Biophysical Journal Volume 98 May 2010 2215–2225 2215

A Multiscale Red Blood Cell Model with Accurate Mechanics,

Rheology, and Dynamics

Dmitry A. Fedosov,† Bruce Caswell,‡ and George Em Karniadakis†*

†
Division of Applied Mathematics and ‡Division of Engineering, Brown University, Providence, Rhode Island

ABSTRACT Red blood cells (RBCs) have highly deformable viscoelastic membranes exhibiting complex rheological response
and rich hydrodynamic behavior governed by special elastic and bending properties and by the external/internal fluid and
membrane viscosities. We present a multiscale RBC model that is able to predict RBC mechanics, rheology, and dynamics in
agreement with experiments. Based on an analytic theory, the modeled membrane properties can be uniquely related to the exper-
imentally established RBC macroscopic properties without any adjustment of parameters. The RBC linear and nonlinear elastic
deformations match those obtained in optical-tweezers experiments. The rheological properties of the membrane are compared
with those obtained in optical magnetic twisting cytometry, membrane thermal fluctuations, and creep followed by cell recovery.
The dynamics of RBCs in shear and Poiseuille flows is tested against experiments and theoretical predictions, and the applicability
of the latter is discussed. Our findings clearly indicate that a purely elastic model for the membrane cannot accurately represent
the RBC’s rheological properties and its dynamics, and therefore accurate modeling of a viscoelastic membrane is necessary.

INTRODUCTION
A healthy human red blood cell (RBC) has a biconcave experiments (9–12). Theoretical models (10,13), which
shape with an average diameter of 7.8 mm. Its membrane attempt to describe this dynamics, indicate that such behav-
consists of a lipid bilayer with an attached cytoskeleton ior depends on membrane elastic properties, shear rate, and
formed by a network of the spectrin proteins linked by short viscosities of the membrane and of the internal/external
filaments of actin. The lipid bilayer is considered to be a fluids. RBC dynamics in capillary flow (14,15) is character-
nearly viscous and area-preserving membrane (1), whereas ized by a transition from biconcave to parachute shapes as
RBC elasticity is attributed to the attached spectrin network, the flow rate is increased. The transition appears to be gov-
as is the integrity of the entire RBC when subjected to severe erned by the RBC membrane elastic and bending properties,
deformations in the capillaries as small as 3 mm. The RBC as found in recent simulations (16).
membrane encloses a viscous cytosol (hemoglobin solution) To capture realistic behavior of RBCs, mechanical models
whose viscosity is several times larger than that of blood must address:
plasma under physiological conditions.
1. Membrane viscoelasticity with a viscous contribution of
Mechanical and rheological characteristics of RBCs and
the lipid bilayer and with an elastic contribution of the
their dynamics are governed by membrane elastic and viscous
spectrin network;
properties, bending resistance, and the viscosities of the
2. Membrane bending resistance;
external/internal fluids. Various RBC properties have been
3. Separate external/internal fluids with distinct viscosities.
measured in a number of experiments, which include micro-
pipette aspiration (2,3), RBC deformation by optical tweezers Several RBC models have been recently developed at
(4,5), optical magnetic twisting cytometry (6), and three- the continuum level (17,18) and at the mesoscopic level
dimensional measurements of membrane thermal fluctua- (16,19–23). Continuum models (17,18) of deformable parti-
tions (7,8). The micropipette aspiration and optical tweezers cles use immersed boundary or boundary integral methods.
techniques tend to deform the whole RBC membrane, to yield They employ the same external/internal fluids and often
values of the macroscopic shear modulus of healthy cells in use simplified treatments of the membrane (e.g., purely
the range of 2–12 mN/m. The optical magnetic twisting elastic membrane) that fail to capture the viscoelastic proper-
cytometry and measurements of membrane thermal fluctua- ties of real RBCs. In addition, continuum models omit some
tions probe locally for membrane characteristics and furnish phenomena at the mesoscopic scales such as membrane
measurements of local rheological properties (e.g., the com- thermal fluctuations.
plex modulus). These experiments show that the mechanical Detailed mesoscopic modeling of RBCs at the spectrin
response of the membrane is clearly viscoelastic. level (19,20) is much limited by the demanding computa-
RBCs in shear flow respond by tumbling at low shear tional expense. In these works, the RBC membrane was
rates and tank-treading at high shear rates, as observed in modeled as a network of elastic springs with fixed connec-
tivity yielding a purely elastic membrane. To improve
Submitted October 27, 2009, and accepted for publication February 1, 2010. computational efficiency, several mesoscopic coarse-grained
*Correspondence: [email protected] models have been developed (16,21–23) using a similar
Editor: Reinhard Lipowsky. network membrane representation with many fewer vertices
 2010 by the Biophysical Society
0006-3495/10/05/2215/11 $2.00 doi: 10.1016/j.bpj.2010.02.002
2216 Fedosov et al.

than at the spectrin level. Noguchi and Gompper (16) used rather than an individual atom. DPD particles interact
multiparticle collision dynamics (24) and employed two through pairwise soft potentials, whose description can be
coupled networks to represent a membrane: one with a fixed found in the Supporting Material.
connectivity to mimic the spectrin cytoskeleton, and the
other with dynamic flipping between bonds to model the vis- Red blood cell model
cous lipid bilayer. Their model takes into account the sepa- The membrane model is built as a set of vertex points {xi},
ration between external and internal fluids and incorporates i ˛ 1.Nv (DPD particles) that form a two-dimensional trian-
the main RBC membrane properties; however, it is not clear gulated network, similar to that of Discher et al. (19) and Li
whether the model can simultaneously reproduce appropriate et al. (20). The vertices are connected by Ns edges forming Nt
RBC mechanics, rheology, and dynamics. Dupin et al. (21) triangles. The potential energy of the system is defined as
coupled a single elastic RBC network to a fluid described
by the lattice Boltzmann method (25). Despite very prom- Vðfxi gÞ ¼ Vinplane þ Vbending þ Varea þ Vvolume : (1)
ising results, their model does not account for membrane
The in-plane elastic energy mimics the elastic spectrin
viscosity and thermal fluctuations, and the same external
network, and is given by
and internal fluids are employed. Lim et al. (23) employed  
the area-difference-elasticity model (26), which uses two " #
X kB Tlm 3xj2  2xj3 kp
coupled elastic networks, to capture the full stomatocyte- Vinplane ¼   þ ; (2)
discocyte-echinocyte sequence of RBC shapes that may j˛1.Ns 4p 1  xj ðn  1Þln1
j

