffusion is subdominant (Brannick et al., 2014),
∂t φ + ∇ · (φv) = 0, (5.4)
and writing vα = δxα /δt, we obtain the variation of the order parameter and,
after differentiation, its derivatives
δφ = −φ∇α δxα − ∇α φδxα .

δ∇β φ = −∇β φ∇α δxα − φ∇β ∇α δxα − ∇β ∇α φδxα − ∇α φ∇β δxα .
(5.5)
δ∇2 φ = −∇2 φ∇α δxα − 2∇β φ∇β ∇α δxα − φ∇α ∇2 δxα − ∇α ∇2 φδxα
−2∇β ∇α φ∇β δxα − ∇α φ∇2 δxα .
The work necessary to induce a general deformation δxα of an interface with

free energy (5.1) reads

∂L ∂L ∂L 2
δF = δLdV = δφ + δ∇β φ + δ∇ φ dV. (5.6)
∂φ ∂∇β φ ∂∇2 φ
Introducing expressions (5.5) in (5.6), and after several straightforward inte-

grations of those terms containing second and third gradients of δxα , one finds
5.1. Phase-field method 85

∂L ∂L ∂L 2
δF = − + ∇α φ + ∇α ∇ φ δxα dV
∂φ ∂(∇α φ) ∂(∇2 φ)

∂L ∂L 2 ∂L
− − ∇β +∇ φ∇α δxα dV (5.7)
∂φ ∂(∇β φ) ∂(∇2 φ)

∂L ∂L ∂L
− ∇α φ + ∇ α ∇β φ − ∇β ∇α φ ∇β δxα dV.
∂∇β φ ∂∇2 φ ∂∇2 φ
In the first term of the right hand side of (5.7) the divergence ∇α L can be
recognized. Analogously, the second term contains the expression of the functional
derivative of L, given by (5.2). Hence, comparing with (5.3), the stress tensor is
identified,

δL ∂L ∂L ∂L
σαβ = L−φ δαβ − ∇α φ − ∇ β 2
∇α φ + ∇α ∇β φ. (5.8)
δφ ∂(∇β φ) ∂(∇ φ) ∂(∇2 φ)
From (5.8), it can be shown that the divergence of the stress tensor reduces to
∇β σαβ = −φ∇α μ, (5.9)
which provides an expression for the local force density of the interface in terms
of the chemical potential. The expressions obtained so far are valid for any free
energy of the form (5.1), so that the elastic properties of the interface are solely
determined by this free energy.
The dependence of the force density on the chemical potential can be un-
derstood from a thermodynamic perspective, taking into account that the stress
represents the internal reaction to an external pressure (strictly speaking, the
pressure and stress tensor are related by Pαβ = −σαβ ). Hence, equation (5.9) may
be related with the Gibbs-Duhem equation, which reads

V dP = Ni dμi (5.10)
i
where Ni is the amount of matter of the species i and taking into account that
φ ∼ N/V , leads to dP = φdμ. The force density arise as the free energy change
per unit volume, φδμ, due to the transport of matter concentration φ for a change
in the chemical potential δμ.
5.2 Elastic properties of the phase-field interface
In the Helfrich theory, the physical meaning of each elastic modulii is clear, as
each modulus is explicitly associated to the elastic deformation that penalizes (ie
in (5.1), the surface tension represents the energetic cost of having an interface
of surface dA, and the bending modulus corresponds to the cost of an interfa-
cial bending given by the curvature C). In the phase-field representation of the
interface free energy, however, the information about the geometrical properties
of the interface deformation is implicitly contained in the gradients of the order
parameter, and the elastic properties of the interface cannot be directly identi-
fied. It is therefore necessary to establish a connection between the phase-field
coefficients, g and c, of the phase-field free energy (5.1), and the interface elastic
parameters expressed in the Helfrich energy (2.16). The comparison between both
free energies for simple geometries, such as spheres or cylinders, is a useful method
for obtaining a mapping between the Helfrich and phase-field representations. In
this Section, we derive the expressions of the elastic coefficients of (2.16) from the
parameters of the free energy (5.1), as first proposed by Gompper and Zschocke
(1991), and Gompper and Zsckocke (1992). Thereby, the method allows to specify
the elastic properties of the phase-field free energy and conciliate this expression
with the classic elastic description.
Let us suppose a generic interface with free energy (2.16), with contributions of
surface tension and bending, but disregard the pressure difference term assuming
that the interfacial surface is open. Note that this free energy can be locally
understood as an expansion in terms of the radius of curvature 1/R, since C ∼ 1/R
and G ∼ 1/R2 . Hence, the surface tension is associated to the zeroth order, 1/R0 ;
the spontaneous curvature corresponds to the first order of the expansion, 1/R1 ;
and the bending and saddle splay modulii are associated with the second order,
1/R2 . The elastic energy per area (2.16) for a sphere of radius R is given by
Fe
(s)
κ κ 2κ + κG
= γ + c20 − c0 + . (5.11)
A 2 R R2
For simplicity, the first term is rewritten γ̄ = γ + κc20 /2. For a cylinder, the
Gaussian curvature vanishes and then the free energy reduces to
(c)
Fe κ κ
= γ̄ − c0 + . (5.12)
A R 2R2
The inclusion of the cylinder is important to identify the bending modulus,

which appears coupled to the saddle-splay modulus in the spherical geometry.
5.2. Elastic properties of the phase-field interface 87
The equilibrium condition for the free energy (5.1) determines the relaxed planar
interface profile φ0 (z) in the normal direction z,
δF [φ]
= f − 2g(φ0)∂z2 φ0 − g (∂z φ0 )2 + 2c∂z2 φ0 = 0, (5.13)
δφ
where g = ∂g/∂φ and f = ∂f /∂φ. Multipliying by ∂z φ0 , and integrating across

the interface,
∞
f (φ) = [2g∂z2 φ0 ∂z φ0 + g (∂z φ0 )3 + 2c∂z2 φ0 ∂z φ0 ]dz. (5.14)
−∞
Note that, formally, (5.13) is the stationary condition of the Euler-Lagrange

equation and then (5.14) is its first integral. For the specific geometry of spheres
and cylinders, the free energy (5.1) reads
2
∞
F (c,s)[φ] 1 d
= d+1 drr d f (φR ) + g(φR)(∂r φR )2 + c ∂r2 φR + ∂r φR ,
A R 0 r
(5.15)
where d = 1 and 2 for cylinders and spheres, respectively, and φR is the phase-field
profile for these two configurations. If the radius R is large, the phase-field profile
φR can be expressed in terms of the relaxed planar solution, φ0 (z), by identifying
the position of the interface at z = r − R. This approximation, known as locally
flat interface, allows to expand the profile in powers of 1/R,
φ1 (r − R)
φR (r − R) = φ0 (r − R) + + ... (5.16)
R
By introducing this expression in (5.14), and using this to remove f (φR ) from
(5.15), to leading order the free energy for spheres and membranes now reads,
2
F (c,s) [φ] ∞
r d 2 2 2 d
= dr 1 + 2g(φ0)(∂r φ0 ) + 3c(∂r φ0 ) + c ∂r φ0 + ∂r φ0
A 0 R r+R
∞
1
− 2 δ(d − 2)2c dr(∂r φ0 )2 .
R 0
(5.17)
The expansion of this expression in powers of 1/R leads to

∞ ∞
F (c,s)[φ] 2 2 2 2r
= dr[2g(φ0)(∂r φ0 ) + 4c(∂r φ0 ) ] + dr [2g(φ0)(∂r φ0 )2 + 4c(∂r2 φ0 )2 ]
A R
∞ 2 0 0 ∞
r 2c
+ 2
[2g(φ0)(∂r φ0 )2 + 4c(∂r2 φ0 )2 ] + δ(d − 1) dr 2 (∂r φ0 )2 .
0 R 0 R
(5.18)
Comparing this expression with the free energy of cylinders (5.12) and spheres
(5.11), and identifying the coefficients of the corresponding terms of the expansion,
one obtains
+∞
γ̄ = s(z)dz,

−∞
+∞
−κc0 = 2 zs(z)dz,
+∞ −∞ (5.19)
κ= 2c(∂z φ0 )2 dz,
+∞ −∞
κG = (z 2 s(z) − 4c(∂z φ0 )2 )dz,

−∞
where we have recovered the planar profile notation as a function of the normal
coordinate z and we have introduced the function
s(z) = 2g(φ0)(∂z φ0 )2 + 4c(∂z2 φ0 )2 . (5.20)
The interpretation of this expression is found in the elasticity theory frame-

work. The lateral stress profile of an object is calculated from the stress tensor
via
se (z) = σt (z) − σn (z), (5.21)
where σn and σt are the projections of the stress tensor in the normal and tangen-
tial directions to the interface, respectively. By computing the stress tensor (5.8)
of an interface normal to the direction ẑ and characterized by a free energy (5.1),
it is found that (5.20) represents the phase-field representation of the lateral stress
profile of the interface. Accordingly, the elastic modulii arise as moments of the
lateral stress profile, as stablished in the Helfrich theory (Helfrich, 1973).
The particular microstructure and chemical composition of the interface dic-
tate the interactions and internal tensions of the interface, ultimately determining
its elastic behaviour. For instance, in the case of two inmiscible fluids, such
as the water-oil coexistence, the unique contribution to the free energy corre-
sponds to the surface tension, associated solely to the cost of having a surface in
which both species of molecules interact. The benefit of homotypic interactions

when a molecule is surrounded by others of the same species is lost in the inter-
face, where the van de Waals interactions between oil and water molecules are
weaker. Accordingly, the system tries to minimize this contact surface, but the
bending of the surface does not have any energetic cost. Other systems, such as
monoloyers formed by amphiphilics, present a more complicated interface internal
physics, reflected in an extremely rich phenomenology. In the presence of water
and oil, amphiphiles form several different structures ranging from micelles to
lamellar (plane) arrays of amphiphiles separating regions of each fluid (Gompper
and Schick, 1994). The principles driving this kind of organization respond to the
polar nature of the amphiphile, with a hydrofilic head, which prefers to interact
with water, and the hydrophobic tail, which prefers the oil. The system evolves to
minimize the contact between the hydrophobic tail and water. In these systems,
the lateral tensions between the amphiphiles induce a more complex elastic be-
haviour, and all the terms in (2.16) contribute to the interface elasticity. Still, in
the low curvature regime (when the interface thickness ξ is much smaller than the
typical radius of curvature, ξ/R 1; this regime applies to most systems, since
ξ is of the order of the lipid length), the dynamics of the interface is generally
dominated by the surface tension term, since it corresponds to the leading term
of (2.16).
Lipid bilayers present an even more complex internal structure than mono-
layers, and are known to induce a strong reduction in the surface tension of the
interface, sometimes of up to 5 orders of magnitude (Gompper and Schick, 1994),
as a result of the internal balance of the lateral stresses between lipids. Accord-
ingly, the dynamics of the membrane is driven by bending (the subsequent non-
vanishing term in the curvature expansion of the Helfrich energy (2.16), assuming
membrane symmetry). Since the scope of this Chapter is the modeling of biologi-
cal lipid membranes, hereafter we focus on tensionless interfaces. It is noteworthy
that the elimination of the surface tension contribution in phase-field models is
highly non-trivial, as it requires of a subtle balance of the lateral tensions of the
interface.
5.2.1 Cell membrane model
The theory presented thus far offers the necessary ingredients to build a model
for cell membranes. The simplest approach to describe the cell is to follow the
Ginzburg-Landau spirit, considering two domains, the inner fluid (eg cytosol) and
the outer aqueus environment (eg plasma), and associate each medium to one of
the equilibrium phases of the order parameter. The coefficients associated to the
order parameter gradients determine the elastic properties of the interface, and
hence they must be chosen to capture the two main characteristics of membranes:
resistance to bend and vanishing surface tension. The first condition is achieved
by c > 0, given that a second derivative is necessary to introduce a bending
contribution. The second condition requires a negative constant value of g, or
alternatively an inhomogeneous function g = g(φ). However, the choice of the free
energy coefficients to obtain a vanishing surface tension is delicate. The coefficients
determine the equilibrium profile of the order parameter, and in general its solution
requires of numerical integrations. The profile enters in the elastic parameters
calculation, (5.19). Thus, the elastic properties depend on the coefficients via
the equilibrium profile which in turn can be very sensitive to the values of the
coefficients. A priori, there is not a unique solution for g and c that produces the
prescribed elastic properties of the membrane.
We consider the particular case fb (φ) = φ2 − 2φ4 + φ6 , gb (φ) = 2(3φ2 − 1),
and cb = 1, which is of particular interest because its equilibrium equation has
analytical solution and therefore it allows a more fine control of the interfacial be-
haviour. The interface free energy reads (Du et al., 2004; Campelo and Hernández-
Machado, 2006)

κ∗
Fb [φ] = (φ2 − 2φ4 + φ6 + (3φ2 − 1)2 (∇φ)2 + 4 (∇2 φ)2 )dV, (5.22)
2
where the subscript b indicates that this model corresponds to a bending free
energy. The free energy can be rewritten as

κ∗
Fb [φ] = (ψ[φ])2 dV, (5.23)
2
where we have introduced the functional ψ = −φ + φ2 − 2 ∇2 φ and additionally

the parameter κ∗ is incorporated to control the bending rigidity of the interface..
From this expression the chemical potential as defined in (5.2) reads
μb = δFb [φ]/δφ = κ∗ [(3φ2 − 1)ψ − 2 ∇2 ψ]. (5.24)
The relaxed profile is obtained by solving the equilibrium condition, μb = 0.

Although this equation does not have a unique solution, the trivial one ψ = 0
represents the minimal energy solution, given that the bending energy is always
positive. Other potential solutions may arise as metastable solutions. This equi-
librium profile of the model (5.22), ψ = 0, can be analytically integrated, leading
√
to φ0 = tanh(z/ 2), connecting the bulk phase -1 with the phase +1. If this
expression is introduced in (5.19) with the energy coefficients from (5.22), the
Figure 5.1: (A) Scheme of a lipid bilayer. The multiple interactions between the different
chemical groups of the bilayer (including repulsion between polar groups, hydrophobic
attraction, cohesive and repulsive effects between tails, etc) leads to a complicate stress
profile in real cell membranes. However, they are characterized by a balance between
internal tensions, so that the surface tension vanishes in this system. (B) Example of
lateral stress profile sexp(z) for a lipid bilayer, based on the results of Hu et al. (2013)
obtained from MARTINI simulations of DMPC bilayers. (C) Lateral stress profile of
the membrane phase field model, sb (z) (blue line), and for a tension interface, st (z)
(red line). The phase-field model for membranes does not reproduce the exact lateral
stress profile of a realistic membrane but effectively concentrates the interactions in two
contributions, a central term of repulsion and two symmetric attractions, and their bal-
ance recovers the tensionless nature of the membrane. A physical interpretation of this
simple profile is that the central term corresponds to the entropic repulsion between the
lipid tails, whereas the the lateral attractions would represent the attractions between
head and tails, following the spirit of the model of Petrov and Bivas (1984). In compar-
ison, the lateral stress profile of a tension interface, involves a single term at the frontier
between the two phases, penalizing the presence of the interface.
√
resulting elastic parameters are γ = c0 = 0, κ = (23 /3 2)κ∗ and κG = 0, con-
sistent with a symmetric membrane. Hence, the model describes the cell as two
fluid domains separated by an interface with the elastic properties characteristic
of membranes.
From the bending free energy (5.22) we can study the interface properties that
determine its elastic response. The stress tensor is computed from (5.8), obtaining
b
σαβ = −(κ∗ /2){[−φ2 + 6φ4 − 5φ6 − 22 (3φ3 − φ)∇2 φ − (3φ2 − 1)2 (∇φ)2
−62 φ2 (∇φ)2 + 22 φ∇2 ∇2 φ − 2 (∇2 φ)2 ]δαβ + 22 (3φ2 − 1)∂α φ∂β φ−
24 ∂α ∂γ φ∂β ∂γ φ − 24 [∂α φ∂β ∇2 φ + ∂β φ∂α ∇2 φ] + 24 ∂α ∂β ∂γ φ∂γ φ + 44 ∂α ∂β ∇2 φ}.
(5.25)
In Figure 5.1 C, the lateral stress profile sb (z) for the membrane free energy
(5.22) is shown, in addition to the stress profile st (z) for the classic surface tension
Figure 5.2: (A) Bulk potential of the curvature free energy (5.22), with three phases of
equal energy. (B) Linear dispersion of the decay of a phase when subjected to a weak
sinusoidal perturbation. The φ±1 phases are stable, whereas the φ0 phase is only stable
for small domains of length < .
interface which applies, for instance, for a water-oil phase separation, given by
the coefficients c = 0 (not bending penalization) and g(φ) = g0 > 0. In this
last case, the model includes an energetic cost for the existence of the interface.
This area penalty is the surface tension. The stress through the interface are
represented by a unique positive term, implying that the pressure is negative
and thus the interfacial molecules are compressed, trying to minimize the surface
area. In the case of the membrane, the lateral stress of the phase-field model
does not capture a realistic profile, such as the one shown in Figure 5.1 B, but
it condensates the information in two terms of repulsion and attraction. One
could interpret the forces at the middle of the interface as the entropic repulsion
between tails, whereas the two lateral attractive regions may correspond to the
attraction between lipid and heads. In general terms, the profile recovers the
behaviour proposed in the simple microscopic model studied in Section 2.2.1. In
this approach, the balance between attractive and repulsive interactions means
that the interface is not energetically penalized for having a certain surface (ie, it
has zero surface tension) but instead it is penalized for having a non zero curvature,
as the first non-vanishing term in the expanded free-energy corresponds to 1/R2 .
The parameter that determines this energetic cost is identified as the bending
rigidity of the membrane.
The membrane model should include both the bending contribution and the
compression of the membrane. The area-compression effect is modeled by directly
imposing a constant surface area, as first proposed by Helfrich. This is done by
adding a Lagrange multiplier in the bending free-energy, so that the complete
membrane free energy is

3
Fmem [φ] = Fb [φ] + √ γA 2 (∇φ)2 dV. (5.26)
2 2
where γA represents the Lagrange multiplier for the area conservation. If the
interface is small, then |∇2 φ0 | behaves as a δ-function,

3 x
lim √ sech4 √ = δ(x), (5.27)
→0 4 2 2
and the volume integral converges to a surface integral over the interface, given
by the coordinates x ,

3 2 →0
√ |∇φ| dV −−→ δ(x − x )dV = dS. (5.28)
2 2
Although the last term in (5.26) has the expression of a surface tension, note
that γA is shape dependent and it varies with the deformation. The expressions of
the chemical potential μmem and stress tensor σ mem for the complete membrane
model must be obtained.
Finally, it should be noted that a term φ6 is present in (5.22). The system actu-
ally presents three equilibrium phases, φeq = 0, ±1, evoking the case of amphiphilic
systems, where often a third phase is introduced accounting for the presence of
the lipid-rich domain (Gompper and Schick, 1994). Although the three phase
description is actually more realistic (since it considers explicitly the presence of
lipids), the two phase model described above is more consistent with the Helfrich
theory, in the sense that it treats the membrane as a sheet characterized by its
local curvature, disregarding the microstructure of the membrane. By means of
a linear stability analysis, it can be shown that the phase φeq = 0 is not macro-
scopically stable. Figure 6.3 A shows the form of the bulk potential fb , with three
equilibrium phases. The stability analysis reveals that the phase φeq = 0 presents
a different behaviour with respect to φeq = ±1. The exact derivation requires of
a detailed description, but we briefly outline the main steps here. Let us sup-
pose a planar interface separating two of the stable phases. A sinusoidal small
perturbation is introduced, φ0 = φeq + ξeiqx , and one assumes that it will decay
as φ = φ0 ert , an ansatz generally valid in the linear regime ξq 1. Introducing
these expressions in the model, the growth rate r(q) is obtained, as shown Figure
6.3 B. If r < 0 the interface is stable and the system will decay to the initial
planar interface. For r > 0 the system is unstable. Note that in these calculations
an explicit time dependence is required, though it has not been presented yet.
The time evolution is discussed in section 5.3 and we do not make here further
comments on this, but let us concentrate on the stability of the system for each
equilibrium phase. The phases ±1 are both stable for the entire range of wave-
lengths of the perturbation, λ = 2π/q. The phase φ0 = 0 is, however, stable for a
short range λ < . Thus, φeq = 0 is not a thermodynamic stable phase and it can
only be present as a microemulsion, and thus being irrelevant in our membrane
description.
5.2.2 Generalized membrane model
The model presented in this Section corresponds to a symmetric homogeneous

membrane. However, the phase-field methodology allows to extend the model
to cover more complex membranes. Two cases are considered here: asymmetric
membranes with a non zero spontaneous curvature that affects to the membrane
balance, and multicomponent membranes in which the elastic properties vary
along the membrane surface due to an inhomogeneous composition.
Asymmetric membranes
Cells often present asymmetric membrane composition, and benefit from the con-
trol of this property in a number of ways, such as the case of pearling and tubula-
tion in the Golgi apparatus (Campelo and Hernández-Machado, 2007a, 2008). The
phase-field free energy (5.22) can be modified in order to account for a membrane
asymmetry, effectively captured in the Helfrich model by a positive spontaneous
curvature, c0 . The resultant free-energy reads (Campelo and Hernández-Machado,
2007b)
2
κ∗
FSC = (−φ + φ3 − ∇2 φ) − cφ0 (1 − φ2 ) dV. (5.29)
2
The term (1 − φ2 ) represents a δ-function centered at the interface position,

so that the interface is forced to accomodate its surface to the spontaneous cur-
√
vature c0 = cφ0 / 2. In principle, this spontaneous curvature can be spatially
inhomogeneous, c0 (x).
Multicomponent vesicles
The different lipid species present in the membrane often pack forming monospe-
cific aggregates, with important implications in the cell functioning as they are
related with membrane trafficking and signaling (Simons and Vaz, 2004). The
experiments carried out by Baumgart et al. (2003) with lipidic vesicles showed
that the bilayer lipid composition can be controlled to form these aggregates, and
modulate the vesicle shape. Depending on the specific lipids of the domain, the
vesicle presents inhomogeneous membrane properties. The morphology assumed
by the vesicle responds to the balance of this particular membrane composition.
The problem was theoretically approached by Wang and Du (2008), who devel-
oped a model in which both the bending rigidity and the spontaneous curvature
depend on the local domain,

κi
F = (C − c0,i )2 dA + Γi dl, (5.30)
2
where i refers to the lipid species of the domain. The last term in (5.30) rep-
resents a line tension term accounting for the border between the different lipid
domains. Wang and Du (2008) considered the problem of two coexistent species
and introduced a phase field formulation based on two coupled order parameters,
φ and η. The main field φ represented the vesicle surface, whilst the auxiliary
field η formed a perpendicular surface to the vesicle, so that the regions where
both surfaces superpose define the domain of one of the species. Furthermore, the
parameter coefficients of the free energy of φ depend on the the field, κ(η) and
c0 (η), with two constant values at η = +1 and η = −1 and a smooth transition
in the interface. Thus, they propose a free energy of the form

κ(η) 2
Fmc
φ
= (−φ + φ3 − ∇2 φ) − c0 (η)(1 − φ2 ) dV + Γ l(φ, η)dV, (5.31)
2
and added the usual constraints of constant total area and volume for the order
parameter φ. The functional l(φ, η) accounts for the line element of the frontiers
of the lipid domains,
l(φ, η) = L|∇φ|2 |∇η|2, (5.32)
with normalization constant L. Minimization of (5.31) leads to the minimal shapes

of the vesicle. The model nicely reproduces the shapes obtained in Baumgart
et al. (2003), and highlights the subtle control of membrane shape mediated by
the formation of local aggregates with suitable elastic properties.
Alternative membrane models
The model presented here exploits the balance of stresses along the smooth inter-
face in order to obtain a tensionless membrane, whose only elastic contribution
is then the bending. The goal of this approach is that the information of the
interface geometry (eg the local curvature) is intrinsically contained in the gra-
dients of the order parameter, so that we only need to specify the dynamics of
the order parameter and forget any other treatment of the surface membrane. A
different approach has been proposed by Biben et al (Biben and Misbah, 2003;
Biben et al., 2005). The basis of this model can be conceived as intermediate
between phase-field and explicit methods. They proposed a bending free enegy of
the form

κ |∇φ|
F= dV (Ĉ − C0 )2 , (5.33)
2 2
where Ĉ is the total curvature of each isosurface of constant φ. Thus, Ĉ is defined

in the entire volume rather than in the membrane surface, albeit the presence of
the delta-like function |∇φ| reduces the volume integral to a surface integral in the
limit of thin interface, so that in this limit Ĉ(x)δ(x − xm ) → C at the membrane
position xm . The mean curvature is numerically computed from the normal vector
to the isosurface, Ĉ = −∇ · n̂, where n̂ = ∇φ/|∇φ|. Hence, although in this
scheme the interface is not tracked either, the mean curvature must be specifically
computed, implying a number of stability numerical problems that require of a
very fine tuning of the model. For instance, ∇φ → 0 in the bulk, given rise to
numerical divergences of the normal vector far from the interface. This model
of specific curvature calculation reproduces the desired elastic behaviour of the
membrane, but it deviates from the original spirit of phase-field models.
5.3 Membrane dynamics and hydrodynamic cou-

pling
The theory of Ginzburg-Landau provides a basis for the energetic characterization

of membranes. However, the time dependence of the interface is obviously a critical
ingredient in the modeling of the membrane phenomenology. In non-equilibrium
dynamics it is usual to assume that, in absence of external forces, each small
volume element is locally in thermodynamic equilibrium so that the whole system
evolves towards a global equilibrium. In the case of membranes, the separation
in time scales between the rearrangement and diffusion of lipids, ∼ 10−9 s, and
typical cell deformation times and mechanic response, ∼ 10−3 s, ensures that the
so-called local equilibrium approach applies correctly.
In the framework of the Cahn-Hilliard theory (Cahn and Hilliard, 1959), the
dynamic evolution of the order parameters is dictated by a diffusive equation,
5.3. Membrane dynamics and hydrodynamic coupling 97

∂φ δFmem
= ∇ · M∇ . (5.34)
∂t δφ
It is not difficult to show that the total amount of order parameter is constant
in the system. We will call this dynamic model simply phase-field model (PF). If
the interface is locally in equilibrium, this implies that the total amount of each
equilibrium phase is also constant. This remark is specially relevant in the present
context, given that most membranes are impermeable (there is no transport of
water through them) so, for closed systems such as cells, the conserved evolution
ensures that the amount of fluid both inside and outside the cell frontiers remains
constant, and the volume constraint of the Helfrich free energy (2.16) is directly
fulfilled. Alternative models which consider a non-conserved dynamics, such as
an Allen-Cahn dynamic equation (Du et al., 2004, 2005), explicitly introduce the
volume conservation by adding the correspondent Lagrange multiplier.
The Cahn-Hilliard equation (5.34) dictates the dynamics of the interface but, in
many systems, hydrodynamical effects of the aqueous enviroment are also crucial
in the membrane evolution. A paradigmatic example is the study of red blood
cells and lipid vesicles while flowing along capillaries forced by an external flow.
To model the interaction of the membrane with the surrounding fluid, it is usual
to incorporate the Navier-Stokes equation to describe the dynamics of the fluid,
and both equations are coupled describing the interaction membrane-fluid. The
complete Navier-Stokes phase-field model (NS-PF) is
∂φ
+ v · ∇φ = M∇2 μmem . (5.35)
∂t

∂v
ρ + (v · ∇v) = −∇P + fmem + η∇2 v + fext . (5.36)
∂t
where fext is an external forcing driving the flow, such as a pressure difference
(ΔP )/L between the edges of a channel of length L, or a gravity force ρg. The
advection term v · ∇φ in (5.35) describes how the fluid pushes the membrane, and
the response exerted by the membrane is given by the force density fmem , which
affects to the surrounding fluid. The elastic force density fmem is obtained from
the divergence of the stress tensor (5.8),
fmem = ∇ · σmem = −φ∇μmem . (5.37)
The complexity of these equations and the geometries that are usually studied
in membrane problems avoid an analytic integration of the model, and thefore nu-
merical methods are required. The Navier-Stokes equation can be integrated by
different methods which must incorporate the coupling with the order parameter.
Among others, new hybrid formulations have been developed for immerse bound-
ary methods (Du and Li, 2011; Shao et al., 2013) and lattice-Boltzman method
(Kendon et al., 2001).
5.4 Macroscopic equations

The validity of a diffuse-interface mesoscopic model is justified as long as it recovers
the correct macroscopic equations in the limit → 0, the so-called sharp-interface
limit. This procedure has been extensively applied in phase-field models to obtain
the set of equations that describe the dynamics of the interface (Folch et al., 1999;
Biben et al., 2005), allowing the study of some important interfacial behaviour
such as the Saffman-Taylor instability (Hernández-Machado et al., 2003). The
scope of this Section is to characterize the macroscopic equations of the model
that enable us to study of membrane elasticity by means of a linear stability
analysis.
5.4.1 Sharp-interface limit
The sharp-interface limit consists in considering a macroscopic length scale, much

larger than , and derive the equations dependent of the macroscopic variables,
such as the total curvature of the interface, from the mesoscopic dynamic equation
for the order parameter. It is a singular limit, as must be small but strictly
nonzero. We first focus on the dynamic equation of the order parameter without
hydrodynamic coupling, (5.34), which accounts for the dynamics of a membrane
immersed in a diffusive medium, such as gel. We then discuss the application to
the complete NS-PF model in which membrane dynamics is critically influenced
by the hydrodynamics of the flows generated around the membrane.
We only outline the sharp-interface limit here, but further details of the method
can be found in Appendix B. The space is separated into two regions, the (macro-
scopic) outer region, and the (microscopic) inner region of the interface. Both
regions are described by the coordinates r and ω, respectively, related by r = ω/.
Thus, the fast coordinate in the inner region allows to resolve the details of the
interface. In the macroscopic scale, the interface is sharp and then → 0. The
starting point is a flat interface in equilibrium, subjected to a weak perturba-
tion. The interface is characterized by the general interface (5.1). The interface
must be symmetric, c0 = 0, so that the flat interface represents an equilibrium
configuration; for strictly nonsymmetric interfaces the method should perturb an
equilibrium configuration, such as a spherical membrane of radius 1/c0 , requiring
5.4. Macroscopic equations 99
a more complex analytical treatment. Once the fast coordinates are introduced,
as explained in Appendix B, the dynamic equation reads

1 1 2 C 2 2
∂τ φ − v∂ω φ = M ∂ − ∂ω + ∂s + ∂u μ. (5.38)
2 ω
Membrane equilibrium equation
The sharp-interface limit allows the derivation of the macroscopic equilibrium

equation for the membrane, imposing the condition μ = const. (Note that this
condition directly gives v = 0 in (5.38). The problem reduces then to the determi-
nation of this constant in terms of the macroscopic variables. The starting point
is a flat interface in equilibrium subjected to a weak perturbation. The phase-field
profile of the perturbed interface can be expanded, φ = φ0 + φ1 + 2 φ2 + ..., where
φ0 is the equilibrium profile. The chemical potential can be analogously expanded,
accounting for the expanded differential operators (see Appendix B), leading to

μ0 = f0 − g0 (∂ω φ0 )2 − 2g0 ∂ω2 φ0 + 2c∂ω2 φ0 .

μ1 = f1 − g1 (∂ω φ0 )2 − 2g1 ∂ω2 φ0 + 2Cg0 ∂ω φ0 − 2g0 ∂ω2 φ1
(5.39)
+2c∂ω4 φ1 − 4cC∂ω3 φ0 .
...
where f = ∂f /∂φ and f0 = f (φ0 ). The equilibrium value, μ0 = 0 provides the

equilibrium phases φeq and defines the equation for φ0 . The next contribution μ1
is non zero, and its value is evaluated by multiplying by ∂ω φ0 and integrating over
the normal direction,
+∞ +∞

μ1 dω∂ω φ0 = dω[f1 − g1 (∂ω φ0 )2 − 2g1 ∂ω2 φ0 + 2Cg0 ∂ω φ0 − 2g0 ∂ω2 φ1
−∞ −∞
+ 2c∂ω4 φ1 − 4cC∂ω3 φ0 ]∂ω φ0 .

