Article

Cite This: ACS Biomater. Sci. Eng. 2017, 3, 2962-2973 pubs.acs.org/journal/abseba

Numerical Simulation of Real-Time Deformability Cytometry To

Extract Cell Mechanical Properties
M. Mokbel,† D. Mokbel,†,‡ A. Mietke,¶,§ N. Trab̈ er,‡ S. Girardo,‡ O. Otto,‡,∥ J. Guck,‡ and S. Aland*,†,⊥
†
Institute of Scientific Computing, TU Dresden, Zellescher Weg 12-14, 01069 Dresden, Germany
‡
Biotechnology Center, TU Dresden, Tatzberg 47-49, 01307 Dresden, Germany
¶
Max-Planck-Institute for Cell Biology and Genetics, Pfotenhauerstrasse 108, 01307 Dresden, Germany
§
Max-Planck-Institute for the Physics of Complex Systems, Nöthnitzer Strasse 38, 01187 Dresden, Germany
∥
Center for Innovation Competence, University of Greifswald, Fleischmannstrasse 42-44, 17489 Greifswald, Germany
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

⊥
Faculty of Informatics/Mathematics, HTW Dresden, Friedrich-List-Platz 1, 01069 Dresden, Germany
Downloaded via UNIV NOTTINGHAM on January 11, 2020 at 07:33:36 (UTC).

ABSTRACT: The measurement of cell stiffness is an important part of

biological research with diverse applications in biology, biotechnology and
medicine. Real-time deformability cytometry (RT-DC) is a new method to
probe cell stiffness at high throughput by flushing cells through a
microfluidic channel where cell deformation provides an indicator for cell
stiffness (Otto et al. Real-time deformability cytometry: on-the-fly cell 725
mechanical phenotyping. Nat. Methods 2015, 12, 199−202). Here, we
propose a full numerical model for single cells in a flow channel to
quantitatively relate cell deformation to mechanical parameters. Thereby
the cell is modeled as a viscoelastic material surrounded by a thin shell
cortex, subject to bending stiffness and cortical surface tension. For small
deformations our results show good agreement with a previously developed
analytical model that neglects the influence of cell deformation on the fluid flow (Mietke et al. Extracting Cell Stiffness from Real-
Time Deformability Cytometry: 728 Theory and Experiment. Biophys. J. 2015, 109, 2023−2036). Including linear elasticity as
well as neo-Hookean hyperelasticity, our model is valid in a wide range of cell deformations and allows to extract cell stiffness for
largely deformed cells. We introduce a new measure for cell deformation that is capable to distinguish between deformation
effects stemming from cell cortex and cell bulk elasticity. Finally, we demonstrate the potential of the method to simultaneously
quantify multiple mechanical cell parameters by RT-DC.
KEYWORDS: finite-element simulation, RT-DC, cell mechanics, elastic moduli, cell stiffness

1. INTRODUCTION The observed deformation of initially spherical cells in the

The stiffness of cells is a powerful biophysical marker for cell
state. Information on cell elasticity can, for instance, be used to
distinguish between different cell phenotypes or between
healthy and diseased cells.3−7 Therefore, the measurement of
a cell’s elastic response is an important part of biological
research including medical applications in cell sorting and
diagnostics8−16
The heterogeneity of cells in medical and biological samples
often enables meaningful statistical analysis of cell mechanical
properties only for larger population of cells. Recently, several
microfluidic techniques have been introduced that achieve the
required high-throughput measurement rates.1,16−18 One of
these methods is real-time deformability cytometry (RT-DC),1
where suspended animal cells are advected by a shear flow
through a microfluidic channel at a constant speed. In this
process, cells are deformed due to strong velocity gradients
within the channel cross-section.1 Shortly after channel entry,
cells assume a stationary shape that is imaged by a high-speed
camera. A sketch of the measurement setup is depicted in
Figure 1 (left).
flow channel depends on cell stiffness, but is also influenced by
flow speed and relative cell size. Additionally, the internal cell
architecture plays an important role and it is currently unknown
to which extent the cell bulk and the cortex each contribute to
the effective mechanical response of the cell. To disentangle
mutual contributions of cell size and cell stiffness to cell
deformation, Mietke et al.2 presented an analytic study. There,
the analytical solution for the flow around a sphere within an
axisymmetric channel is used to calculate the corresponding
surface stresses and to predict the expected shape deformation
assuming elastic shell or bulk properties. A comparison between
predicted and experimentally measured cell deformations
permits conclusions on the elastic moduli of the corresponding
Special Issue: Multiscale Biological Materials and Systems: Integra-
tion of Experiment, Modeling, and Theory
Received: September 16, 2016
Accepted: January 11, 2017
Published: January 11, 2017