develop under special conditions. Under physiological

conditions, healthy RBCs retain their biconcave shape, where lj is the length of the spring j, lm is the maximum
which can be properly reproduced by a single network. In spring extension, xj ¼ lj/lm, p is the persistence length, kBT
addition, RBCs simulated with the area-difference-elasticity is the energy unit, kp is the spring constant, and n is a power.
model were only considered in statics, and thus further Note that the spring forces in membrane are a combination of
model development is required to capture RBC rheology conservative elastic forces that can be expressed in terms of
and dynamics. Pivkin and Karniadakis (22) used dissipative the energy potential above, and dissipative forces, which are
particle dynamics (27) to simulate a coarse-grained RBC rep- to be defined below. The first term in Eq. 2 is the attractive
resented by a single elastic spring network—which is the wormlike chain (WLC) potential, and the second term
starting point of this work. defines a repulsive force for n > 0 to be called the power
The RBC model in this article is constructed by a network force (POW), so that we abbreviate this spring model as
of viscoelastic springs combined with bending energy and WLC-POW. A nonzero equilibrium spring length is defined
constraints for surface-area and volume conservation. The by the balance of these two forces.
model is multiscale, as the RBC can be represented on the The bending energy represents the bending resistance of
spectrin level, where each spring in the network corresponds the lipid bilayer, and is defined as
X   
to a single spectrin tetramer with the equilibrium distance Vbending ¼ kb 1  cos qj  q0 ; (3)
between two neighboring actin connections of ~75 nm. On j˛1.Ns
the other hand, the RBC network can also be highly
coarse-grained with the equilibrium spring lengths of up to where kb is the bending constant, qj is the instantaneous angle
~500–600 nm. Theoretical analysis of the hexagonal between two adjacent triangles having the common edge j,
network yields its macroscopic elastic and dynamic proper- and q0 is the spontaneous angle.
ties explicitly in terms of its microscopic parameters, and The area and volume conservation constraints, which
thus completely eliminates adjustment of the model parame- account for area-incompressibility of the lipid bilayer and
ters. The multiscale RBC model is able to reproduce realistic incompressibility of the inner cytosol, respectively, are ex-
RBC mechanical and rheological properties and their pressed as
dynamics. The model is general enough to be used with other  2  
ka A  Atot
0
X kd Aj  A0 2
simulation methods such as lattice Boltzmann, Brownian Varea ¼ þ ; (4)
dynamics, the immersed boundary method, multiparticle 2Atot
0 j˛1.Nt
2A0
collision dynamics, etc.  2
kv V  V0tot
Vvolume ¼ ; (5)
MODEL FRAMEWORK 2V0tot

We employ dissipative particle dynamics (DPD) (27), where ka, kd, and kv are the global area, local area, and
a mesoscopic simulation technique, to model the motion of volume constraint coefficients, respectively. The terms A
the RBC membrane and of the internal and external fluids. and V are the total area and volume of RBC, while A0tot and
A DPD system consists of N point particles, where each V0tot are the specified total area and volume, respectively.
particle corresponds to a collection of atoms or molecules Note, that the above expressions define global area and

Biophysical Journal 98(10) 2215–2225

Multiscale Red Blood Cell Model 2217

volume constraints, and the second term in Eq. 4 incorporates on Eq. 6 and a given m0. After this modification, a new
the local dilatation constraint. Detailed description and network that is virtually free of irregularities appears.
discussion of the RBC model can be found in Fedosov (28).
Membrane viscoelasticity
Mechanical properties To incorporate viscous dissipation of the lipid bilayer into
To obtain a relationship between macroscopic elastic proper- the RBC membrane, the spring definition is modified by
ties (shear, area-compression, and Young’s moduli) of the adding viscous contribution through dissipative and random
network and model parameters, we extend the linear analysis forces. Such a term fits naturally in the DPD method, where
of Dao et al. (29) for a regular hexagonal network having interparticle dissipative interactions are an intrinsic part of
the above energies (see the Supporting Material for details). the method. We follow the general framework of the fluid
The membrane shear modulus is given by particle model (31) to define dissipative FD ij and random
pffiffiffi ! FRij forces for each spring as
3 kB T x0 1 1  
m0 ¼  þ FDij ¼ gT vij  gc vij  eij eij ; (8)
4plm x0 2ð1  x0 Þ3 4ð1  x0 Þ2 4
(6)  
pffiffiffi
3kp ðn þ 1Þ pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffitr dWij

þ ; FRij dt ¼ 2kB T T S
2g dWij þ 3g g C T 1  eij ;
4ln0 þ 1 3
(9)
where l0 is the equilibrium spring length and x0 ¼ l0/lm. The
corresponding area-compression and Young’s moduli are where gT and gC are dissipative parameters, vij is the relative
found as velocity of spring ends, tr[dWij] is the trace of a random
matrix of independent Wiener increments dWij, and
K ¼ 2m0 þ ka þ kd ;
dWSij ¼ dWSij  tr½dWSij 1=3 is the traceless symmetric part.
4Km0 (7) Note that the last equation imposes the condition 3gC > gT
Y ¼ : and these forces satisfy the fluctuation-dissipation balance
K þ m0
providing consistent temperature of the RBC membrane in
The bending coefficient kb of Eq. 3 can be expressed in equilibrium. The membrane shear viscosity hm is related to
terms of the macroscopicpbending ffiffiffi rigidity kc of the Helfrich the dissipative parameters gT, gC (see (28) for details) as
model (30) as kb ¼ 2kc = 3. Note that this expression may pffiffiffi C
not represent actual bending resistance of the RBC mem- pffiffiffi 3g
hm ¼ 3gT þ : (10)
brane as bending of the membrane also results in local 4
in-plane deformations. Thus, this relation characterizes Our experience indicates that gT accounts for a large portion
contribution of the defined bending energy with respect to of viscous contribution, and therefore gC is set to gT/3 in all
the macroscopic bending model with zero in-plane shear simulations.
modulus.
The necessary model parameters can be calculated from RBC-solvent boundary conditions
the equations above for given macroscopic RBC properties,
thereby circumventing manual adjustment. An equilibrium The internal and external fluids are modeled by a number
simulation of such a RBC reveals that the cell surface may of free DPD particles. External/internal fluid separation
develop local bumps manifested as stress anomalies in a (nonmixing) is enforced by bounce-back reflections of fluid
membrane triangulation because all springs have the same particles at a moving membrane surface. In addition, the
equilibrium length, although a network on a nondevelopable no-slip boundary conditions at the membrane surface are
surface cannot be constructed with triangles having the same enforced through an appropriate choice of a dissipative force
edge lengths. The degree of local stress artifacts depends between fluid particles and membrane vertices. We refer the
on the regularity of triangulation and the ratio of the mem- reader to the Supporting Material and to Fedosov (28) for
brane modulus of elasticity to the bending rigidity expressed more details on boundary conditions.
2
Föppl-von
tot
ffi Kármán number k ¼ YR0 /kc, where
by thepffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Scaling of model and physical units
R0 ¼ A0 =ð4pÞ:
To fully eliminate these anomalies, a stress-free model is The dimensionless constants and variables in the DPD model
obtained by computational annealing. For each spring, the must be scaled with physical units. The superscript M
equilibrium spring length l0i is adjusted to be the edge length denotes that a quantity is in model units, while P identifies
after triangulation for i ¼ 1,.,Ns. The maximum spring physical units (SI units). We define the length scale as
extension is then set individually to lmi ¼ l0i  x0,
where x0 ¼ 2.2; see Fedosov (28). In addition, the spring DP0
rM ¼ m; (11)
parameters are calculated individually for each spring based DM
0