(5.40)
The right hand side of this expression includes terms depending solely on φ0
whereas others also depend on φ1 . We focus first on the later. By using the

relation m1 = m0 φ1 for any polinomial function m, (5.40) can be rewritten into
+∞ +∞
μ1 dω∂ω φ0 = dω∂ω φ0 [f0 − g0 (∂ω φ0 )2 − 2g0 ∂ω2 φ0 − 2g0 ∂ω2 + 2c∂ω4 ]φ1
−∞ −∞
+∞
+ dω∂ω φ0 [2Cg0 ∂ω φ0 − 4cC∂ω3 φ0 ].
−∞
(5.41)
Let us consider now the expression for μ0 . Applying the operator ∂ω , we obtain
the equation
[f0 − g0 (∂ω φ0 ) − 2g0∂ω − 2g0 ∂ω + 2c∂ω3 ]∂ω φ0 = 0. (5.42)
Thus, ∂ω φ0 , known as Goldstone mode and related to the translational in-

variance of the interface, is an eigenvector of the linear operator in brackets in
(5.42). Integrating by parts (5.41) and introducing (5.42), the term associated to
φ1 vanishes and then (5.40) reduces to
+∞ +∞
μ1 dω∂ω φ0 = dω[2Cg0(∂ω φ0 )2 − 4cC∂ω3 φ0 ∂ω φ0 ]. (5.43)
−∞ −∞
Integrating by parts, this equation reads
+∞
μ1 Δφeq = C dωsφ (ω), (5.44)
−∞
where we have introduced the lateral stress profile (5.20), and Δφeq is the gap
between the two bulk phases. The left hand side of this equation can be rewritten
in terms of the pressure, given that in the macroscopic scale p1 (0) = φeq μ1 (0). In
the right hand side, the expression of the surface tension (5.19) can be identified,
leading to
Δp1 = γC, (5.45)
which corresponds to the Young-Laplace equation that provides the pressure dif-
ference sustained across the interface.
By means of this procedure, the subsequent terms of the chemical potential
can be evaluated, providing high-order contributions to the pressure difference.
The explicit calculations are simple but long, and only the main steps are shown
here. By using the equation ∂w μ1 = 0 as with the Goldstone mode, evaluation of
μ2 reduces to:
5.4. Macroscopic equations 101
+∞ +∞
2
μ2 dω∂ω φ0 = (C − 2G) dωω[2g0(∂ω φ0 )2 − 4c∂ω3 φ0 ∂ω φ0 ]. (5.46)
−∞ −∞
The term of the right hand side of this expression vanishes, since s(ω) is
stricly symmetric. The term Δp2 corresponds to the pressure contribution of the
spontaneous curvature and accordingly it vanishes for a symmetric membrane.
From the algebraic calculations shown in Appendix B, and considering the final
expression (B.20), the subsequent term is given by
+∞ +∞
3
μ3 dω∂ω φ0 = (C − 3GC) dωω 2[2g0 (∂ω φ0 )2 − 4c∂ω3 φ0 ∂ω φ0 ]
−∞ −∞
+∞ +∞ (5.47)
2 1 2 2
− Δs C dω2c(∂ω φ0 ) − C(C − 4G) dω2c(∂ω φ0 ) .
−∞ 2 −∞
The first term in the right hand side of (5.47) vanishes due to the symmetry of
the equilibrium profile, and identifying the bending rigidity from (5.19), we obtain
1
Δp3 = − κC(C 2 − 4G) − κΔs C. (5.48)
2
Considering all the contributions and disregarding for simplicity the term as-
sociated to the Gaussian curvature modulus, the macroscopic equation is
1
Δp = Δp0 + Δp1 + ... = γC − κC(C 2 − 4G) − κΔs C. (5.49)
2
This equation corresponds to the equilibrium equation for a symmetric mem-
brane, as first derived by Ou-Yang and Helfrich (1987).
Dynamic equations
In the previous section we restrict our analysis to the situation of equilibrium,

without considering the dynamics described by equation (5.34). The sharp-interface
limit can be extended to this more general case, providing the macroscopic equa-
tions that describe the dynamic behaviour of the interface. The derivation of the
complete dynamic model is necessary to study, among others, the stability and
relaxational properties of the interface.
Following the formalism presented in the previous section, the space is sepa-
rated in two domains, the inner corresponding to the interfacial region and the
outer, which corresponds to the bulk. The variables of equation (5.34) are ex-
panded in terms of . In the inner region, the order −2 is given by ∂ω2 μ0 = 0,
and hence μ0 = m0 + n0 ω. Since μ0 must be finite in the limit ω → ±∞, n0
vanishes and we fix μ0 = 0. The dynamic equation at order −1 reads
−v0 ∂ω φ0 = M∂ω2 μ1 . (5.50)
Integrating this equation accross the interface profile,
∂ω μ1 (+∞) − ∂ω μ1 (−∞) = −v0 Δφeq . (5.51)
Note that v represents the interface velocity, so that in the inner region it is con-
stant by definition. Introducing the matching condition ∂ω μ1 (±∞) = ∂r μ0 (±0) =
0 and we obtain v0 = 0. We found that μ1 is a constant that can be evaluated in
the same manner than in the previous section. The subsequent order, 0 , is given
by
−v1 ∂ω φ0 = M(∂ω2 μ2 − C∂ω μ1 ∂s2 μ0 ). (5.52)
Integration by parts and application of the matching conditions lead to
−Δφeq v1 = M(∂ω μ2 (+∞) − ∂ω μ2 (−∞)) = M(∂ω μ1 (+0) − ∂ω μ1 (−0)). (5.53)
Considering |φeq | = 1 and introducing the permeability B = M/(2|φeq |2 ), this

expression can be rewritten,
v = −B(∇p+ + ∇p− ). (5.54)
This is the continuity equation that describes the interface velocity as a func-
tion of the pressure gradients at the interface.
In the outer region, at first order μ0 = 0 due to the matching conditions with
the inner region. The subsequent order, however, is given by ∇2 μ1 = 0 which can
be rewritten us ∇2 p = 0. This expression represents the Laplace equation in the
bulk.
Taking into account the three equations obtained, the complete macroscopic
model reads
5.5. Linear stability analysis 103
Δp = γC − (1/2)κC(C 2 − 4G) − κΔs C,

v = −B(∇p+ + ∇p− ), (5.55)
∇2 p = 0.
Some remarks should be pointed out here. Although we have included the
equilibrium equation (5.49) for completeness, the sharp interface limit specifies
that the dynamics is dominated by the first contribution to the pressure difference.
Thus, if the interface has a nonzero surface tension, its dynamics will be generally
dictated by surface tension. For tensionless interfaces, in which the first non-
vanishing term is Δp3 , bending governs the dynamics of the interface.
5.5 Linear stability analysis

The macroscopic model (5.55) describes the dynamics of the membrane for the
PF model. In deterministic dynamics, a linear stability analysis can be performed
to obtain information of the properties and relaxational behaviour of the mem-
brane, analogous to the study of the fluctuation spectrum in stochastic systems,
as stated by the Onsager’s theory of linear relaxation processes. We perform a lin-
ear stability analysis to study the response of a flat interface in equilibrium when
subjected to a sinusoidal perturbation (Hernández-Machado and Jasnow, 1988).
The relaxational time of the membrane is calculated, and the method is applied
to measure the elastic properties of the membrane. The flat interface is weakly
perturbed, and the membrane position is then given by y = h(x, t) = ξ0 eiqx+ωt ,
where h is the distance of the membrane with respect to the equilibrium position,
q is the wavelength of the perturbation and ω is the relaxation rate. If ω > 0, the
perturbation will grow and the membrane is unstable; if ω < 0, the membrane is
stable and it recovers the initial configuration in a typical relaxation time 1/ω.
For the pressure field we assume the ansatz p = p0 + Aeiqx+ky+ωt , where 1/k is
the distance from the interface in which the pressure converges to the bulk value.
For sufficiently small amplitudes, the curvature of the interface can expressed as
C ≈ −∂x2 ξ = q 2 ξ. Finally, the velocity of the interface is given by v = ∂t h.
Introducing all these considerations into equation (5.55), we find
A = (γq 2 + κq 4 )ξ,
ωξ = −BA2k, (5.56)
0 = −q 2 + k 2 .
And from here, the dispersion relation is obtained
ωφ (q) = −2B|q|(γq 2 + κq 4 ). (5.57)

where we have introduced ωφ = ω to specify that this result corresponds to the

PF model, (5.34). This represents the relaxational rate for a membrane in a
diffusive medium, such as a gel. In the case of a membrane surrounded by a liquid,
when the fluxes generated by the membrane movement affect the dynamics of the
own membrane, we must solve the complete NS-PF model. The derivation of
the macroscopic model of the NS-PF model is complex, but we can assume that
the dispersion relation in this case should correspond to the general expression
(2.27) (Theissen et al., 1998). For a flat membrane the term associated to volume
conservation is not present and the relaxation rate, ωη , reads
1
ωη (q) = − (γq + κq 3 ). (5.58)
4η
In order to validate the complete NS-PF model, and confirm that it captures
the correct membrane dynamics given by expression (5.58), we numerically study
the relaxation of an interface. The numerical method is described in detail in
Chapter 6, but we include this Section here for completeness. The procedure is
analogous to the derivation outlined above. A flat interface is perturbed with an
initial sinusoidal profile h(x, 0). The interface position h(x, t) is tracked during the
evolution and, from here, the relaxation rate ωη can be easily obtained. The initial
amplitude must be small compared to the wavelength of the perturbation, qξ 1.
In Figure 5.3 we show the dispersion relation for interfaces dominated by surface
tension and bending (membranes), comparing the results for the PF model, given
by (5.57), and the NS-PF model, (5.58). In these simulations, the size of the box
is maintained constant but sweeping along and extensive range of q; we fix the
reference mode q0 as that corresponding to the domain length (ie the minimum q
studied). The relaxation rates are normalized by the relevant relaxation rate w0
of the mode q0 ; for instance, for the PF model of surface tension, ω0 = 2Bγq03 ,
and analogously for the other three cases. The dimensionless curves obtained are
therefore universal for each model and type of interface. The models reproduce the
expected behaviour, though we observe a certain deviation from the theoretical
prediction for the longest modes. This is likely related with the penetration length
of the perturbation, 1/k, which scales linearly with the wavelength λ = 2π/q, so
that the interface effectively interacts with the boundaries of the system in the
limit of low q. By fitting the numerical results to the relaxation rate predicted by
the linear theory, the effective elastic modulii of each interface can be obtained.
Our results show a relatively good accuracy, considering that finite-system effects
are present, obtaining γφ /γth = 1.02 and κη /κth = 0.88 for the PF model, and
γη /γth = 1.17 and κη /κth = 0.85 for the NS-PF model. In these expressions the
subscript th indicates the theoretical value of the elastic modulus.
5.6. Conclusions 105

Figure 5.3: Dispersion relation for interfaces dominated by surface tension and bending
(membranes), in the PF (A) and NS-PF (B) models. The simulations are perfomed in a
box of constant dimensions, and hence q0 represents the lowest mode. w0 corresponds to
the relaxation rate of the mode q0 for each case; eg for the PF model of surface tension,
ω0 = 2Bγq03 , and analogously for the other three cases. Bold grey lines correspond to
the theoretical prediction for the relaxation rate given by expressions (5.57) and (5.58),
respectively. Both models agree with the theoretical prediction, though the curves
deviate at low q, probably as a consequence of a finite-system effect (see main text).
The fitting of the numerical results to the theoretical curves provides a measure of the
effective value of the elastic parameters. We obtain γφ /γth = 1.02 and κη /κth = 0.88
for the PF model, and γη /γth = 1.17 and κη /κth = 0.85 for the NS-PF model.
5.6 Conclusions
Based on the Helfrich theoretical framework, we have presented the phase-field
formalism as a powerful tool for approaching complex phenomena related with
dynamics and morphology of biological membranes. Phase field models have been
applied to the study of different interface problems, but only recently for mem-
brane modeling. The phase-field theory makes use of an order parameter which
has two stable phases, and the interface connects both phase domains by a smooth
profile. Thereby, one needs to solve the dynamics of the field, avoiding the complex
treatment of the moving boundary condition of the interface.
We have shown the link of the phase-field model with the elastic theory of
membranes, providing a flexible methodology for the study of membranes with
different properties. The membrane model presented here effectively describes
the internal stresses of the membrane, which in turn determine their elastic prop-
erties. Phase-field methods allow a fine tuning of the interface properties, as
demonstrated in this Chapter, a basic property when dealing with membranes
given the particular elastic characteristics of these structures. The phase-field can
be coupled to a velocity field describing the hydrodynamics of the surrounding
fluid, since in many situations hydrodynamics plays a critical role in membrane

dynamics.
The robust physical basis of the method allows to derive the general equi-
librium equation of lipidic membranes. The study of the relaxational behaviour
of a flat membrane, analogous to the study of the fluctuation spectrum in real
membranes, serves to validate the model. This procedure can be applied in the
future to study membrane fluctuations in more complex geometries, offering an
interesting method for exploring the interplay between geometry and elasticity.
Chapter 6
Numerical scheme
In the previous Chapter we have introduced the basis of the phase-field meth-
ods for the physical representation of amphiphilic membranes. Equations (5.35)
and (5.36) are generally impossible to solve analytically and numerical methods
must be invoked. Since, in general, at the membrane and cell scale the Reynolds
member is usually ∼ 10−3 − 10−2 , models often consider the Stokes equation. In
our case, however, we integrate the complete Navier-Stokes equation by means
of a lattice-Bolztmann method (LBM), a discretized version of the Bolztmann
equation. It is well known that in the hydrodynamic limit, the macroscopic equa-
tions derived from the Bolztmann equation recover the conservation equations for
the continuous hydrodynamic fields. The convergence of the Bolztmann equation
to the Navier-Stokes equation can be proven by a Chapman-Enskog expansion
(Duenweg and Ladd, 2009). The LBM makes use of this fact to perform numeri-
cal integrations of the discretized Bolztmann equation, and then the macroscopic
fields are recovered. LBM is a fast and robust numerical method for fluid dynam-
ics (Ladd and Verberg, 2001) which has been used in many soft matter systems
and complex fluids (Cates et al., 2009, 2005; Benzi et al., 2009; Gonnella et al.,
1997). Our model extends the classic LBM to incorporate the phase-field method
(Kendon et al., 2001).
6.1 Lattice-Boltzmann method

The Bolztmann theory assumes that the hydrodynamic variables like the density
or the momentum depend on a set of distribution functions of the velocity. The
dynamics of these distribution functions is dictated by the Boltzmann equation,
and they should obey the conservation laws of the system such as mass or mo-
mentum conservation. Hereafter, we concentrate on the discrete formulation of
this theory, the lattice-Boltzmann method (Succi, 2001).
107
108 Chapter 6. Numerical scheme
6.1.1 Single-relaxation time lattice-Boltzmann
In the lattice-Bolztmann scheme, the space is discretized in a lattice, and the

nodes are connected by the velocity vectors ci . The time is also discretized, so
that the velocity distribution functions are fi (x, t). The evolution of this function
is given by the single relaxation time (Bhatnagar et al., 1954) model,
1
fi (x + ci , t + 1) − fi (x, t) = − (fi − fieq ) + Fi . (6.1)
τ
The dynamics of the distribution function f can be separated in two steps. The
right hand side of the equation represents a collision step in which the function
relaxes to the equilibrium one, in a time scale given by τ . The left hand side
of (6.1) represents the propagation of the distribution function to the neighbour
nodes. The term Fi represents an external forcing, such as a pressure difference or
the gravity. The hydrodynamic variables (density and momentum) are recovered
from the distribution functions by

f = ρ.
i i (6.2)
i fi ci = ρv.
The conservation laws enforce that the equilibrium distributions should obey
eq
f = ρ.
i eqi (6.3)
i fi ci = ρv.
¯, where we use this notation here for the sake of clarity,

And the stress tensor σ̄
is defined from the equilibrium distributions as

fieq ci ci = ρvv + σ̄
¯. (6.4)
i
The velocities ci connect with the neighbour nodes but we need to specify the
particular velocity subset. We use the D3Q19 model (Desplat et al., 2001) which
√
includes 1 velocity of modulus c = 0, 6 of velocity c = 1 and 12 of c = 2. So
far, the expression of the distribution function has not been specified. They are
calculated as a expansion in terms of v,

1 1
fieq = ρων 1 + 2 v · ci + 4 vv
¯ : ci ci , (6.5)
cs 2cs
where we have introduced the tensor vv¯ = vv − (u2 /3)v, ν specifies the velocity
subset and the weights ων are constants of the model. For the D3Q19 model ν=0,
6.2. Phase-field integration 109
√
1, and 2, and ων = 12/36, 2/36, and 1/36, respectively, and cs = 1/3 is the
speed of sound. Analogously, the external forcing term,

1
Fi = 4ων 1 − [fext · ci + v · ci − v · fext ], (6.6)
2τ
where fext is the external forcing vector. Finally, the viscosity is related with the
relaxational time by η = (2τ − 1)/6.
6.1.2 Boundary conditions
It is obvious that boundaries are crucial if one is studying the behaviour of RBCs
and vesicles while flowing through thin capillaries. We impose the stick condition
(ie zero fluid velocity at the interface) by means of the bounce-back method
developed by Ladd (1994). The method consists in reflecting the distribution
functions on the solid nodes that come from the fluid after each propagation
step. This formulation implies that the distribution function is bounce backed
to the fluid node instead of been absorbed by the solid (wall) node, and the zero
velocity condition is achieved at the middle point between the fluid and solid nodes
(Stratford and Pagonabarraga, 2008).
In this Thesis, we perform two-dimensional simulations, fixing Ly = 1. The
solid boundaries are placed in the x̂ direction, so that the principal fluid direction
is ẑ. We thus apply periodic boundary conditions in the ẑ direction.
6.2 Phase-field integration

The advection-diffusion equation that dictates the dynamics of the order param-
eter is solved by a simple finite-element method,
(φ(x, t + 1) − φ(x, t))/Δt = −v(x, t) · ∇φ + M∇2 μ(x, t). (6.7)
Both the implementation of the advective term and the discrete calculation of
the different derivatives are detailed below.
6.2.1 Advection
The advection term propagates the order parameter with the fluid fluxes. In a
discrete scheme, this propagation is however affected by numerical dissipation,
which occurs for wavelengths q lower than 2π/Δx, where Δx is the spatial grid
unit. For this reason, it is convenient to use a specific technique to minimize

numerical dissipation, such as the upwind-biased schemes which haven proven
to be very robust (Li, 1997). The advection term v · ∇φ is calculated from the
gradients of φ given by

1
N
N
∂j φ = a−n φj−n + a0 φj + an φj−n ; vj > 0.
Δx n=1 n=1
N (6.8)
1
N
∂j φ = an φj−n + a0 φj + a−n φj−n ; vj < 0.
Δx n=1 n=1
N is the order of the scheme, depending on the precission required. We use

a third-order upwind sheme, N=2. The parameters aN =2 are then adjusted to
minimize the dissipation, a0 = 0.927865, a2 = −0.213933 and a−2 = 0.286067.
6.2.2 Derivative calculations

For computing the different derivatives of the order parameter and the chemical
potential, we make use of the velocity set geometry ci instead of the usual discrete
expressions. Let us consider the order parameter at some node of the lattice at
the position x. Assuming smooth spatial variations in φ, it can be expanded in a
Fourier series
1
φ(x + ci ) = φ(x) + ∇α φ(x)cαi + ∇α ∇β φ(x)cαi cβi + O(∇3 φ). (6.9)
2
Multipliying this expression by cαi and suming over all the neighbours,

φ(x + ci )cαi = ∇α φ(x) cαi cαi + O(∇3 φ). (6.10)
i i
where we have made use of the symmetry conditions

α
c = 0,
αi iα β (6.11)
i ci ci ci = 0.
The expression of the gradients yields

φ(x
i + ci )cαi
∇α φ(x) = α α
. (6.12)
i ci ci
Applying the same symmetry argument, if equation (6.9) is directly sumed

over all the neighbours, the laplacian reads

2 2 ( i φ(x + ci ) − Ni φ(x))
∇ φ(x) = α α . (6.13)
i ci ci
where Ni is the number of velocities, eg. Ni = 19 in the D3Q19 model.
6.2.3 Area conservation

In the previous Chapter, we explained that the area conservation can be modelled
by introducing a Lagrange multiplier in the free-energy. The area is computed
from the square of the gradient of the order parameter,

3
A= √ |∇φ|2dV. (6.14)
4 2
However, one needs to calculate the value of the multiplier that ensures that
the area will be maintained constant. This problem can be solved by several
methods but here we study two: (1) the strictly imposed condition in the phase-
field equation; (2) an effective method which penalyzes deviations from a reference
area.
Strict constraint
We shall consider the dynamic equation of the order parameter,
∂t φ = v · ∇φ + M∇2 [μ0 − γ∇2 φ] (6.15)
where μ0 is the chemical potential of bending, (5.24). Applying the gradient

operator over both sides of the equation and multiplying by ∇φ,
∇φ · ∂t (∇φ) = ∇φ · ∇(v · ∇φ) + M∇φ · ∇∇2 [μ0 − γ∇2 φ] (6.16)
But, in this expression the first term can be rewritten as ∇φ · ∂t (∇φ) =

(1/2)∂t |∇φ|2 . Hence, given that we want to maintain a constant area ∂t A = 0, in-
tegrating over the complete domain and introducing the last relation, one directly
obtains

(−∇φ · ∇(v · ∇φ) + M∇φ · ∇∇2 μ0 ) dV
γ= . (6.17)
M ∇φ · ∇∇4 φdV
This method ensures minute deviations from the initial one, ΔA/A0 < 10−7 ,
but it has an expensive numerical cost as it requires computations of high order
derivatives.
Figure 6.1: (A) Evolution of the total gradient square of order parameter for two values
of α, the parameter that defines the dynamics of the Lagrange multiplier. The initial
condition is a rectangle and it is allowed to relax during the first 300 steps before the
Lagrange multiplier is switched on. α controls the amplitude of the oscilations around
the reference value, as well as the timescale of the oscillations decay. The profile can
be fitted to a curve of the form ΔA(t) = A(t) − A0 = (ΔA0 )e−rt+iωt , where we use
A because the |∇φ|2 is related to the surface area, as given by (5.28). Here, A0 is
proportional to the desired area reference value. (B) Example of the oscillation decay,
in a timescale given by 1/r. (C) Dependence of the oscillation amplitude ΔA with α.
Oscillations are attenuated by higher values of this parameter, showing a exponential
dependence of the form ∼ α2 . (D) Dependence of the decay rate k with α, showing a
∼ α−1 behaviour. The curve demonstrates that 1/α represents the time scale in which
the area converges to the reference value.
Penalty approach
An effective way of conserving area is by penalizing the deviations from a certain

reference area A0 . This can be performed by introducing the following dynamic
equation for the multiplier evolution:
∂γ
= α(A(t) − A0 ) (6.18)
∂t
The parameter α controls the dynamics of the Lagrange multiplier and the
precision of the fixed constraint. We perform a simple test in order to understand
its behaviour and comparison with the strict constraint, as shown in Figure 6.1.
The initial condition is a square domain Ω in two dimensions, so that we initialize
φ(x) = 1 if x ∈ Ω and φ(x) = −1 otherwise. The order parameter evolves freely
during the first t0 =300 steps, and afterwards the Lagrange multiplier is switched
on. We analyze the evolution of the magnitude |∇φ|2, integrated over the whole
domain, which is the conserved quantity in our scheme as it is proportional to
the area of the φ = 0 isosurface, as given by (6.14). For the sake of simplicity
we refer here directly to the area A. The first plot (A) in Figure 6.1 shows
that once the mutiplier is activated at t0 , the actual area oscillates around the
desired value A0 = A(t0 ), and the amplitude of the oscillation is damped, as
can be seen in plot (B). This amplitude is clearly controlled by α∗ = αΔt, and
we perform a systematic study over an extensive range of this parameter to set
its precision and time scale. The oscillation amplitudes are fitted to a curve
ΔA(t) = A(t) − A0 = (ΔA0 )e−rt+iωt . Hence, 1/r represents the timescale in
which the actual area converges to the desired one, A0 . The minimization of the
oscillations in the actual area, ΔA, is important even if they take place in a very
short timescale, since if large area variations are allowed the cell could deform into
prohibitted shapes. In plots (C) and (D) of Figure 6.1 we show the dependence
of the area amplitude ΔA and decay rate r with the control parameter α∗ . Both
magnitudes decrease for higher values of α, although eventually higher α lead to
numerical instabilities. The time scale in which the area converges to the reference
value must be much smaller than the other time scales of the system, such as the
elastic time τκ and the viscous time τη = l/v, where l and v are a typical lenth and
velocity of the system. The results show that for high values of α∗ there is a clear
separation in the timescales of the Lagrange multiplier and the cell deformation,
ensuring that at the cell time scale the area is conserved with high precision.
In our simulations we fix α∗ = 10−4 , which fulfills the requirements of low area
oscilations, fast time convergence to the reference area and it proves numerical
stability over the entire range of parameters studied.
Figure 6.2: (Left) Equilibrium profiles of the φ (red) and ψ (blue) implementations of
a flat interface, for = 1 and Δx = 1. The resulting profiles are different, ∗φ = 0.6 and
∗φ = 0.93. (Right) Dependence of the effective width ∗ with the spatial mesh resolution
/Δx. In the continuum limit, the φ-implementation recovers the analytic solution and
converges to the ψ-implementation.
6.3 Numerical realization of tensionless interfaces
The simulation of cell membranes requires of an exquisite control of the elastic

properties of the interface. Interfaces show an extreme sensitivity to the value of
the surface tension. Interfaces with low but finite value of the surface tension, such
as lipid monolayers, are still dominated by this contribution rather than bending.
Only in the case of bilayers, characterized by a strong decrease of the surface
tension with respect to monolayers, membrane dynamics is controlled by bending.
Accordingly, within the phase-field framework the elimination of any potential
tension contribution is crucial to describe the appropriate membrane behaviour.
In Chapter 5, the phase-field free energy (5.22) has been shown to present a stricly
zero surface tension. However, the precision lose when the model is discretized
introduces certain problems in the control of the interface properties that must
be adressed.
The minimization in a discrete mesh of the free energy (5.22) for a flat interface
leads to an equilibrium profile of effective width ∗ = 0.6, when fitted to φ0 =
√
tanh(x/ 2∗ ), thus the actual and desired interfacial widths do not match. If
this profile is introduced into expressions (5.19), it is straightforward to find that
the resulting interface has nonzero surface tension. As explained in the Chapter
5, only the solution ∗ = entails a tensionless interface. For obtaining this
equilibrium profile, we introduce the field ψ = −φ + φ2 − 2 ∇2 φ and express the
chemical potential (5.24) in terms of this field. The ψ-implementation forces the
solution ψ = 0 which corresponds to the required equilibrium profile (see Figure
6.4. Several cells 115
6.2A). In the continuum limit, Δx/ → 0, the φ-implementation must recover

the tensionless interface. This is tested in Figure 6.2B, where the dependence
of the interface width ∗ with the spatial resolution of the mesh is shown for
increasing widths . Only for ≈ 3Δx (which corresponds to roughly 11 nodes
within the range −0.9 < φ < 0.9, in comparison with the 3 nodes for = 1) both
implementations lead to a similar equilibrium profile.
The implications of the profile precision in the elastic properties of the inter-
face can be inferred from Figure 6.3, where we plot the energetic contributions of
surface tension and bending for different interfacial widths. The effective value
of the effective modulii is obtained from the integration in the grid of the expres-
sions (5.19). For /Δx = 1, the φ-implementation corresponds to an interface
dominated by surface tension whereas the ψ-implementation leads to an interface
dominated by bending, though still a small tension is present. A considerable in-
crease in the interfacial width is required to recover the tensionless interface in the
φ-implementation. Given that phase-field models require of an interface as thin
as possible, we conveniently adopt the ψ implementation and set = 1. Note also
that the remnant surface tension might be partially controlled by the Lagrange
multiplier, as changes in area will be corrected by this last contribution.
6.4 Several cells
The presence of several cells in the domain has two main consequences concerning
the numerical implementation of the model. First, the area conservation condition,
as imposed by the Lagrange multiplier, (6.18), must be modified to restrict the
integration of |∇φ|2 to a local domain of the cell, so that the areas of each cell
can be identified. Second, when flowing freely in a channel, cells can potentially
collide. In the phase-field scheme, collision between two cells will lead to fussion,
obtaining a large vesicle. To avoid this effect, we introduce a short-range repulsion
which prohibits fussion. We have introduced a new field, c(x), which takes the
value i if φ(x) > −0.98 for the i-cell of the suspension, and 0 otherwise. The field
c follows the field φ, and it is updated just after the later. From this field, both
the area calculation and the repulsion between cells can be introduced.
Individual area calculation.
The scheme is exactly the same as presented before, though the integral for com-
puting the area of the i cell, Ai , given by (6.14), is restricted to the domain
x ∈ {c = i}. Each cell has then a specific Lagrange multiplier associated, γi .
Figure 6.3: Effective bending and surface contributions for an interface in the φ and ψ-
implementation, expressed relatives to the expected theoretical value in the continuum
limit. The tension term is expressed in units of energy by multiplying by a typical drop
radius, R = 15 in code units. The elastic modulii are measured as the discrete version of
expressions (5.19). The plot shows that in the ψ a low spatial resolution is required to
obtain a negligible surface tension, and therefore this scheme is much more convenient
than the φ-implementation.
Repulsion between cells.
We introduce a weak repulsion to avoid cell fussion when two cells are too close.
The repulsion is incorporated as a new term to the free energy, and it takes the
form

Erep (xi ) = W e−|xi−xj |/l δ(c(xi ) − c(xj ))[1 − δ(c(xi ))]. (6.19)
j
where j runs over the two closest neighbour nodes, and we fix W = 0.2κ and
l = 1.5Δx. We check the dependence of the cell behaviour with respect to these
parameters, ensuring that the repulsion avoids the fussion but does not introduce
a spureous interaction that pushes the cells away.
6.5. Parameter steering 117
6.5 Parameter steering

The simulations performed in this Thesis are based on LUDWIG (Desplat et al.,
2001), an implementation of the lattice-Boltzmann scheme presented in this Chap-
ter. The parameter steering is determined by both physical requirements and
numerical stability restrictions. As commented in Section 6.3, simulation of mem-
branes requires of a subtle control of the interface elasticity. Thus, must be
sufficiently large to resolve correctly the gradients of φ in the interface. We have
shown that = 1 is a reasonable choice. Hence, the scale of the cell length l is
fixed, since there must be a clear separation, l/ 1. We use l = 50, which
represents a compromise between the previous condition and a sufficiently small
domain so that the simulation time is reasonably short.
We are interested in the viscous (or Stokes) regime, in which inertial effects
are negligible. The Reynolds number must be small,
ρvL
Re = 1 (6.20)
η
and in particular we fix Re < 0.05, ensuring that our results do not depend
on the particular value of Re. This condition fixes that the value of η must be
high, and we set η = 1 in code units (this is the highest value in LUDWIG, as
larger values could potentially lead to problems of numerical stability). The low
Reynolds condition fixes a value of the maximum flow velocity, in turn determining
the maximum external forcing f which is our actual control paramenter. Since
we are interested in capturing a realistic elastic response of the cell membrane
with respect to the surrounding flow, from the maximum velocity the value of the
bending rigidity κ is also fixed. We then obtain a typical relaxation time of the
cell shape τκ = ηl3 /κ. We need to impose that the dynamics of the interface,
controlled by the order parameter mobility M, is much faster than the dynamics
of shape change. Hence, the time of interface relaxation, τ = 2 /M, must follow
τκ /τ 1. This provides a condition for the mobility value, which we fix at
M = 0.1 obtaining the desired separation between both dynamics, τκ /τ = 105 .
Part IV
Red Blood Cells flow in confined

systems
119
Chapter 7
Blood circulation
The recent development of microfluidics has allowed an important improvement

in our understanding of blood circulation and rheological properties. The study
of blood flow at the single-cell scale enables the identification of the individual
elastic mechanisms driving blood behaviour, which formerly should be infered from
macroscopic observations of the whole blood (Thurston, 1972). The dynamics
and elastic mechanics of RBCs in confined systems is a subject of fundamental
interest due to its enormous applications in biomedical engineering, as it affects to
the hemorheology during blood handling and storage, or to cell manipulation in
pathology diagnosis. It also represents an interesting problem from a theoretical
perspective, as the highly non-linear rheology of suspensions of deformable objects
is still poorly understood. In this Chapter, we first study the main applications
and current state of the microfluidic technology in order to identify the problems
of interest within this field. We then overview our current knowledge, both from
experiments and theory, about the behaviour of RBC flow.
7.1 Motivation
7.1.1 Blood properties
The proper functionality of blood is very sensitive to the mechanic properties of

the fluid, as perfusion only occurs when blood shows certain rheological properties
(Baskurt et al., 2007). Altered blood, for instance by abnormal RBC concentra-
tions or stiffening of the cells, can lead to a reduction in the oxygen delivered.
The control of these properties is therefore important for the organism function-
ing. For these reasons, among many others, the study of blood properties has
been a subject of prime interest for a long time (Thurston, 1996).
121
122 Chapter 7. Blood circulation
Rheology
Rheology is the field that studies the flow behaviour and deformability of different
materials. Fluids show a continuous deformation when externally forced, and they
do not recover the initial state once the force is removed. On the contrary, solids
respond to external forces by a certain deformation, after recovering the initial
shape when the forces do not apply. Elastic solids are those which their deforma-
tion is proportional to the applied force. Some materials present an intermediate
behaviour, combining solid-like and liquid-like response depending on the type or
intensity of external forcing. These materials are especially interesting within the
rheology frame, since their rheological properties usually show a strong sensitivity
to this forcing.
The external force that deforms the material, expressed per unit area, is usually
called stress and involves several components, most relevants being the shear stress
(which acts parallel to the material surface) and the normal stress (which in a
fluid is recognized as the pressure). If we consider the rate at which the shear
deformation is applied, a definition for the viscosity of the liquid arises from the
ratio between shear rate and shear stress. Most macroscopic viscometers use the
local relation between shear stress and rate to measure the viscosity (Degré et al.,
2006). However, although this standard procedure provides important information
about the bulk behaviour of the fluid, it is of limited interest for understanding
the flow in very confined systems, when the rheological behaviour can be severely
affected.
Viscosity is constant in Newtonian fluids, but it depends on the applied stresses
and tube geometry in non-Newtonian (or complex) fluids. In the case of blood, it
is known that blood viscosity decreases for high shear rates, presenting a shear-
thinning behaviour.
Hemorheology
Hemorheology is the study of the flow and deformation properties of blood. In

vivo, blood flow is determined by a multitude of factors, including RBC deforma-
bility, hematocrit, elasticity of venules and arteries, blood pressure, etc. Clinical
and experimental evidence suggest that blood fluidity is strongly related to tissue
perfusion, and thus processes affecting the properties of blood, such as damage of
the RBC membrane by a particular disease or ageing, impair the normal rate of
oxygen received by the body cells. Thereby, understanding blood flow is determi-
nant in medicine for identifying symptoms of many diseases associated with the
deficient functioning of circulation.
7.1. Motivation 123
The rheological behaviour of blood is dictated by a complex combination of dif-

ferent coupled effects (Baskurt et al., 2007). The hematocrit is critical to increase
blood viscosity with respect to plasma viscosity. The degree of disturbance of flow
streamlines, and thus the resistance to flow of blood, is severely increased for high
hematocrits. Many mammals are able to specifically regulate their hematocrit in
response to physical requirements (eg a well known example is horses during de-
manding exercise), due to the existance of RBC reservoirs in the spleenic system,
but this mechanism is largely absent in humans. The elasticity of the capillar walls
permits the modulation of the blood pressure, regulating the external forcing driv-
ing the flow. The rigidity and deformability of RBCs is also known to alter blood
properties, and blood with high concentrations of echinocytes or spherocytes show
a much higher resistance to flow, impairing perfusion capability. Blood viscosity
also increases if RBCs form aggregates. RBC aggregation is promoted by certain
molecules allocated at the cell surface, and aggregation is of importance in certain
diseases, such as malaria, which benefits from alterations in normal blood viscos-
ity. Hemorheology is likewise important ex vivo, especially in the improvement of
blood handling and storage for blood transfusion or renal dialysis, among others.
Designing and optimization of pipe devices used for blood manipulation are based
on a detailed knowledge of blood behaviour.
Hemorheology is mainly based on experiments performed in laboratories, and
the importance of the rheological parameters for in vivo flow conditions is still
uncertain. For instance, most viscosity measurements make use of cylindric rota-
tional viscometer, which certainly differs from the geometries found in the circu-
latory system. It has been suggested that alterations in the rheological properties
of in vivo blood do not correspond to the predictions based on ex vivo stud-
ies (Baskurt and Meiselman, 2003). Accordingly, a better understanding of the
fundamental mechanisms dictating hemodynamics in pressure-driven flows is fun-
damental to characterize and predict the correct rheological behaviour of blood in
the human circulatory system.
7.1.2 Cell manipulation
The central role of blood in the body running has been exploited from the begin-
nings of the medicine to detect abnormalities in the state of health. Quoting Toner
and Irimia (2005): “Blood is a treasure of information about the functioning of
the whole body. Every minute, the entire blood volume is recirculated throughout
the body. [...] Cells of the immune system are transported quickly and efficiently
through blood, to and from every place in the body. As a result, blood harbors a
massive amount of information about the functioning of all tissues and organs in
the body. Consequently, blood sampling and analysis are of prime interest for both
medical and science applications, and hold a central role in the diagnosis of many
physiologic and pathologic conditions, localized or systemic. However, tapping
into this wealth of information, for clinical and scientific applications, requires not
only the understanding of the biology involved but also adequate technologies”.
In the last ten years, the microfluidic technology has undergone an important
and rapid development, especially in the designing of fine lab-on-a-chips. These
devices represent an exquisite toolbox for blood handling at the microscale, allow-
ing single-cell distinction and manipulation (see Figure 7.1). The technology can
be used both for scientific research, to improve our understanding about blood
properties, or for applications in medicine. The insights gained from scientific
studies are, in turn, necessary to improve devices design, and thus medicine ben-
efits from this interplay between science and engineering.
The main applications of this technology concern cell manipulation for pathol-
ogy diagnosis and blood preparation for different uses, such as plasma-cell sep-
aration (Dimov et al., 2011). Ideally, specific chips of basic functioning could
provide instantaneous diagnostics and prognostics at the doctor office, avoiding
the tedious process of blood test which usually takes several days (mainly spent
in the transport to the laboratory) and requires large blood samples. It also will
be extremely useful for medical development in regions lacking a suitable medi-
cal infrastructure, such as many areas of Africa where malaria has a catastrophic
effect.
Typically, microfluidic devices take advantage of the different properties of
RBCs, platalers and leukocytes, including size, deformability, density, and mem-
brane electric properties. Both RBCs and leykocytes are target cells in pathology
tests. Devices are designed to exploit the different response of each cell species
(as well as the plasma) to a number of physical fields, and select the object of in-
terest. Among others, mechanical forces, electric and magnetic fields, biochemical
interactions, and optical trapping have been incorporated to chip devices, often
combining more than one technique (Li et al., 2013). However, regardless of the
nature of the field, lab-on-a-chips require of a fine microscale control of the flow.
The most relevant active methods include (Toner and Irimia, 2005)
• Dielectrophoresis (DEP). This technique benefits from the sensitivity of