© 2017 American Chemical Society 2962 DOI: 10.1021/acsbiomaterials.6b00558

ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering
Figure 1. Left: RT-DC setup adapted from Mietke et al.2 Cells are flown through a narrow channel and assume a steady shape in the end portion of
the channel where the deformations are measured by a camera snapshot. Right: Simplified structure of a eukaryotic cell. Cellular components
determine cell mechanical properties.
cells. Because the analytical model neglects the interaction
between the deformed cell shapes and changes in the
surrounding hydrodynamic flow profiles, it is only valid for
small deformations. Still, the theoretical results of Mietke et al.2
are regarded as a highly important step toward combining
speed with precision in cell mechanics, which can open the
door to high-throughput screening and sorting of large cell
populations based on the precise mechanical properties of each
individual cell.19
The goal of the present paper is to go beyond the analytical
approach in Mietke et al.2 by introducing a full numerical
model that is valid in a wider parameter range. This will allow
to probe and analyze the limits of validity of the analytical
model in more detail and to extract cell mechanical parameters
for stronger deformed cells at finite Reynolds number. To
achieve this, the model will include linear elasticity as well as a
neo-Hookean hyperelastic model for both, the cell bulk as well
as the cell membrane, combined with a cortical tension model.
The neo-Hookean model is based on the statistical
thermodynamics of cross-linked polymer chains which roughly
correspond to the cell cortex and cytoskeleton. The neo-
Hookean hyperelastic model is known to be reasonably
accurate when the maximum strain is on the order of 100%20
and has been used for incompressible elasticity of cell
membranes,20,21 cell bulk,22 and whole tissues.23,24
Current numerical methods for biological cells in flow are
mainly designed for red blood cells (RBCs), which are relatively
soft because of the absence of a nucleus and most interior
organelles. Immersed boundary (IB) methods are widely used
for such cells and can naturally include membrane elasticity21 as
well as surface tension and bending stiffness.25,26 Very recently
IB methods have been extended to incompressible viscoelas-
ticity27 and to coupled membrane and bulk elasticity.28 The
main problems of the IB method are the handling of variable
viscosity,29,30 and the stiffness occurring from the coupling
between flow and membrane advection. Efficient parallelization
of an adaptive grid IB method is not straightforward and
continues to be the subject to active research.31
An alternative approach are particle methods. Early attempts
of dissipative particle dynamics account for cell bulk elasticity
and simulate RBCs as solid elastic bodies.32 Recently, a lot of
studies on RBC dynamics used the multiparticle collision
dynamics model, where cells are described as elastic capsules
with bending rigidity. This numerical scheme is very flexible
and allows to simulate many-body problems of vesicles and red
blood cells;33 however, the model is partly phenomenological
and to the best of our knowledge has not yet been extended to
include cell bulk elasticity.
Boundary Element Methods (BEM) as used by Pozrikidis34
is another approach, which can couple the Stokes equations to
2963
Article
thin shell theory for the cell membrane. Additional cell bulk
elasticity has not been included in the BEM framework.
All of the above methods belong to the class of interface
tracking methods where an explicit representation of the
interface by individual points or a grid is used. In contrast to
this, interface capturing methods describe the interface
implicitly by an auxiliary field. Elastic contributions are
traditionally not included in such methods, although very
recently some interesting progress in this direction has been
made for bulk elasticity in phase-field,35 Level-Set,36 and
Volume-Of-Fluid methods.37
Grid based fluid-structure interaction as in the standard
benchmark38 is usually not considered for flowing cells, because
of its limitation to small displacements. However, in the special
case of RT-DC a comoving grid can be employed to keep the
cell in the center of the computational domain. The
representation of the fluid and cell domain by two fitted
grids enables a very accurate description of the cell surface
which makes the approach very efficient. In contrast to most of
the discussed numerical methods, cell bulk elasticity, membrane
elasticity and cortical surface tension can all be easily included
in one model. We will follow this idea and present an
axisymmetric body-fitted numerical approach for single cells in
a flow channel.
The numerical results will be validated with the analytical
results from Mietke et al.2 for small deformations and can be
used to analyze the errors that are made in there due to the
neglect of cell deformation on the fluid flow. Using hyperelastic
material laws the numerical model can be used in RT-DC to
extract elastic parameters for the larger deformations, that are
experimentally observed. Comparison to experiments will show
that experimental cell shapes can be reproduced in particular if
the cell is assumed to be an elastic shell. By combination of bulk
and membrane elasticity, the model allows us to discriminate
between both effects and to extract both elastic parameters
from cell deformation images.
2. MECHANICAL CELL MODEL
The basic structure of eukaryotic cells is to a large extent
conserved across different types of animal cells. An
impermeable lipid bilayer membrane surrounds an intracellular
region filled with fluid and organelles. The membrane is
attached to the cell cortex, a thin polymer network at the cell
surface that behaves elastically at short time scales and is subject
to surface tension forces arising from pulling myosin motor
proteins.39 Coarse graining the elasticity imposed by intra-
cellular organelles and cytoskeleton leads to a representation of
the cell as a viscoelastic bulk material,32 enclosed by a
membrane subject to shear and stretch elasticity, bending
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