Biophysical Journal 98(10) 2215–2225

2218 Fedosov et al.

where rM is the model unit of length, D0 is the cell diameter, a

and m stands for meters. The energy per unit mass (kBT) and
N- N+
the force unit (N denotes Newton) scales are given by

2 DT
M Y P DP0
ðkB TÞ ¼ M ðkB TÞP ;
Y DM 0
(12)
M Y P DP0 P
N ¼ M MN ; DA
Y D0
Stretch
where Y is the membrane Young’s modulus. The timescale is
defined as b Experiment
Nv = 500
P P M
a 20 Nv = 1000
D0 h Y Nv = 3000
t ¼ s; (13) 18 Nv = 9128
DM
0 h
M YP
x Nv = 27344
x
16 x x
x x
x x
where h is the characteristic viscosity (e.g., internal/external 14 x x
x DA

diameter ( m m)
x
fluids or membrane) and a is a chosen scaling exponent 12 x
x
x
similar to the power-law exponent in rheology. 10
x
x
8x
x x x x
6 x x x
RESULTS x x x x x x x x
4
x
In this section, the RBC model is compared against several 2
DT
available experiments that examine RBC mechanics, rhe-
0
ology, and dynamics. First, stretching simulations of 0 50 100
force (pN)
150 200

modeled RBCs are performed and compared with RBC

deformation by optical tweezers (5). Second, the rheological FIGURE 1 Schematic RBC deformation (a) and stretching response of
properties of the modeled membrane are validated against the stress-free RBC model for different coarse-graining levels (b) compared
with the experiments of Suresh et al. (5). Nv ¼ 27,344 corresponds to the
optical magnetic twisting cytometry experiments (6) and spectrin-model.
against experimental measurements of membrane thermal
fluctuations (7,32). In addition, RBC rheological characteris-
tics are tested in a creep test and cell extensional recovery in spondence between simulations and experiments is achieved
comparison with those in the literature (6,33). Finally, RBC for m0 ¼ 6.3 mN/m and Y ¼ 18.9 mN/m independently of the
dynamics in shear and Poiseuille flows is simulated and level of coarse-graining. These are also in agreement with the
compared to RBC shearing experiments (9–12) and theories RBC mechanical properties that Suresh et al. (5) extracted
(10,13), and to experiments of RBCs in a tube flow (14,15). from the optical tweezers experiments by finite element
simulations. The small discrepancy between simulated and
Stretching test experimental transverse diameters is probably a consequence
of the optical measurements being performed from only
The RBC membrane network is characterized by Nv vertices,
a single observation angle. Numerical simulations showed
which define the level of membrane representation or coarse-
that RBCs subjected to stretching tend to rotate in the y-z
graining from the spectrin-level (Nv ¼ 27,344) to the highly
plane, and therefore measurements from a single observation
coarse-grained network of Nv ¼ 500. Fig. 1 a shows a sketch
angle may result in underprediction of the maximum trans-
of the RBC membrane under deformation. The total stretch-
verse diameter. However, the simulation results remain
ing force fM is applied to N– and Nþ vertices (drawn as small
within the experimental error bars.
black spheres, N– ¼ Nþ ¼ 3Nv) along the negative and the
positive directions, respectively. These vertices cover a
Membrane rheology from twisting torque
near-spherical area on the RBC surface with 3 ¼ 0.02, which
cytometry
corresponds to the contact diameter of the attached silica
bead with diameter 2 mm used in experiments (5). Note Twisting torque cytometry (TTC) is the numerical analog of
that the viscous properties of the membrane and of the the optical magnetic twisting cytometry (OMTC) used in the
suspending medium do not affect final stretching, because experiments (6), in which a ferrimagnetic microbead is
RBC response is measured after convergence to the equilib- attached to RBC top and is subjected to an oscillating
rium stretched state is achieved for given force. Fig. 1 b magnetic field. In simulations, a microbead is attached to
compares the simulated axial and transverse RBC diameters the modeled membrane, and is subjected to an oscillating
with their experimental counterparts (5). Excellent corre- torque as shown in Fig. 2 a. In analogy with the experiments,

Biophysical Journal 98(10) 2215–2225

Multiscale Red Blood Cell Model 2219

oscillating torque per unit volume, but it is shifted by a phase angle f depend-
a ing on the frequency. The phase angle can be used to derive
components of the complex modulus according to linear
rheology as
0 DT
displacement g ðuÞ ¼ cosðfÞ;
Dd
(14)
00 DT
g ðuÞ ¼ sinðfÞ;
Dd
where g0 (u) and g00 (u) are the two-dimensional storage and
loss moduli (G0 and G00 in three dimensions), and DT and
Dd are the torque and bead displacement amplitudes. Note
that under the assumption of no inertial effects, the phase
angle satisfies the condition 0 % f < p/2.
Fig. 3 presents components of the complex modulus
compared with the experimental data of Puig de Morales-
Marinkovic et al. (6). A good agreement of the membrane
moduli in simulations with the experimental data is found
for the bending rigidity kc ¼ 4.8  1019 J and the mem-
b w = 33.1 Hz brane viscosity hm ¼ 0.022 Pa $ s. Note that this corresponds
1 300
Torque Displacement to a bending rigidity that is two times larger than the widely
accepted value of 2.4  1019 J. In general, simulations for
Torque per unit volume (Pa)