the membrane to changes in the charge of the environment. When cells are
exposed to non-uniform electric fields, they polarize and their translational
motion or orientation can be driven by control of the external field. The
typical strategy is to create electric traps in which target cells are attracted.
The induced polarization strongly depends on the electric properties and
composition of the external surface of the membrane, so that DEP pro-
vides a very high selectivity in the cell separation, not only at species level
7.1. Motivation 125
but also it can distinguish between different states of activation of similar

cells. Nonetheless, the multiple variables affecting to the output of DEP
experiments represents an important drawback for this technique, as it diffi-
cults the interpretation of the results and comparison between experiments.
Moreover, even if weak electric fields in principle do not affect to the mem-
brane balance, DEP represents a (weakly) invasive technique and potential
conformational changes in the cell membrane must be considered.
• Magnetic interactions. RBCs contain a high density of hemogoblin in

their cytosolic volume. Hemoglobin is an iron-bearing molecule whose fer-
romagnetic properties could be exploited to differentiate RBCs from other
cells. However, RBCs in oxygenated blood present a similar diamagnetic
behaviour to leukocytes, and only in oxygen-depleted blood a weak param-
agnetic behaviour has been detected. High magnetic fields are required to
achieve a good accuracy in cell separation.
• Optical traps. Laser tweezers devices can be incorporated to chips, for

instance creating an optical lattice in which cells are trapped. This technique
offers many possibilities for cell manipulation as it achieves a very high
precision, but devices capable of dealing with millions of cells (as required
in small blood samples) would be really challenging.
• Biomarkers. The addition of certain bioagents at specific points of the chip

can be used to selectively destroy subpopulations of target cells, if exposed
to a toxinc enviroment, or attach them to the surface. For instance, leuko-
cytes are more resistant to solutions of ammonium chloride than RBCs, and
therefore the lysis of the later can be used to separate both cells. Although
these methods also achieve a high precision, agents usually act over all the
cells and the method could affect the healthy state of the target cells.
On the other hand, the use of mechanical forces, such as filters and junctions
of channels, represents a passive method. The size is the most direct observable
for differentiating cells, and the designing of filters that selectively permit the pass
of RBCs, whereas they are thin enough to hold back leukocytes, is the simplest
device that can be fabricated for cell separation. However, separation based on
mechanical restriction has a limited efficiency and low purity of the sample. Still,
methods in this direction have been able to separate small tumor cells (Mohamed
et al., 2004). A different method consists in taking advantage of the lateral trans-
lation that cells show depending on their specific size and shape. Although with
limited examples in cells yet, the method has proven good accuracy for separating
particles of a few microns difference, analogous to the RBC and leukocytes case.
This method uses the interactions of the cells with the channel geometry, so that
complex geometries such as unduloids have been fabricated (Carlo et al., 2007),
combined with passages through thin bottlenecks (Abkarian et al., 2008).
Figure 7.1: (A) Example of lab-on-a-chip for RBC handling. The chip is designed to
exploit the channels geometry that permit the cells/plasma separation. Then specific
biomarkers act over the target cells and the outcome sample is collected. Credits: Ivan
Dimov. (B) Sketch of use of channel geometry to manipulate RBCs. In this example,
the channel is designed to focus the cells to a very specific position in the channel.
Reproduced from Carlo et al. (2007).
7.2 RBC behaviour in shear and parabolic flows
RBCs and vesicles present a rich phenomenology when flowing in thin tubes. The
problem has been especially studied for giant unilamelar vesicles (GUVs), enclosed
bilayers of typically 10-30μm size and circular or elliptical shape, since it has been
generally accepted that these simple structures capture the basic physics of RBCs
and serve as model systems (Dimova et al., 2007). The use of GUVs offers many
advantages as they are more stable under chemical changes of the environment,
and a number of features, such as their internal viscosity, can be manipulated
without affecting their stability. Conversely, RBCs require of a delicate treatment
and the number and range of the parameters studied decreases. However, recent
studies have highlighted the importance of the differences between RBCs and
GUVs in terms of shape and membrane elasticity in blood flow (Abkarian et al.,
7.2. RBC behaviour in shear and parabolic flows 127
2007). Theoretical studies have concentrated on the dynamics of an isolated RBC

or vesicle in linear shear flow, as it represents a simpler problem and invites to a
more extensive analytical treatment than the Poiseuille (or parabolic) flow, which
presents a non uniform stress profile.
7.2.1 Shear flow
Membrane motion
RBCs and vesicles in unbounded shear flows present three main regimes of motion,
determined by two or three basic parameters (Vlahovska et al., 2009). Since most
experiments performed are based on the study of vesicles, here we use this term
to refer equally to vesicles an RBCs. Namely, these regimes are: (i) tank-treading
(TT), when the vesicle deforms into a prolate shape and mantain a fixed position
and orientation, though its membrane is subjected to continuous rotation; (ii)
tumbling (TB), when the vesicle performs a periodic rotation, and (iii) vacillating
breathing (VB), when the vesicle shows oscillations on its inclination with respect
to the flow, also known as trembling.
The dynamics is controlled by two main parameters, the viscosity ratio between
the inner and outer region of the vesicle, λ = ηin /ηout , and the capillary number
which characterizes the shear rate of the force relative to the membrane rigidity,
Ca = η γ̇R3 /κ. Additionally, the deformability of the vesicle is also relevant,
as rigid vesicles only present TT and TB motions (Keller and Skalak, 1982).
A phase diagram for a deformable vesicle in an unbounded flow, obtained by
Kaoui et al. (2009), is shown in Figure 7.2. At low viscosity ratios, typically
λ < 5, only TT is observed. Vesicles exhibit VB motion at a reduced range
5 < λ < 6, whereas for higher viscosity ratios only TB is observed. The transition
between each motion regime depend on the capillary number, as at low capillaries
VB dissapears and TB is observed at lower viscosity ratios λ > 4. The TB
behaviour at high λ is not surprising, as it is known that solid rods in shear flows
rotate with a constant frequency (Jeffrey and Pearson, 1965). Therefore, the solid
limit (κ → ∞, λ → ∞) is seemingly governed by the viscosity contrast, whilst
membrane rigidity plays a secondary role. This phase diagram is drifted towards
higher values of λ when the confinement is increased. These results have been
predicted by analytic models (Keller and Skalak, 1982) and numerical simulations
(Kaoui et al., 2009; Noguchi and Gompper, 2005), showing a nice agreement with
experimental results (Deschamps et al., 2009).
Experiments dealing directly with RBCs have shown the particularities of these
entities. RBCs exhibit a new motion, known as swinging (SW) (Abkarian et al.,
2007), which can be described as a combination of tank treading motion with a
Figure 7.2: Phase diagram of the dynamic motion of vesicles in simple shear flow.
At intermediate and high capillary number, tank treading, vacillating-breathing and
tumbling are sequentially found for increasing viscosity contrasts. At low capillary
number, VB disappears and TB is observed at lower λ. For λ = 1, the motion is
dominated by TT regardless of the value of the other control parameters. Reproduced
from Kaoui et al. (2009).
periodic change in the inclination of its major axis. SW is observed at low viscosity
ratios λ ∼ 4, in the intermediate regime between TB and TT. The mechanism
dictating this motion is still unclear and both the presence of the spectrin network
(and thus the existance of resistance to shear), and the particular biconcave shape
of RBCs have been identified as important features with respect to GUVs.
It has been recently demonstrated that in confined systems, vesicles require
of a larger viscosity ratio to recover the TB motion (Kaoui et al., 2012). This is
likely related with the presence of walls, which confine the motion and reduce the
freedom to rotate.
Wall-induced migration
It is well known that neutrally buoyant particles perform crosstream migrations

in the inertial regime (Re 1), driven by both wall repulsions or shear-gradients
(Wang and Skalak, 1969). This result explains the classic observation that par-
ticles flowing in tubes are located at an off-center position, the so-called Segré-
Silverberg effect. The lateral position in the tube is determined by the balance
between the shear-driven shift from the axis and the repulsion from the wall. In
the viscous regime (Re 1), however, the linearity of the Stokes equation de-
termines a symmetry under flow-reversal that inhibits this crosstream migration.
7.2. RBC behaviour in shear and parabolic flows 129
To break this symmetry, the objects must be deformable or of asymmetric shape

(Olla, 1999).
The migration of vesicles when placed close to a wall in shear flows has at-
tracted interest from theoricists. The problem can be analyzed by the method of
images: the flow driven by the vesicle towards the wall is reversed by its image
vesicle at the symmetric position. By this method an analytic expression for the
repulsion velocity can be obtained (Olla, 1997, 1999), dependent on the distance
to the wall, size of the particle, and capability to deform. It is also expected that
high viscosity ratios promote migration. The models show good agreement with
the experimental data (Abkarian et al., 2002).
7.2.2 Parabolic flow
Single-cell behaviour
In spite of its prime interest for the understanding of microcirculation, the be-
haviour of RBCs in parabolic flow has received limited attention in comparison
with shear flow, at least from a theoretical perspective. At low confinements (when
the channel width is much larger than the RBC diameter, and thus at the cell
scale RBCs effectively interact with a constant gradient velocity profile) RBCs
are known to maintain their discocytic shape, exhibiting little or no deformations.
In thinner channels, when both the width of the channel and the cell diameter
are comparable, RBCs present a complex and fascinating behaviour (Abkarian
et al., 2007). While flowing in channels of < 30μm, RBCs exhibit an asymmetric
morphology known as slipper. Additionally, they can also assume a symmetric
shape, termed parachute, situated at the center of the channel. Whilst both con-
figurations are often seen during experiments in microchannels (sometimes simul-
taneously, see Figure 7.3 A), the conditions necessary to observe one or the other
shape have not been properly determined (Abkarian et al., 2008; Tomaioulo et al.,
2009). Actually, why RBCs remains an asymmetric (slipper) shapes when flowing
in symmetric flows remains as an unresolved question (Noguchi and Gompper,
2005; Kaoui et al., 2009), despite its importance to understand its implications
in the capability of RBCs to flow along very thin conduits. Some authors argue
that the slipper configuration enhances flow efficiency, reducing the lag in speed
between RBC and the flow (Kaoui et al., 2009), whereas others defend that slip-
pers do not serve any specific purpose and they just represent a consequence of
the increasing flow (Secomb et al., 2007).
It is remarkable that, whereas RBCs are usually observed at off-center posi-
tions, spherical vesicles present a considerably different behaviour. The deforma-
bility of the lipidic membrane allows vesicles to perform crosstream migrations,
but they stabilize at axial positions in the tube (Coupier et al., 2008), instead of
the lateral position observed in RBCs.
Collective flow
The flow of RBCs in tubes and channels is critically controlled by the hemat-
ocrit, as RBCs show a very different behaviour at low and high concentrations.
The interactions between RBCs, involving hydrodynamic interactions, purely ge-
ometrical constraints, or aggregation, play a fundamental role in the collevtive
dynamics of the RBC suspension.
At low concentrations, vesicles and hard spheres flowing in thick tubes mi-
grate from the center line and reach a stable trajectory at ∼ 0.6R from the axis,
forming an annulus of high density at this radial distance, the so-called Segrè-
Silverberg effect (Segré and Silberberg, 1962). At high concentrations, however,
RBCs distribute along the tube core, avoiding the region close to the wall (see
Figure 7.3 D). The transition from the single-cell to the high hematocrit behaviour
is still poorly understood (Narsimhan et al., 2013), in spite of its important in the
rheological behaviour of the fluid. The Faehreus-Lindqvist effect (Fahraeus and
Lindqvist, 1931), characterized by a dependence of the blood viscosity with the
channel thickness, is perhaps the most important example. In the range between
roughly 300μm and 10μm of tube diameter, the effective viscosity decreases up
to 4-5 times. The effect occurs as a consequence of the strong repulsions from
the walls that force the blood cells to concentrate on the central region of the
channel. The formation of layers free of cells close to the walls allows a rapid flow
in these regions, enhancing the overall fluidity. At high confinements, the walls
proximity enforces a more concentrated distribution of cells in the center and con-
sequently broader layers of free flow are present. In larger channels the free layers
are proportionally thinner until their effect becomes eventually negligible.
Although in the narrowest channels (<10μm) RBCs flow ordered in a single
row at low concentrations, and thus interactions between cells are disregardable,
at intermediate channels (∼ 20μm) RBCs present a more complex behaviour and
collective effects must be considered.
7.3 Scope of Part IV

This Part of the thesis is devoted to the study of RBC and vesicle suspensions.
We concentrate on parabolic flows, the relevant ones in blood circulation, and in
the limit of highly confined microchannels, which has been scarcely studied by
theoretical models. We focus on the understanding and identification of some of
7.3. Scope of Part IV 131
Figure 7.3: (A) and (B) Shape evolution of a single RBC during downstream flow, for a
channel width of 10μm. At similar flow conditions, both a parachute (A), and a slipper
(B), are developed. Reproduced from Tomaiuolo and Guido (2011). (C) Flow of RBCs
at low concentrations for a slightly thicker channel. Both parachutes and slippers are
observed simultaneously. Reproduced from Oishi et al. (2012) (D) Collective flow of
RBCs. Even at relatively low concentrations (top image; 3% hematocrit), if cells flow
separated by short distances they interact and their behaviour changes, focusing towards
the core of the channel. At higher concentrations (bottom image; 23% hematocrit), cells
are concentrated on the central region and a layer free of cells is formed close to the
walls, the so-called Faehereus-Lindqvist effect. Reproduced from Lima et al. (2009).
the most basic and fundamental aspects driving the dynamics of RBC flow, dealing
with some of the most relevant questions that arised in the previous sections.
Thereby, the main aims of this part are:
• characterization of the interplay between membrane elasticity, cell deforma-

bility and flow properties as the basis to explain the complex rheological
behaviour of the suspension. Specifically, we focus on how the stiffening of
the membrane affects RBC deformability and suspension viscosity,
• identification of the control parameters of RBC flow at the microscale. In

particular, we concentrate on RBC focusing, due to its relevance in the
designing of chip devices, as shown in Figure 7.1,
• comparison between the single-cell behaviour and the collective properties

when several cells interact, and the identification of the mechanisms driving
the different behaviour at these conditions.
The subsequent Chapters, 8, 9, 10 and 11, are dedicated to elucidate these

important questions, and perform a systematic study of the different problems,
exploring wide range of the parameters space.
Chapter 8
Rheology and deformability of

RBCs
Blood is a typical example of complex fluid, in which the non-linear rheological

behaviour is introduced by the presence of RBCs. However, compared to classic
complex fluids, blood is a singular case because its compounds are considerably
large (typical polymers and micelles are of a few hundred of nanometers, compared
to 8 microns of RBCs) and characterized by a remarkable deformability. For these
reasons, two main sources of non-linearity can be identified, the collective effect
induced by the interaction between cells, and the purely elastic effect intrisinc
of each cell. As explained in Chapter 1, changes in the elastic properties of the
RBC membrane is a common consequence of a number of pathologies, and its
impact on the affected blood is still unclear. In macroscopic rheology, however,
it is difficult to separate the collective and the elastic effects. Thereby, the main
aim of this Chapter is to identify the rheological properties of blood that can be
solely associated to the deformability of RBCs, and explore the role of membrane
elasticity in blood fluidity.
In this Chapter, we study the behaviour of a RBC flowing in a thin tube at

fixed confinement, sweeping over a wide range of capillary number. The main
morphological regimes are identified, and we discuss the relevance of the RBC
capability to deform and orient in the suspension fluidity. Before these results
are presented, we first describe the main characteristics and specific values of the
system and parameters of the simulations we have carried out. These conditions
likewise apply to the results of Chapters 9, 10 and 11.
133
134 Chapter 8. Rheology and deformability of RBCs
8.1 Simulation of RBC and vesicle flow

In Part III of this Thesis we present a phase-field model which provides the tech-
nology necessary to simulate cell membranes. In the subsequent chapters of this
Part IV we make use of this model to study different aspects of the elastic response
of RBCs in channel flow. Some preliminar considerations, concerning the relevant
physical aspects on the modelization of blood, must be highlighted here.
RBCs properties
Stricly speaking, RBCs elasticity results from the interplay between the bilayer
and the cytoskeleton. Thus, a complete model accounting for RBC elastic prop-
erties should include the terms of shear and compression. The purely lipidic
description of Helfrich is often associated to vesicles, and the term RBC is applied
to models which include resistance to shear (Noguchi and Gompper, 2005). In
Part II of this Thesis we have shown that the cytoskeleton is severely stressed
under certain high-curvature deformations such as spicules, but its contribution is
subdominant for cell deformations close to the discocyte. We consider that dur-
ing blood flow RBC deformations are weak enough (involving low curvatures) to
disregard the cytoskeleton contribution, and accordingly the Helfrich description
accurately describes the RBC elastic behaviour.
RBCs and vesicles are enclosed membranes characterized by their fixed volume
and area, and the bending rigidity of their membrane, κ. Since the density of
the cytosol and the blood plasma is similar to that of water, we consider that
the density is homogeneous. The viscosity of the cytosol is roughly ∼ 5 times
larger than that of the plasma, but studies have proven that the viscosity contrast
does not play a relevant role in the cell dynamics at confined systems (Tahiri
et al., 2013), and for simplicity we assume a homogeneous viscosity, which we
express relative to that of plasma, ηin = ηout = η0 . Therefore, λ = 1 and only
tank-treading motion is expected. Blood plasma is known to present a slightly
viscoelastic behaviour, but we disregard this contribution for simplicity. The
viscosity of the lipid membrane is known to be considerably higher than that of
plasma ∼ 102 η0 , but the volume fraction of lipids in the system is small enough to
consider that the hydrodynamic dissipation in the bulk is prevalent, and membrane
viscosity does not play a relevant role in the aspects of the dynamics we focus on
(Vlahovska et al., 2009).
In order to compare the role of shape, and distinguish the particularities of
RBCs with respect with classic vesicles (GUVs), we generate a set of shapes of the
same diameter a, including a nearly spherical shape, of reduced volume vred = 0.97,
and three deflated vesicles of 0.69, 0.55 and 0.48 (note that, in spite of the 2D
8.1. Simulation of RBC and vesicle flow 135
approach, we maintain the nomenclature reduced volume for consistency with the
literature). The last one corresponds to the usual discocyte shape of RBCs. The
shapes are obtained from the minimization of an initial ellipsoid with the desired
area and perimeter, by switching off the flow v = 0 in equation (5.35). During the
first steps, the Lagrange multiplier is set to zero, while the equilibrium profile is
formed. We then restore the multiplier in order to conserve the perimeter. Since
the time scales of the interface dynamics and shape deformation are very different,
we assume that the lose in perimeter during the first steps is disregardable, though
we check the actual values of the area and perimeter at the final state. Once the
system reaches the steady state, the equilibrium shape of the cell is obtained. The
vesicles used as equilibrium configurations in this Thesis are depicted in Figure
8.1.
Figure 8.1: Morphologies studied in this Thesis, for decreasing νred =0.97, 0.69, 0.55
and 0.48, from left to right. The last one corresponds to the 2D profile of the disco-
cyte morphology, representative of the healthy RBC shape. The shapes are obtained
by minimizing the bending energy (5.22) for an initial ellipsoid, maintaining constant
volume and area.
Flow characteristics
In Part I, we show that the periodic nature of blood flow dictated by the heart
is attenuated along the circulatory system due to pressure damping. The blood
pressure in the capillaries is nearly constant, and we consider a flow driven by a
time-independent and homogeneous pressure gradient.
We carry out simulations in a two dimensional domain, which qualitatively
capture the actual RBC behaviour (Kaoui et al., 2009) and enable a more extensive
sweep of the capillary number avoiding the constraints of the computational time
for tree dimensional domains. The fluid is forced in the channel axis direction, z,
and periodic boundary conditions are applied in this direction. Strictly speaking,
this situation corresponds to a regular array of RBCs, but we keep the cell to cell
distance sufficiently large so that the cell does not interact with its images. This
flow might be understood as the section of a rectangular flow in 3D, with walls
Ly Lx (often known as plane Poiseuille). We consider that the coupling of the

shape in the normal direction, y, is subdominant with that induced by the close
walls in x̂.
At narrow capillaries and channels, the concentration of cells reduces down
to ∼ 10% Fung (1997). The interactions between RBCs consequently decrease
and the dynamics of the cells is dominated by the interactions with the walls.
The RBC effectively behaves as hydrodynamically isolated from its neighbours
cells. This effect requires tubes of similar width b to that of the cell, roughly
b ∼ 10μm, which includes a large proportion of the microvessels that compound
the circulatory system, as well as many of the typical widths of the channels
currently used in microfluidic devices. We explore the range of confinements
a/b = 0.8 − 0.3. In addition, consistent with Table 1.1, the Reynolds number is
maintained below Re < 0.05, corresponding to the viscous regime present in the
thinnest capillaries.
The interplay between RBC elasticity and flow is captured by the capillary
number, Cκ , defined as the ratio between the elastic (2.27) and viscous τη = b/v̄z
relaxational times,
τκ η0 v̄z a2 a
Cκ = = . (8.1)
τη κ b
where v̄z is the mean velocity in the direction of the flow. This new definition
relates with the previous one by Cκ = Ca(a/b), containing more information of the
confinement of the system, given by (a/b). Hereafter we only use this definition of
capillary number, Cκ . The change in the capillary value can be understood as the
different response of a specific cell under different flow velocities or, alternatively,
the behaviour of cells of different softness under the same flow conditions. We
will explore the range 0.2 < Cκ < 120. For RBCs, typical values for the bending
rigidity and cell diameter can be fixed at κ = 2 · 10−19 J and a = 8μm.
Rheological measures
Traditionally, the rheological properties of complex fluids have been experimen-

tally determined by means of rotational viscometers (Thurston, 1972). These
devices, characterized by a cylinder-in-cylinder or cone-plate geometry, operate
maintaining a constant shear rate (fixed by the rotational speed) or a constant
stress, and thus the effective viscosity is extracted from the response of the fluid
in terms of the measured shear stress or plate velocity. However, the behaviour
of complex fluids is highly sensitive to the flow properties, and hence the fluid
response in pressure-driven flows in narrow channels and tubes may substantially
8.2. RBC behaviour 137
differ from that observed in rotational viscometers at larger scales (Baskurt and
Meiselman, 2003). In order to study the rheological behaviour of complex fluids
below the microscale, an alternative procedure consists in using micro-PIV tech-
niques, in which small beads are added to the flow, and the tracking of these beads
allows the reconstruction of the velocity field. The shear stress and rate can be
then computed, and the shear viscosity arises from the ratio of shear stress to rate
(Degré et al., 2006). However, in very narrow channels, the resolution necessary
requires of a high control of the flow conditions (Pommer and Meinhart, 2005),
making the previous approach inappropriate. An alternative measure of the vis-
cosity, usually termed effective (or apparent) viscosity, is obtained from the ratio
of the pressure drop to the flow rate in a channel. Chip viscometers based on this
idea have been increasingly used in microfluidics (Guillot et al., 2006; Srivastava
et al., 2005; Nghe et al., 2010; Gachelin et al., 2013), due to their simplicity, low-
cost and fast measure processing. For practical purposes, we base our rheological
study in the latter approach to obtain the effective viscosity, facilitating a direct
comparison with these new chip viscometers.
The effective (or apparent) viscosity of the whole suspension (ie liquid and
cells) is computed from the relation of the applied force and the outcome flow
given by the mean velocity v̄z ,
f0 2
ηef f = b. (8.2)
12v̄z
This expression arises as the viscosity necessary to, assuming a Poiseuille flow
and given the imposed forcing f0 , obtain a mean velocity v̄z . The high inhomo-
geneity of the suspension of study (originated on the large length of the object
that introduces the elasticity relative to the channel width) suggests to deal with
averaged quantities.
8.2 RBC behaviour

The sequence of RBCs flowing in a parabolic flow at steady state, for confinement
a/b = 0.71, is shown in Figure 8.2. We maintain fixed κ, the bending rigidity,
and increase the intensity of the external forcing (and therefore the mean flow
velocity also increases). In all cases the RBC was initially placed symmetrical at
the channel center, with its main axis normal to the flow direction. At low Cκ
the RBC flows pushed by the fluid with slight or no deformation from the initial
discocytic shape due to its relative rigidity. This initial stage is called discocyte
regime. As fluid forces become important, the RBC starts to deform coupling
its shape with the flow. It simultaneously acquires an asymmetric shape and
Figure 8.2: Red blood cell morphologies in a Poiseuille flow, with vred = 0.48 and
confinement a/b = 0.71, for increasing capillary number, Cκ . The coloured regions
represent the three main morphological regimes, namely the discocyte (yellow), the
slipper (red), and the parachute (blue); this convention will be used all along the Thesis.
The dotted line represents the channel axis, and the crosses are the center of mass of each
RBC. At low capillaries, RBCs maintain a discocytic shape. For increasing capillary,
they start to deform and rotate, migrating from the channel axis. The lateral positioning
in the channel is accompanied by an allignement with the flow profile, exhibiting the
slipper shape. The center of mass of the cell moves towards a forward position, and
the leading part of the cell becomes thicker. Further increasing Cκ , this configuration
becomes unstable and the RBC moves back to the axis, now acquiring a parachute like
shape.
eventually migrates from the center line, displaced to a lateral position in the
channel. Cells initially placed close to the wall migrate in the opposite direction,
stabilizing at a close but outer (closer to the wall) position than those starting from
the center, forming a band of stable lateral positions where cells are localized. This
band narrows for increasing Cκ . In this intermediate regime, the so-called slipper
morphology, the RBC is characterized by an asymmetric shape and orientation
with respect to the local flow profile, with the leading lobe of the cell being bulkier
than the trailing one, leading to the usual denomination of slipper or slipper-bowl
shape. At high capillaries, the off center positions become unstable and the RBC
returns to its axial position, but the strong fluid flow forces it to bend and to
acquire the parachute like shape, in which the RBC is again normal to the flow
direction but both edges of the cell are bent, coupled to the parabolic profile of
the flow.
Energetic contributions
The three morphological regimes previously introduced can be understood in terms

of the RBC deformation energy, as shown in Figure 8.3A, where the dimensionless
bending energy Fb /κ, expression (5.22), is plotted as a function of Cκ . The defor-
mation energy is expressed as the excess energy with respect to the reference state,
the equilibrium discocyte. In the discocyte regime, the bending energy gradually
increases from the equilibrium one, as the cell is forced to bend and couple its
surface to the incoming flow. The bending energy peaks before the RBC shows
a marked migration towards an off center position, where it aligns its axis with
the local flow profile and this configuration allows a relaxation of the bending
energy by a recovery of a more planar shape, the slipper-bowl. We define the
limit between the discocyte and slipper regimes at the maximum bending energy
that preludes the migration of the cell, Cκ = 4. Slippers are characterized by a
low deformation energy, especially at Cκ < 40. For higher Cκ , the RBC develops
a thicker leading lobe to stabilize the horizontal inclination, and this asymmet-
ric deformation is increasingly penalized by the elastic energy. Parachutes are
highly distorted shapes, penalized by a high deformation energy. The transition
between typical slipper and parachute energies is sharp, and we place the limit of
the parachute regime just before the first high-energy configuration found, corre-
sponding to the cell g in Figure 8.2, at Cκ ∼ 98. Cells g and h of Figure 8.5 could
be arbitrarily identified as slippers attending to morphological criteria, given that
they still maintain an important asymmetry. A classic symmetric parachute, cell
i, is found at Cκ = 111.
Figure 8.3B shows the ratio between the bending energy and the surface in-
compressibility contributions to the membrane elastic energy (5.26). At low Cκ ,
the RBC deformation induced by the flow is weak and the stretching imposed to
the membrane is negligible. The RBC elastic energy is therefore dominated by
the bending contribution. At intermediate Cκ , RBCs undergo larger deformations
and the incompressibility contribution becomes comparable to the bending one.
The formation of slippers far from the axis roughly defines when the incompress-
ibility becomes dominant. For higher Cκ , the incompressibility plays a relevant
role in the cell dynamics. If it is removed (effectively performed in our simulations
by setting the Lagrange multiplier to zero), the object lengthens and undergoes
extreme deformations. The incrompressibility arises as a crucial ingredient in the
preservation of a compact shape, and as a consequence, RBCs flowing are high
confinement are subjected to strong membrane tensions.
RBC morphologies
In this section we deal in detail with the morphological characterization of each

regime. The parameters studied to describe each morphology are the shape dis-
tance to the discocyte, and the shape asymmetry with respect to the cell axis.
The shape distance provides a measure of the amplitude of the deformation from
the reference shape, fixed here as the relaxed RBC discocyte. It is defined as

Δs = i |ri − ri0 |/Δ0s , where i = 1, .., N are the sequence of N points of the cell
contour, and r and r0 represent the position vector of the deformed and relaxed
shapes, respectively (see Figure 8.4A, inset). Δ0s corresponds to a reference value
of the distance between the equilibrium shape and the ellipse with equivalent
perimeter and area. The shape asymmetry ΛA = (A1 − A2 )/(A1 + A2 ), where
A1 and A2 are the areas of the regions separated by the cell axis, measures the
asymmetry of the deformed shape. Two asymmetries can be calculated, either
with respect to the major (disk diameter in 3D) or the minor (disk thickness in
3D) axis (see Figure 8.4B, inset).
Discocytes .– RBCs maintain a discocytic shape close to its equilibrium mor-
phology for capillaries Cκ 5. For Cκ > 2 the RBC deforms into a slightly
asymmetric shape, bending and coupling its shape to the external flow. RBCs
maintain, however, a symmetric position in the channel and remain moving nor-
mal to the flow direction (see Figure 8.2). The effective viscosity in this regime is
high compared to that of the solvent, corresponding to the plateau when RBCs
are close to the rigid limit. RBCs occupy a large proportion of the channel section
and thus force the solvent to flow through thin layers between the RBC and the
walls. Part of the shear-thinning behaviour, corresponding to the region when the
RBC deforms and curves, also takes place within the discocyte regime. The range
of Cκ characteristic of this regime requires velocities < 5μm/s, below the typical
ones affordable in experiments, and therefore these shapes are rarely observed in
experiments at high confinement.
Slippers .– By gradually increasing the flow velocity, the fluid pressure is large
enough to force the RBC to deform and bend, adapting its shape to the flow
profile, and losing its symmetric discoid shape. Eventually, RBCs focus to a
lateral position, and depending on the initial height their flowing within a band of
permitted trajectories, at roughly hcm /R ≈ 0.22. In the capillary range Cκ ≈ 10−
90, the RBC stays off-center adopting the slipper morphology with very limited
changes on its shape. Although the term slipper has been applied for different
morphologies in the literature, the energetic criteria used here corresponds to the
slipper-bowl shape that RBCs adopt when laterally displaced. The asymmetric
position is accompanied by a tank-treading motion, characterized by a continuous
rotation of the membrane while flowing. The switch from the axial to the stable
lateral position requires of a small increase in the capillary, from Cκ ∼ 7 to ∼ 12.
Far from the central region, where the curvature of the flow profile is lower, the
RBC recovers a symmetric shape, closer to the equilibrium one (Figure 8.4 A),
similar to a bowl. It gradually aligns its main axis with the flow, reaching a nearly
horizontal orientation at high capillaries. In the 3D analogy, the RBC will flow
with its disk parallel to the walls. In the slipper regime, the asymmetry along the
major axis reduces to nearly zero (Figure 8.4 B). However, the center of mass is
gradually displaced to a forward position as the leading lobe becomes thicker for
Figure 8.3: (A) Bending energy of representative morphologies of RBCs in parabolic

flow (first term in expression (5.26)), expressed as excess energy from the equilibrium
discocyte, Fb0 . Discocytes are obviousy the shapes with less energy (as they are the
closest to the equilibrium shape). However, when they start to deform their energy
considerably increases. In this sense, the lateral migration allows relaxation of the
elastic energy as the RBC recovers more symmetric, discocytic-like shapes (see below).
Parachutes are much severely penalized shapes. The inset shows a zoom up of the energy
curve at low Cκ , where the relaxation of the bending energy when the cell assumes a
slipper shape can be observed. (B) Ratio between the incompressibility contribution
(second term in eqn 4.4, associated to the Lagrange multiplier that ensures membrane
inextensibility), and the bending energy. At low capillaries, the cell is nearly rigid and
the membrane does not undertake relevant stretching. At higher capillaries, when the
bending forces are much weaker than the fluid forces, the incompressibility represents
the dominant contribution to the elastic deformation.
increasing capillary number, as shown in Figure 8.4 B, where the asymmetry along
the minor axis increases whereas the asymmetry along the major axis remains at
low values. The effective viscosity in this regime continues decreasing with the
migration and it reaches a value ∼ 1.05η0 when the RBC is already on its lateral
position deformed in a slipper-bowl shape. Experimentally, isolated slippers are
observed at higher values of Cκ , close to the transition to the parachute (Abkarian
et al., 2008).
Parachutes .– For Cκ > 90, the increasing asymmetry along the minor axis
leads to a shape in which the leading lobe is much thicker than the trailing one,
and eventually this configuration becomes unstable and the RBC returns to a
centered position. The RBC acquires the symmetric parachute shape, a severely
deformed morphology with respect to the equilibrium discocyte. The transition
from slippers to parachutes is rather abrupt and we find parachutes similar to that
observed in experiments (centered and highly curved, eg cell i in Figure 8.2) at
Cκ > 105, in good agreement with previous numerical (Cκ ∼ 112 in Kaoui et al.
(2009)) and experimental results (Cκ ∼ 110 in Tomaioulo et al. (2009)), assuming
Figure 8.4: Morphological

characterization of RBCs in a parabolic flow. (A) Shape
distance Δs = i |ri − ri |/Δ0s , where i = 1, .., N are the sequence of N points of the
0
cell contour, r is the position vector and the 0 refers to the equiblibrium configuration.
This parameter is normalized with the shape distance to the ellipse of same area and
perimeter, Δ0s . The shape distance between slippers and the equilibrium discocyte
is lower than discocytic shapes that remain more centered and are forced to bend.
Parachutes represent, as suggested by the previous plot, large deformations from the
equilibrium shape. (B) Asymmetry of each shape, defined as ΛA = (A1 − A2 )/(A1 + A2 )
where A1 and A2 are the areas at both sides of the minor or major axis. The axis is
defined as the segment between the two most distant points of the cell contour (major
axis) and the normal to this segment (minor axis). Interestingly, the asymmetry along
the major axis of the RBC almost vanishes in the slippers. However, the asymmetry
along the minor axis increases as the cell becomes thicker in the leading edge, until it
eventually transforms into a parachute.
a typical value of the bending rigidity κ ∼ 2·10−19 J). The effective viscosity within
this regime is very close to that of the solvent, as viscous forces are much larger
than elastic ones and the RBC is not able to significantly perturb the surrounding
flow.
8.3 Effective viscosity of RBC suspensions
The effective viscosity for an isolated RBC is shown in Figure 8.5. At low Cκ , the
system presents a certain memory of the initial conditions and the steady shape
will depend on its initial configuration. This effect implies that for each Cκ , there
is a set of possible shapes, orientations and positions of the RBC flowing in the
channel. We plot in Figure 8.5 an average, effective viscosity, which we obtain
from averaging over a set of 21 different initial configurations, including 3 different
heights along the channel section and 7 different RBC inclinations. Further details
on this phenomena are provided in Chapter 10.
8.4. Flow description 143
At low Cκ , the viscosity presents a newtonian behaviour, in which the RBC

behaves as a rigid object (without deformation) pushed by the flow. When the
external forcing becomes comparable with respect to the elastic membrane forces,
the RBC starts to deform and the fluid viscosity decreases, showing a strong shear-
thinning behaviour. The viscosity continues decreasing with the focusing of RBCs
to a thin lateral band, but its decay gradually decreases when this band narrows,
and eventually the cells are localized at a unique position. The effect of the flow
reduces to a increasing orientation of the cell, entering in the asymptotic region in
which the viscosity converges to that of the solvent. For higher Cκ the elastic forces
of the RBC are largely exceeded by the flow forcing and the elastic contribution
to the viscosity is small. The off center position permits the cell to align with
the flow profile, acquiring gradually an horizontal inclination, minimizing the cell
cross-section exposed to the incoming flow and thus reducing the dissipation.
Centered cells present a normal orientation which is less favourable. Thereby, the
alignment of the cell requires a previous migration towards a lateral position, and
both mechanisms play an important role to explain the shear-thinning behaviour
of the suspension.
In Figure 8.5 (inset), the effective viscosity is shown for different bending
rigidities, to test if Cκ represents an appropriate parameter to characterize the
rheology of the fluid. For each bending rigidity, we sweep the external forcing f0
covering in all the cases the same range of Cκ . Thus, for each capillary number
analyzed, the ratio f0 /κ is constant. Accordingly, the curves should collapse
indicating that Cκ captures the relevant mechanisms controlling blood viscosity,
as shown in Figure 8.5.
8.4 Flow description
In this section, we describe in detail the flow characteristics and stress profile for
the three main regimes. In Figure 8.6 A, we show the colormap of the velocity field
in the flow direction, vz , from which we have substracted the contribution of the
unperturbed Poiseuille vz0 in the absence of cell. The velocity is normalized by the
maximum velocity of the unperturbed flow, vz0,max . Within the slipper regime, the
flow is considerably disrupted, and at the region occupied by the cell the velocity
reduces down to a 50% of the imposed value. Only close enough to the walls
the velocity converges to the reference one, forced by the stick condition at the
boundary. The velocity profile is therefore more planar (compared to the parabolic
profile of the unperturbed Poseuille) at the center region. In the slipper regime,
flow disruption is weaker, roughly 20% in the region occupied by the cell, and the
perturbation induced by the cell is obviously asymmetric. In the parachute regime,
Figure 8.5: Effective viscosity of a RBC suspension as a function of the capillary number
for different bending rigidities, κ. The value presented here, η̄ef f , is averaged for different
initial conditions of the RBC, see main text for details. The coloured regions correspond
to the three morphological regimes (see Figure 8.2). The curves for different rigidities as
a function of the shear rate show the sensitivity of the viscosity and RBC morphology to
the rigidity of its membrane (inset); however, the curves collapse when the relative effect
between the viscous and elastic forces is considered. The curve presents three regions:
(i) a newtonian plateau at low capillaries when it achieves a value ∼ 3.5 (which depends
on the volume fraction; in these simulations, φves = 0.012). (ii) a region of strong shear
thinning decay, converging to the solvent viscosity; and (iii) an assymptotic region
in which the viscosity converges to that of the solvent. Both the shear rate and the
bending rigidity, which are subsequentially incorporated to the dimensionless number
Cκ , are normalized here with respect to their lowest value used in the simulations.
8.4. Flow description 145
the flow recovers a more symmetric profile, and the velocity decrease at the central
region is 25 − 30%, higher than in the slipper. This fact agrees with the general
observation that centered cells flow slower than those situated at the lateral, as
we study in Chapter 9. The length of the perturbation is also important as it
determines the range of the hydrodynamics interactions present. For instance,
if a cell is situated close to the discocyte shown in Figure 8.6 A, it will interact
with a very perturbed flow, attenuated by the original cell. Thus, the second cell
will be subjected to a effective lower Cκ than the actual imposed, and a different
deformations (such as lower deformations) are likely expected. This effect, called
termed here as screening, will be studied in detail in Chapter 11, when several cells
are interacting. Figure 8.6 A intuitively shows that the length of the perturbed
flow can be associated with an interaction distance, which reduces with Cκ . The
dependence of the hydrodynamic interactions with the flow will be also adressed
in Chapter 11.
The vorticity map for the different regimes is shown in Figure 8.6 B. The vor-
ticity here corresponds to the normal component, ω = ωy , and we also extract the
contribution of the unperturbed Poiseuille and normalize with respect to its max-
imum value. In the discocyte regime, the vorticity is homogeneously distributed,
with maximum values at the edges of the cell closer to the walls. In the slipper
regime, however, the velocity is extremely localized at the trailing lobe of the cell,
and the rest of the domain presents neglible vorticity. The pattern contrasts with
that obtained for parachutes, in which the vorticity is extensively distributed in
the regions surrounding the cell. Accordingly, the development of slippers seems
to reduce the formation of vortices with respect to the centered configurations.
η
From the viscous stress tensor, σαβ = −η[(∇α vβ )+(∇β vα )T ], we can obtain the
η
shear stress profile, in our geometry given by the component σxz . This magnitude
relates with the viscosity of the suspension, as explained in Section 8.1. The
distribution of shear stresses for the three main regimes is shown in Figure 8.7
A, for the same conditions as in Figure 8.6. Analogously to the previous plots,
we extract the contribution from the unperturbed Poiseuille and normalize with
the maximum value at each configuration. In the discocyte regime, shear stresses
concentrate on the inner region of the cell, and in the region between the cell and
the wall where the fluid is pushed by both the membrane and the wall. In the
slipper regime, the stress reduces considerably and it is concentrated around the
trailing lobe of the cell, being of relative low value in the rest of inner region of
the cell. In contrast, the centered parachute presents high shear stresses in the
interior fluid, particularly at both cell lobes.
Finally, in Figure 8.7 the map of normal membrane force, fnmem = n̂α fαmem ,
is shown, normalized by the external forcing f0 . n̂ is the normal vector to the
membrane surface, computed from n̂ = ∇φ/|∇φ|. The results show that in the
symmetric configurations, discocyte and parachute, the membrane presents an ho-

mogeneous tension distributed along their contour, whereas in the case of the slip-
per, the membrane force is concentrated on the two extremes of the cell, whereas
the lateral edges parallel to the major axis are exposed to very reduced forces. The
magnitude of the force is high compared to that of the external forcing, roughly
50 times higher. It is also interesting that maximum forces in the parachute are
slightly higher than those found in the slipper, in spite of the increase in the exter-
nal forcing by a factor of 2. This fact suggests that cell membrane at the slipper
regime presents lower elastic stresses than the parachute, in accordance with the
results from the deformation energy obtained in Section 8.2.
8.5 Discussion
The sequence of RBC morphologies depicted here has been also found in exper-
imental studies with RBCs (Abkarian et al., 2008; Tomaioulo et al., 2009), with
qualitative good agreement between the shapes found in our simulations and those
observed in the laboratory, as shown in Figure 8.8. Likewise, our results agree
with the shapes found in previous numerical studies in which the membrane elas-
ticity incorporates an in-plane contribution (Noguchi and Gompper, 2005; Shi
et al., 2012), ie both shear and area-compression, suggesting that the importance
of these contributions is limited in this system and the cell dynamics is effectively
dominated by the bending elasticity and area incompressibility. This hypothesis
is in good agreement with some recent results, in which the dynamic behaviour
of RBC flow has been analyzed for healthy cells, and for others in which the
cytoskeleton was severely stiffened by addition of aminade (Forsyth et al., 2011).
Aminade produces rigidity of the spectrin mesh, but it does not affects the proper-
ties of the bilayer. Healthy and stiffened cells were shown to present virtually the
same behaviour, concluding that the influence of the cytoskeleton in this system
is limited.
The relative softness of RBCs, which for typical in vitro flow velocities at highly
confined channels lie in the advanced slipper and parachute regimes, implies that
their resistance to flow at normal conditions is reduced. However, in vivo flow is
usually characterized by lower velocities and then the fluidity of blood is drastically
dependent on the cell deformability. A potential stiffening of their membrane,
as occurs in several pathologies (including malaria) or during the development of
echinocytes, could lead to severe alterations in their morphology and rheology. For
instance, measurements of the bending rigidity during morphological changes have
reported increases of roughly 2 and 5 times in the bending rigidity of echinocytes
and spherocytes, respectively (Park et al., 2010b). This change could cause the
8.5. Discussion 147
Figure 8.6: (A) Velocity of a RBC flowing in a channel; from top to bottom, in the
discocyte, slipper, and parachute regimes, for Cκ =, respectively. The contribution of
the original Poiseuille in the absence of cell, vz0 , has been substracted, and it has been
normalized with respect to the maximum velocity. (B) Vorticity field, corresponding
to the only non zero component in this geometry, ω = ω0 . The contribution of the
unperturbed Poiseuille has been removed.

η
Figure 8.7: (A) Viscous shears stress, σxz of a RBC flowing in a channel; from top to
bottom, in the discocyte, slipper, and parachute regimes, for Cκ =, respectively. The
0 , has been substracted,
contribution of the original Poiseuille in the absence of cell, σxz
and it has been normalized with respect to its maximum value. (B) Normal contribution
of the membrane force, fnmem = n̂α fαmem , for the different configurations. The force is
normalized with the external forcing f0 that drives the flow.
8.5. Discussion 149
Figure 8.8: Comparison between slippers (left) and parachutes (right) observed in exper-
iments and found in our simulations, for the same confinement a/b = 0.8. Experimental
snapshots reproduced from Tomaioulo et al. (2009).
shift from parachute to slipper or from slipper to discocyte, affecting to the overall
rheology of the suspension. Experimentally, the effect of RBC rigidity has been
studied in the macroscopic rheology of RBC suspensions (Thurston, 1972), but it
has received little attention in the case of single RBCs at high confinement, in spite
of its importance in the development of new devices for cell targeting. Although
the suitable experiments are complex, hopefully future experiments could through
some light into this issue.
The three morphologies exhibited by RBCs when flowing isolated along a thin
channel are also found for higher concentrations, where RBCs are closely placed
each other and they present collective behaviour. McWhirter et al. (2009) studied
the aggregation of several RBCs at similar conditions to the ones used here. They
identified three phases, namely (i) the disordered-discocyte, in which the cells
maintain the discocyte shape with random orientation; (ii) the zigzag slipper,
where cells cluster in two stripes in the center of the channel, accomodating their
shape to the neighbour cells; and (iii) the alligned parachute, formed by an array
of centered parachutes. The random orientation of the cells in the disordered-
discocyte phase, found at low flow velocities, might relate with the dependence we
observe with the initial conditions in the discocyte regime. This disordered phase is
stable for rather high capillaries; this may respond to the screening effect between
cells, which atenuates the effect of the flow perturbation on each cell. The zigzag-
slipper phase, in which the observed slippers present an horizontal inclination, is a
highly stable configuration observed for an extensive range of capillaries, even for
values in which parachutes are also found if the cell-to-cell distance is increased.
One could speculate that, being a lower energy configuration with respect to the
parachute, the collective effect favour a longer life span of the slipper configuration,
whereas when RBCs flow isolated this shape eventually becomes unstable. The
alligned parachute forms at high shear rates and large distances between cells; the
shapes observed are similar to the isolated case, although they are not forced to
deform as much severely.
8.6 Conclusions
An isolated RBC is found to present a non-linear rheological behaviour, charac-
terized by a shear-thinning decay at high shear rates due to a combined effect
of cell deformation and orientation with the flow. For a given flow, the effective
viscosity is entirely determined by the membrane rigidity.
RBCs present three main morphologies, strongly dependent on the rigidity of
their membrane. In our model the rigidity is expressed relative to the incoming
flow. Very rigid cells maintain their relaxed shape, with weak deformations from
the discocyte. For intermediate rigidities the cell acquires an asymmetric slipper
shape, orienting its axis with the flow. For soft cells return to an axial position and
acquire a parachute shape. Why cells develop this sequence and the advantages
of the lateral position of the slipper is discussed in the ensuing Chapter.
Chapter 9
On the mechanisms of RBC

deformation and migration
In Chapter 8 we find that RBCs flowing in confined channels assume asymmet-

ric shapes, called slipper or slipper-bowl, at intermediate values of the capillary
number. However, why RBCs adopt this morphology in detritment of a symmet-
ric shape remains as an intriguing phenomenon, in spite of the numerous studies
that have faced the problem. The lateral position of the cell can be exploited
in microfluidics to separate different type of cells or particles, as we will see in
Chapter 10, and the understanding of the mechanisms driving RBC deformation
is therefore important.
The asymmetric position of RBCs when flowing in symmetric flows, even if

their initial configuration was symmetric, indicates that there exists some kind of
instability. In order to adress this question, we perform simulations comparing the
behaviour of axisymmetric RBCs, which are limited to a centered position in the
channel, with unconstrained RBCS (that can move freely along the channel pro-
file), which adopt the slipper morphology, identifying the benefits and drawbacks
of each configuration.
Axisymmetry is introduced imposing reflective boundary conditions at the

channel axis, and therefore the dynamics of the phase and velocity fields is sym-
metric with respect to the axis. Accordingly, the cell must be initiliazed obeying
this symmetry, with its center of mass strictly located at the axis. The imposed
dynamics ensures that when the external forcing is applied, the cell will remain
centered. By sweeping over a range of capillary number, we compare this configu-
ration with the free cell, imposing that both share the same initial configuration.
We explore the differencess between symmetric and asymmetric cells in the vis-
cosity, the absolute elastic energy and dissipated energy.
151
152 Chapter 9. On the mechanisms of RBC deformation and migration
Figure 9.1: Morphologies of RBCs flowing in a confined channel, a/b = 0.70, for free
cells that can assume asymetric shapes (up) and strictly axisymmetric shapes (bottom).
Each shape correspond to a letter a, b, .., e, from left to right. The respective Cκ for
each configuration is shown in the bottom scale.
For convenience, we fix a system with slightly lower confinement than in Chap-

ter 8, a/b = 0.70, and lower volume fraction. In Figure 9.1 some examples of the
morphologies for each case are shown. The case of the asymmetric RBC is similar
to the results found in Chapter 8: at low Cκ , cells remain centered but already
showing a certain asymmetry. At intermediate Cκ RBCs assume the slipper shape
at a lateral position in the channel, and they orient their axis with the flow. At
high Cκ , the cell returns to the axis and assumes a parachute shape. The sequence
of axisymmetric shapes is, as expected, considerably different. During an exten-
sive range of Cκ , RBCs exhibit a very similar shape, with and increasingly bent
profile for higher values of Cκ , but maintaing both edges of the cell parallel. Only
for Cκ > 50 cells assume parachute shapes, with a cup-like profile in which the
edges of the cell are uncoupled.
9.1 Effective viscosity
The effective viscosity for both the symmetric and asymmetric cells is shown in
Figure 9.2, measured from the expression provided in Chapter 8, (8.2). The shear-
thinning behaviour is recovered in both cases, although asymmetric cells present
9.1. Effective viscosity 153
lower resistance to flow along the entire range of Cκ explored, only converging at
very high capillaries when both configurations are similar. The shear-thinning is
more pronounced in asymmetric RBCs when the cell displaces from the center, at
the beginning of the slipper regime, and reduces slowly when the cell is situated
far from the center. On the contrary, axisymmetric shapes show a smoother decay,
converging to the solvent viscosity only at very high capillaries. Largest deviations
in the effective viscosity are observed at the slipper regime, demonstrating that
the lateral position and orientation of the slipper actually enhance the suspension
fluidity with respect to symmetric configurations. Nonetheless, even at low cap-
illary number there is a considerable difference in the viscosity, suggesting that
asymmetry, even if weak deformations or migrations are involved, is an important
mechanism to decrease the viscosity.
The velocity of the RBC center of mass for each configuration reveals that
asymmetric cells not only enable a higher flow rate of the suspension but also move
faster than centered cells, especially in the advanced slipper regime, as shown in
Figure 9.2 (inset), where we have expressed the velocities in the co-moving frame
of the suspension. This agrees with the well-known fact that particles flowing off-
center lead the flow whereas those situated close to the axis lag the flow (Matas
et al., 2004). At low Cκ , the velocity of both configurations is virtually the same,
suggesting that fast RBC velocities are only achieved when cells are situated far
from the axis.
Figure 9.2: Effective viscosity of a RBC suspension as a function of Cκ for asymmetric

and axisymmetric cells. (Inset) Velocity of the RBC center of mass with respect to the
mean velocity of the whole suspension. The velocity is normalized by v̂z .
Figure 9.3: Bending energy of a RBC as a function of Cκ for axisymmetric and sym-
metric cases, expressed as excess energy from the relaxed discocyte shape.
9.2 Elastic energy
In Chapter 8 we have observed that the lateral position of the cell within the
slipper regime allows the cell to recover a symmetric shape, closer to the resting
discocyte, and this implies a relaxation of the deformation energy stored. In Fig-
ure 9.3 this behaviour is compared with the energy landscape of an axisymmetric
cell. Centered cells bend and deviate from the original discocyte monotonously.
At low Cκ , the deformation energy increases similarly to the asymmetric case,
being slightly lower as a consequence of the penalization to the asymmetry. How-
ever, the energy relaxation found for slippers at intermediate Cκ is not observed
in the axisymmetric case as the centered position forces a bent profile of the cell,
inhibiting any mechanism of shape recovery. Hence, the elastic energy of axisym-
metric cells increases monotonously with Cκ . Only when free cells return to a
center position at high capillary, within the parachute regime, the deformation
energy of both configurations is similar. Asymmetric cells strongly benefit from
their capability to deform and adopt symmetric shapes by a low deformation en-
ergy, in contrast to axisymmetric cells, which are severely penalized by their more
curved shape.
9.3 Energy dissipation
The flow is driven by an external forcing analogous to a pressure gradient. The

energetic cost of maintaining this forcing can be calculated from the Navier-Stokes
equation. At low Re, the equations reads,
9.3. Energy dissipation 155
∂vβ
ρ = ∇α [η∇α vβ + σαβ
mem
] + ρfβext . (9.1)
∂t
Under conditions of mechanical equilibrium the left hand side vanishes, and
after multiplying by the velocity field vβ and integrating over the whole volume,

ρ vβ fβext dV = −η vβ ∇α ∇α vβ dV − vβ ∇α σαβ
mem
dV. (9.2)
The left hand side can be readily identified as the power introduced to the
system which maintains the fluid flow,

Wext = ρ vβ fβext dV, (9.3)
The first term on the right side of (9.2) can be integrated by parts, obtaining
the viscous dissipation,

Wη = η0 (∇α vβ )2 dV. (9.4)
The remaining term can be rewritten in terms of the force exerted by the
membrane, obtaining the elastic power of the membrane,

Wmem = vβ fβmem dV. (9.5)
Hence, equation (9.2) is formulated as
Wext = Wη − Wmem . (9.6)
Note the that we use the term dissipation and power according to the sign of
each term on the left relative to the sign of the external power. However, Wmem
is computed from the force exerted by the membrane, and thus it represents a
reactive term; it is therefore expected to be negative.
The formal equivalence between (9.4) and (9.5) is proven by introducing the
η
hydrodynamic stress tensor σαβ = −η∇α vβ . Although Wmem is calculated from
the integral over the whole domain, fmem vanishes far from the interface. Hence,
whereas Wη and Wext are extensive variables that depend on the size of the sys-
tem, Wmem only depends on the membrane, regardless of the domain size. We
therefore rescale the former with the volume fraction in order to obtain compara-
ble variables. In this Section we analyze how the external energy is distributed in
the asymmetric and axisymmetric case. All the powers are expressed in a dimen-
sionaless form relative to the reference elastic power, given by the ratio between
the energy scale of the system, κ, and the elastic relaxation time of the membrane,
tκ = η0 a3 /κ, leading to Wκ0 = κ2 /η0 a3 .
9.3.1 External power

From the definition of the external power, it is obvious that fluid suspensions with
a higher flow rate require of higher external power than those with lower flow
rates, if the external forcing is maintained constant. As expected, the comparison
between the external power for asymmetric and axisymmetric shapes, Figure 9.4,
shows that high powers are required for asymmetric cells over the entire range
of Cκ , but specifically in the slipper regime when the different behaviour of each
configuration is more accentuated.
In Figure 9.4 B, we test the energy balance of (9.6) in our simulations, compar-
ing the external power with the sum of the contributions to the energy dissipation.
We observe a good agreement between both terms at low capillaries, but a con-
sistent though slight deviation at high Cκ , with an overestimation of the external
power of roughly 7%. We associate this error with the numerical precission due to
the mesh discretization. We check this hypothesis by comparing (9.3) and (9.4)
for a single fluid (in the absence of membrane), in which the newtonian flow is
given by the Poiseuille profile and it is straightforward to prove that both expres-
sions coincide. In this case, we still observe a deviation of 4%, which must be
necessarily due to numerical precission. The difference of 3% in the presence of
the cell might be explained by errors in the estimation of the membrane force.
9.3.2 Elastic membrane power

The elastic power, ploted in Figure 9.5 A, shows a markedly different behaviour
between the asymmetric and axisymmetric RBCs. In both cases the power is
negative, reflecting its reactive nature. At low Cκ , the power is similar in both
configurations and increases in magnitude as the cell bends. For asymmetric
shapes, however, the power peaks in the transition to the slipper at Cκ ∼ 5,
meaning that the relaxation of the elastic energy of the membrane has anologously
associated a strong reduction of the dissipated energy of the membrane. During
the entire slipper regime the elastic energy remains below 5% of the total power.
The evolution of the elastic power for axisymmetric cells is more complex. At low
Cκ , it increases similarly to the asymmetric case, but this increasing continues
up to Cκ ∼ 7, when it peaks and starts to decay towards values similar to that
of the slipper, at high Cκ ∼ 100. The change in the tendency is accompanied
9.3. Energy dissipation 157
Figure 9.4: (A) External power introduced to the system, (9.3), as a function of Cκ ,
for both the asymmetric and axisymmetric case. The power is higher for asymmetric
shapes due to its higher flow rate, and therefore differences are more pronounced at
the slipper regime. (B) Energy balance given by equation (9.6). The external power as
defined in (9.3) is compared to the sum of the contributions of the viscous dissipation
9.4 and the elastic power (9.5). Differences around 7% are observed at high capillary
number, which we associate to numerical precission (see main text).
by a subtle morphological transition. For Cκ < 7, axisymmetric shapes show

weak deformations, bending their profile but maintaining parallel leading and
trailing edges, reminiscent of a boomerang shape. For Cκ > 7, centered cells lose
this symmetry, their edges are no longer parallel and deform into more cup-like
shapes, with a central core at the center region of the disk, eventually assuming
the classic parachute shape. Examples of these morphologies are shown in the
inset of Figure 9.5; cell a, still weakly bent and with parallel edges, shows a
high dissipation. Parachutes b, c, and especially d, with a subtle different profile,
present lower dissipation. Although, a priori, it is difficult to understand how this
change in the morphology triggers the decay in the dissipated elastic energy, the
regime of real parachutes shows a very different behaviour than the boomerang-
like shapes. Indeed, the regime at high Cκ in which centered cells equate their
dissipated energy with that of the slippers, agrees when free cells return to the
center and assume parachute shapes.
The elastic power presents a different behaviour if expressed relative to the

total power, instead of in absolute value, as shown in Figure 9.5 B. In this case,
centered vesicles present a homogeneous decay in the elastic power from the max-
imum value at low Cκ . Asymmetric shapes, on the contrary, present a marked
reduction of the dissipation at low Cκ , and the contribution to the total dissipation
is low (almost negligible) at intermediate and high values of the capillary number.
Figure 9.5: (A) Elastic power as a function of Cκ for asymmetric and axisymmetric
RBCs. Both contributions are negative, reflecting that it represents a reactive term.
For asymetric shapes, the formation of the slipper is associated to a strong reduction of
the elastic power. In the case of axisymmetric cells, the reduction of the power is more
difficult to identify in terms of the cell behaviour, but we find the peak corresponds to
the transition from the boomerang to the real parachute morphologies (as can be seen
in the inset of (B)). (B) Elastic power relative to the total power Wmem + Wη .
9.3.3 Viscous dissipation
The total viscous dissipation corresponds to the remaining term of (9.6), and is
shown in Figure 9.6 A. Dissipation increases homogeneously in both cases, but
the increment is much sharper for asymmetric shapes as a result of the damping
observed in the elastic power. Hence, once the elastic power is minimum, most
of the external energy introduced is dissipated by the fluid. For axisymmetric
shapes, however, the balance is more homogeneous and the viscous dissipation
shows a smoother increase. In addition to the total viscous dissipation, it is
interesting to study where this energy is preferentially dissipated. In Figure 9.6
B, we plot the dissipation in the inner fluid of the cell. Thus, the integral of (9.4)
is reduced to the region {φ > 0}, and expressed relative to Wtot . The results
suggest that at low Cκ dissipation is concentrated in the interior of the cell (note
that the volume fraction is low ∼ 0.08 but the inner dissipation corresponds to a
0.7 of the total). The inner dissipation, however, decreases reducing to 0.1 at high
Cκ . The comparison between symmetric and axisymmetric shapes does not show
substantial differences in their behaviour, suggesting that the distribution of the
viscous dissipation is not a relevant mechanism driving RBC migration towards
asymmetric position.
Figure 9.6: (A) Viscous dissipation as a function of Cκ . (B) Viscous dissipation of the
internal fluid of the cell, relative to the total viscous dissipation.
9.4 Conclusions
There has been an intense debate about the significance and advantages of the
asymmetric shapes and positions that RBCs acquire while flowing in narrow tubes.
Secomb (2011) attributed this phenomenon to the reduction of the flow resistance,
as demonstrated in Figure 9.2. Kaoui et al. (2009), on the contrary, proposed
that the formation of slippers responds to a reduction of the velocity lag between
the cell and the flow. Although we agree with these observations, it is difficult to
recognize them as driving mechanisms of the migration. There is a number of other
aspects, of similar physical meaning, that distinguish the behaviour of symmetric
and slipper shapes. For instance, as shown in Chapter 8, slippers also benefit from
a low vorticity. The formation of vortices can prelude a flow instability, and hence
slippers can be understood as more stable configurations against hydrodynamic
instabilities.
The results presented in this Chapter suggest that the minimization of the
viscous dissipation does not play a key role in driving the migration of cells towards
off-center positions. However, the cell benefits from this lateral position by a
strong relaxation of its deformation energy and the reduction of the elastic power.
Minimization of the elastic stresses induced by the flow deformation seems to be
a strong advantage of the slipper morphology.
Chapter 10
RBC focusing
The manipulation of single cells in microfluific devices requires of a sutile control of

the flow behaviour, as exposed in Chapter 7. Among others, the lateral position of
cells and particles can be exploited to separate objects of different characteristics
(eg shape, rigidity) in pressure-driven flow without the need of any other agent,
such as an electromagnetic field.
The equilibrium position of a particle in the tube depends on several factors
such as flow velocity or confinement (Matas et al., 2004). It is well known that
isolated particles flowing in thick tubes move laterally (ie cross stream migration)
towards a specific off center position, as first observed by Segré and Silberberg
(1962) for rigid spheres in a Poiseuille flow, and extended later on to other objects
(Jeffrey and Pearson, 1965; Han et al., 1999), and different channel geometries
(Tachibana, 1973). This effect, however, is driven by an inertial drift which forces
the object to leave the center (Wang and Skalak, 1969). The dynamics of the
object is not critically determined by its specific properties and the off center fo-
cusing has been also observed for red blood cells and other soft entities (Carlo
et al., 2007). At very narrow microchannels, when inertial effects are negligible,
the absence of this drift should prevent from a lateral migration during the down-
stream evolution, and hard spheres and spherical vesicles does not focalize. At
this regime, however, the deformability and geometry of the object arise as crucial
mechanisms of symmetry breaking (Olla, 1999), and RBCs are known to present
a much more complex behaviour than spherical objects (Abkarian et al., 2008).
Whilst the dynamics of particles and RBCs in the presence of inertia has been
extensively studied in the literature, the limit of low Reynolds number, which re-
quires of a more accurate description of the geometry and elasticity of the object,
has been scarcely tackled in spite of its relevance in microfluidics.
In this Chapter, we explore the effect of channel geometry and flow velocity to
control RBC focusing and lateral position. The formation of regular trains, an in-
161
162 Chapter 10. RBC focusing
teresting problem in microfluidic engineering, is also studied in detail. Finally, the

relevance of the object shape is also analyzed, in order to identify the differences
in the dynamic behaviour between RBCs and other cells or particles.
10.1 Effective viscosity and focusing
Our results show that ηef f depends on the initial condition of the RBC, especially
at low Cκ when the dispersion Δηef f = (ηef f − η̄ef f )/η̄ef f , where η̄ef f is the average
viscosity, can be as high as 25%. The dispersion of the effective viscosity is shown
in Fig 10.1 A, for a number of different configurations. In experiments, the effect of
thermal noise, that allows random exploration of different orientations, may lead to
a more uniform measurement. Therefore, in order to obtain a robust measurement
of the viscosity, we average over a set of 7 different inclinations and 3 different
lateral positions. The dispersion of each configuration is shown in Figure 10.1 B
(inclinations) and C (lateral positions). According to their particular position and
alignment with the flow, RBCs will offer variable resistance to flow. RBCs aligned
parallel to the incoming flow and flowing close to the walls (where the shear stress
decreases) present smaller contribution to the viscosity than those centered and
with normal orientation, which induce a severe perturbation of the surrounding
flow. Remarkably, in this region RBCs oriented symmetrically with respect to
the normal flow direction (eg, orientation of θ = 45◦ and 135◦) show a similar
viscosity. This could suggest that the relevant parameter controlling the value
of the viscosity, within this rigid limit, is the ratio between the section occupied
by the cell with the respect to the total section of the channel. The dispersion
reduces for intermediate Cκ and it is negligible at high values.
The memory of the RBC to its initial configuration can be separated into two
different contributing phenomena: alignment of the RBC with the flow and focus-
ing to a final position. Due to the channel symmetry with respect to its axis, RBCs
focus at two symmetric lateral bands, depending on which channel region was the
cell located with respect to the axis at the initial condition. Initially centered
RBCs migration is determined by numerical noise. Henceforth, we concentrate
in one of the channel regions (ie from the axis to one of the walls), bearing in
mind that symmetric phenomena occur in the opposite region. At very low Cκ ,
RBCs do not orient their axis with the flow, and they flow maintaining the initial
position, without showing migration across streamlines. The increasing external
flow forces the cell to rotate and align with the flow, but still showing a depen-
dence with the initial position along the channel section. The dispersion of the
viscosity decreases as the range of cell inclinations is reduced, but still maintaing
the contribution due to the different position along the tube Further increase in-
10.1. Effective viscosity and focusing 163
Figure 10.1: Effective viscosity of a suspension of RBCs as a function of the capillary

number, Cκ . Examples of the RBC morphology, as well as the steady position and
orientation for different initial configurations, are shown in the snapshots. (A) The
viscosity presents a certain sensitivity to the initial conditions of the RBC, especially
at low capillary numbers. Each dashed line represents the viscosity for a specific ini-
tial configuration. The mean effective viscosity, η̄ef f (red line), is therefore averaged
over the different configurations, covering: (B) initial inclinations (with respect to the
channel axis), with normal RBCs presenting a higher resistance to flow compared to
aligned RBCs; (C) initial heights in the channel, with centered RBCs opposing more
resistance than RBCs flowing close to the walls. The viscosity dispersion is defined
as Δηef f = (ηef f − η̄ef f )/η̄ef f , where ηef f is the effective viscosity for each particular
initial configuration. In all cases, within the slipper regime () the final configuration
is nearly independent from the initial one. The dependence is stronger at intermediate
() and especially low (◦) capillaries. The schemes on the right represent the RBC
inclination θ and center of mass height hcm .
Figure 10.2: (A) Evolution during the downstream flow of the center of mass height,
hcm , of RBCs initially placed at different lateral positions, within the slipper regime,
Cκ = 13.2. RBCs close to the wall are repulsed and migrate towards the center; likewise,
centered RBCs migrate towards the walls. All RBCs stabilize at intermediate positions,
forming a band of stable trajectories. In this case the confinement has been increased
with respect to the standard one, here being a/b = 0.49, in order to observe larger
and more perceptible migrations. (B) Width of the band of stable trajectories as a
function of Cκ . The width of the band can be controlled by increasing the flow velocity
(confinement a/b = 0.71). High velocities induce a narrowing of the band until the RBC
eventually occupies a unique position in the channel. Within the parachute regime,
RBCs develop parachute and slipper morphologies at fixed flow conditions depending
on their initial position: RBCs initially placed close to the wall are repulsed and assume
a slipper shape, whereas cells placed at the channel core deform into a parachute.
duces a migration towards an off center region, then the RBC distribution narrows
forming a thin band. The viscosity dispersion decreases with the focusing and,
for sufficiently high flow velocities, the final position and orientation of the RBC
is unique regardless of its initial condition, leading to a well-defined value of the
effective viscosity.
10.2 RBC focusing and alignment