0.75
200 various Young’s moduli and bending rigidities showed the
dependence of g0 (u) ~ kc0.65Y00.65. In Fig. 3, only the
Displacement (nm)

0.5
100
membrane bending rigidity is varied, as the Young’s
0.25 modulus was obtained in the RBC stretching tests above.
The loss modulus appears to be independent of the RBC
0 0 elastic properties and is governed by the membrane
viscosity. The modeled g00 exhibits the exponent 0.85 with
−0.25
−100 respect to frequency, while the exponent 0.64 was observed
−0.5
in the experiments. This discrepancy may result from both
f
−200
−0.75
−1
10
−1 −300
g’ − experiment (OMTC)
0 2 4 6 8 10 12 g’’ − experiment (OMTC)
Dimensionless time − tw g’ −kc= 2.4×10−19 J ω0.85
−19
−2 g’ −kc= 4.8×10 J
FIGURE 2 A setup of the TTC (a) and the characteristic response of a 10
g’’ − ηm= 0.04 Pa⋅s
g’ and g’’ (Pa/nm)

microbead subjected to an oscillating torque (b).

g’’ −ηm= 0.022 Pa⋅s
g’’ −ηm= 0.01 Pa⋅s
the modeled RBC is attached to a solid surface, where the −3
10
wall-adhesion is modeled by keeping stationary 15% of
vertices on the RBC bottom, although other vertices are
free to move. The adhered RBC is filled and surrounded
−4 0.65
by fluids having viscosities much smaller than the membrane 10 ω
viscosity, and therefore, only the membrane viscous contri-
bution is measured. The microbead is simulated by a set of
vertices on the corresponding sphere subject to a rigid
−5
body motion. The bead attachment is modeled by including 10 −2 −1 0 1 2 3
several RBC vertices next to the microbead bottom into the 10 10 10 10 10 10
rigid motion. Frequency (Hz)
A typical bead response to an oscillating torque measured FIGURE 3 Two-dimensional storage and loss (g0 and g00 ) moduli of the
in simulations is given in Fig. 2 b. The bead displacement RBC membrane obtained from simulations for different membrane viscosi-
has the same oscillating frequency as the applied torque ties and bending rigidities in comparison with the experiments (6).

Biophysical Journal 98(10) 2215–2225

2220 Fedosov et al.

simulation and experimental errors because the data is fitted example, the MSD of microbeads in a viscoelastic fluid
over only two orders of magnitude in frequency. According can be well approximated by the generalized Stokes-Einstein
to the obtained exponent, the timescale in Eq. 13 utilizes the relation, where C ¼ 6p and L is the bead radius. This inter-
values of a ¼ 0.85 and h ¼ hm. In addition, RBC rheological pretation was chosen by Amin et al. (32) for microbeads
measurements allow us to define a characteristic time tc of attached to the RBC surface. However, the Stokes-Einstein
the RBC membrane found as the g0 and g00 intersection, relation cannot be valid in this case, as the membrane elastic
which uniquely relates the membrane elastic properties and properties are not taken into account. Several other models
viscous dissipation. From the intersection in Fig. 3, we (35) attempt to incorporate effects of the elastic and bending
obtain tc ¼ 1/uc z 0.1 s, in agreement with the RBC char- properties, but there is no agreement as to whether a partic-
acteristic relaxation time found in RBC recovery experi- ular model yields quantitatively accurate results for RBC
ments (33). rheology.
Note that simulation results for higher frequencies (espe- Fig. 4 a shows RBC spectral density. Theoretical predic-
cially g0 ) may be influenced by inertial effects. Absence of tions for viscoelastic vesicles (35) yield the asymptotic
inertial effects was verified by decreasing the mass of the scaling of the spectral density obtained from MSD with
attached microbead. Rheological data for high frequencies respect to frequency, when the tracked beads are much
may be computationally expensive to obtain, because smaller than the membrane, as
much smaller timesteps are required to ensure numerical
stability. An additional difficulty appears when g00 domi-
−2
nates the storage modulus, resulting in a substantial drop a 10
in the bead-displacement amplitudes for the fixed torque. Re(<D r 2(w)>), k = 2.4 ´ 10−19J
c
2 −18
Re(<D r (w)>), k = 1.2 ´ 10 J

Spectral density (mm2/Hz)

This may greatly affect reliability of experimental measure- −3
c

ments at high frequencies, while in simulations bead 10

displacements of several nanometers can still be success-
−3
fully detected. −4 x 10
10 10

8
Membrane thermal fluctuations −5 MSD (mm2) w−1.85
10 6
RBC membrane thermal fluctuations were measured in exper- 4
iments using diffraction phase microscopy (7), where instan- −6 2
10
taneous RBC-height maps were obtained, and using a micro-
rheology technique (32) through dynamical tracking of 0 1 2 3
−7
Time (s)
microbeads attached to the RBC surface. These measure- 10 −1 0 1 2 3
ments can be interpreted in terms of the three-dimensional 10 10 10 10 10
complex modulus G* with the components G0 and G00 , Frequency (Hz)
analogously to the two-dimensional version described in the
1
previous section. To mimic the experiments, the attached b 10
G’ experiment
RBC is surrounded by a fluid of viscosity ho ¼ 1  103 G’’ experiment
Pa $ s and is filled with a fluid of viscosity hi ¼ 5  103 G’ simulation 0.85
Pa $ s. The RBC cytosol is a hemoglobin solution whose 0
10 G’’ simulation w
viscosity was measured in Cokelet and Meiselman (34) to
G’ & G’’ (Pa)

be in a range of 3–10  103 Pa $ s for the physiologically

relevant concentrations of hemoglobin. The membrane −1
10
viscosity is set to hm ¼ 0.022 Pa $ s.
Membrane thermal fluctuations are measured on the RBC
top through the mean-square displacement (MSD) hDr2(t)i −2
of several membrane vertices. Theoretical developments in 10
microrheology (35) provide a relation between MSD and
G* as
−3
10 −1
 kB T 10 10
0
10
1
10
2
10
3
G ðuÞ ¼ ; (15)
CLðhDr2 ðtN Þi þ iuhDr 2 ðuÞiÞ Frequency (Hz)
pffiffiffiffiffiffiffi FIGURE 4 RBC spectral density (a) and the components of complex
where i ¼ 1, hDr2(u)i is the unilateral Fourier transform
modulus G* (b) obtained from MSD (inset plot). Dotted lines are drawn
of hDr2(t)i – hDr2(tN)i, C is a constant, and L is a length to the eye to illustrate the power with respect to frequency. Experimental
scale. Note that the corresponding C and L depend on the results (32) are obtained by dynamical tracking of microbeads attached to
physical problem and selected theoretical model. As an the RBC surface.