RBC focusing has been extensively studied in the inertial regime, but it is not well
understood in the viscous one. In this section we study how the focusing of RBCs
to a defined, off center position can be obtained by tuning the capillary number
and the distance between the channel walls.
10.2.1 Focusing
At low Cκ , RBCs flow occupying the entire channel section, without any favoured
position. If the external forcing increases, however, RBCs concentrate on a narrow
10.2. RBC focusing and alignment 165
Figure 10.3: (A) Orientation of the RBC as a function of the capillary number for
different initial conditions, θ0 . Within the discocyte regime, the RBC maintains its
initial orientation, with slight deformation. For Cκ > 2, all RBCs rotate and orient
their axis with the imposed flow profile. (B) Temporal evolution of the inclination of
a RBC at Cκ = 4.5. The rotation towards the equilibrium inclination occurs in a fast
time scale compared to the migration and relaxation time of the cell. In the inset some
snapshots of the evolution are shown.
band off the channel axis, as shown in Figure 10.2. In Figure 10.2A, at Cκ = 10,
the height in the channel during the downstream flow for RBCs placed at differ-
ent initial position is shown. RBCs initially placed close to the wall experience
a repulsion and migrate towards the center, stabilizing at hcm /R = 0.5. RBCs
initially placed in the channel axis also migrate from the center and reach a equi-
librium position at hcm /R = 0.25. The rest of RBCs stabilize at intermediate
positions, forming a band of stable trajectories. The channel width has been in-
creased to a/b = 0.49 to allow larger and more perceptible migrations. The time
scale of the migration is typically half of the deformation time τκ . The increase
in Cκ induces a narrowing of the band of stable trajectories, as shown in Figure
10.2B, and eventually RBCs are found in a unique lateral position regardless of
their initial condition, for Cκ > 40.
10.2.2 Alignment
RBCs at very low capillary Cκ < 2 flow maintaining their initial orientation, as can
be seen in Figure 10.3A. However, beyond the critical value Cκ = 2 RBCs lose this
dependence and rotate towards a fixed orientation, which in turn depends on the
lateral position along the channel. RBCs flowing close to the wall assume slightly
higher inclinations than those more centered, as a result of the non uniform cur-
vature of the flow profile. The focusing of RBCs takes place for higher capillaries.
The sensitivity to the initial conditions of the orientation and the lateral position
is seemingly uncoupled. Accordingly, the dispersion on the final configurations
Figure 10.4: Effect of channel confinement in the RBC behaviour at different Cκ . All the
RBC are initially placed at the channel axis. (A) RBC morphologies found at different
confined channels; cells shown here are initially placed at the channel axis, normal to
the flow direction. The shear rate is maintained constant for each value of Cκ . The
effect of the walls is important to induce lateral migration, as confined RBCs exhibit
the characteristic slipper morphology whereas the RBC flowing at the thicker tube still
retain a discocytic shape. Less confined RBCs, which may present weaker interactions
with the walls, require of higher Cκ to deform and migrate. (B) Effective viscosity ηef f
for a suspension of RBCs as a function of the channel confinement and capillary number,
at constant volume fraction. ηef f , specially at high capillaries, strongly depends on the
confinement. For confined RBCs which have migrated, the viscosity is low and uniform.
However, if the channel is thick enough the RBC does not migrate and its center position
and normal orientation imply a higher viscosity.
can be separated into three different situations: (i) for Cκ < 2, RBCs flow main-
taining their initial distribution and orientation, deforming its shape to the local
flow profile; (ii) for 2 < Cκ < 7, RBCs show a fixed orientation aligned with the
flow, but they still retain their memory to the initial position along the channel
section; and (iii) Cκ > 7 RBCs focalize to two symmetric lateral positions in the
channel, aligned with the flow. The temporal characterization of the rotation of
the RBC towards its stable orientation is shown in Figure 10.3B. RBCs rotate in
a much shorter timescale than the migration, typically ∼ 0.1τκ . In the inset, some
snapshots of the process are depicted.
10.2.3 Effect of walls confinement
The focusing of RBCs can be achieved by both increasing the flow velocity or the
degree of confinement of the channel. By varying the distance between the walls
b, but maintaining a constant shear rate γ̇, we identify how the closer presence
of the walls affects to the RBC morphology. We focus on intermediate and high
10.2. RBC focusing and alignment 167
Figure 10.5: Focusing of RBCs as a function of the wall confinement and the capillary
number. For each Cκ , the up and bottom values correspond, respectively, to the outer-
most (ie closest to the wall) and innermost (closest to the center) equilibrium positions;
all the rest possible positions are found in between these two. The wall confinement
favours the focusing and typically the effect is reduced below a/b < 0.5. The lateral
position of the RBC is found further away from the axis in thicker channels, and in the
limit in which the width of the channel is equal to the cell diameter, RBCs flow nearly
centered. The capillary number also induces an increasing focusing (especially in the
case Cκ = 109.5), though its effect is weaker.
capillaries in which the presence of the walls is more relevant. In Figure10.4A,

a phase diagram for different capillaries and confinements is presented. Mov-
ing along a row of constant capillary number, the increasing confinement induces
higher migrations and lower inclinations. This effect is especially noticeable at in-
termediate capillaries, in which high confined RBCs have migrated and acquired
the slipper morphology whereas less confined RBCs still remain centered and re-
tain a discocyte shape. The relative distance to the wall arises as an important
factor to trigger the migration, and at thick channels the capillary required to ob-
serve slippers and parachutes might be extremely high. Interestingly, the diagram
highlights the similar effect of moving along the vertical and horizontal axis. This
suggests a coupling between the effect of confinement and capillary, implying that
increasing shear rates induce stronger interactions between the cell and the walls,
leading to a larger repulsive force from the axis, and this effect can be emphasized
by a closer distance of the RBC to the walls. The lateral position and focusing
of RBCs for different wall confinements is shown in Figure 10.5. The confinement
positively affects focusing, and for all capillaries Cκ > 20 RBCs are localized at a
defined lateral position if a/b > 0.5. The distance from the axis of this position
increases for thicker channels, from hcm = 0.05R to 0.3R in the range of channels
studied. At the narrowest channels, the RBC is placed close to the center, likely
due to the geometrical constraints but also to the strong wall repulsion. The in-
crease in Cκ reinforces the focusing effects, and especially for Cκ = 109.5 focusing
has extended to thicker channels and RBCs are localized at positions further from
the channel axis. Note, however, that the effect is relatively weak for the two
other cases Cκ = 18.5 and 43.2.
The larger inclination and migration observed in RBCs flowing along the nar-
rowest channels is accompanied by lower values of the effective viscosity, as shown
in Figure10.4 B. In the case of high capillary, the viscosity is similar for all the
channels in which the RBC has been forced to migrate and align. However, if the
confinement is low and the RBC still flows along the centerline, it will displace
a larger amount of external fluid and therefore its viscosity will be also larger.
For decreasing confinements the effective viscosity increases, until it eventually
stabilizes when the wall influence is negligible and the RBC behaviour is similar
to that in an unbounded flow. The confinement necessary to recover this un-
bounded behaviour also depends on the capillary number, as expected given the
existing coupling between confinement and capillary number. At lower values of
the capillary, all the RBCs exhibit a less marked migration and alignment, and
the differences between the viscosity of each initial condition are less accentuated.
10.3. Vesicle shape 169
10.3 Vesicle shape
We explore the relevance of the object geometry by studying the case of three
vesicles with different reduced volume and compare with the RBC case. The
relaxed shapes of the vesicles are obtained from minimizing several initial ellip-
soids with different area an perimeter, νred = 0.97, 0.69, 0.55, 0.48, corresponding
to the equilibrium shapes from a circular to a discocyte shape. This minimiza-
tion is performed in the absence of fluid. The vesicles are then placed in the
channel and the fluid is switched on. Due to its symmetric shape, circular vesi-
cles present less degrees of freedom to adapt and orient with the external flow,
showing slight deformations even at high forcements. They do not migrate out of
the axis, remaining at the center line, but their center of mass moves to a more
forward position at high capillaries, acquiring at intermediate capillaries an al-
most triangular shape which reminds a parachute, the so-called bullet, as shown
in Figure 10.6 A. We check the evolution of a circular vesicle when initially placed
close to the wall, finding that it rapidly migrates towards the center recovering
the symmetric shape, as opposed to the discocytes. Deflated vesicles present an
intermediate behaviour between the circular and the discocyte vesicles. For all
the cases studied the equilibrium position is strictly asymmetric and therefore the
vesicles eventually deform and migrate, but lower values of the capillary require
of more deflated vesicles to acquire asymmetric shapes. The off center position
of objects in the inertial regime does not depend on their specific geometry, and
hard spheres, soft beads and cells behave similarly migrating towards roughly the
same equilibrium position. The repulsive force from the axis does not manifest a
critical dependence on the specific properties of the object. A potential explana-
tion could be that deflated shapes present an asymmetric distribution of normal
stresses, implying a effective repulsive drift, which is on the contrary balanced in
circular vesicles because of their symmetry.
In the rigid limit, discocytes present a considerably higher resistance to flow

than circular vesicles, in spite of its lower volume. This may respond to the
smoother streamlines of the flow when passing round the object, whereas the flat-
ter face of the discocyte impose a sharper change in the flow direction. However,
the capability to deform, migrate and orient implies that discocytes present a
larger gap in the effective viscosity (Figure 10.6 B). The mechanism of migration
and orientation permits a sharp relaxation and at high shear rates the circular
vesicle has higher effective viscosity than the discocyte. All the deflated vesicles
behave very similarly, especially at high capillaries, with a marked shear-thinning
behaviour which increases for lower reduced volumes. Conversely, the viscosity
of the spherical vesicle is rather constant at high capillaries, when the vesicle has
adopted the so-called bullet shape (ie the equivalent of the parachute for spherical
shapes).
Figure 10.6: Effect of the reduced volume on vesicles and RBCs behaviour and suspen-
sion rheology. (A) Vesicle morphologies for different reduced volumes. Circular vesicles,
which are symmetric, do not migrate and remain centered flowing at the channel cen-
ter line. The increasing deflation of the vesicles (ie, lower reduced volumes) induces a
higher degree of migration and asymmetry. Discocytes exhibit the largest degree of mi-
gration towards the wall and orientation with the flow profile. (B) Effective viscosity as
a function of the capillary number for different reduced volumes. The ability to migrate
and orient of deflated vesicles enables them to reduce their resistance to flow, and this
property implies a sharper shear-thinning behaviour than in the case of circular vesicles.
10.4 Discussion and conclusions

The results presented in this Chapter describe the extensive phenomenology of
the circulation of RBCs at microchannels, highlighting the subtle dependence of
the cell dynamics with the flow velocity, wall confinement and intracell distance.
The focusing and alignment of RBCs play an important role in the rheological
behaviour of the suspension. At low Cκ , when RBCs flow with variable orientation
and position, the effective viscosity of the suspension shows high sensitivity to the
particular configuration. RBCs flowing normal to the flow direction oppose higher
resistance than those flowing paralel to the flow. Additionally, RBCs flowing close
to the walls are also characterized by lower contributions to the viscosity. The
dispersion of the effective viscosity may be prevented by an average measure over
RBCs with different positions and orientations, to reproduce a more homogeneous
suspension.
We have verified that the dependence of the RBC morphology to its initial
condition is not due either to a small relaxation process or to an underlying finite
10.4. Discussion and conclusions 171
size correlation between cells. Figure 10.7 shows the effective deformation time,
1
tef f
κ , computed as the effective relaxation time of the membrane energy , as a
function of Cκ . The relaxation time increases when decreasing Cκ , but it is not
diverging in the regime of small Cκ where we observe strong memory effects.
We have carried out simulations 20 times longer than the highest tef κ
f
measured,
thus largely exceeding the expected relaxation time scale, and observed the same
dependence on the RBC initial configuration. Additionally, we have checked that
increasing the channel length from L = 4a to 12a we do not observe any change
in the final RBC morphology at small Cκ , ruling out that the memory reported is
due to finite size effects. Therefore, the origin of the sensitivity of RBCs to their
initial configuration at small Cκ remains unclear. It could be due to the elastic
nature of the RBC and its ability to slightly deform, adapting to the position
dependent flow.
Figure 10.7: Effective deformation time τκef f , measured from the evolution of the de-
formation energy, as a function of Cκ . We consider that the RBC has reached a steady
shape when the energy achieves a value at 1% of the final value. The curve is restricted
to the regime in which RBCs rotate and orient with the flow, not presentig depen-
dence with the initial inclination. The evolution of the typical time scale of evolution is
smooth and unlikely to diverge at lower Cκ . Deviations between the theoretical defor-
mation time τκ and the effective value measured may respond to the definition of the
relevant length of the cell, which we have fixed for simplicity as the cell diameter a.
The focusing of RBCs in two narrow bands at symmetric lateral positions of

the channel is primarily controlled by the capillary number. At low Cκ , RBCs
maintain their initial height in the channel during the flow, and thus cross stream
migration is not observed. Supposing a suspension in which initially RBCs are
1
We have verified that this relaxation time is comparable to the characteristic time in which
the RBC migrates laterally in the channel or reorients, corresponding to the quantities plotted
in Figure 10.3.A and Figure 10.2 B, respectively.
uniformly distributed along the channel section, the downstream evolution does
not change the cell distribution. For Cκ > 10, RBCs exhibit a marked migration
until they reach an equilibrium off center position. In terms of the RBCs spatial
organization, the flow gives rise to a converging distribution of cells, an effect
often known as tubular pinch. The off-center migration of particles and cells in
the inertial regime results from a balance between the wall repulsion and a drift
from the axis towards the wall, as demonstrated analytically for rigid spheres
(Ho and Leal, 1974; Asmolov, 1999). This drift is caused by inertial effects, so
that the physical origin of the migration observed in the viscous regime must
be different, and may be found in the asymmetry and deformable properties of
the object, explaining the diverging behaviour observed for RBCs and spherical
vesicles. These results highlight the relevance of the specific properties of the
deformable object, as opposed to the inertial regime in which RBCs, vesicles and
particles are known to share a similar behaviour (Carlo et al., 2007).
The channel confinement determines the off center equilibrium position of the
RBC within the slipper regime: cells flowing in thick channels (eg a/b ∼ 0.5)
migrate towards a position of ∼ 0.3R, whereas vesicles flowing at the narrowest
channels (eg a/b ∼ 0.9) are constrained to positions closer to the axis ∼ 0.1R.
This lateral position is considerably lower than the classic Segré & Silverberg
inertial result of 0.6R, but the difference with our results might not relate with
the hydrodynamic regime but it is likely explained by the geometric constraints
imposed by the large size of the cell in comparison with the channel width, since
recent experimental results at large confinements in the inertial regime also found
equilibrium positions in the range 0.2 − 0.4R (Carlo et al., 2009).
The wall effect on the RBC is, additionally, coupled to the flow velocity. Less
confined cells require of higher Cκ (even if the shear rate is maintained constant)
to show lateral migration. Therefore, focusing at thick channels is achieved for
increasingly higher Cκ , implying that slipppers and especially parachutes are rarely
found. The coupling between capillary and confinement has been experimentally
observed for single RBCs. Abkarian et al. (2008) present a phase-diagram in which
advanced parachutes are obtained at the highest confinements and flow velocities,
whilst slippers are restricted to lower values of confinement and velocity. For a
fixed thick channel, they also observe more sparse positions of the RBC at low
flow velocities. Their results might be interpreted as a transition from parachutes
to slippers when the channel width is increased up to a/b < 0.4, suggesting that
they are at the high Cκ limit.
Chapter 11
Collective flow of RBCs
Thus far, we have studied the elastic behaviour of a single cell flowing in a very
narrow channel. This system captures the physics of RBCs suspensions at very
confined channels, when cells flow aligned in a single row and the dynamics of
the cell is dominated by its interaction with the wall rather than with its neigh-
bours. This case facilitates the identification of the physics driving the single-cell
behaviour. However, while flowing in thicker channels, where cells typically flow
at higher concentrations, RBCs do interact, and collective effects substantially
change the flow properties. In this Chapter we explore the transition between the
single cell case to situations in which cells present hydrodynamic interactions, al-
though still far from the high concentration limit of 45% hematocrit characteristic
of blood in large arteries, when the bulk properties of blood are also different.
The study is oriented to understand the basic principles that govern important
macroscopic properties of blood flow, such as the Faehreus-Lindqvist effect. The
Chapter is structured in two parts. In the first, we specifically investigate the
relevance of interactions between cells with respect to the interactions with the
walls. We exploit the periodic boundary conditions in the flow direction to study
the case of regular arrays, when the domain length is short and the cell interacts
with its image. Besides, we study the effect on cell deformation when an isolated
group of three closely placed cells are exposed to a parabolic flow. In the second
part we explore different aspects of RBC flow at moderately narrow channels (of
roughly 15 and 30μm width,) when several RBCs are interacting, focusing on the
spatial ordering of cells when both confinement and shear rate are modified, and
the subsequent effects in the suspension rheology. Most results are conducted to-
wards the comparison with the isolated case, in order to separate the physics of
the single cell situation from the regime dominated by collective effects.
Although RBCs are known to aggregate and cluster due to electrochemical
attractions mediated by membrane composition (Merrill et al., 1966), we restrict
173
174 Chapter 11. Collective flow of RBCs
our analysis to purely hydrodynamic interactions. However, a weak short-range

repulsion is included in order to avoid cell fussion, as described in Chapter 6.
Figure 11.1: (A) Flow disruption induced by a RBC in a Poiseuille flow. The colormap
represents the value of the velocity component vz . Cκ affects the flow disturbance that
the cell induces: rigid RBCs (up, Cκ = 0.8) disturb a larger region of the surrounding
flow than flexible RBCs (bottom, Cκ = 15.1). (B) Penetration length, lp of the flow
disruption caused by the RBC for different Cκ . lp is the typical length of the flow
distortion caused by the cell in the surrounding flow with respect to the unperturbed
Poiseuille, but see the main text for a explicit definition. lp decreases rapidly with
Cκ . For Cκ > 10 it falls down to lp < 0.1a, meaning that RBCs induce a limited
perturbation of the flow at intermediate and high capillaries. (C) Penetration distance
lp for closely placed RBCs, as a function of the distance between cells L. Even if the flow
disruption shows a strong decay at high capillaries, RBC behaviour is highly sensitive
to the presence of neighbour cells. The dashed line separates the single-cell behaviour
(when RBCs are able to migrate off center and assume slipper morphologies) from the
train configuration (when they order in regular arrays of centered, symmetric cells). For
Cκ = 19.5, RBCs only recover the single-cell behaviour if the distance between them
is higher than 1.5a (inset). For lower values, RBCs are hydrodynamically coupled and
migration is inhibited. The penetration length lp shows that the coupling between RBCs
is large if the distance is much lower than the length of the channel (wall confinement
here is a/b = 0.71).
11.1 Hydrodynamic interactions between RBCs
11.1.1 Regular arrays

While flowing along confined channels at high concentrations, RBCs often order
in regular trains, as first noted by Gaehtgens et al. (1980). The formation of these
11.1. Hydrodynamic interactions between RBCs 175
ordered configurations is important in the designing of microfluidic devices as it

increases the control of cell manipulation. From the theoretical point of view, the
organization in trains also offers an interesting way to study the hydrodynamic
interactions between neighbouring cells, and how it affects to the RBC dynamics.
To study these trains of RBCs, we take advantage of the periodic boundary condi-
tions in the ẑ direction that, if the length scale of the hydrodynamic interactions
is below the length of the channel Lz , induce a interaction of the cell with its
image. Effectively, the system reproduces a regular array of identical RBCs sep-
arated by a distance L, which agrees with the domain length Lz . In this section
the competition between the wall-induced effects and interactions between RBCs
is analyzed.
The membrane stiffness dictates the flow disruption induced by the RBC. Rigid
cells induce stronger perturbations of the incoming flow than softer ones, as seen

in Figure 11.1 A. We define the flow amplitude A(z) = (an a0n +bn b0n )/ [(a0n )2 +
(b0n )2 ] of the imposed Poseuille, where (an , bn ) and (a0n , b0n ) are the coefficients of
the Fourier decomposition of the actual vz (x, z0 ) and imposed vz0 (x, z0 ) velocity

profiles, respectively, for the first n = 20 modes, v(x, z0 ) = [an sin(2πnx/b) +
bn cos(2πnx/b)]. By fitting the flow amplitude to a Gaussian decay A(z) =
A0 exp(−((z − zcm )/lp )2 ), a typical distance of the flow distortion is obtained,
the penetration length lp . Figure 11.1 B displays the penetration legth for dif-
ferent Cκ , at very long tubes Lz a when RBCs do not interact. The results
show that the flow disruption decay is strong and for Cκ > 10 the flow deviates
from the reference one only at very close distances from the RBC, ∼ 0.1b. This
could suggest that the coupling between RBCs is only relevant if they are placed
extremely close.
However, even if deviations from the imposed flow are small when RBCs are
distant, interactions strengthen for lower distances between cells, favouring RBC
collective behaviour. If RBCs are initially placed very close to each other, even at
high Cκ , they do not migrate towards the walls but flow maintaining a centered
position, forming an ordered array of RBCs aligned normal to the channel axis,
as shown in Figure 11.1 C (inset, RBCs on the right) for Cκ = 19.5. RBCs
bend, coupling their surface to the flow profile. If the distance between cells is
increased, eventually slippers are recovered, initally showing a slight distortion and
then fully exhibiting the single-cell behaviour. The distance necessary to separate
this ordered train configuration from the single-cell behaviour depends on the
capillary number, low Cκ requiring of larger cell-to-cell distances. For Cκ = 19.5,
the RBC behaves as hydrodynamically isolated for lp > 1.4a, but the critical
distance decays to 0.5 for Cκ = 97. The formation of the trains originates on the
transversal confinement that the RBC feels due to the presence of its neighbours.
The flow disturbance generated by the RBC is symmetrically compressed by both
the preceding and the rear cells of the train, inhibiting the symmetry breaking
of the flows generated around the RBC that prelude the migration in the case of
isolated cells. The transversal confinement thereby constraint the RBC shape to
a centered, symmetric morphology.
Figure 11.1 C presents the dependence of the penetration length with the
separation between RBCs L, at wall confinement a/b = 0.71. The results suggest
that the penetration length for isolated RBCs is typically of the order of the
channel width, though it depends on Cκ . For Cκ = 19.5, at distances L much lower
than the channel width the wall effect is subdominant and the RBC behaviour
is controlled by the transversal confinement, forming trains of RBCs. Only for
a/L > 0.66 RBCs behave as hydrodynamically isolated, recovering the single-cell
behaviour, exhibiting lateral migration and assuming slipper shapes.
The formation of trains of RBCs has important implications in the rheology of
the suspension. The normal orientation of the cells implies a larger resistance to
flow than aligned RBCs (ie slippers), and the solvent is repelled from the channel
core towards the walls where it flows free of cells disturbance. This however
implies that the effective viscosity of the suspension increases, as shown in Figure
11.2 A, where the effective viscosity of the suspensions shown in Figure 11.1 C
at constant Cκ is plotted. A sweep in Cκ reveals that the differences in the
viscosity are accentuated at lower values, as a direct consequence of the stronger
interactions between more rigid cells. This effect is shown in Figure 11.2 B, where
the effective viscosity as a function of Cκ is shown for three different distances
between RBCs. At high Cκ , the distance between RBCs in all cases is larger
than the critical distance of the single-cell regime. For lower values of Cκ , the
differences of the viscosity is primarily due to the different volume fraction. At
some point, marked by the dashed line, the distance between RBCs is lower than
the increasing critical length and RBCs switch to the train configuration, inducing
a considerable increase of the effective viscosity. Accordingly, the presence of
collective behaviour between RBCs critically determines both the morphological
and rheological behaviour of the suspension.
11.1.2 Screening between cells

The study of the cell morphologies in ordered trains demonstrates the relevance
of the screening between RBCs. The close distance between cells disrupts the
Poiseuille flow in the entire domain, and thereby cells interact with an attenuated
flow and they are not forced to migrate. An alternative case of interest is to
consider a channel in which a few RBCs are placed close enough to ensure that
they will interact, but at the same time the tube is sufficiently large so that the
Poiseuille flow is recovered. The analysis of the morphologies adopted by the RBCs
11.1. Hydrodynamic interactions between RBCs 177
Figure 11.2: (A) Effective viscosity for a regular array as a function of the intracell
distance, a/L, at constant capillary number Cκ = 18. Each value of the viscosity cor-
responds here to the shapes shown in Figure 10.2, for initially centered RBCs. The
increase in the viscosity can be decomposed into two different contributions: (i) the
change in the volume fraction, from 0.40 to 0.13, (ii) the effect of the interactions be-
tween RBCs at low L, which allow the formation of train of cells at the channel core that
present higher flow resistance. (B) Effective viscosity as a function of Cκ for different
cell-to-cell distances. If the channel is not sufficiently long, at some point, marked with
a dashed line, the penetration length lp exceeds the distance between cells L and the
RBC interact with its image (forming an array of RBCs due to the periodic boundary
conditions) and this fact triggers a considerable increase in the effective viscosity. The
change between the single-cell and train like configurations is marked by a dashed line,
and the specific cell shapes are shown in the inset. At low distances between RBCs,
they maintain a centered position but bending their surface. Only for large distances
between cells slippers are observed, RBC on the left.
will provide information about the screening effect between them in response to
the external, unperturbed, flow. We fix the initial configuration as a group of
three aligned RBCs with orientation θ0 = 45◦ , separated by a distance d. At the
imposed flow conditions, an isolated RBC shows the usual slipper-bowl shape.
Examples of steady state conditions are shown in Figure 11.3, for d/a = 0.45 and
0.95, and capillary numbers Cκ = 16 and 31. Cells flow maintaining a nearly
constant distance, but their morphological response differs from that of isolated
cells.
The results show that only the leading cell (in the flow direction) deforms
into a slipper, as it interacts with a pure Poiseuille flow. The two following cells
present a different deformation, more curved and distorted. The interpretation
of these shapes is not straightforward, as they could represent both advanced or
retarded morphologies with respect to the slipper (ie parachutes or discocytes,
respectively), depending on which cell is undergoing a stronger deformation due
to the flow. However, Figure 11.3 d offers interesting information, as large d
and Cκ correspond to the single-cell limit, and all the cells indeed show slipper
shapes. Hence, in the other plots screening might be larger, and cells interact
with a disrupt flow which present a lower effective capillary number. Thus, one
could hypothesize that cell morphologies subjected to the screening of a neigh-
bour cell represent retarded stages of the morphological sequence of Figure 11.3.
Accordingly, if more RBCs are added to the trailing end of the group, they will
sequentially deform into shapes closer to the discocyte (ie less distorted). In the
limit of infinite number of cells, all of them are expected to show only slightly
distorted shapes, and this is the idea behind the results presented in the previous
section.
The morphologies depicted in Figure 11.3 intuitively show that the capillary
number reduces the screening effect. Whilst at the lower Cκ the two non-leading
cells are highly deformed, at Cκ = 31 the flow-induced deformation is more dom-
inant and the three RBCs adopt a slipper morphology (d) or only the trailing
one is deformed for closer cells (c). The distance between cells is obviously also
important, as the screening range is related with the penetration length, and cells
separated a larger distance will be exposed to a lower screening. This effect is
quantified in Figure 11.4, where the shape distance (as defined in section 8.2) is
shown for different distances and Cκ . We only consider the morphological defor-
mation of the trailing cell. Note that shape distance for a typical slipper is roughly
Δs ∼ 0.2. The results show that at low Cκ , RBCs are remarkably distorted (far
from the slipper value characteristic of the single-cell regime) with a weak attenu-
ation with the separation between cells, suggesting that screening is present even
for cells separated a distance ∼ a, and thus interactions between cells are domi-
nant. At higher Cκ , on the contrary, there is a strong decay of the shape distance,
11.2. Several RBCs in flow 179
Figure 11.3: RBCs flowing in a channel at confinement a/b = 0.71. In all cases three cells
where initially placed at the channel axis with a distance d between cells and orientation
θ0 = 45◦ . Both the distance and the capillary number minimize the interactions between
cells, and in (d) all the cells develop a slipper morphology, converging to the single-cell
limit. In b, the screening is of the leading cell is low and the middle cell shows and
intermediate shape, whereas the trailing cell benefits from the combined screening and
it assumes a shape similar to those found at lower Cκ .
converging to the isolated slipper value. This transformation suggests that at this
capillary, for separation between cells ∼ 0.5a, wall effect is still subdominant with
respect to interactions between cells, but for separations ∼ 1a the wall influence
completely overcomes the hydrodynamic interactions between cells.
11.2 Several RBCs in flow
The analysis performed in the previous sections highlights the sensitivity of cell
behaviour to hydrodynamic interactions with other cells, and the competition
between these interactions and the wall effects dictates a different RBC flow prop-
erties when several cells are flowing at high and moderate concentration. For
instance, in the inertial regime, the limit of single-cell behaviour is characterized
by the Segré-Silverberg effect, when cells migrate towards a specific lateral posi-
tion, whereas at higher concentrations the collective behaviour dominates and cells
are located at the tube core, the Faehereus-Lindqvist effect. The transition be-
tween both regimes is not well understood, in spite of its relevance for chip-design
as cell focusing will be severely affected. In the subsequent sections we explore
the interactions between RBCs in parabolic flow at relatively low concentrations,
identifying the new mechanisms introduced that explain the main differences with
the single-cell case in terms of spatial ordering and focusing.
Figure 11.4: Shape distance of the trailing RBC (the left one of the group in Figure
11.3) as a function of the distance between cells. Whereas for Cκ = 16 results suggest
that screening is relevant in all the cases, for Cκ = 32 there is a clear switch from the
screened to single-cell configurations, indicated that wall effect has became dominant
for distances ∼ a.
We carry out simulations with the same parameters as in Chapters 8, 9, and

10. Several RBCs (ranging from 3 to 12 cells, depending on the simulations) are
initiliazed in the domain, with different configurations (eg all the cells with the
same orientations, disordered center positions, etc) as discussed below. We explore
the same range of capillaries as the previous section, but extend the confinement
including thicker tubes.
11.2.1 Order vs disordered configurations
Our simulations show that RBCs in flow, especially at low capillaries, depend on
the initial condition, especially if this corresponds to a very symmetric configura-
tion. This fact might be expected according to the results of section 10.1, and it
has been reported elsewhere in the literature (McWhirter et al., 2008). Examples
of a very symmetric and a disordered configurations are given in Figure 11.5. We
quantify the equivalence between configurations by means of the pairwise corre-
lation function and the orientation of the cells. The pairwise correlation function
in the x-direction, G(dx ), for n RBCs, reads
Figure 11.5: (Top) Initial condition with an ordered configuration, at a/b = 0.53, for
n = 8 RBCs. (Bottom) Initial condition with a disordered configuration, at a/b = 0.53,
for n = 8 RBCs. RBCs are placed at random position and orientation within a local
domain.

1
G(dx ) = δ(dx − di,j
cm ) . (11.1)
nρ i j
where ρ = (n − 1)/Lx, and dx expresses distances in the x-direction, so that

cm = xcm − xcm is the projection onto x of the distance between the center of
di,j i j
mass of cells i and j, and δ(x) is a function that takes the value δ(x) = 1/Δx
if x ∈ [−Δx/2, Δx/2], and 0 otherwise. We fix Δx = 0.05a. This function
provides information about the structuring and spatial organization of the cells in
the channel section.
In Figure 11.6 A we compare both quantities for the symmetric and two dis-
ordered configurations. The pairwise correlation function shows a significant dif-
ferent behaviour between both cases: the symmetric configuration orders in two
homogeneous rows, and the peaks of the correlation function are well defined at
distance 0 (corresponding to the cells of the same row) and ∼ 1.1a (correspond-
ing to the axisymmetric row at the opposite side of the channel). For disordered
configurations, however, the profile is not as well defined and presents a third
peak at an intermediate position, corresponding to cells at the axis or with higher
inclination, which are therefore off the row and contribute to the correlation func-
tion at intermediate distances. The curves are, however, considerably similar
between both disordered configurations. In fact, in our simulations we observe
that an ordered initial configuration in which cells are initiallized at three dif-
ferent heights is enough to break the symmetry, and results are consistent with
more disordered configurations. The ordered configuration is particular because
the system presents a long-term memory at high confinement, and therefore it is
not representative of the general behaviour of other configurations. Accordingly,
one must be concerned of avoiding initially symmetric configurations, but results
are expected to be homogeneous for any other initial configuration.
The cell orientation shows similar values, as shown in Figure 11.6 B. The
orientation dispersion is defined as,
1
N
2
< (Δθ) >= (θi − < θ >)2 . (11.2)
N i=1
where N is the number of cells in the domain. The orientation dispersion is

negligible in the case of ordered configurations, because the symmetry implies
that all cells behave very similarly. Disordered cells, as expected, present a much
higher dispersion which reduces with Cκ with the gradual alignment of the cells.
Both disordered configurations present similar values of the dispersion. The mean
orientation (inset) shows that the value of the mean orientation is more similar
between disordered and ordered configurations than the dispersion, though still
the symmetric configuration presents a lower orientation as a result of the perfect
alignment of all the cells, whereas disordered configurations are characterized by
higher absolute value as well as dispersion in the measure of the orientation, as a
result of the limited ordering that the flow induces.
11.2.2 Migration and orientation in ordered configurations
Although ordered configurations, such as the one shown in Figure 11.5, represent
just a particular and very specific case of RBC circulation, their homogeneity can
be exploited to understand some basic mechanisms introduced by interactions
between RBCs. We study the mean orientation and lateral position of the cells
in ordered configurations as a function of the capillary number and separation
between cells, as plotted in Figure 11.7 A, where the value in the single-cell limit
is also provided for comparison. RBC orientation presents two different behaviours
depending on the capillary. At Cκ = 12.3, RBCs orient their axis with the flow,
assuming planar slippers. Depending on the distance between cells, they are free to
completely orient with the profile (in the single-cell limit), or they present higher
orientations if the spatial constraints do not permit a complete alignment. For
higher values of Cκ = 30.8 and 43.40, RBCs acquire a very horizontal inclination
(around 15◦ ), and assume a more curved shape, coupling their profile.
Figure 11.6: (A) Steady pairwise correlation function, G(dx ), for an ordered and two
disordered initial configurations, at Cκ = 30.5. The ordered initial condition forces a
symmetric ordering of the cells during flow. Initially disordered configurations show
as more homogeneous profile, with the development of a third peak at a intermediate
position. (B) Dispersion in RBC orientation < (Δθ)2 > (see main text) for both ordered
and disordered configurations as a function of Cκ . The dispersion is negligible for
ordered configurations, as all cells orient simultaneously showing the same behaviour; as
expected, disordered orientations present a much higher dispersion, though it reduces for
high Cκ when a certain alignment is induced. (Inset) Mean value of the RBC orientation,
averaged over the whole suspension. Although ordered and disordered configurations
still differ, the measure is mean value is more similar between both cases than the
dispersion.
Figure 11.7: (A) Mean orientation of RBCs in an ordered configuration as a function of

the distance d in the z-direction. At low Cκ , RBCs show a higher orientation when are
placed closed, but the orientation is nearly constant, regardles of the intercell distance,
for higher capillary number. Dashed lines represent the single-cell behaviour. (B) Mean
position of the center of mass of RBCs in an ordered configuration, as a function of the
distance d in the z-direction. Although the lateral position in the channel is not affected
for d/a ∼ 1, at closer distances RBCs experience a substantial repulsion and they are
found closer to the walls.
This configuration is similar to the slipper-zigzag described in McWhirter et al.

(2008). The orientation of the cells is nearly independent of the distance between
them. This zig-zag configuration is stable over an extensive range of Cκ , and at
the conditions described we did not found parachutes even for Cκ = 120. Only at
configurations of larger confinement, a/b > 0.65, and relatively symmetric initial
conditions (in which the center of mass of the RBCs is placely at similar heights),
parachute trains are observed. If these conditions are not fullfilled, RBCs prefer
to order in the zig-zag phase. Remarkably, both configurations (slippers zig-zag
and trains of parachutes) are observed for the same flow conditions, depending on
the symmetry of the initial condition.
The mean lateral position of the RBCs, Figure 11.7 B, shows that the closer
presence of neighbour cells induces a repulsion towards the walls, and RBCs are
located considerably outer than at sinle-cell conditions. For instance, at Cκ = 30.8,
the lateral position displaces from 0.1 to 0.27.
11.2.3 Spatial ordering

The spatial organization of RBCs during downstream flow shows several differ-
ences with respect to the single-cell behaviour. We explore the effect of confine-
ment by comparing two channels, of confinement a/b = 0.53 and a/b = 0.40.
Note that they correspond to widths slightly lower than 2a and 3a. The initial
condition was a disordered configuration in which cells are initialized at three
different channel heights (one at the axis and two symmetric lateral positions).
Examples of typical steady configurations are shown in Figures 11.8 and 11.9, for
distances between cells d/a = 0.54 and d/a = 0.735. In both cases, cells show
a considerable horizontal inclination, although at the closer distance the geomet-
rical constraints do not permit a perfect alignment, as already described in the
previous section. RBCs organize forming a well defined structure, with two rows
formed in the narrower channel and three rows in the broader one, in spite of
the three-row structure of the initial condition in both cases. The presence of
cells at the channel core is interesting, as in the single-cell regime this position is
unstable. However, cells that during the downstream flow are located in the in-
termediate region cannot migrate outwards due to the presence of other cells, and
they remain trapped close to the axis. The analysis of the lateral pairwise corre-
lation function demonstrates that both channels present a different organization,
as shown in Figure 11.10 (only shown for the case d/a = 0.40). For a/b = 0.53,
the system presents two well defined peaks at dx = 0 and dx = 1.2a, confirming
the formation of two rows. For a/b = 0.40, however, the profile of the correlation
function suggests the formation of four stable positions, the two expected external
rows as well as two intermediate positions which are symmetric with respect to
the axis. Accordingly, RBCs trapped in the core region also seem to avoid the
axis. The position of these intermediate rows is slightly closer to the outer rows
than to the axis. We have also analyzed the pairwise correlation function in the
transversal direction z, G(dz ) (not shown here). The profile of the correlation
function is relatively homogeneous, suggesting that transverse ordering is weak or
subdominant with respect to the wall-induced lateral organization.
11.2.4 Focusing
The focusing and alignment of cells have been studied for a single cell through-
out the Part III of this Thesis, showing that an increase of the capillary number
provides a more defined localization of the particle lateral distribution. This ob-
servation is due to a stronger repulsion from the wall and off-center drift, leading
to a fine band of equilibrium lateral positions. In addition, in previous sections
the different dynamics of RBCs at higher concentration has been described. Con-
cerning spatial organization, two phenomena have been described: (i) RBCs can
be located at the core axis, showing a more regular distribution of cells along the
channel profile; and (ii) interaction between cells forces lateral positions closer
to the wall than in the case of isolated cells. The competition between these
Figure 11.8: RBC during flow in a channel of confinement a/b = 0.53, at Ck = 41, for
distances d/a = 0.54, (top) and d/a = 0.735 (bottom). n=10 RBCs are simulated in
each domain with periodic boundary conditions in the flow direction. RBCs are more
disordered (ie showing more variation in orientation and position) for lower intercell
distants, as the result of less degrees of freedom to orient and accomodate.
Figure 11.9: RBC during flow in a channel of confinement a/b = 0.40, at Ck = 40.8,
for distances d/a = 0.54 (top) and d/a = 0.735 (bottom). n=10 RBCs are simulated in
each domain with periodic boundary conditions in the flow direction.
Figure 11.10: Pairwise correlation function G(dx ) for RBCs flowing at channels at con-
finement a/b = 0.53 (A) and a/b = 0.40 (B), for Ck = 40.8 and distance between cells
d/a = 0.735. The correlation function clearly describes the organization in two rows in
the narrow channel, whereas in the broad channel intermediate positions at the channel
core are permitted, although the channel axis is avoided.
mechanisms will dictate the overall organization of RBCs in channel flow.

Figure 11.11 shows three steady configurations of RBC flow at confinement
a/b = 0.40 for three increasing capillary numbers, Cκ =25.3, 30.4, and 40.8. For
the lower capillary, RBCs are found with low but not strictly horizontal orien-
tations, and distributed along the channel. For the higher values of Cκ RBCs
flow with its axis oriented with the channel and more ordered in three different
positions.
Figure 11.12 A compares the focusing effect for single-cell and several cells.
Isolated cells focus to two lateral positions, symmetric with respect to the axis.
The focusing results from the balance between the increasingly stronger repulsion
from the wall and drift off the axis. Cells interacting with neighbours are found at
the core channel and also at regions close to the wall which are forbidden in the
previous case. This change is understood from the repulsion between cells, given
that some of them are forced to remain in the center by the outer ones, which in
turn are pushed towards the center by the wall drift; at the same time, cells from
the center force the outer cells towards the wall, a drift not present in the single-
cell case. The balance of these interactions explains the RBC distribution and
the different behaviour from the single-cell case. If Cκ is increased, the stronger
repulsion from the wall forces the outer cells towards more centered positions, and
hence the width of the RBC distribution narrows. A band of fluid free of cells is
formed close to the walls.
Figure 11.11: RBC during flow in a channel of confinement a/b = 0.40, intercell distance
d/a = 0.75, and Cκ =25.3, 30.4, and 40.8, from top to bottom, respectively. n=10 RBCs
are simulated in each domain with periodic boundary conditions in the flow direction.
Figure 11.12 B shows that the presence of neighbour cells also induces a more
horizontal inclination of the cells compared to the isolated case. This effect is
difficult to explain and further research is required to understand this more pro-
nounced orientation. However, a plausible hypothesis is that the lateral interac-
tions between cells promote the ordering in parallel rows, and thus this horizontal
inclination is governed by the lateral confinements induced between the cells rows.
Compare these configurations with that shown in Figure 7.3 D, where cells are also
found to flow downstream with an horizontal inclination located at the channel
core.
11.2.5 Rheology
The dynamics of several RBCs at moderate concentrations have proven to differ

in several aspects from the single-cell case, and this should affect the rheological
behaviour of the suspension. We compute the effective viscosity for three con-
figurations (one ordered and two disordered), at volume fraction φv = 0.14 and
Figure 11.12: (A) Focusing of RBCs in a channel of confinement a/b = 0.40, intercell
distance d/a = 0.75, and Cκ = 40.8. The shading areas represent the permitted tra-
jectories in the case of single-cell (dark shading) and when collective effects are present
(light shading). In the case of several cells, each point represents the outermost position
for the n=10 cells in the domain. The increasing Cκ induces the focusing towards more
centered positions, giving rise to the formation of a layer free of cells close to the walls.
(B) Mean orientation of RBCs at the flow conditions described for (A). Cells flowing at
higher concentrations exhibit more horizontal inclinations that those flowin isolated.
confinement a/b = 0.45. The viscosity measures, shown in Figure 11.13, are sim-
ilar for the three cases. The viscosity curves recover the expected shear-thinning
behaviour, though two main differences are found with respect to the single-cell
case: the magnitude of the effective viscosity obtained and the Cκ required to
observe the shear-thinnig decay.
On one hand, the effective viscosity value obtained is larger than that at the
single-cell case, and several contributions are expected to explain this difference.
The presence of several cells, at least at low Cκ , leads reduces the alignment
capability of the cells, and therefore they oppose higher resistance to flow than
aligned isolated cells. Additionally, the higher concentration may play a relevant
role here.
On the other hand, the shear-thinning decay is found at considerable larger Cκ
than in the isolated case. This suggests that the relevant mechanisms that induce
the viscosity decay, such as lateral ordering and orientation, require of higher flow
velocities than in the single-cell case. The presence of neighbour cells reduces the
alignment with the flow, as shown in eg (Figure 11.11) at Cκ = 25, where cells are
found with a diversity of orientations. At this Cκ , a single cell flows completely
aligned, as can be checked in Figure 10.3. The capillary required to align all the
cells in the conditions of Figure 11.11 is ∼ 50, indicating that the stronger flow
Figure 11.13: Effective viscosity for a RBC suspension at concentration φv = 0.14 and
confinement a/b = 0.45, as a function of the capillary number. Three initial conditions,
one ordered and two disordered, are calculated, obtaining similar results. The curve
shows the expected shear-thinning behaviour.
perturbation when severall cells are present attenuates the alignment induced by
the flow. Hence, only high flow rates are able to induce complete alignment and
the viscosity decay is drifted towards higher Cκ .
11.3 Discussion and conclusions

Hydrodynamic interactions between cells modify several aspects of RBC flow in
confined channels. Even when just a few cells are flowing, if they are closely placed,
the flow disruption introduced by the leading cell implies a different morphological
response of the rest of the cells of the group, and this screening is critically deter-
mined by the distance between cells (an increase of 0.5a in the distance between
cells is enough to switch from the single-cell to the interacting regimes) and the
capillary number (for higher Cκ the screening effect is subdominant as the flow
disruption of the cells is attenuated).
The study of regular arrays of RBCs, when a single row of flowing cells is
formed, allows to compare the influence of the wall and the neighbouring cells in
a controlled manner. Results show that at intermediate Cκ , at the slipper regime,
the typical distance for which RBCs behave as hydrodynamically isolated is of the
order of the channel width, whereas for distances below this threshold value the
characteristic migration and asymmetric shapes are not observed and cells prefer
to order in the center of the channel.
11.3. Discussion and conclusions 191
For thicker channels, several cells can occupy the channel section and thus
lateral interactions between cells also compete with the wall. We study the flow
of RBCs at moderate concentrations ∼ 10 − 20%. At low Cκ , RBCs flow with
a disordered distribution, showing a variety of orientations. Further increasing
Cκ cells orient and flow with very aligned, horizontal inclinations. We observe
that parachute shapes only develop in very confined channels and with initial
conditions in which cells are situated at similar heights. Otherwise, cells prefer
to order in two rows of slippers that is known in the literature as slipper zig-zag.
The collective disruption of the incoming flow allows cells to retain slipper shapes
even at extremely high capillaries, in detritment of parachute deformations which
are penalized by higher elastic energies.
The channel width determines the lateral ordering of the RBCs. In narrow
channels ∼ 2a RBCs order in two symmetric rows, similar to the single-cell be-
haviour, although the repulsion between cells typically force an outer lateral po-
sition of the cell center of mass. In channels of width ∼ 3a, RBCs order in 3-4
rows; some cells are trapped at the channel core, repulsed towards the axis by
those flowing along outer streamlines. Therefore, at these conditions, RBCs can
be found flowing close to the channel axis, and the cell distribution is more homo-
geneous along the channel section than in the low concentration regime. Lateral
focusing is hence lose, although high Cκ have been shown to nduce a higher repul-
sion from the wall, so that the outer position of the cell displaces inwards. The
width of the band of cell distribution is then given by the competition betwen the
wall repulsion, which drifts cells towards the axis, and the repulsion between the
cells in the core and those flowing at the lateral. This principle likely represents
the basis of the Faehreus-Lindqvist effect, and further increase of Cκ will lead to
a cell focusing in the channel central region.
The transition from the single-cell to the collective behaviour is found here
at volume fractions ∼ 0.1, in contrast to the results at thicker tubes where it is
typically found at ∼ 0.2 (Han et al., 1999), but the much higher Reynolds and
capillary numbers might attenuate the interactions between RBCs, implying that
collective effects become dominant for closer distances between cells.
The different elastic and flow behaviour of RBCs at the collective regime has
important consequences in the rheology of the suspension. In addition to the
expected increase of the effective viscosity as a consequence of the higher volume
fraction, the shear-thinning decay requires of higher flow velocities. The collective
disruption and retarded alignment and ordering of the cells at these conditions
might be fundamental to explain this drift in the viscosity curve.
Part V
Conclusions
193
Chapter 12
Conclusions and future

perspectives
12.1 Conclusions
Understanding the elastic response of RBCs under certain situations that they
experience during their functional live span is the guiding thread throughout this
Thesis. From a theoretical perspective, and based on both analytical and nu-
merical calculations, we have studied the role of membrane microstructure (ie
the internal balance between the lipid bilayer and the spectrin cytoskeleton) in
RBC shape control. Besides, we have developed a phase-field model for membrane
modeling, accounting for both the membrane elasticity and hydrodynamics of the
surrounding fluid, and explaining in detail its derivation from the classic theory
of elasticity of membranes. This model can be applied to a number of problems
related with membrane elasticity. In particular, we have made use of this model
to study the deformability of RBCs in channel flow, and the importance of RBC
elasticity in the rheological properties of blood.
In Part I, we have established the biophysical framework on which this Thesis
is based on. Due to their simplicity an anucleated nature, RBCs have been the
scope of most studies about membrane elasticity. Appart from their intrinsic
interest in the field of membranes, RBCs are also the main component of blood,
and the understanding of their deformability and flow behaviour is fundamental
in different biomedical areas. Our research is based on the Helfrich theory of
membranes, which assumes that the membrane elasticity is characterized by its
resistance to bend. Although the Helfrich model is the reference theory within the
membrane field, it fails to explain some phenomena observed in the experiments,
such as the formation of echinocytes. The model has been extended to account for
the effect of the coupling between the leaflets of the bilayer, in the so-called ADE
195
196 Chapter 12. Conclusions and future perspectives
model, and the elastic contribution of the underlaying cytoskeleton. Nevertheless,

the importance of each membrane component in the macroscopic response of the
cell is still an open subject of active debate.
Membrane elasticity and the disco-echinocyte transition
In Part II we concentrate on the disco-echinocyte morphological transition that

RBCs undergo when exposed to certain agents, such as incubation in the presence
of amphiphiles, or ATP depletion. In these conditions, RBCs lose their healthy
biconcave shape and develop numerous spicules and bumps along their contour,
becoming more spherical. The observed altered shapes result from the imbalance
of the conformational structure of the membrane, and for instance the formation
of bumps is driven by the expansion of the outer leaflet with respect to the inner
one.
We have carried out a theoretical study in which we consider the elastic con-
tributions of the lipid bilayer and the cytoskeleton, which allows us to identify
the role of each microstructure in the disco-echinocyte transition. We consider
a gradual increase of the internal asymmetry of the membrane, and by means of
numerical calculations we obtain the minimal shape for each membrane asymme-
try. We can then specify the equilibrium shapes of the RBCs as a function of the
membrane structure. The study is based on a Cassini oval parametrization of the
cell profile.
Our results show that the expansion of the outer leaflet triggers the devel-
opment of bumps, since higher membrane curvatures allow the relaxation of the
bilayer energetic storage. In the absence of the cytoskeleton, the discocyte de-
sestabilizes for minute deviations of the membrane asymmetry. The cytoskeleton
presents a strong resistance to the formation of bumps, as it is severely deformed
and its elastic energy increases sharply. The balance between both contributions
dictates the development of undulations and bumps, as well as the specific shape
of these structures. The results provide an energetic scale of the shape transi-
tion, showing that the separation between each morphology is of the order of
hundreds of kB T , ensuring a high stability of the altered shapes. The study also
highlights the importance of the relaxed state of the cytoskeleton. We have com-
pared the area expansion required to induce each morphological stage from our
results (assuming a discocytic shape) with previous from the literature (in which
a elliptic relaxed shape is considered). The discocytic conformation leads to a
gradual transition, with crenated shapes developed for a homogeneous increase
of the asymmetry. For elliptic relaxed shapes, the transition is sharper, and a
considerable increase of the asymmetry is necessary to desestabilize the discocyte
whereas the development of spicules require of minute changes. We can state that
the main conclusions of this Part are:
• The interplay between the bilayer and the cytoskeleton determines the stable
shape of the RBC. When subjected to a expansion of the outer leaflet, the
bilayer tends to curve and develop bumps, and the cytoskeleton opposes
resistance to these deformations, preserving compact shapes.
• The cytoskeleton is a key ingredient in the stabilization of the discocyte

structure against changes in the membrane internal asymmetry, even if dis-
cocytic relaxed shapes are considered. In the absence of the cytoskeleton
the discocyte is unstable under very small increases of the asymmetry.
• The morphological transition has a marked hierarchy, with a considerable

energetic separation between the shapes sequentially found. In this picture,
the discocyte represents a ground state of the morphologies ensemble.
• The quantitative comparison of the development of the different morpholo-

gies with respect to the area expansion supports the hypothesis of a disco-
cytic relaxed shape of the cytoskeleton, in detritment of an elliptic reference
conformation.
The theoretical study allows us to explain the experiments of AFM-induced

shape transformations carried out by Kathryn A. Melzak and José Luis Toca-
Herrera. A series of experiments shows that echinocytes can be perturbed by a
AFM tip, and the cells deform into less crenated shapes or discocytes. The method
allows to control and modify the shape of altered cells at the single-cell scale. Sup-
ported by our theoretical results, we hypothesize that the AFM tip punctures the
lipid bilayer, breaking locally the hydrophobic potential around the tip perime-
ter and hence permitting the rearrangement of lipids from the inner to the outer
leaflet, relaxing the shape stress energy towards lower-energy configurations.
Membrane phase-field model
Part III is devoted to the derivation of a phase-field model for membrane mod-
eling. Phase-field methods have been widely used to the study of interfaces, an
their robust physical basis has served to explain a wealth of dynamic instabil-
ities. The classic framework focuses on tension-driven interfaces, in which the
only elastic contribution is the surface tension. The modelization of more com-
plex interfaces, such as membranes, in which different elastic effects potentially
contribute, requires of further and detailed analysis.
We have first connected phase field models with the theory of elasticity, deriv-
ing the general expressions for the chemical potential and stress tensor in terms
of the order parameter. We have identified the elastic modulii of the interface
in terms coefficients of the phase-field energy and equilibrium profile of the order
parameter, showing that the elastic parameters are sequentially found as moments
of the lateral stress profile, in accordance with the Helfrich theory. These calcu-
lations allow us to consider a specific phase-field model which indeed captures
the physics of a membrane, with vanishing surface tensions and thus obtaining a
bending-driven interface.
We have incorporated the phase-field model to a dynamic formalism. We
consider a diffusive dynamics in the form of a Cahn-Hilliard equation. We have
extended this theory to account for the hydrodynamics of the surrounding fluid,
presenting a complete phase-field Navier-Stokes model. The equation describing
the membrane dynamics is coupled to the Navier-Stokes equation. Besides, we
have performed a sharp-interface limit analysis which provides us the macroscopic
equations of the phase-field model. The hydrodynamics has not been considered
in this study. On one hand, we recover the well-known Ou Zhang-Helfrich equi-
librium equation for a symmetric membrane. On the other hand, we obtain the
complete macroscopic dynamic model. From the set of equations of obtained,
we have studied the relaxation of a flat membrane subjected to a sinusoidal per-
turbation, and this procedure provides information about the elasticity of the
interface. We have used this method to test the accuracy of both the PF and
PF-NS models, performing numerical simulations. The results serve to validate
the model, obtaining a good agreement with the relaxational behaviour predicted
by the theory.
In Chapter 6, we have described the numerical method implemented to per-
form simulations of the model. We make use of a lattice-Boltzmann method for
integrating the Navier-Stokes equation, coupled to a standard finite-differences
method for the integration of the phase-field equation. Given the particulari-
ties of bending interfaces and their delicate control, we have explained the main
numerical problems that must be adressed to correctly simulate these interfaces.
We thus might point that
• We have derived a phase-field model for membrane modeling, proving its

consistency with the classic theory of membranes and obtaining the expres-
sions of the elastic stress tensor and lateral stress profile.
• The model recovers the equilibrium equation of the membrane, as proved

by means of a sharp-interface limit analysis. The method also allows us to
obtain the macroscopic equations of the model.
• We have tested numerically our model by means of a linear stability analysis

of a flat interface, obtaining the desired membrane dynamics.
• The complete method is numerically integrated by a lattice-Boltzmann scheme,

which must be tunned carefully in order to capture the delicate behaviour
dictated by bending interfaces.
RBC flow
In Part IV of this Thesis, we have studied the elastic behaviour of RBCs flowing
in confined microchannels, both at low concentrations in very thin channels, when
they display single-cell behaviour, and at higher concentrations at intermediate
channels, when collective effects dominate the dynamics. The study has been
conducted to understand and identify the elastic mechanisms that contribute to
the rheological behaviour of blood. We have performed simulations making use of
the phase-field Navier-Stokes model derived in Part III.
In Chapters 8 and 9, we have focused on the deformability and rheological
properties of single RBCs. The parameter control is the capillary number, which
balances the elastic forces of the membrane and the fluid forcing. Our results
show that for increasing capillary, RBCs develop a sequence of different mor-
phologies, namely discocytes, slippers and parachutes. The shapes obtained show
a nice agreement with experimentally observed RBCs at similar conditions. The
deformability and capability to orient of the cell modulates the rheological be-
haviour of the suspension. We obtain a shear-thinning behaviour of the effective
viscosity. The viscosity decay is mainly associated with the transition to the slip-
per, highlighting the importance of cell deformation and orientation. The analysis
of the energetic contributions has shown that membrane incompressibility is cru-
cial for preserving compact shapes, and at intermediate and high capillary RBCs
are subjected to strong membrane tensions. We have investigated the origins of
the slipper morphology, showing that the RBC benefits from a relaxation of its
deformation energy in comparison with cells flowing at the channel axis. The
fluidity of the suspension also benefits from this lateral positioning. The main
conclusions of this study therefore are,
• RBCs present a rich behaviour determined by the interplay between their

elastic membrane and the flow. Sequentially, discocytes, slippers and parachutes
are found for increasing capillary number.
• The suspension presents a characteristic shear-thinning behaviour, origi-

nated on the elasticity and orientation of the cell with the flow.
• Asymmetric shapes are low-energy configurations which also enhance the

fluidity of the suspension compared to cells flowing at the channel axis.
• The similar shapes obtained by our model, which lacks an in-plane contri-
bution of the cytoskeleton, and previously observed morphologies both in
experimental and numerical studies in which the cytoskeleton was consid-
ered, suggests that the cyskeleton does not play a key role in this type of
deformations.
In Chapter 10, we have investigated the control of RBC focusing and spatial
organization, motivated by the enormous interest of this subject in lab-on-a-chip
devices and microfluidics. We have shown that for low capillary numbers RBCs
flow along the entire section, but increasing the capillary number the cell distri-
bution narrows to a lateral band. We have explored the effect of capillary number
and confinement, showing that both effects are coupled and reinforce the focus-
ing. For sufficiently high capillary, the band converges towards a unique lateral
position. For thicker channels, larger values of the capillary are required to in-
duce focusing. However, in these channels the stable lateral position is found
further from the axis. This position is likely determined by the balance between
the repulsive force from the wall and a drift from the axis. Additionally, we have
studied the dependence of cell shape, obtaining that more spherical vesicles prefer
the channel axis.
• RBC focusing is controlled by the flow velocity and confinement of the chan-
nel, and both effects are coupled. Higher flow velocities and narrower chan-
nels induce the focalization of cells towards a more defined position in the
lateral channel.
• The shape of the object is a key parameter controlling the object flow. RBCs
and deflated vesicles flow at different lateral positions along the channel sec-
tion, with more spherical vesicles prefering the center and RBCs occupying
outer positions.
In Chapter 11, we have explored the effect of hydrodynamic interactions be-

tween RBCs when they flow maintaining short distances between cells, and the
competition of these interactions with the wall effect. If cells are sufficiently dis-
tant, wall-effects dominate and the cell exhibits single-cell behaviour, such as the
lateral migration and orientation. The presence of a close neighbour could inhibit
this effect and then RBCs behave collectively. Afterwards, we have studied the
behaviour of several cells flowing in a channel. RBCs are found flowing close to
the axis, as they are repulsed by the neighbour flowing at an outer position, and
at the same time the cells situated in the center push the outer ones towards the
walls. The focusing effect described in Chapter 9 dissapears. However, we have
12.2. Future perspectives 201
found that increasing the flow velocity also induces a focusing, now towards a more
concentrated distribution of cells in the channel core as wall repulsion increases.
As a summing-up of this study, we can state that
• Hydrodynamic interactions between RBCs strongly modify the flow behaviour

of the cell, as the screening between cells attenuates cell deformation andit
can inhibit lateral migrations. Which effect is dominant is determined by the
distance between cells and the capillary number.
• When several cells are present, parachutes are only found at very confined
channels and close distances between cells. For thicker channels and higher
distance between cells, they prefer to organize in paralel rows of slippers, a
configuration stable even at very high flow velocities.
• For channels of typical width three times the cell length, cells order forming
several rows of RBCs with horizontal inclination, parallel to the walls, and
exhibit weak deformation from the discocyte.
• The collective effect forces the occupation of the channel core, and the charac-
teristic lateral focusing of the single-cell behaviour is not observed. However,
the higher repulsion from the walls for higher capillaries forces the narrowing
of the band of trajectories occupied, as cells are repulsed towards the axis.
12.2 Future perspectives
The complexity of cell membranes invariably causes that in the process of elluci-
dating a certain aspect of their functioning, many other questions arise, of similar
complexity and importance. Some of these questions can be actually studied with
the phase-field method developed in this Thesis.
Stability of cell membranes
In Chapter 5, we have carried out a stability analysis for the simplest membrane
configuration, a symmetric flat membrane. This method, however, can be ex-
tended to more complex geometries and membranes. For instance, the relaxation
of membranes at cylindric or spherical morphologies is a subject of interest as
it is not completely understood (Shiba and Noguchi, 2011). The effect of a non
zero spontaneous curvature is also a problem of complex analytical treatment, and
the phase-field model offers the appropriate framework to study and adress these
problems.
Phase-field modelization of the cytoskeleton
One of the major limitations of the phase-field model presented here is that it does
not account for the in-plane contribution of the cytoskeleton. Ideally, the model
could be refined to incorporate terms of shear and stretching, as has been devel-
oped in other fields, such as fracture dynamics (Pons and Karma, 2010). Although
the derivation of such a model seems, a priori, a considerable challenge, it would
represent a formidable tool for studying numerous new problems. For instance,
the disco-echinocyte transition could be study even introducing dynamic effects,
such as the competition in the time scales of membrane asymmetry change and
cell deformation. This study could clarify important aspects of membrane dam-
age during blood storage. A different problem would be to study the rheological
behaviour of echinocytes in flow, as blood under crenation is known to present a
considerable different behaviour.
New aspects of RBC flow
The study of RBCs flow presented in this Thesis has adressed the main aspects
at the cell scale or the effects when a few RBCs are interacting. However, many
other problems of interest can be potentially study. The convergence to the high
concentration limit, at hematocrit ∼ 45%, could reveal new mechanisms and as-
pects of RBC deformability. At high concentration, cells often aggregate due to
the presence of certain molecules on their membrane, forming new structures that
severely alter the suspension rheology (Fedosov et al., 2011). In addition, new
components of blood can be added, such as hard spheres of small size, roughly
4 − 5μm, which mimic the behaviour of platalets or beams. This system is inter-
esting because it is known that during blood flow, these objects avoid the arteriole
core when RBCs concentrate, and flow at the layer close to the wall. This effect,
known as cell-margination, is of great interest due to its applications for drug
delivery (Kumar and Graham, 2012).
Chapter 13
Resumen en castellano
Las membranas celulares son estructuras de una enorme complejidad, compues-

tas por un gran número de lı́pidos y proteı́nas. La membrana tiene un papel
fundamental en la vida celular, pues separa el interior de la célula del medio ex-
terno y define un gran número de funciones y actividad proteı́nica. Es, además,
responsable de la respuesta celular frente a perturbaciones mecánicas, y por ello
las membranas poseen una propiedades elásticas únicas entre los materiales. Por
todo ello, el estudio de las membranas ha atraı́do a los fı́sicos desde hace muchos
años, y un gran numero de fenómenos en este campo han sido explicados gracias
a la contribución de modelos fisicos.
En este contexto, los glóbulos rojos son células muy particulares entre las
diferentes especies de células humanas, por ser las únicas que carecen de núcleo
y orgánulos internos, de forma que sus propiedades morfológicas y mecánicas es-
tán completamente determinados por su membrana. Por este hecho, el estudio
de membranas celulares se ha centrado a menudo en glóbulos rojos. Estas célu-
las, además, tienen un papel muy importante en el funcionamiento del cuerpo
humano, pues son el componente principal de la sangre y son los encargados del
transporte de oxı́geno al resto de células del cuerpo. En concreto, cuando los
glóbulos atraviesan los capilares más finos, de sección menor que su propia área,
se ven forzados a deformarse fuertemente y alargarse. La capacidad para realizar
estas deformaciones y poder atravesar los capilares muestra una extrema sensibil-
idad a las propiedades elásticas de la membrana.
Esta Tesis está centrada en el estudio teórico de las propiedades mecánicas y

elásticas de los glóbulos rojos. Nuestro estudio abarca desde la respuesta mor-
fológica de la célula ante cambios en la microestructura de la membrana, hasta el
estudio del flujo de glóbulos en canales donde las células se deforman e interaccio-
nan con células vecinas.
203
204 Chapter 13. Resumen en castellano
13.1 Introducción
13.1.1 Introducción biológica

La membrana celular está formada por dos componentes principales: una bicapa
lipı́dica y un citosqueleto de espectrina que se encuentra bajo la bicapa. La bicapa
lipı́dica está formada por dos pelı́culas de lı́pidos, que se ensamblan en dirección
contraria, dejando las colas lipı́dicas (de naturaleza hidrofóbica) en la región in-
terna, y las cabezas (hidrofı́licas) en contacto con el lı́quido externo. La bicapa
es fluı́da en el plano de la membrana, debido a una rápida difusión lateral de
los lı́pidos, y contiene un gran número de proteı́nas transmembrana que se en-
cuentran flotando, en el llamado mosaico fluı́do (Singer and Nicolson, 1972). El
citoesqueleto, por el contrario, es una malla de espectrina que presenta resistencia
a realizar esfuerzos laterales. Se trata de una estructura de dos dimensiones que
se encuentra anclada a la cara interna (citosólica) de la membrana. La interacción
de ambas estructuras determina la respuesta elástica de la membrana celular.
Esta membrana es, por tanto, el elemento principal de los glóbulos rojos. La
célula se compone además de un citosol rico en proteı́nas, donde la hemogoblina
se concentra. Los glóbulos presentan una forma caracterı́stica, conocida como
discocito, similar a un disco plano pero con una zona cóncava en el centro. El
perfil de esta célula es, por tanto, un eliptocito bicóncavo. El diámetro de la
célula es de unas 8 micras, y el grosor varı́a entre las 1 y 2 micras entre sus valores
máximo y mı́nimo.
La sangre es un fluı́do formado por una gran variedad de componentes, pero
su elemento principal son los glóbulos rojos, llegando a constituir un 95% de las
células sanguı́neas. Otros componentes de importancia son las plaquetas y los
leucocitos. La concentración de células es tı́picamente del 45%, siendo el resto un
fluı́do llamado plasma sanguı́neo similar al agua, aunque con abundantes proteı́nas
y otras moléculas. La elasticidad y deformabilidad de los glóbulos confiere a la
sangre unas complejas propiedades reológicas.
13.1.2 Modelos fı́sicos de membranas

La fuerte separación entre la escala del grosor de la membrana, de unos 4nm, y la
longitud tı́pica de la célula, unas 8 micras, sugiere que un tratamiento apropiado es
considerar la membrana como una lámina de dos dimensioneso. Ası́, su elasticidad
puede aproximarse por la de una lámina elástica, de acuerdo con la teorı́a general
de la elasticidad. Fueron Canham (1970) y Helfrich (1973) los primeros en pro-
poner una energı́a elástica para describir las membranas celulares. En concreto,
su teorı́a asume que la única contribución relevante proviene de la resistancia a
13.2. Resultados 205
doblarse. Además, la bicapa lipı́dica presenta una fuerte incompresibilidad lateral,

de forma que el área se mantiene constante. Por último, la membrana es imper-
meable al paso de agua, de forma que para membranas el volúmen es constante.
Teniendo esto en cuenta, propusieron una energı́a de la forma

κ 2
Fb = (C − c0 ) dA + κG GdA + γdA + ΔpdV, (13.1)
2
donde C y G son la curvatura total y Gaussiana, respectivamente, κ y κG son los

módulos de flexión y Gaussiano. c0 es la curvatura espontánea, que refleja una
posible asimetrı́a interna de la membrana. Este modelo, sin embargo, no tiene en
cuenta algunas particularidades de la arquitectura de las membranas. Por un lado,
no considera que la bicapa está formada por dos hojas separadas y acopladas. Por
otro, no tiene en cuenta la contribución a la elasticidad en el plano de la membrana
por parte del citoesqueleto, que se descompone en una contribución de esfuerzo
normal y otro de cizalla (Evans and Skalak, 1980). La teorı́a de Helfrich ha sido
extendida, incorporando estas nuevas contribuciones, de forma que un modelo de
membrana general serı́a

κ κN L π
Emem = (C − c0 )2 dS+ 2
(ΔA − ΔA0 )2
2 ARBC d
S
(13.2)
K 2 (λ1 − λ2 )2
+ (λ1 λ2 − 1) + μ dS0 .
S0 2 2λ1 λ2
donde el primer término se corresponde al modelo de Helfrich y la contribución de

diferencia de área (Seifert et al., 1991; Waugh et al., 1992), dada por la diferencia
entre las áreas de la capa exterior y la interior, ΔA0 . Los paramétros elásticos del
citoesqueleto vienen dados por el módulo de cizalla, μ, y el de compresión, K.
13.2 Resultados
13.2.1 Elasticidad de la membrana y transición del

discocito al equinocito
En los Capı́tulos 3 y 4, nos hemos centrado en el estudio de la transición discocito-

equinocito. Se trata de una secuencia de formas que los glóbulos rojos desarrollan
cuando se ven afectados por la acción de diferentes agentes. Entre otros muchos,
es bien sabido que al ser expuestos a la presencia de ciertos lı́pidos, o bien a la
reducción de ATP, los globulos se deforman y adquieren morfologı́as más esféricas,
con numerosas espı́culas y protuberancias en su superficie. Estas deformaciones

se deben a cambios conformacionales en la microestructura de la membrana. Por
ejemplo, en el caso en el que las células se incuban en presencia de lı́pidos, es-
tos son incorporados a la membrana, y se sitúan en la capa externa. En estas
condiciones, la capa externa acumula una mayor concentración de lı́pidos y trata
de expandir su área para poder acomodarlos. Por eso tiende a curvarse y formar
esa protuberancias, posibilitando la expansión de la capa externa con respecto a
la interna. La transición ha sido estudiada experimentalmente y un gran número
de agentes y mecanismos han sido identificados. Sin embargo, la transición no ha
sido caracterizada cuantitativamente y ciertos aspectos de las transiciones no son
bien comprendidos.
En el Capı́tulo 4 hemos presentado un estudio teórico en el que analizamos la
respuesta mecánica y morfológica de los globulos cuando son sujetos a un incre-
mento en su asimetrı́a de la membrana. Consideramos el modelo de membrana
(13.2), basándonos en una parametrización de la superficie de la célula en términos
de los óvalos de Cassini, una familia de curvas que son conocidas por presentar
una sección biconcava muy similar a la del discocito. La expresión analı́tica de
estas curvas viene dada por
(x2 + z 2 + a2 )2 − 4a2 x2 = c4 . (13.3)
de donde se define la biconcavidad, ≡ a/c. Para = 1 la curva se corresponde

con un lemniscato de Bernuilli, mientras que para = 0 se recupera un cı́rculo.
En nuestro caso, recuperamos la forma de una célula aplicando simetrı́a axial
en torno a x = 0. La modelización de las espı́culas y protuberancias se realiza
superponiendo superficies especı́ficas.
Nuestros resultados prueban que la morfologı́a de la célula muestra una sen-
sibilidad extraordinaria al equilibrio entre la bicapa lı́pidica y el citoesqueleto.
Para asimetrı́as bajas, el discocito es la forma de energı́a mı́nima, pero cuando
cuando la asimetrı́a se va incrementando, la célula prefiere adquirir formas más
convexas, con las que poder acomodar el exceso de área en la capa externa. El
citoesqueleto, por el contrario, se opone a este tipo de deformaciones y prefiere
mantener formas más cercanas al discocito. La competición de ambas estruc-
turas dictamina la forma de equilibrio. Si la asimetrı́a de la membrana induce
suficiente almacenamiento de energı́a en la bicapa, finalmente el citoesqueleto se
verá forzado a deformarse y las protuberancias van apareciendo sequencialmente.
Primero, aparecen ondulaciones en el contorno de la célula, lo que se denomina
discoequinocito I. Después, protuberancias aparecen también en la cara principal
del disco, distribuı́das a lo largo de la superficie. Simultáneamente el glóbulo se
va volviendo más esferico. Nuestro estudio se limita a estos primeros estadios de
la transición.
Nuestro estudio permite cuantificar la cantidad de exceso de área en la capa
externa necesaria para inducir el cambio morfológico. Obtenemos valores cercanos
al 0.63% para desestabilizar el discocito, pero este valor se reduce al 2.0% en la
ausencia del citoesqueleto. Por tanto, el citoesqueleto es una estructura básica
para controlar la estabilidad del discocito con respecto a cambios en la asimetrı́a de
la membrana. Las escalas energéticas de la transición muestran que las diferencias
entre las diferentes morfologı́as son del orden de cientos de kB T , evidenciando la
gran estabilidad de estas formas. En concreto, el discocito representa el estado
fundamental de la transicón, y está muy favorecido energéticamente respecto a los
equinocitos, que presentan un alto grado de estrés del citoesqueleto.
El estudio teórico explicado previamente nos permite interpretar los resultados
realizados por Kathryn A. Melzak y José Luis Toca-Herrera. En estos experimen-
tos, se utiliza la punta de un AFM para inducir una transición morfológica en
glóbulos rojo. En concreto, se seleccionan equinocitos que han sido formados me-
diante reducción de los niveles de ATP, y tras la perturbación con el AFM se
observa la transición hacia el discocito. Proponemos que en los experimentos el
AFM perfora la bicapa lipı́dica, y el contacto con la punta del AFM rompe la bar-
rera hidrofóbica permitiendo el salto de algunos lı́pidos de la cara externa, donde
están en alta concentración, a la interna, donde se encuentran más relajados. El
mecanismo permite la relajación de la energı́a de curvatura y la célula vuelve hacia
el discocito, que representa una morfologı́a de mucha menor energı́a.
13.2.2 Métodos de interfase difusa para modelado de mem-

branas
Los métodos de interfase difusa (en inglés usualmente conocidos como phase-field
models) son muy útiles a la hora de tratar problemas dinámicos de interfases.
Se han aplicado ampliamente al estudio, entre otros, de inestabilidades entre dos
fluı́dos, en los que la interfase está caracterizada por la tensión superficial. Sin
embargo, las membranas son objetos complejos en los que la tensión superficial
es insignificante y la dinámica de la membrana está dominada por la flexión. Los
modelos de interfase difusa se basan en utilizar un parámetro de orden, φ, que
tiene dos fases estables ±1 y varia suavemente de una a otra. La interfase viene
caracterizada por un grosor . El método se basa en resolver directamente la
dinámica de la membrana, y de ahi extraer la forma y evolución de la interfase.
Ası́, se evita el problema de aplicar condiciones de contorno para una interfase
móvil.
Hemos desarollado un modelo de interfase difusa que captura las propiedades
elásticas de las membranas. Para ello, se propone un modelo de energı́a libre de

la forma

κ∗
Fb [φ] = (φ2 − 2φ4 + φ6 + (3φ2 − 1)2 (∇φ)2 + 4 (∇2 φ)2 )dV. (13.4)
2
La incompresibilidad de la membrana se introduce mediante un multiplicador

de Lagrange que fija el área de la célula. Hemos probado que esta energı́a converge
a la de Helfrich (13.1) para los valores particulares de γ = c0 = κG = 0. Hemos
derivado, a partir de la energı́a libre, las expresiones para el potencial quı́mico
μ = δF /∂φ y el tensor de esfuerzos en función del parámetro de orden. Esta
última adquiere la forma,