Biophysical Journal 98(10) 2215–2225

Multiscale Red Blood Cell Model 2221

 
Re Dr 2 ðuÞ  kc1=3 u5=3 ; (16) the cell due to a varying cell circumference along the stretch-
ing axes. Recent experiments (36) also showed that the RBC
where Re denotes the real part. Similar exponents were found response in a creep test depends strongly on strain rates
for actin-coated vesicles in the experiments of Helfer et al. applied. Moreover, the whole cell stretching experiments
(35). Our simulation results for RBCs show that subject a RBC to relatively high strains, in which nonlinear
Re(hDr2(u)i)) ~ kc0.26, close to the theoretical prediction effects may be present, and therefore the application of linear
of 1/3. Furthermore, the power of the spectral density at rheological models to extract RBC relaxation time may not
high u is found to be 1.85 in simulations, in agreement be correct.
with 5/3 in Eq. 16. In light of this, local membrane deformation techniques
Fig. 4 b presents a comparison of experimental and simu- are more favorable to measure bulk rheological properties
lation results for the complex modulus. The timescale expo- of a RBC membrane as the applied loads are local and
nent used here is the same as in the TTC simulations a ¼ have much smaller magnitudes compared to the whole cell
0.85, whereas h ¼ ho þ hi þ hm in Eq. 13. This is consistent deformations. For example, in OMTC (6) it is likely that
with the exponent of G00 found to be 0.85. For the calculation the applied strains are nearly uniform and RBC deformations
of G* (Eq. 15), we assumed that C ¼ 3p and L ¼ 0.25 mm, in remain within the linear regime. Another complication that
agreement with those used in the experiments (32). Although often arises in the rheology of viscoelastic materials is that
the behavior of the loss (viscous) modulus G00 with respect to a spectrum of relaxation times exists in which a single relax-
frequency provides a reasonable agreement between the ation time may be only relevant within a certain range of
experimental and simulation data, the behavior of the storage experimental conditions. The complexity of a RBC response
modulus G0 is clearly different. In simulations, G0 shows under different experimental conditions requires develop-
a plateau at low frequencies indicating that, in this regime, ment of accurate and realistic cell models that would allow
the RBC membrane is nearly elastic—which is typical for for a proper quantitative interpretation of experiments.
viscoelastic solid materials and is consistent with the TTC The numerical setup for the creep test followed by the cell
results described in the previous section. However, G0 found recovery is analogous to the RBC stretching in Stretching
in the experiments shows a significant decrease at low Test with the observation of a time-dependent RBC deforma-
frequencies, which is typical for viscoelastic fluids. This tion. Thus, the total stretching force f is applied to N– and
disagreement may be due to differences in measurement Nþ vertices (Fig. 1 a) in the negative and positive directions,
techniques, because in simulations, three-dimensional respectively, while the axial cell diameter is monitored.
RBC-membrane thermal fluctuations are measured directly Internal/external fluid viscosities are set to hi ¼ 5  103
at the network surface, whereas in experiments, displace- Pa $ s and ho ¼ 1  103 Pa $ s, respectively, whereas the
ment of the attached microbeads is tracked. The scatter at membrane viscosity is varied. The simulations showed that
high frequencies in simulations is due to data underresolu- the characteristic timescale depends on the RBC membrane
tion at short times. and fluids properties as Y0.75 and h0.75, where h ¼ ho þ
hi þ hm. Hence, the timescale exponent in Eq. 13 is set
to a ¼ 0.75, which is consistent with the TTC and mem-
Creep test and cell recovery
brane thermal fluctuations simulations described above. The
A creep test is the standard rheological technique to measure assumption of linear dependence (a ¼ 1) of the timescale
time-dependent deformation of a material under constant with membrane properties made in experiments (33) appears
load or stress, whereas a recovery test provides the dynamic to be a rather crude approximation.
relaxation of a material after deformation. The dynamic Fig. 5 a shows RBC creep tests for different membrane
stretching and recovery of a whole RBC was performed by viscosities. Comparison between the experimental data and
micropipette aspiration (33) and by optical tweezers (4). simulations suggests that the RBC membrane viscosity lies
The creep test was also done locally on a RBC membrane in the range of 0.02–0.06 Pa $ s. Fig. 5 b illustrates differ-
using OMTC (6), where an attached ferrimagnetic microbead ences in the creep response for distinct total applied forces.
was subjected to a constant magnetic field. The RBC charac- This demonstrates the complexity of the RBC membrane
teristic relaxation time obtained from these experiments response, which depends on total strains and strain-rates
shows qualitative agreement at best, although quantitative (36). Results of RBC recovery test after stretching can be
agreement is poor, such that the values of the obtained relax- found in the Supporting Material.
ation times may differ by a factor of 3–5.
Several issues, such as those presented by geometry,
RBC dynamics in shear flow
stress-and-strain magnitudes, and simplified models for
extracting the parameters of interest, may contribute to quan- Experimental observations (9–12) of RBC dynamics in shear
titative disagreement. The whole cell stretching or recovery flow show RBC tumbling at low shear rates and tank-
measures RBC rheological membrane properties on average, treading at high shear rates. This behavior is related to
because membrane strains and stresses are not uniform along existence of a RBC minimum energy state shown in the

Biophysical Journal 98(10) 2215–2225

2222 Fedosov et al.

a 1.2 RBC swinging around the tank-treading axes with certain

frequency and amplitude.
1
A simulated RBC is suspended into a solvent placed
between two parallel walls moving with constant velocities
in opposite directions. Fig. 6 a shows tumbling and tank-
0.8
treading frequencies with respect to shear rates in comparison
with experiments (11,12). Comparison of the simulated
J(t)