δF ∂F ∂F ∂F
σαβ = F −φ δαβ −∇α φ +∇α φ∇β 2
+∇α ∇β φ . (13.5)
δφ ∂(∇β φ) ∂(∇ φ) ∂(∇2 φ)
y permite demostrar que la densidad de fuerza elástica de la membrana, en tér-

minos del parámetro de orden, viene dada por la expresión
fmem = ∇ · σmem = −φ∇μmem . (13.6)
Hemos incorporado la energı́a libre a un modelo dinámico de tipo Cahn-

Hilliard, incluyendo también los efectos hidrodinámicos del fluı́do que rodea a
la membrana. La hidrodinámica viene dada por la ecuación de Navier-Stokes, de
forma que tenemos dos ecuaciones acopladas, una que describe la dinámica de la
membrana y otra para el fluı́do. El modelo completo es, por tanto,
∂φ
+ v · ∇φ = M∇2 μmem . (13.7)
∂t

∂v
ρ + (v · ∇v) = −∇P + fmem + η∇2 v + fext . (13.8)
∂t
Un estudio del lı́mite de interfase abrupta, → 0, nos permite obtener las

ecuaciones del modelo macroscópico. En equilibrio, el modelo recupera la ecuación
clásica para una membrana asimétrica, c0 = 0, que viene dada por
Δp = γC + (1/2)κC(C 2 − 4G) − κΔs C. (13.9)

Fuera del equilibrio, obtenemos la ecuación que describe la dinámica de la

membrana, (5.55). A partir de estas ecuaciones, hemos realizado un análisis de
estabilidad lineal que nos permite estudiar el comportamiento relajacional de la
membrana, y sirve a su vez como test para comprobar que nuestro modelo captura
la dinámica correcta predicha por la teorı́a, como se muestra en la Figura 5.3.
13.2.3 Flujo de glóbulos rojos en canales confinados
El modelo de interfase difusa presentado en el Capı́tulo 5 ofrece las herramientas

necesarias para simular el comportamiento de glóbulos rojos fluyendo en canales.
Este problema es de enorme interés en biomedicina por diversas razones. Por un
lado, entender el comportamiento reológico de la sangre es clave para mejorar
nuestro conocimiento sobre el funcionamiento de la microcirculación, ası́ como
mejoras en el almacenamiento y manejo de sangre para transfusiones o análisis
médicos. Por otro, gracias al reciente desarollo en microfluı́dica, el manejo de una
única célula posibilita la diagnosis de enfermedades en pequeños dispositivos.
Nuestro modelo de interfase difusa es utilizado para realizar simulaciones por
medio de un método conocido como lattice-Boltzmann. El esquema y la imple-
mentación numérica se han descrito en detalle en el Capı́tulo 6, concentrándonos
en las particularidades de la simulación de membranas.
En los Capı́tulos 8 y 9, nos hemos centrado en el estudio de las propiedades
elásticas y reológicas de un glóbulo fluyendo aislado en un canal confinado. Asum-
imos que en este régimen, de baja concentración de células, los efectos inducidos
por las paredes son dominantes con respecto a las interacciones hidrodinámicas
entre células. El sistema está caracterizado por el número capilar, que define el
balance entre las fuerzas viscosas del fluı́do y las elásticas de la membrana de la
célula,
τκ η0 v̄z a2 a
Cκ = = . (13.10)
τη κ b
Situando un glóbulo en el canal, y para un incremento del número capilar,

obtenemos una secuencia de morfologı́as bien diferenciadas. A bajos capilares
Cκ < 4, las células fluyen mostrando formas muy similares al discocito de referen-
cia, que denominamos discocitos (discocytes). Una célula inicializada en el centro
se mantendrá en esa posición, mostrando un pequeño acoplamiento con el flujo y
una ligera asimetrı́a. Para capilares mayores, Cκ > 10, las células migran hacia
una posición lateral en el canal, estabilizándose en una posición intermedia entre
el canal y la pared. En ésta posición adquieren una forma que se denomina slipper.
Esta configuración es muy estable y se observa en el rango de capilares 10-90. Por
encima de ese valor, el slipper se vuelve inestable y la célula vuelve a una posición
axial, donde adquiere una forma caracterı́stica denominada parachute. El estudio
de las contribuciones elásticas en cada régimen muestra que la migración lateral
y el desarrollo del slipper supone una relajación en la energı́a de deformación con
respecto a las configuraciones más centradas. Ası́ mismo, muestra que el término
de incompresibilidad es dominante a altos capilares, evidenciando que las células
están sujetas a altas tensiones en la membrana. Ası́ mismo, hemos estudiado el
comportamiento reológico de la suspensión. La viscosidad efectiva muestra un
comportamiento de shear-thinning, en el que la viscosidad decae al aumentar el
ritmo de cizalla. Este decaimiento, asociado a una mayor fluidez de la suspen-
sión. Este cambio se debe a la posición lateral y consiguiente alineamiento del
glóbulo durante el régimen de slipper, ası́ como la gradual pérdida de rigidez de
la membrana que provoca una menor distorsión del flujo.
En el Capı́tulo 9, hemos estudiado explicı́tamente las ventajas de la posición
asimétrica del slipper con respecto a posiciones simétricas en el canal. El por
qué los glóbulos prefieren posiciones y formas asimétricas cuando fluyen en flujos
simétricos es una cuestión abierta y de gran interés en la comunidad. Nuestros
resultados muestran que la posición asimétrica permite una relajación de la energı́a
de deformación, y una impotante reducción de los esfuerzos en las membrana. En
este sentido, las posiciones centrales se ven mucho más penalizadas. Además,
la vorticidad se cuando la célula se encuentra en una posición lateral disminuye
drásticamente y se focaliza en el lóbulo trasero del glóbulo.
En el Capı́tulo 10, nos hemos centrado en el estudio de la focalización de los
glóbulos hacia ciertas posiciones especı́ficas a lo largo de la sección del canal, moti-
vados por su interés a la hora de separar células en pequeños chips o la posibilidad
de diagnosticar enfermedades. Nuestros resultados muestran que la focalización de
las células a una posición especı́fica se puede controlar mediante el acoplamiento
del flujo y el confinamiento. De esta forma, mientras que para bajos capilares las
células pueden encontrarse fluyendo a lo largo de toda la sección del canal, para
capilares intermedios y altos, las células se concentran en dos bandas laterales
(simétricas respecto al eje), que con el aumento del capilar se van haciendo más
finas hasta que, eventualmente, todas las células fluyen en una misma trayectoria,
independientemente de si su configuración inicial es cercana a la pared o al eje.
El confinamiento es también relevante: en canales anchos, es necesario un capilar
mayor para inducir este efecto de focalización, pero a la vez las células miran hacia
posiciones más exteriores, cercanas a la pared. Finalmente, hemos estudiado la
dependencia de este comportamiento con la forma, comparando el caso de glóbu-
los rojos con vesı́culas más esféricas. Los resultados muestran que las vesı́culas
esféricas migran hacia posiciones centradas, fluyendo a lo largo del eje. Por tanto,
la asimetrı́a es un factor importante a la hora de explicar la posición lateral.
13.3. Conclusiones 211
En el Capı́tulo 11, nos hemos centrado en el estudio del comportamiento de

glóbulos fluyendo a concentraciones mayores, en los que la interacción con células
vecinas puede jugar un factor relevante y dominar respecto a la interacción con
la pared. Estudiando el efecto de apantallamiento entre células, observamos que
la perturbación que una célula ejerce sobre el flujo entrante hace que la célula
vecina interaccione de forma efectiva con un flujo caracterizado por fuerzas vis-
cosas menores, de forma que sufre un menro deformación. Por esta razón, celulas
fluyendo en formaciones alineadas o a pequeñas distancias muestran deformaciones
mucho menores que aquellas fluyendo aisladas, y la migración lateral se ve par-
cialmente inhibida. Solo cuando la distancia entre células aumenta lo suficiente,
el efecto de apantallamiento de atenúa y se recupera el comportamiento de célula
aislada. Este cambio de comportamiento marca el régimen en el que los efectos
de la pared son dominantes respecto a las interacciones entre celulas.
Finalmente, hemos estudiado estos efectos cuando muchos glóbulos fluyen en
un canal. Los resultados muestran que el comportamiento de los glóbulos cambia
significativamente. Para canales de anchura tres veces la de la célula, los glóbulos
se concentran en el centro del canal, formando 3-4 filas. Las células situadas
cerca del eje se ven atrapadas por las más externas, y están forzadas a fluir en
el centro (una posición inestable para el caso de una célula aislada). A su vez,
las células internas repelen hacia la pared a las externas, y estas se encuentran
localizadas más cerca de la pared que en el caso de una célula aislada. Ası́, la
focalización lateral se pierde, y las células se concentran ahora en el núcleo del
canal. Aumentando el capilar, la repulsión de las paredes también aumenta y la
localización de las células focaliza a una distribución más centrada.
13.3 Conclusiones
El objetivo principal de esta Tesis ha sido estudiar desde una perspectiva teórica el
comportamiento mecánico de los glóbulos rojos en diferentes situaciones, concen-
trándonos en el entendimiento de las propiedades elásticas macroscópicas a partir
de la microestructura de la membrana celular.
La investigación original de esta Tesis se ha divido en tres partes. En la
Parte II hemos estudiado la respuesta morfológica de los glóbulos cuando se ven
afectados por cambios en la asimetrı́a de la bicapa lipı́dica. Hemos encontrado que
el desarrollo de protuberancias que dan pie a la formación del equinocito responde
a la relajación de la bicapa cuando se ve sometida a un exceso de área en su capa
externa. El citoesqueleto, en cambio, opone resistencia a la formación de estas
estructuras, manteniendo un forma más compacta y siendo por tanto un elemento
clave en la estabilización del discocito.
En la Parte III, nos hemos centrado en la derivación de un modelo de interfase

difusa para el modelado de membranas. Hemos demostrado que nuestro modelo
captura la fı́sica relevante de las membranas y recupera las ecuaciones de equilibrio
y el comportamiento relajacional predichos por la teorı́a.
Finalmente, en la Parte IV, hemos utilizado el método de interfase difusa para
estudiar el comportamiento de glóbulos rojos en flujos en canales. Inicialmente nos
hemos centrado en el caso de una célula a alto confinamiento, observando que las
células prefieren posiciones asimétricas a lo largo del canal, donde se orientan con
el flujo, y que se ven sujetas a fuertes tensiones en la membrana. Después hemos
estudiado también el caso en el que muchas células esán presentes, comprobando
la fuerte dependencia en las interacciones con otras células y como los efectos de
migración y alineación se ven atenúados.
Appendix A
Notes on differential geometry
In this Appendix we include some basic concepts of differential geometry of sur-

faces that could be useful for the understanding of some of the calculations in the
main text.
A.1 Mean curvature

The curvature and area element of a shape can be obtained from the first and
second fundamental forms of the manifold. Considering the parametrization:
r = (rcos(θ), rsin(θ), z(r, θ)) (A.1)
and defining rr = dr/dr, rθ = dr/dθ, then
E = rr · rr ,
F = rr · rθ , (A.2)
G = rθ · rθ ,
using standard notation. The first fundamental form reads:
I = Edr 2 + 2F drdθ + Gdθ2 . (A.3)
In a similar way, for the second fundamental form,
L = rrr · n̂,
M = rrθ · n̂, (A.4)
N = rθθ · n̂,
213
214 A. Notes on differential geometry
where n̂ is the normal to the surface, and from here:
II = Ldr 2 + Mdrdθ + Ndθ2 . (A.5)
The total curvature and surface area element can be easily obtained:
NE − 2MF + LG
C= , (A.6)
(EG − F 2 )
√
dA = EG − F 2 drdθ. (A.7)
Unfortunately, even if these expressions are analytical (note that they only
depend on the derivatives zr , zθ , zrr , zrθ , zθθ ), the surface integral of eqn (4.2) has
no analytical solution and the integration must be numerically computed.
A.2 Extension ratios

The extension ratios represent the normal strain of a line element. Within the
finite strain theory framework, they characterize the deformation of a manifold
state. They are obtained from the
from the reference S0 (x) to the current S(X)
Green’s strain tensor,

1 ∂X γ ∂Xγ
αβ ≡ − δαβ , (A.8)
2 ∂xα ∂xβ
and from the eigenvalues of this tensor i ,
√
λi = 2i + 1. (A.9)
The Green’s strain tensor can be easily calculated if the current state can be
analytically expressed in terms of the reference state. Once the extension ratios are
obtained, eqn (4.4) is numerically computed. In a isotropic dilation λ1 = λ2 and
the second term vanishes; alternatively, if the area element is conserved λ1 = 1/λ2 ,
the first term of the energy is null.
A non-trivial issue concerns the mapping from the reference to the actual
state S0 :→ S. In our model, the connection between manifolds is determined by
the condition A(r) = A0 (r0 ), thus equaling the areas from the polar axis in both
configurations. This implies that the central point is less sensitive to deformations
than the external region, a reasonable hypothesis at these early stages of the
transition. Obviously, this method is only valid as far as the area of the reference
state is equal to that of the cell.
A.3. Differential operators in curvilinear coordinates 215
A.3 Differential operators in curvilinear coordi-

nates
In Part III of this Thesis, we introduce the curvilinear coordinates in order to deal
with the sharp-interface limit of a slightly curved interface. These coordinates rep-
resent a naı̈ve parametrization of the membrane surface, but they are rarely used
in the literature and hence it is convenient to specify its metrics and differential
operators. The vector basis is chosen as follows. At each position of the surface,
the normal vector n is defined. From the tangential plane we choose two vectors
forming an orthonormal trihedron, t2 = n × t1 . We define a vector position n
that can be decomposed into the vector position in the surface and the normal
projection, r = R(s, u) + rn(s, u), where r is the coordinate in the direction of n
and (s,u) are the coordinates in the directions of t1 and t2 , respectively. Hence,
(r, s) are orthogonal coordinates on the membrane surface, and describe arclengths
of the curve formed by the intersection of the membrane surface and the planes
(r, s) = const and (r, u) = const. From here, we obtain |∂R/∂u| = |∂R/∂s| = 1,
and t1 = ∂R/∂u and t2 = ∂R/∂s by definition. The particular choice of this co-
ordinates reduces the classic Bonnet-Kowalewski trihedron to the Frenet formulas
(do Carmo, 1976; Biben et al., 2005), obtaining
∂n
= −c1 t1 − τ1 t2 ,
∂s
∂t1 ∂t2
= c1 n, = τ1 n,
∂s ∂s (A.10)
∂n
= −c2 t2 − τ2 t1 ,
∂u
∂t1 ∂t2
= −τ2 n, = c2 n.
∂s ∂u
where c1 and c2 are the curvatures and τ1 and τ2 the torsions in the (u, s) directions,
respectively.
Combining the expressions given before, the base vector can be calculated,
∂r
Er ≡ = n,
∂r
∂r ∂R ∂n
Es ≡ = +r = (1 − rc1 )t1 − rτ1 t2 ,
∂s ∂s ∂s (A.11)
∂r ∂R ∂n
Eu ≡ = +r = (1 − rc2 )t2 + rτ2 t1 .
∂u ∂u ∂u
From this basis the metric can be readily computed as gαβ = Eα Eβ , leading to
216 A. Notes on differential geometry
⎛ ⎞
1 0 0
gαβ =⎝ 0 (1 + rc1 )2 + r 2 τ 2 −2rτ + r 2 (c1 + c2 )τ ⎠ . (A.12)
0 −2rτ + r 2 (c1 + c2 )τ (1 + rc1 )2 + r 2 τ 2
We have introduced τ = τ1 = τ2 as can be easily proved from ∂s Eu = ∂u Es .

From here, the determinant of the metric reads
g = det gαβ = [(1 − rc1 )(1 − rc2 ) − r 2 τ 2 ]2 . (A.13)
We are interested in the expression of the differential operators, in particular,

those which appear in the phase-field free energy, ie the gradient and the laplacian.
The gradient can be computed from its formal definition ∇ = E α ∂α , obtaining
1
∇ = n∂r + {[t1 [(1 − rc1 )∂s + rτ ∂u ] + t2 [rτ ∂s + (1 + rc1)∂u ]}. (A.14)
1 − rC + r 2 G
where we have introduced the total curvature C = c1 + c2 , and the Gaussian

curvature G = c1 c2 − τ 2 . Analogously, the general expression of the Laplacian is
given by
1 √
∇2 = √ ∂α gg αβ ∂β . (A.15)
g
Introducing (A.12) and (A.13) in (A.15), the final expression of the Laplacian
in curvilinear coordinates reads

2 C − 2Gr (1 − rc2 )2 + r 2 τ 2
∇ = ∂r2 − ∂r + ∂2
1 − Cr + Gr 2 [(1 − rc1 )(1 − rc2 ) − r 2 τ 2 ]2 u
(A.16)
(1 − rc1 )2 + r 2 τ 2 2 2
+ ∂ ∂ +··· ,
[(1 − rc1 )(1 − rc2 ) − r 2 τ 2 ]2 u s
where, for the sake of clarity, we have neglected the contribution of terms of the
form ∂u C∂u , ∂s ∂u ,..., as they are irrelevant for the purposes of this Thesis.
Appendix B
The sharp-interface limit
The sharp-interface limit is a classic method for determining the macroscopic

equations of diffuse interface models (Folch et al., 1999). It exploits the separation
between the length scale of the interface, given by , and the typical length of the
interface, which can be defined from the total curvature as l ∼ 1/C. The limit
applies only in the low-curvature limit, C 1.
B.1 The inner and the outer region
The interface is described by the curvilinear coordinate (r, s, u), described in Ap-
pendix A. r is normal to the interface at each point (s, u) on the surface. The
space is separated into two different regions. The interfacial region, called here
inner region, is described by a fast coordinate ω = r/. Thus, the interfacial
region is zoomed up in order to resolve the details of the smooth interface, al-
though in the outer region the interface is effectively sharp in the limit → 0.
The starting point is a flat interface in equilibrium, with relaxed profile φ0 . If a
small perturbation is induced, the deviations from the equilibrium profile can be
expanded in terms of ,
a(ω, s, u) = a0 (ω, s, u) + a1 (ω, s, u) + 2 a2 (ω, s, u) + ...,

(B.1)
A(r, s, u) = A0 (r, s, u) + A1 (r, s, u) + 2 A2 (r, s, u) + ...,
where capital letters denote variables in the outer region, and lower case letters
indicate variables in the inner region. In the limit ω → ±∞ both regions meet,
and the matching conditions read:
217
218 B. The sharp-interface limit
a0 (ω, s, u) = A0 (0± , s, u),

a1 (ω, s, u) = a1 (0± , s, u) + ω∂r a0 (0± , s, u),
2 (B.2)
a2 (ω, s, u) = a2 (0± , s, u) + ω∂r a1 (0± , s, u) + ω2 ∂rr
2
a1 (0± , s, u),
··· .
The dynamic equation (5.34) is analogously decomposed in the two regions,

with the corresponding inner and outer variables. The differential operators must
be also expanded in terms of . We also assume that the interface motion is much
larger than the characteristic times of order parameter diffusion, and hence the
time is also rescaled τ = t, where τ is the time in the inner region. Hence, in this
quasiestatic approximation we can write,
1
∂t φ(r) = ∂τ φ(ω) − ∂τ ω∂ω φ(ω), (B.3)

and defining the normal velocity of the interface v = −∂τ ω the dynamic equation
reads
1
∂τ φ − v∂ω φ = M∇2 μ, (B.4)

where the laplacian must be expanded, as shown below. The method does not
allow for finding explicit solutions of the fields, but provide a set of equations that
in the limit → 0 represent the macroscopic equations of the model.

Figure B.1: Scheme of the sharp interface method. The interface has a normal direction
at each point given by the coordinates r in the outer region and the fast coordinate
ω = r/ in the inner region. In the inner region, each isosurface φ = const has an
associated mean curvature C (u, s), whereas the isosurface φ = 0 is associated to C.
B.2. Differential operators 219
B.2 Differential operators

The differential operators in curvilinear coordinates have been derived in Appendix
A. However, here we follow a different procedure which provides a more intuitive
understanding of the sharp-interface methodology. An scheme of the interfacial
isosurfaces and curvatures is provided in Figure B.1. As previously stated, the
differential operators must be also expanded in terms of . In the inner coordinates,
∇φ = (−1 ∂ω φ, ∂s φ, ∂u φ), and thus at leading order only the normal coordinate
contributes. This means that the variations of φ along the coordinate directions
(s, u) are disregardable with respect to variations in ω. For simplicity, in the
subsequent calculations we deliverately neglect some derivatives of the tangential
coordinates, which do not contribute after the expansion, in order to simplify the
expressions obtained. Some useful identities are (Du et al., 2005)
n̂ = ∇r,
C = −∇ · n̂ = −∇2 r, (B.5)
G = −(1/2)[2tr((∇α ∇β r)2 ) − (tr(∇α ∇β r))2 ].
for the normal vector and total curvature, respectively. Note that, because the
normal vector is unitary, (∇r)2 = 1, and consequently
nα ∇β nα = (1/2)∇β (nα nα ) = 0. (B.6)
We suppose that at each point of the isosurface φ = const, there is a local

coordinate system given by the coordinate r(x), normal to the surface, and the
tangential coordinates (u(x), s(x)). We suppose that there exists a solution for
the order parameter profile of the form φ = φ(r, u, s). We can write the gradient
and laplacian as
∇φ = ∂r φ∇r + ∂s φ∇s + ∂u φ∇u.

2
∇ φ= ∂r2 φ(∇r) + ∂r φ∇2 r + ∂s2 φ(∇s)2 + ∂s φ∇2 s + ∂u2 φ(∇u)2 + ∂u φ∇2 u.
(B.7)
In this expression, the normal vector n̂ and total curvature C can be directly
identified. The terms related with the tangential derivatives are more difficult to
interpret, though for instance ∇s is given by
1
∇s = √ [(1 − rc1 )t1 + rτ t2 ]. (B.8)
g
It can be proved that these terms correspond to those appearing in the last
terms of (A.16), although it will not be derived here. For simplicity, we only
consider the terms associated to the higher derivative, eg ∂s2 in the Laplacian.
Introducing these considerations, the expression for the Laplacian can be rewritten
as
∇2 φ = ∂r2 φ − C ∂φ + ∂s2 φ + ∂u2 φ + · · · . (B.9)
Here, C corresponds to the local total curvature of the isorfuce at each point
of the space (r, s, u). However, in the interface region it is convenient to write the
local curvature of the isosurface φ = const expressed in terms of the curvature of
the isosurface φ = 0, which we denote C, given that in the sharp interface limit
C → C. The relation between the curvature of two surfaces is given by

2rG/C − 1
C =C . (B.10)
1 − rC + r 2 G
Introducing this expression in (B.9), the expression for the Laplacian is given
by

2 2rG − C
∇ φ= ∂r2 φ − ∂φ + ∂s2 φ + ∂u2 φ + · · · . (B.11)
1 − rC + r 2 G
Note that this expression agrees with that found in Appendix A, (A.16). At
low curvatures, the function associated to ∂r can be expanded, obtaining

∇2 φ = ∂r2 φ− C + r(C 2 − 2G) + r 2 (C 3 − 3GC + ...) ∂φ+∂s2 φ+∂u2 φ+· · · . (B.12)
If the fast coordinate ω is introduced, the expression reads

∇2 φ = −2 ∂r2 φ−−1 C + r(C 2 − 2G) + 2 r 2 (C 3 − 3GC + ...) ∂φ+∂s2 φ+∂u2 φ+· · · .
(B.13)
The derivation of the fourth derivative, ∇4 = ∇2 ∇2 , is more complicated. By
taking the derivative of the expression of the Laplacian (B.7), one obtains
∇2 ∇2 φ = ∂r4 + 2∇2 r∂r3 + 2∇α r∇α ∇2 r∂r2 φ + (∇2 r)2 ∂r2 φ + ∇2 ∇2 r∂r φ. (B.14)
B.2. Differential operators 221
In this expression, several terms can be readily identified in terms of the total
curvature (eg C 2 = (∇2 r)2 ), but particularly the physical meaning of ∇α r∇α ∇2 r
is not straightforward.
For convenience, we perform below some algebraic manipulations that will
be useful in the derivation of the equilibrium condition in Section 5.6. Let us
consider first the fourth term in the right hand side of (B.14). Multiplying by ∂r φ
and (∇α r)(∇α r) = 1, and integrating by parts

∇α r∇α r∂r φ∂r2 φ(∇2 r)2 dx
(B.15)
1
=− [(∇2 r)3 + 2∇α r∇α ∇2 r∇2 r](∂r φ)2 dx.
2
The interpretation of this expression requires to consider the identity C 2 −

4G = 2tr(∇2 r)2 − (∇2 r)2 , as can be obtained from identities (B.5). Then, it is
straightforward to show the equivalence:
C (C 2 − 4G ) = −∇2 r(2tr(∇α ∇β r)2 − (∇2 r)2 ) =

− 2∇α (∇2 r∇β r∇α ∇β r) + 2(∇α ∇2 r∇β r∇α ∇β r) (B.16)
+ 2∇2 r∇α r∇α r∇2 r + (∇2 r)3 = (∇2 r)3 + 2∇α r∇2 r∇α ∇2 r,
where in the last equality we have used (B.6). The comparison between equa-
tions (B.15) and (B.16) demonstrates that the fourth term in (B.14) relates with
−(1/2)C (C 2 − 4G ).
Considering now the third term in the right hand side of (B.14), multiplying
by ∂r φ and (∇β r)(∇β r) = 1,

2 1
∇β r∇β r∇α r∇α ∇ r∂r2 φ∂r φdx
=− [∇β r∇β ∇2 r∇2 r+
2 (B.17)
∇β r∇α ∇β r∇α ∇2 r + ∇β r∇α r∇α ∇β ∇2 r](∂r φ)2 dx.
Note that the second term in the expression in brackets vanishes by (B.6).
From the expression for the gradient operator projected over the surface S, ∇Sα =
∇α − nα nβ ∇β , the Laplace-Beltrami operator over the surface reads
ΔS = ∇Sα ∇Sα = ∇2 + C nα ∇α − nα nβ ∇α ∇β . (B.18)

Manipulating this expression, and using nα = ∇α r, leads to,
ΔS C = ∇2 C + C nα ∇α C − nα nβ ∇α ∇β C
(B.19)
= −∇2 ∇2 r + ∇2 r∇α r∇α ∇2 r + ∇α r∇β r∇α ∇β ∇2 r.
Hence, our calculations show that the third and fifth terms of (B.14) are equiv-
alent to the surface variations of the curvature, ΔS C . As a summing up, intro-
ducing the fast coordinate ω and multiplying equation (B.14) by ∂ω φ we obtain
the relation

2 2
∂ω φ∇ ∇ φdx = {−4 ∂ω4 φ∂ω φ
+ 2[C + ω(2G − C 2 ) + ω 2 (C 3 − 3GC) + ...]∂ω3 φ∂ω φ (B.20)

1
− −1 C(C 2 − 4G)(∂ω φ)2 − −1 ΔS C(∂ω φ)2 }dx.
2
Note that the terms associated with the first derivative, ∂ω φ, correspond to
the highest order considered in the expansion and at this order C = C .
Appendix C
List of publications
• K. A. Melzak, G. R. Lázaro, A. Hernández-Machado, I. Pagonabarraga, J.

M. Cárdenas Dı́az de Espada, and J. L. Toca-Herrera. AFM measurements
and lipid rearrangements: evidence from red blood cell shape changes. Soft
Matter, 8:7716–7726, 2012.
• G. R. Lázaro, K. A. Melzak, J. L. Toca-Herrera, I. Pagonabarraga, and A.
Hernández-Machado. Elastic energies and morphologies of the first stages of
the discoechinocyte transition. Soft Matter, 9:6430–6441, 2013.
• G. R. Lázaro, I. Pagonabarraga and A. Hernández-Machado. Phase field-
theories for mathematical modeling of biological membranes. Membrane me-
chanics: from the molecular to the cellular scale, Special Issue of Chemistry
and Physics of Lipids.
• G. R. Lázaro, A. Hernández-Machado, and I. Pagonabarraga. Rheology of
red blood cells under flow in highly confined microchannels: I. Effect of
elasticity. Soft Matter 2014.
• G. R. Lázaro, A. Hernández-Machado, and I. Pagonabarraga. Rheology of
red blood cells under flow in highly confined microchannels: II. Effect of
focusing and confinement. Soft Matter 2014 b.
• G. R. Lázaro, I. Pagonabarraga and A. Hernández-Machado. Mechanical
and dynamic properties of membrane phase-field models. (submitted)
• G. R. Lázaro, A. Hernández-Machado and I. Pagonabarraga. Asymmetric
shapes and lateral position of red blood cells in confined channels. (in prep.)
223
References
M. Abkarian, C. Lartigue, and A. Viallat. Tank treading and unbinding of de-

formable vesicles in shear flow: Determination of the lift force. Phys. Rev. Lett.,
88:068103, 2002.
M. Abkarian, M. Faivre, and A. Viallat. Swinging of red blood cells under shear
flow. Phys. Rev. Lett., 98:188302, 2007.
M. Abkarian, M. Faivre, R. Horton, K. Smistrup, C. A. Best-Popescu, and H. A.

Stone. Cellular-scale hydrodynamics. Biomed. Mater., 3(3):034011, 2008.
B. Alberts, D. Bray, M. Raff, K. Roberts, and J. D. Watson. Molecular Biology

of the Cell. Garland Science, New York, 1994.
J. Alcaraz, L. Buscemi, M. Grabulosa, X. Trepat, B. Fabry, R. Farré, and D. Nava-

jas. Microrheology of human lung epithelial cells measured by atomic force
microscopy. Biophys. J., 84(3):2071–2079, 2003.
J. A. Allen, R. A. Halverson-Tamboli, and M. M. Rasenick. Lipid raft mi-

crodomains and neurotransmitter signalling. Nature Rev. Neurosci., 8:128–140,
2007.
B. Angelov and I. Mladenov. On the Geometry of the Red Blood Cell, Geometry,
Integrability and Quantization. Coral Press Scientific Publishing, Sofia, 2000.
G. M. Artmann, K.-L. Paul Sung, T. Horn, D. Whittemore, G. Norwich, and

S. Chien. Micropipette apiration of human erythrocytes induces echinocytes
via membrane phospholipid translocation. Biophys. J., 72:1434–1441, 1997.
E. S. Asmolov. The inertial lift on a spherical particle in a plane poiseuille flow

at large channel reynolds number. J. Fluid. Mech., 381:63–87, 1999.
O. K. Baskurt and H. J. Meiselman. Blood rheology and hemodynamics. Seminars

in Thrombosis and Hemostasis, 29(5):435–450, 2003.
225
226 REFERENCES
O. K. Baskurt, M. Hardeman, M. W. Rampling, and H. J. Meiselman. Handbook

of Hemorheology and Hemodynamics. IOS Press., Amsterdam, Netherlands.,
2007.
T. Baumgart, S. Hess, and H. Webb. Imaging coexisting fluid domains in biomem-

brane models coupling curvature and line tension. Nature, 425:821824, 2003.
V. Bennett. The spectrin-actin junction of erythrocyte membrane skeletons.