0.6
hm= 22 h0 dynamics with experiments showed that a purely elastic
hm= 35 h0 RBC with or without inner solvent (circles and squares)
0.4 hm= 60 h0 results in an overprediction of the tank-treading frequencies,
Hochmuth et al., (1979) because the membrane assumes no viscous dissipation.
Marinkovic et al., (2007)
Addition of the membrane viscosity (triangles) reduces
0.2
the values of the tank-treading frequencies and provides a
good agreement with experiments for the membrane viscosity
0
0 0.2 0.4 0.6
hm ¼ 22  103 Pa $ s. Note that for all cases, a finite inter-
Time (s) mittent region is observed and it becomes wider for a nonzero
membrane viscosity. This result is consistent with the
b 1.2

1
a ho= 0.005 Pa s = hi h m = 0.022 Pa s
80

Tank-treading
Intermittent
Tumbling
0.8

f = 4.67 pN 60
J(t)

0.6
Frequency (rad/s)

f = 14.02 pN
f = 46.7 pN

0.4 40

0.2
20 h =h =0
him = 0m
all viscosities
0 Tran-Son-Tay et al. (1984)
0 0.2 0.4 0.6 Fischer (2007)
Time (s) 0
0 50 100 150 200
g (s -1)
FIGURE 5 The normalized creep compliance (f ¼ 7 pN) for different
membrane viscosities (a) and applied forces (b) with hm ¼ 0.022 Pa $ s.
The upper figure also shows the creep test by micropipette aspiration (33) b ho= 0.005 Pa s = hi h m= 0.022 Pa s
and by OMTC (6). 40
hi= hm= 0
Swinging angle and amplitude (degrees)

35 hm= 0
experiments by Fischer (9), where a RBC relaxed to its orig- all viscosities
30
inal state marked by several attached microbeads after some
time of tank-treading motion. Hence, the RBC has to exceed 25
a certain energy barrier in order to transit into a tank-treading
20
motion in shear flow. average angle
Theoretical predictions (10,13) attempt to capture RBC 15
dynamics in shear flow depending on the shear rate and
the viscosity contrast defined as l ¼ (hi þ hm)/ho. According 10
amplitude
to the theories, for a small l < 3, a RBC tumbles at low shear 5
rates and tank-treads at high shear rates. Near the tumbling-
to-tank-treading transition there exists a narrow intermittent 0
0 50 100 150 200
region where theories predict an instability such that RBC g (s -1)
tumbling can be followed by tank-treading and vice versa.
FIGURE 6 Tumbling and tank-treading frequency (a) of a RBC in shear
In the case of a large viscosity contrast (l > 3), the theories
flow and swinging average angle and amplitude (b) for different cases: 1),
predict a well-defined tumbling regime followed by an ho ¼ 5  103 Pa $ s, hi ¼ hm ¼ 0 (circles); 2), ho ¼ hi ¼ 5  103 Pa $ s,
intermittent region, although stable tank-treading may not hm ¼ 0 (squares); and 3), ho ¼ hi ¼ 5  103 Pa $ s, hm ¼ 22  103 Pa $ s
be present. The tank-treading state is also characterized by (triangles).

Biophysical Journal 98(10) 2215–2225

Multiscale Red Blood Cell Model 2223

experiments, but it disagrees with the theoretical predictions. tensor characterizes RBC shapes. As an example, the gyra-
Similar results for the intermittent region were reported in tion tensor of a biconcave RBC in equilibrium has two large
simulations of viscoelastic vesicles (37). Moreover, an eigen-values corresponding to the cell diameter and one
increase in the internal fluid or membrane viscosities results small eigen-value characterizing the RBC height. Thus, the
in a shift of the tumbling-to-tank-treading transition to higher RBC transition to the parachute shape can be identified by
shear rates. Fig. 6 b shows the average RBC tank-treading the smallest eigen-value as it increases from its equilibrium
angle and the swinging amplitude. The values are consistent value to that of an elongated RBC in Poiseuille flow.
with experimental data (10) and appear to be not very sensi- Fig. 7 shows a snapshot of the RBC parachute shape
tive to the membrane viscosity. Note that the swinging (Fig. 7 a) and the biconcave-to-parachute transition for
frequency is equal to twice the tank-treading frequency. different bending rigidities (Fig. 7 b) with respect to the
In conclusion, the RBC model accurately captures mean flow velocity. The plot shows that healthy RBCs
membrane dynamics in shear flow, while the theoretical transit to a parachute shape at the mean flow velocity as
models can predict RBC dynamics at most qualitatively. low as 65 mm/s. Cells with a larger bending rigidity show
The theoretical models assume ellipsoidal RBC shape and the biconcave-to-parachute transition at larger flow rates
a fixed (ellipsoidal) RBC tank-treading path. Our simulations with a nearly linear dependence of the transition on the
showed that a RBC is subject to deformations along the tank-
treading axis. In addition, modeled RBCs show substantial
shape deformations (buckling) in a wide range around the a
tumbling-to-tank-treading transition. A degree of these
deformations depends on the Föppl-von Kármán number k
defined in Mechanical Properties. As an example: if the
RBC bending rigidity is increased by a factor of five, the
aforementioned shape deformations become considerably
smaller, while if the RBC bending rigidity is increased
by a factor of 10, the shape deformations practically subside.
The theoretical models do not take the bending rigidity into
consideration, while experimental data are not conclusive on
this issue. This again raises the question about the magnitude
of bending rigidity of healthy RBCs, as our simulations
(TTC and RBC dynamics in shear flow) indicate that the
RBC bending rigidity may be several times higher than the
widely used value of kc ¼ 2.4  1019 J.

RBC dynamics in Poiseuille flow

RBC dynamics in Poiseuille flow in tubes with a diameter b D=9 mm C = 0.05
comparable with the RBC diameter was the subject of inves- 0.6
tigation in several experiments (14,15). The main feature of healthy
shifted eigen value of the gyration tensor

this flow is the RBC transition from a biconcave shape to 5 x kc

parachutelike shape as the flow rate is increased. Transition 10 x k c
to the parachute shape of a RBC in Poiseuille flow is simu-
0.4
lated in a tube of the diameter 9 mm. Poiseuille flow can be
characterized by the mean flow velocity defined as
Z
v ¼ vðrÞdA=A;
0.2
where A is the area of the tube cross section, and v(r) is the
axial flow velocity.
transition to parachute
To identify the biconcave-to-parachute transition we use
the gyration tensor defined as
0
1 X i   0 200 400 600 800
Gmn ¼ r  rmC rni  rnC ; (17) U ( mm/s)
Nv i m
FIGURE 7 RBC parachute shape (a) and the normalized eigen-values of
where ri values are the RBC vertex coordinates, rC is the the gyration tensor (b) characterizing biconcave-to-parachute transition for
RBC center-of-mass, and m, n can be x, y, or z. The gyration different bending rigidities. kc ¼ 2.4  1019 J.