Biochim. Biophys. Acta, 988(1):107–121, 1989.
R. Benzi, M. Sbragaglia, S. Succi, M. Bernaschi, and S. Chibbaro. Mesoscopic

lattice boltzmann modeling of soft-glassy systems: theory and simulations. J.
Chem. Phys., 131:104903, 2009.
M. Besis. Living Blood Cells and Their Ultrastructure. Springer, New York, 1973.
T. Betz, M. Lenz, J.-F. Joanny, and C. Sykes. Atp-dependent mechanics of red

blood cells. Proc. Nat. Acad. Sci. USA, 106(36):15312–15317, 2009.
P. L. Bhatnagar, E. P. Gross, and M. Krook. A model for collision processes

in gases. i. small amplitude processes in charged and neutral one-component
systems. Phys. Rev., 94:511–525, 1954.
T. Biben and C. Misbah. Tumbling of vesicles under shear flow within an advected-
field approach. Phys. Rev. E, 67:031908, 2003.
T. Biben, K. Kassner, and C. Misbah. Phase-field approach to three-dimensional

vesicle dynamics. Phys. Rev. E, 72:041921, 2005.
J. Brannick, C. Liu, T. Qian, and H. Sun. Diffuse interface methods for multiple
phase materials: an energetic variational approach. arXiv:1402.5375, page 1,
2014.
F. Brochard and J.-F. Lennon. Flicker spectrum of the flicker phenomenon in

erythrocytes. J. Phys. France, 11:1035, 1975.
J. E. Cahn and J. E. Hilliard. Free energy of a nonuniform system: Ii. nucleation

in a two-component incompressible fluid. J. Chem. Phys., 31:688–699, 1959.
F. Campelo and A. Hernández-Machado. Dynamic model and stationary shapes

of fluid vesicles. Eur. Phys. J. E, 20(1):37–45, 2006.
F. Campelo and A. Hernández-Machado. Model for curvature–driven pearling

instability in membranes. Phys. Rev. Lett., 99:088101, 2007a.
REFERENCES 227
F. Campelo and A. Hernández-Machado. Shapes instabilities in vesicles: A phase-

field model. Eur. Phys. J. Special Topics, 143:101–108, 2007b.
F. Campelo and A. Hernández-Machado. Polymer-induced tubulation in lipid

vesicles. Phys. Rev. Lett., 100:158103, 2008.
F. Campelo, A. Cruz, J. Pérez-Gil, L. Vázquez, and A. Hernández-Machado.

Phase-field model for the morphology of monolayer lipid domains. Eur. Phys.
J. E Soft Matter, 35(6):49, 2012.
P. B. Canham. The minimum energy of bending as a possible explanation of the

biconcave shape of the human red blood cell. J. Theor. Biol., 26:6181, 1970.
R. Capovilla and J. Guven. Stresses in lipid membranes. J. Phys. A: Math. Gen.,

35:6233, 2002.
D. Di Carlo, D. Irimia, R. G. Tompkins, and M. Toner. Continuous inertial

focusing, ordering, and separation of particles in microchannels. Proc. Nat.
Acad. Sci. U.S.A., 104(48):18892–18897, 2007.
D. Di Carlo, J. F. Edd, K. J. Humphry, H. A. Stone, and M. Toner. Particle

segregation and dynamics in confined flows. Phys. Rev. Phys., 102:094503,
2009.
M. E. Cates, J.-C. Desplat, P. Stansell, A. J. Wagner, K. Stratford, and I. Pag-

onabarraga. Physical and computational scaling issues in lattice boltzmann
simulations of binary fluid mixtures. Phil. Trans. R. Soc., 363:1917–1935, 2005.
M. E. Cates, O. Henrich, D. Marenduzzo, and K. Stratford. Lattice boltzmann

simulations of liquid crystalline fluids: active gels and blue phases. Soft Matter,
5:3791, 2009.
B. M. Cooke, N. Mohandas, and R. L. Coppel. Malaria and the red blood cell
membrane. Seminars in Hematology, 48(2):173–188, 2004.
G. M. Cooper and R. E. Hausman. The Cell. A Molecular Approach. Sinauer

Associates, Sunderland, USA., 5th edition, 2009.
G. Coupier, B. Kaoui, T. Podgorski, and C. Misbah. Noninertial lateral migration

of vesicles in bounded poiseuille flow. Phys. Fluids, 25:111702, 2008.
G. Degré, P. Joseph, P. Tabeling, S. Lerouge, M. Cloitre, and A. Ajdari. Rheology

of complex fluids by particle image velocimetry in microchannels. Appl. Phys.
Lett., 89:024104, 2006.
228 REFERENCES
J. Deschamps, V. Kantsler, and V. steinberg. Phase diagram of single vesicle

dynamical states in shear flow. Phys. Rev. Lett., 102(11):118105, 2009.
M. Deserno. Mesoscopic membrane physics: concepts, simulations, and selected

applications. Macromol. Rapid. Commun., 30:752–771, 2009.
J.-C. Desplat, I. Pagonabarraga, and P. Bladon. Ludwig: A parallel latticeboltz-

mann code for complex fluids. Comput. Phys. Comms., 134:273–290, 2001.
P. F. Devaux, A. Herrmann, N. Ohlwein, and M. M. Kozlov. How lipid flippases

can modulate membrane structure. Biochim. Biophys. Acta, 1778:1591–1600,
2008.
P. Dimitrakopoulos. Analysis of the variation in the determination of the shear

modulus of the erythrocyte membrane: Effects of constitutive law and mem-
brane modeling. Phys. Rev. E, 85:041917, 2012.
I. K. Dimov, L. Basabe-Desmonts, J. L. Garcia-Cordero, B. M. Ross, A. J. Riccoa,

and L. P. Lee. Stand-alone self-powered integrated microfluidic blood analysis
system (simbas). Lab Chip, 11:845–850, 2011.
R. Dimova, K. A. Rieske, S. Aranda, N. Bezlyepkina, R. L. Knorr, and

R. Lipowsky. Giant vesicles in electric fields. Soft Matter, 3:817–827, 2007.
D. E. Discher, D. H. Boal, and S. K. Boey. Phase transitions and anisotropic

responses on planar triangular nets under large deformation. Phys. Rev. E, 55:
4762–4772, 1997.
M. P. do Carmo. Differential Geometry of Curves and Surfaces. Prentice-Hall,

New Jersey, U.S.A., 1976.
A. M. Dondorp, P. A. Kager, J. Vreeken, and N. C. White. Abnormal blood flow

and red blood cell deformability in severe malaria. Parasitology Today, 16(6):
228–232, 2000.
Q. Du and M. Li. On the stochastic immersed boundary method with an implicit

interface formulation. Discrete and Continuous, 15(2):373–389, 2011.
Q. Du, C. Liu, , and X. Wang. J. Comput. Phys., 198:450, 2004.
Q. Du, C. Liu, R. RYham, and X. Wang. A phase-field formulation of the willmore

problem. Nonlinearity, 18:1249–1267, 2005.
B. Duenweg and A. J. C. Ladd. Lattice boltzmann simulations of soft matter

systems. Adv. Polym. Sci., 221:89, 2009.
REFERENCES 229
D. M. Engelman. Membranes are more mosaic than fluidic. Nature, 438:578–580,

2005.
E. Eremina, J. Hellmers, Y. Eremin, and T. Wriedt. Different shape models for

erythrocyte: Light scattering analysis based on the discrete sources method.
Jour. Quan. Spec. & Rad. Trans., 102:3–10, 2006.
E. Evans and D. Needham. Physical properties of surfactant bilayer membranes:

Thermal transitions, elasticity, rigidity, cohesion, and colloidal interactions. J.
Phys. Chem., 91:42194228, 1987.
E. A. Evans. Bending resistance and chemically induced moments in membrane

bilayers. Biophys. J., 14(12):923, 1974.
E. A. Evans and R. Skalak. Mechanics and Thermodynamics of Biomembranes.

CRC Press, Boca Raton, 1980.
R. Fahraeus and T. Lindqvist. The viscosity of the blood in narrow capillary

tubes. Am. J. Physiol., 96:562–568, 1931.
D. A. Fedosov, W. Pan, B. Caswell, G. Gompper, and G. E. Karniadakis. Pre-

dicting human blood viscosity in silico. Proc. Nat. Acad. Sci. U.S.A., 108(29):
11772–11777, 2011.
J. E. Ferrell, K.J. Lee, and W. H. Huestis. Biochemistry, 24:2849–2857, 1985.
T. M. Fischer. Shape memory of human red blood cells. Biophys. J., 86(5):
3304–3313, 2004.
R. Folch, J. Casademunt, A. Hernández-Machado, and L. Ramı́rez-Piscina. Phase-

field model for hele-shaw flows with arbitrary viscocity contrast. i. theoretical
approach. Phys. Rev. E, 60:1724, 1999.
A. M. Forsyth, J. Wan, W. D. Ristenpart, and H. A. Stone. The dynamic behavior

of chemically stiffened red blood cells in microchannel flows. Microvas. Res.,
80:37–43, 2011.
Y. C. Fung. Biomechanics: Circulation. Springer, New York, 1997.
J. Gachelin, G. Mi no, H. Berthet, A. Lindner, A. Rousselet, and E. Clément.

Non-newtonian viscosity of escherichia coli suspensions. Phys. Rev. Lett., 110:
268103, 2013.
P. Gaehtgens, C. Duhrssen, and K. H. Albrecht. Motion, deformation, and inter-

action of blood cells and plasma during flow through narrow capillary tubes.
Blood Cells, 6:799–812, 1980.
230 REFERENCES
P. G. Gallagher. Red blood cell membrane disorders. Hematology AM. Soc.

Hematology Educ. Program, 13:8, 2005.
G. Gompper and M. Schick. Self assembling amphiphilic systems. Academic Press

Limited, London, 1994.
G. Gompper and S. Zschocke. Elastic properties of interfaces in a ginzburg–landau

theory of swollen micelles, droplet crystals and lamellar phases. Europhys. Lett.,
16:731, 1991.
G. Gompper and S. Zsckocke. Ginzburg-landau theory of oil-water-surfactant

mixtures. Phys. Rev. A, 46(8):4836, 1992.
G. Gonnella, E. Orlandini, and J. M. Yeomans. Spinodal decomposition to a

lamellar phase: effects of hydrodynamic flow. Phys. Rev. Lett., 78:1695, 1997.
N. S. Gov. Active elastic network: Cytoskeleton of the red blood cell. Phys. Rev.
E, 75:011921, 2007.
N. S. Gov and S. A. Safran. Red blood cell membrane fluctuations and shape
controlled by atp-induced cytoskeletal defects. Biophys. J., 88:1859–1874, 2005.
P. Guillot, P. Panizza, J.-B. Salmon, M. Joanicot, A. Colin, C.-H. Bruneau, and

T. Colin. Viscosimeter on a microfluidic chip. Langmuir, 22:6438–6445, 2006.
J. E. Hall. Guyton and Hall: Textbook of Medical Physiology. Elsevier Saunders,

Philadelphia, U.S.A., 11th edition, 2011.
M. Han, C. Kim, M. Kim, and S. Lee. J. Rheol., 43:1157–1174, 1999.
W. Helfrich. Elastic properties of lipid membranes: theory and possible experi-

ments. Z. Naturforsch. C, 28:693703, 1973.
J. Hellmers, E. Eremina, and T. Wriedt. Simulation of light scattering by bicon-

cave cassini ovals using the nullfield method with discrete sources. Jour. Opt.
A: P and Appl. Opt., 8:1–9, 2006.
A. Hernández-Machado and D. Jasnow. Stability of a nonequilibrium steady-state

interface. Phys. Rev. A, 37(2):656, 1988.
A. Hernández-Machado, A. M. Lacasta, E. Mayoral, and E. Corvera Poiré. Phase-

field model of hele-shaw flows in the high-viscosity contrast regime. Phys. Rev.
E, 68:046310, 2003.
B. P. Ho and L. G. Leal. Inertial migration of rigid spheres in two-dimensional

unidirectional flows. J. Fluid. Mech., 65:365–400, 1974.
REFERENCES 231
J. Hoffman. Further musings on some red blood cell problems for perspicacious
physiologists. Blood Cells. Mol. Dis., 27:57, 2001.
M. Hu, D. H. de Jong, S. J. Marrink, and M. Deserno. Gaussian curvature elastic-

ity determined from global shape transformations and local stress distributions:
a comparative study using the martini model. Faraday Discuss., 161:365–382,
2013.
A. Ikai and R. Afrin. Toward mechanical manipulations of cell membranes and

membrane proteins using an atomic force microscope. Cell Biochem. Biophys.,
39:257–277, 2003.
K. Jacobson, O. G. Mouritsen, and R. G. Anderson. Lipid rafts: at a crossroad

between cell biology and physics. Nature Cell Biol., 9:7–14, 2007.
R. C. Jeffrey and J. R. A. Pearson. Particle motion in laminar vertical tube flow.

J. Fluid. Mech., 22:721–735, 1965.
B. Kaoui, G. Biros, and C. Misbah. Why do red blood cells have asymmetric
shapes even in a symmetric flow? Phys. Rev. Lett., 103:188101, 2009.
B. Kaoui, T. Krueger, and J. Harting. How does confinement affect the dynamics
of viscous vesicles and red blood cells? Soft Matter, 8:9246–9252, 2012.
S. R. Keller and R. Skalak. Motion of a tank-treading ellipsoidal particle in a

shear flow. J. Fluid Mech., 120:27, 1982.
V. M. Kendon, M. E. Cates, I. Pagonabrraga, J.-C. Desplat, and P. Blandon. Iner-

tial effects in three–dimensional spinodal decomposition of a symmetric binary
fluid mixture: a lattice boltzmann study. J. Fluid Mech., 440:147–203, 2001.
K. Khairy and J. Howard. Minimum-energy vesicle and cell shapes calculated

using spherical harmonics parameterization. Soft Matter, 7:2138–2143, 2011.
K. Khairy, J. Foo, and J. Howard. Shapes of red blood cells: Comparison of 3d

confocal images with the bilayer-couple model. Cell. Mol. Bioeng., 1:173–181,
2010.
M. M. Kozlov. Membrane shape equations. J. Phys. Cond. Matt., 18:1177–1190,

2006.
A. Kumar and M. D. Graham. Margination and segregation in confined flows of

blood and other multicomponent suspensions. Soft Matter, 8:10536, 2012.
T. Kuriabova and A. J. Levine. Nanorheology of viscoelastic shells: applications

to viral capsids. Phys. Rev. E, 77:031921, 2008.
232 REFERENCES
T. G. Kuznetsova, M. N. Starodubtseva, N. I. Yegorenkov, S. A. Chizhik, and

R. I. Zhdanov. Atomic force microscopy probing of cell elasticity. Micron, 38:
824–833, 2007.
A. J. C. Ladd. Numerical simulations of particulate suspensions via a discretized

boltzmann equation. part 2. numerical results. J. Fluid Mech., 271:285–309,
1994.
A. J. C. Ladd and R. Verberg. Lattice-boltzmann simulations of particle-fluid

suspensions. J. Stat. Phys., 104:1191–1251, 2001.
C. A. Lamontagne, C. M. Cuerrier, and M. Grandbois. Afm as a tool to probe

and manipulate cellular processes. Pflugers Arch., 456(1):61–70, 2008.
L. D. Landau and E. M. Lifshitz. Theory of Elasticity. Butterworth – Heinmann,

Oxford, 1999.
Y. Lange and J. Slayton. Interaction of cholesterol and lysophosphatidylcholine

in determining red cell shape. J. Lipid Res., 23:1121–1127, 1982.
J. C. M. Lee, D. T. Wong, and D. E. Discher. Direct measures of large, anisotropic

strains in deformation of erythrocyte cytoskeleton. Biophys. J., 77:853–864,
1999.
M. Leslie. Do lipid rafts exist? Science, 334:1046–1047, 2011.
A. Lewalle and K. H. Parker. Axisymmetric optical-trap measurement of red

blood cell membrane elasticity. J. Biomech. Eng., 133:011007, 2011.
J. Li, M. Dao, C. T. Lim, and S. Suresh. Spectrin-level modeling of the cy-

toskeleton and optical tweezers stretching of the erythrocyte. Biophys. J., 88:
3707–3719, 2005.
J. Li, G. Lykotrafitis, M. Dao, and S. Suresh. Cytoskeletal dynamics of human

erythrocyte. Proc. Nat. Acad. Sci. USA, 104:4937–4942, 2007.
P. Li, Z. S. Stratton, M. Dao, J. Ritz, and T. J. Huang. Probing circulating tumor

cells in microfluidics. Lab on a chip, 13:602–609, 2013.
Y. Li. Wavenumber-extended high-order upwind-biased finite-difference schemes

for convective scalar transport. J. Comp. Phys., 133(2):235–255, 1997.
G. H. W. Lim, M. Wortis, and R. Mukhopadhyay. Stomatocyte–discocyte–

echinocyte sequence of the huan red blood cell: Evidence for the bilayer cou-
ple hypothesis from membrane dynamics. Proc. Nat. Acad. Sci. USA, 99(26):
16766–16769, 2002.
REFERENCES 233
R. Lima, M. Nakamura, T. Omori, T. Ishikawa, S. Wada, and T. Yamaguchi. 1st

ECCOMAS Thematic Conference on Computational Vision and Medical Image
Processing. Advances in Computational Vision and Medical Image Processing:
Methods and Applications., chapter Microscale flow dynamics of red blood cells
in microchannels: an experimental and numerical analysis. Springer, 2009.
Q. H. Liu, Z. Haijun, J. X. Liu, and O. Y. Zhong-Can. Spheres and prolate

and oblate ellipsoids from an analytical solution of the spontaneous-curvature
fluid-membrane model. Phys. Rev. E, 60:3227–3233, 1999.
S. C. Liu, L. H. Derick, and J. Palek. Alteration of the erythrocyte-membrane

skeletal ultrastructure in hereditary spherocytosis, hereditary elliptocytosis, and
pyropoikilocytosis. Blood, 76:198–205, 1990.
I. López-Montero, N. Rodrı́guez abd S. Cribier, A. Pohl, M. Vélez, and P. F.

Devaux. Rapid transbilayer movement of ceramides in phospholipid vesicles
and in human erythrocytes. J. Biol. Chem., 280(27):25811–25819, 2005.
A. Malevanets and R. Kapral. Mesoscopic model for solvent dynamics. J Chem.

Phys., 110:8605–8613, 1999.
S. Manno, Y. Takakuwa, and N. Mohandas. Modulation of erythrocyte membrane

mechanical function by protein 4.1 phosphorylation. J. Biol. Chem., 280(9):
7581–7587, 2005.
J. F. Marko and E. D. Siggia. Stretching dna. Macromolecules, 28(26):8759–8770,

1995.
S. J. Marrink and D. P. Tieleman. Perspective on the Martini Model. Chem. Soc.

Rev., 42:6801–6822, 2013.
S. J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tieleman, and A. H. de Vries.

The MARTINI Force Field: Grained Model for Biomolecular Simulations. J.
Phys. Chem. B, 111:7812–7824, 2007.
R. Martino and P. Zampirollo. A mathematical analysis of the disk-sphere tran-

sition of the human red cell. Experientia, 34(4):466–467, 1978.
J.-P. Matas, J. F. Morris, and E. Guazzelli. Inertial migration of rigid spherical

particles in poiseuille flow. J. Fluid. Mech., 515:171–195, 2004.
E. D. Matayoshi. Distribution of shape-changing compounds across the red cell

membrane. Biochemistry, 19:3414–3422, 1980.
234 REFERENCES
J. L. McWhirter, H. Noguchi, and G. Gompper. Flow-induced clustering and

alignment of vesicles and red blood cells in microcapillaries. Proc. Nat. Acad.
Sci. USA, 106(15):6039–6043, 2008.
J. L. McWhirter, H. Noguchi, and G. Gompper. Flow-induced clustering and

alignment of vesicles and red blood cells in microcapillaries. Proc. Nat. Acad.
Sci. U.S.A., 106(15):6039–6043, 2009.
G. R. Lázaro K. M. Melzak, J. L. Toca-Herrera, I. Pagonabarraga, and

A. Hernández-Machado. Elastic energies and morphologies of the frst stages
of the discoechinocyte transition. Soft Matter, 9:6430–6441, 2013.
K. A. Melzak, G. R. Lázaro, A. Hernández-Machado, I. Pagonabarraga, J. M.

Cárdenas-Dı́az de Espada, and J. L. Toca-Herrera. Afm measurements and
lipid rearrangements: evidence from red blood cell shape changes. Soft Matter,
8:7716–7726, 2012.
E. W. Merrill, E. R. Gilliland, T. S. Lee, and E. W. Salzman. Blood rheology:

effect of fibrinogen decuded by addition. Circ. Res., 18:437–446, 1966.
H. Mohamed, L. D. McCurdy, D. H. Szarowski, S. Duva, J. N. Turner, and

M. Caggana. Development of a rare cell frationation device: application for
cancer detection. IEEE Trans. Nanobiosci., 3(4):251–256, 2004.
R. Mukhopadhyay, G. Lim H. W., and M. Wortis. Echinocyte shapes: bending,

stretching, and shear determine spicule shape and spacing. Biophys. J., 82:
1756–1772, 2002.
M. Nakao, N. Nakao, and S. Yamazoe. Adenosine triphosphate and maintenance

of shape of the human red cells. Nature (London), 187:945–946, 1960.
V. Narsimhan, H. Zhao, and E. S. G. Shaqfeh. Coarse-grained theory to predict

the concentration distribution of red blood cells in wall-bounded couette flow
at zero reynolds number. Phys. Fluids, 25:061901, 2013.
P. Nghe, P. Tabeling, and A. Ajdari. Flow-induced polymer degradation probed

by a high throughput microfluidic set-up. J. Non-Newtonian FLuid Mech., 165:
313–322, 2010.
H. Noguchi and G. Gompper. Shape transitions of fluid vesicles and red blood cells
in capillary flows. Proc. Nat. Acad. Sci. U.S.A., 102(40):14159–14164, 2005.
M. Oishi, H. Kinoshita, T. Fuiji, and M. Oshima. Continuous and simultaneous

measurement of the tank-treading motion of red blood cels and the surrounding
flow using translational confocal micro-particle image velocimetry (micro-piv)
with sub-micron resolution. Meas. Sci. Technol., 23:035301, 2012.
REFERENCES 235
P. Olla. The lift on a tank treading ellipsoidal cell in a shear flow. J. Phys. II
France, 7:1533–1540, 1997.
P. Olla. Simplified model for red blood cell dynamics in small blood vessels. Phys.
Rev. Lett., 82(2):453, 1999.
Z. Ou-Yang and W. Helfrich. Instability and deformation of a spherical vesicle by

pressure. Phys. Rev. Lett., 59:2486–2488, 1987.
Y. Park, C. A. Best, T. Auth, N. S. Gov, S. A. Safran, G. Popescu, S. Suresh,

and M. S. Feld. Proc. Nat. Acad. Sci. USA, 107(4):1289–1294, 2010a.
Y. Park, C. A. Best, K. Badizadegan, R. R. Dasari, M. S. Feld, T. Kriabova, M. L.

Henle, A. J. Levine, and G. Popescu. Measurements of red blood cell mechanics
during morphological changes. Proc. Nat. Acad. Sci. USA, 107(15):6731–6736,
2010b.
Z. Peng, R. J. Asaro, and Q. Zhu. Multiscale simulation of erythrocyte membranes.

Phys. Rev. E, 81:031904, 2010.
Z. Peng, X. Li, I. V. Pivkin, M. Dao, G. E. Karniakidis, and S. Suresh. Lipid

bilayers and cytoskeletal interactions in a red blood cell. Proc. Nat. Acad. Sci.
USA, 110:13356–13361, 2013.
C. S. Peskin. The immersed boundary method. Acta Numerica, 11:1–39, 2002.
A. G. Petrov and J. Bivas. Elastic and flexoelectric aspects of out-of-plane fluc-

tuations in biological and model membranes. Progress. Surf. Sci., 18:389, 1984.
J. C. Pinder, D. Bray, and W. B. Gratzer. Control of interaction of spectrin and

actin by phosphorylation. Nature, 270:752–754, 1977.
M. S. Pommer and C. D. Meinhart. Proceedings of the 6th International sympo-

sium on particle image velocimetry PIV’05., chapter Shear-stress distribution
surrounding individual adherent red cells in a microchannel measures using
micro-PIV. 2005.
A. J. Pons and A. Karma. Helical crack-front instability in mixed-mode fracture.

Nature, 464:85–89, 2010.
C. Pozrikidis. Boundary Integral & Singularity Methods for Linearised Viscous

FLow. Cambridge University Press, New York, 1992.
C. Pozrikidis. Finite deformation of liquid capsules enclosed by elastic membranes

in simple shear flow. Soft Matter, 297:123–152, 1995.
236 REFERENCES
W. Rawicz, K. C. Olbrich, T. McIntosh, D. Needham, and E. Evans. Effect of

chain length and unsaturation on elasticity of lipid bilayers. Biophys. J., 79(1):
328–339, 2000.
H. R. Reinhart and S. Chien. Red cell rheology in stomatocyte-echinocyte trans-

formation: roles of cell geometry and cell shape. Blood, 67:1110–1118, 1986.
S. B. Schwarz, B. Deuticke, and C. W. Haest. Adenosine triphosphate and main-

tenance of shape of the human red cells. Mol. Membr. Biol., 16:247–255, 1999.
T. W. Secomb. Mechanics and computational simulation of blood flow in mi-

crovessels. Med. Eng. Phys., 33(7):800–804, 2011.
T. W. Secomb, B. Styp-Rekowska, and A. R. Pries. Two-dimensional simulation of

red blood cell deformation and lateral migration in microvessels. Ann. Biomed.
Eng., 35:755–765, 2007.
G. Segré and A. Silberberg. Behavior of macroscopic rigid spheres in poiseuille

ı̈ňĆow. J. Fluid. Mech., 14:136–157, 1962.
U. Seifert and R. Lipowsky. Structure and Dynamics of Membranes, Handbook

of Biological Sciences, vol. 1, chapter The morphology of vesicles. Elsevier,
Amsterdam, 1995.
U. Seifert, K. Berndel, and R. Lipowsky. Shape transformations of vesicles: phase

diagram for spontaneous-curvature and bilayer-couple models. Phys. Rev. A,
44(2):11821202, 1991.
J. Y. Shao, C. Shu, and Y. T. Chew. Development of an immersed boundary

phase field-lattice boltzmann method for neumann boundary condition to study
contact line dynamics. J. Comp. Phys., 234:8–32, 2013.
M. P. Sheetz and S. J. Singer. Biological membranes as bilayer couples. a molecular

mechanism of drug-erythrocyte interactions. Proc. Nat. Acad. Sci. USA, 71:
4457–4461, 1974.
L. Shi, T.-W. Pan, and R. Glowinski. Numerical simulation of lateral migration

of red blood cells in poiseuille flows. Phys. Rev. E, 85:016307, 2012.
H. Shiba and H. Noguchi. Estimation of the bending rigidity and spontaneous

curvature of fluid membranes in simulations. Phys. Rev. E, 84:031926, 2011.
J. C. Shillcock and R. Lipowsky. The computational route from bilayer membranes

to vesicle fusion. J. Phys.: Condens. Matter, 18:1191–1219, 2006.
REFERENCES 237
K. Simons and W. Vaz. Model systems, lipid rafts, and cell membranes. Annu.
Rev. Biophys. Biomol. Struct., 33:269–295, 2004.
S. J. Singer and G. L. Nicolson. The fluid mosaic model of the structure of cell
membranes. Science, 175:720–731, 1972.
N. Srivastava, R. D. Davenport, and M. A. Burns. A nanoliter viscometer for

analyzing plasma and other liquid samples. Annal. Chem., 77:383–392, 2005.
I. Steinbach. Phase-field models in materials science. Mod. Simul. Mater. Sci.

Eng., 17:073001, 2009.
B. T. Stokke, A. Mikkelsen, and A. Elgsaeter. Spectrin, human erythrocyte shapes,

and mechanical properties. Biophys. J., 49:319–327, 1986.
K. Stratford and I. Pagonabarraga. Parallel simulation of particle suspensions

with the lattice boltzmann method. Comput. Math. Appl., 55:1585–1593, 2008.
S. Succi. The Lattice Boltzmann Equation: For Fluid Dynamics and Beyond.
Oxford University Press, Norfolk, 2001.
S. Svetina, D. Kuzman, R. E. Waugh, P. Ziherl, and B. Zeks. The cooperative

role of membrane skeleton and bilayer in the mechanical behaviour of red blood
cells. Bioelectrochemistry, 62:107–113, 2004.
K. Svodoba, C. F. Schmidt, D. Branton, and S. M. Block. Conformation and

elasticity of the isolated red blood cell membrane skeleton. Biophys. J., 63:
784–793, 1992.
M. Tachibana. On the behaviour of a sphere in the laminar tube flows. Rheol.

Acta, 12:58–69, 1973.
N. Tahiri, T. Biben, H. Ez-Zahraouy, and C. Misbah. On the problem of slipper

shapes of red blood cells in the microvasculature. Microvasc. Res., 85:40–45,
2013.
O. Theissen, G. Gompper, and D. M. Kroll. Lattice-boltzmann model of am-

phiphilic systems. Europhys. Lett., 42(4):419–424, 1998.
G. B. Thurston. Viscoelasticity of human blood. Biophys. J., 12(9):1205–1217,

1972.
G. B. Thurston. Advances in Hemodynamics and Hemorheology, chapter Vis-

coelastic properties of blood and blood analogs. JAI Press. INc., 1996.
238 REFERENCES
G Tomaioulo, M. Simone, V. Martinelli, B. Rotoli, and S. Guido. Red blood cell

deformation in microconfined flow. Soft Matter, 5:3736–3740, 2009.
G. Tomaiuolo and S. Guido. Start-up shape dynamics of red blood cells in micro-
capillary flow. Microvasc. Res., 82:35–41, 2011.
M. Toner and D. Irimia. Blood-on-a-chip. Annu. Rev. Biomed. Eng., 7:77–103,

2005.
I. Tsafrir, D. Sagi, T. Arzi, M. A. Guedeau-Boudeville, V. Frette, D. Kandel, and

J. Stavans. Pearling instabilities of membrane tubes with anchored polymers.
Phys. Rev. Lett., 86:1138–1141, 2001.
I. Tsafrir, Y. Caspi, M. A. Guedeau-Boudeville, T. Arzi, and J. Stavans. Budding

and tubulation in highly oblate vesicles by anchored amphiphilic molecules.
Phys. Rev. Lett., 91:138102, 2003.
C. Uzoigwe. The human erythrocyte has developed the biconcave disc shape to
optimise the flow properties of the blood in the large vessels. Medical Hypotheses,
67(5):1159–1163, 2006.
G. van Meer, D. R. Voelker, and G. W. Feigenson. Membrane lipids: where they

are and how they behave. Nature Rev. Mol. Cell Biol., 9:112–124, 2008.
P. M. Vlahovska, T. Podgorski, and C. Misbah. Vesicles and red blood cells in

flow: From individual dynamics to rheology. C. R. Physique, 10:775–789, 2009.
H. Wang and R. Skalak. Viscous flow in a cylindrical tube containing a line of

spherical particles. J. Fluid. Mech., 38:75–96, 1969.
X. Wang and Q. Du. Modeling and Simulations of Multi-Component Lipid Mem-

branes and Open Membranes via Diffuse Interface Approaches. J. Math. Biol.,
56(3):347–371, 2008.
R. Waugh. Elastic energy of curvature-driven bump formation on red blood cell

membrane. Biophys. J., 70:1027–1035, 1996.
R. E. Waugh, J. Song, and B. Zeks. Local and nonlocal curvature elasticity in

bilayer membranes by tether formation from lecithin vesicles. Biophys. J., 61:
974–982, 1992.
J. G. White. Effects of an lonophore, a23187, on the surface morphology of normal

erythrocytes. Am. J. Pathol., 77:507–518, 1974.
P. Wong. A basis of echinocytosis and stomatocytosis in the disc-sphere transfo-

rations of the erythrocyte. J. Theor. Biol., 196:343–361, 1999.
REFERENCES 239
P. L. Yeagle. The Membranes of Cells. Academic Press, San Diego, CA, USA.,
2nd edition, 1993.
J. Yu, D. A. Fischman, and T. L. Steck. Selective solubilization of proteins

and phospholipids from red blood cell membranes by nonionic detergents. J.
Supramol. Struct., 1:233–248, 1973.
P. R. Zarda, S. Chien, and R. Skalak. Elastic deformations of red blood cells. J.

Biomech., 10:211–281, 1977.
K. Zeman, H. Engelhard, and E. Sackman. Bending undulations and elasticity of

the erythrocyte membrane: Effects of cell shape and membrane organization.
Eur. Biophys. J., 18:203–219, 1990.

GRL Thesis

Cargado por

Información del documentohacer clic para expandir la información del documento

Copyright:

Formatos disponibles

GRL Thesis

Cargado por

Información del documento

Título original

Derechos de autor

Formatos disponibles

Compartir este documento

Compartir o incrustar documentos

Opciones para compartir

¿Le pareció útil este documento?

¿Este contenido es inapropiado?

Copyright:

Formatos disponibles

GRL Thesis

Cargado por

Copyright:

Formatos disponibles

Red Blood Cell mechanics:

from membrane elasticity

Aquesta tesi doctoral està subjecta a la llicència Reconeixement- NoComercial –

Esta tesis doctoral está sujeta a la licencia Reconocimiento - NoComercial – CompartirIgual

This doctoral thesis is licensed under the Creative Commons Attribution-NonCommercial-

Guillermo Rodrı́guez Lázaro

Programa de Doctorat en Fı́sica

Barcelona, 7 de Mayo de 2014.

I Physical description of cell membranes 7

2 Physical approach to membranes 25

II Membrane elasticity and RBC morphological transi-

3 The disco-echinocyte transition 45

4 Membrane elasticity and RBC morphology 55

III A phase-ﬁeld method for membrane modeling 79

5 Phase-ﬁeld modeling of biological membranes 81

5.4 Macroscopic equations . . . . . . . . . . . . . . . . . . . . . . . . 98

5.4.1 Sharp-interface limit . . . . . . . . . . . . . . . . . . . . . 98

5.5 Linear stability analysis . . . . . . . . . . . . . . . . . . . . . . . 103

5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6 Numerical scheme 107

6.1 Lattice-Boltzmann method . . . . . . . . . . . . . . . . . . . . . . 107

6.1.1 Single-relaxation time lattice-Boltzmann . . . . . . . . . . 108

6.1.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . 109

6.2 Phase-ﬁeld integration . . . . . . . . . . . . . . . . . . . . . . . . 109

6.2.1 Advection . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6.2.2 Derivative calculations . . . . . . . . . . . . . . . . . . . . 110

6.2.3 Area conservation . . . . . . . . . . . . . . . . . . . . . . . 111

6.3 Numerical realization of tensionless interfaces . . . . . . . . . . . 114

6.4 Several cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

6.5 Parameter steering . . . . . . . . . . . . . . . . . . . . . . . . . . 117

IV Red Blood Cells ﬂow in conﬁned systems 119

7 Blood circulation 121

7.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

7.1.1 Blood properties . . . . . . . . . . . . . . . . . . . . . . . 121

7.1.2 Cell manipulation . . . . . . . . . . . . . . . . . . . . . . . 123

7.2 RBC behaviour in shear and parabolic ﬂows . . . . . . . . . . . . 126

7.2.1 Shear ﬂow . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

7.2.2 Parabolic ﬂow . . . . . . . . . . . . . . . . . . . . . . . . . 129

7.3 Scope of Part IV . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

8 Rheology and deformability of RBCs 133

9 On the mechanisms of RBC deformation and migration 151

10 RBC focusing 161

11 Collective ﬂow of RBCs 173

11.2.2 Migration and orientation in ordered conﬁgurations . . . . 182

12 Conclusions and future perspectives 195

13 Resumen en castellano 203

A Notes on diﬀerential geometry 213

B The sharp-interface limit 217

C List of publications 223

In recent years, the application of physical techniques and rapid technological

Elasticity of cell membranes

Thesis aims and structure

Attending to these objectives, the Thesis is structured in four separated parts.

based on the Helfrich theoretical framework. In Chapter 5, we dedicate special

Physical description of cell

1.1 Cell membrane

1.1.1 The lipid bilayer

Lipids represent up to 50% of the total mass of the membrane in mammalian

In this expression, nAT P is the ATP concentration and = (ΔEAT P − ΔEsa −