Biophysical Journal 98(10) 2215–2225

2224 Fedosov et al.

bending rigidity kc. These results are consistent with the the dependence of cell response on the applied stresses. In
numerical simulations of Noguchi and Gompper (16). particular, such tests cannot accurately provide a meaningful
In addition, note that the biconcave-to-parachute transition membrane characteristic relaxation time as attempted in
results in a reduction of flow resistance by 10–15%. a number of experiments. The experiments also attempt to
fit linear relaxation models, where using the power-law
model would be preferred. The comparison of the simulation
SUMMARY AND DISCUSSION
results with the experimental cell response yields the
We presented a RBC model that accurately captures RBC membrane viscosity to be between 20 and 60 times that of
mechanics, rheology, and dynamics. The membrane skeleton water.
is constructed as a network of interconnected viscoelastic The RBC dynamics was studied in Couette and Poiseuille
springs that provide RBC elasticity analogously to the spec- flows. The RBC dynamics in shear flow yields two regimes:
trin network, and viscous dissipation similarly to that in the tumbling at low-shear rates and tank-treading at high-shear
lipid bilayer. This network also incorporates the membrane rates with a narrow intermittent region of mixed dynamics.
bending rigidity to mimic bending resistance of the lipid The comparison of the simulation results with the theoretical
bilayer. In addition, local and global area constraints ensure predictions showed that RBC dynamics in shear flow can be
the membrane incompressibility of real RBCs, while the predicted at most qualitatively by the existing theories
volume constraint ensures the incompressibility of the inner because they do not take into account the RBC deformations
solvent. and bending rigidity. In addition, the simulated RBCs showed
Independently of a RBC coarse-graining level (Nv), the strong membrane deformations around the tumbling-to-tank-
model mechanical properties were found to be in excellent treading transition, which greatly depend on the relative
agreement with those obtained in optical tweezers experi- membrane elastic properties and bending rigidity. The obser-
ments. This was achieved without ad hoc tuning or adjustment vation of RBC shapes near the transition in experiments
of the model parameters, because they can be analytically would be of great interest. Finally, RBC dynamics in Pois-
derived from the imposed macroscopic membrane properties. euille flow showed RBC transition to the parachutelike shape,
In addition, the proposed stress-free model allowed us to elim- which occurs at the mean flow velocity of ~65 mm/s.
inate existing artifacts of the network triangulation and to set Current RBC models are commonly validated to capture
a realistic Föppl-von Kármán number k of a RBC. RBC mechanical response, but they are not able to appropri-
RBC membrane rheology was first probed by TTC, ately describe membrane rheology and dynamics. The model
showing good agreement with the experiments for the of this work exhibits a number of RBC membrane properties
bending rigidity kc ¼ 4.8  1019 J and the membrane simultaneously, and thus provides an adequate and realistic
viscosity hm ¼ 0.022 Pa $ s. Our results in Fig. 3 indicate representation of a RBC. The model is general enough and
that the RBC membrane bending rigidity may be 2–3 times can be used in other simulation methods, such as lattice
larger than a widely used value of 2.4  1019 J. This may be Boltzmann, Brownian dynamics, immersed boundary
verified by new OMTC experiments that employ ferrimag- method, multiparticle collision dynamics, etc.
netic microbeads of different sizes. The TTC simulations
also showed that a characteristic time of the RBC membrane
can be defined by the intersection of g0 and g00 curves in SUPPORTING MATERIAL
Fig. 3 and is equal to ~0.1 s consistent with experiments. Three figures, one table, and 20 equations are available at http://www.
The second rheological test monitored membrane thermal biophysj.org/biophysj/supplemental/S0006-3495(10)00228-6.
fluctuations, resulting in reasonable agreement with the
analogous experiments. In simulations, three-dimensional We thank Petia Vlahovska and Manouk Abkarian for useful discussions.
membrane fluctuations were measured directly at the Computations were performed at the National Science Foundation’s
National Institute for Computational Science facility.
network; however, in experiments, the motion of attached
to the lipid bilayer microbeads was observed, which may This work was supported by the National Science Foundation under grant
No. CBET-0852948 and by the National Institutes of Health grant No.
account for the discrepancies. In addition, membrane undu- R01HL094270.
lations measured in experiments may be influenced by poten-
tial metabolic activities (e.g., ATP) suggested by recent
experiments (38) and theories (39,40). However, no depen-
REFERENCES
dence of membrane fluctuations on ATP was found in Evans
et al. (41). This issue is not fully resolved and is currently not 1. Fung, Y. C. 1993. Biomechanics: Mechanical Properties of Living
addressed in our model. However, it suggests a topic for Tissues, 2nd ed. Springer-Verlag, New York.
future work. 2. Waugh, R., and E. A. Evans. 1979. Thermoelasticity of red blood cell
membrane. Biophys. J. 26:115–131.
The third rheological test applied time-dependent stretch-
3. Discher, D. E., N. Mohandas, and E. A. Evans. 1994. Molecular maps of
ing and relaxation of a whole RBC. The simulations show red cell deformation: hidden elasticity and in situ connectivity. Science.
complex behavior due to the nonuniform cell strains, and 266:1032–1035.

Biophysical Journal 98(10) 2215–2225

Multiscale Red Blood Cell Model 2225

4. Hénon, S., G. Lenormand, ., F. Gallet. 1999. A new determination of 23. Lim, H. W. G., M. Wortis, and R. Mukhopadhyay. 2002. Stomatocyte-
the shear modulus of the human erythrocyte membrane using optical discocyte-echinocyte sequence of the human red blood cell: evidence
tweezers. Biophys. J. 76:1145–1151. for the bilayer-couple hypothesis from membrane mechanics. Proc.
5. Suresh, S., J. Spatz, ., T. Seufferlein. 2005. Connections between Natl. Acad. Sci. USA. 99:16766–16769.
single-cell biomechanics and human disease states: gastrointestinal 24. Malevanets, A., and R. Kapral. 1999. Mesoscopic model for solvent
cancer and malaria. Acta Biomater. 1:15–30. dynamics. J. Chem. Phys. 110:8605–8613.
6. Puig de Morales-Marinkovic, M., K. T. Turner, ., S. Suresh. 2007. 25. Succi, S. 2001. The Lattice Boltzmann Equation for Fluid Dynamics
Viscoelasticity of the human red blood cell. Am. J. Physiol. Cell and Beyond. Oxford University Press, Oxford, UK.
Physiol. 293:597–605.
26. Miao, L., U. Seifert, ., H. G. Döbereiner. 1994. Budding transitions of
7. Popescu, G., Y.-K. Park, ., M. S. Feld. 2007. Coherence properties of fluid-bilayer vesicles: the effect of area-difference elasticity. Phys. Rev.
red blood cell membrane motions. Phys. Rev. E Stat. Nonlin. Soft Matter
E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics. 49:5389–5407.
Phys. 76:031902.
27. Hoogerbrugge, P. J., and J. M. V. A. Koelman. 1992. Simulating micro-
8. Park, Y.-K., M. Diez-Silva, ., S. Suresh. 2008. Refractive index
scopic hydrodynamic phenomena with dissipative particle dynamics.
maps and membrane dynamics of human red blood cells parasitized
by Plasmodium falciparum. Proc. Natl. Acad. Sci. USA. 105: Europhys. Lett. 19:155–160.
13730–13735. 28. Fedosov, D. A. 2010. Multiscale modeling of blood flow and soft
9. Fischer, T. M. 2004. Shape memory of human red blood cells. Biophys. matter. PhD thesis, Brown University, Providence, RI.
J. 86:3304–3313. 29. Dao, M., J. Li, and S. Suresh. 2006. Molecularly based analysis of
10. Abkarian, M., M. Faivre, and A. Viallat. 2007. Swinging of red blood deformation of spectrin network and human erythrocyte. Mater. Sci.
cells under shear flow. Phys. Rev. Lett. 98:188302. Eng. C. 26:1232–1244.
11. Tran-Son-Tay, R., S. P. Sutera, and P. R. Rao. 1984. Determination of 30. Helfrich, W. 1973. Elastic properties of lipid bilayers: theory and
red blood cell membrane viscosity from rheoscopic observations of possible experiments. Z. Naturforsch. [C]. 28:693–703.
tank-treading motion. Biophys. J. 46:65–72. 31. Espanol, P. 1998. Fluid particle model. Phys. Rev. E Stat. Phys. Plasmas
12. Fischer, T. M. 2007. Tank-tread frequency of the red cell membrane: Fluids Relat. Interdiscip. Topics. 57:2930–2948.
dependence on the viscosity of the suspending medium. Biophys. J. 32. Amin, M. S., Y.-K. Park, ., G. Popescu. 2007. Microrheology of red
93:2553–2561. blood cell membranes using dynamic scattering microscopy. Opt.
13. Skotheim, J. M., and T. W. Secomb. 2007. Red blood cells and other Express. 15:17001–17009.
nonspherical capsules in shear flow: oscillatory dynamics and the 33. Hochmuth, R. M., P. R. Worthy, and E. A. Evans. 1979. Red cell exten-
tank-treading-to-tumbling transition. Phys. Rev. Lett. 98:078301. sional recovery and the determination of membrane viscosity. Biophys.
14. Suzuki, Y., N. Tateishi, ., N. Maeda. 1996. Deformation of erythro- J. 26:101–114.
cytes in microvessels and glass capillaries: effects of erythrocyte
34. Cokelet, G. R., and H. J. Meiselman. 1968. Rheological comparison
deformability. Microcirculation. 3:49–57.
of hemoglobin solutions and erythrocyte suspensions. Science. 162:
15. Faivre, M. 2007. Red blood cells and vesicles. PhD thesis, Université J. 275–277.
Fourier Grenoble, Grenoble, France.
35. Helfer, E., S. Harlepp, ., D. Chatenay. 2000. Microrheology of
16. Noguchi, H., and G. Gompper. 2005. Shape transitions of fluid vesicles biopolymer-membrane complexes. Phys. Rev. Lett. 85:457–460.
and red blood cells in capillary flows. Proc. Natl. Acad. Sci. USA.
102:14159–14164. 36. Yoon, Y.-Z., J. Kotar, ., P. Cicuta. 2008. The nonlinear mechanical
response of the red blood cell. Phys. Biol. 5:036007.
17. Pozrikidis, C. 2005. Numerical simulation of cell motion in tube flow.
Ann. Biomed. Eng. 33:165–178. 37. Kessler, S., R. Finken, and U. Seifert. 2008. Swinging and tumbling of
elastic capsules in shear flow. J. Fluid Mech. 605:207–226.
18. Eggleton, C. D., and A. S. Popel. 1998. Large deformation of red blood
cell ghosts in a simple shear flow. Phys. Fluids. 10:1834–1845. 38. Tuvia, S., S. Levin, ., R. Korenstein. 1998. Mechanical fluctuations of
the membrane-skeleton are dependent on F-actin ATPase in human
19. Discher, D. E., D. H. Boal, and S. K. Boey. 1998. Simulations of the
erythrocyte cytoskeleton at large deformation. II. Micropipette aspira- erythrocytes. J. Cell Biol. 141:1551–1561.
tion. Biophys. J. 75:1584–1597. 39. Gov, N. S., and S. A. Safran. 2005. Red blood cell membrane fluctua-
20. Li, J., M. Dao, ., S. Suresh. 2005. Spectrin-level modeling of the cyto- tions and shape controlled by ATP-induced cytoskeletal defects. Bio-
skeleton and optical tweezers stretching of the erythrocyte. Biophys. J. phys. J. 88:1859–1874.
88:3707–3719. 40. Zhang, R., and F. L. H. Brown. 2008. Cytoskeleton mediated effective
21. Dupin, M. M., I. Halliday, ., L. L. Munn. 2007. Modeling the flow of elastic properties of model red blood cell membranes. J. Chem. Phys.
dense suspensions of deformable particles in three dimensions. Phys. 129:065101.
Rev. E Stat. Nonlin. Soft Matter Phys. 75:066707. 41. Evans, J., W. Gratzer, ., J. Sleep. 2008. Fluctuations of the red blood
22. Pivkin, I. V., and G. E. Karniadakis. 2008. Accurate coarse-grained cell membrane: relation to mechanical properties and lack of ATP
modeling of red blood cells. Phys. Rev. Lett. 101:118105. dependence. Biophys. J. 94:4134–4144.

Biophysical Journal 98(10) 2215–2225