stiffness and surface tension. Figure 1 (right) illustrates this where E is the Young’s modulus of the cell bulk, u1 is the
picture of a cell and the corresponding forces. displacement field within Ω1, and B is the left Cauchy−Green
We focus on the simulations of single cells on the time scale deformation tensor. The latter two quantities are obtained from
of milliseconds, in agreement with the time span of a cell in an the evolution equations
RT-DC channel. While on larger time scales (e.g., minutes)
∂tu1 + (v1 − vcell) ·∇u1 = v1 in Ω1 (4)
cells can dissipate elastic stresses by changing their internal
structure, this is should not be possible within milliseconds,
which makes cell bulk and membrane elasticity the dominant ∂tB + (v1 − vcell) ·∇B = (∇v1)T B + B(∇v1) in Ω1 (5)
mechanical element.39 where u1 = 0, B = I prescribe the undeformed reference
We assume that the cell cortex can be described as a shell of configuration as initial condition.35 At the interface Γ the
thickness h, much smaller than the cell radius r. In such a shell following jump conditions are specified to ensure continuity of
the bending energy Ebend is small compared to the stretching velocities and balance of forces
energy Estretch, in fact Ebend/Estretch ≈ (h/r)2 ≪ 1. Hence,
contributions from bending rigidity are usually neglected in the δEtension δE bend δE
[v]Γ = 0, [S]Γ ·n = − − − stretch
description of eukaryotic cells. Here, we will test this δΓ δΓ δΓ
assumption by investigating the impact of bending rigidity on where the interface normal n is defined to point into Ω1 and
deformed cell shapes at energy scales that can be realistically the jump operator is [f ]Γ = f1−f 0. The interfacial forces are
expected for the cell membrane. given by the first variation of the interfacial energies with
The computational domain is divided into two regions: the respect to changes in Γ. The surface tension energy and
fluid domain Ω0 and the cell domain Ω1. Both domains are bending energy are given by
separated by an interface Γ which corresponds physically to the
cell membrane and cortex. Movement of all quantities is cb 2
considered relative to the (time-dependent) velocity of the cell
Etension =
Γ
∫
γ dA , E bend =
Γ 8
∫ κ dA
barycenter, vcell, which allows to keep the cell in the center of where cb is the joint bending stiffness of the bilayer and cell
the fluid domain. Moreover, we assume incompressibility of the cortex, γ is the effective surface tension due to myosin
cell domain, motivated by the large water content of the cells contraction and κ = ∇·n is the total curvature (twice the mean
and the very short measurement times of RT-DC (≈ 1 ms) curvature). The resulting surface tension and bending forces
which render a significant water exchange with the environment are40
unlikely. Analytical studies with compressible cells (Poisson
ratio ν∈[0.3,0.5]) have been shown to slightly decrease the δEtension δE bend c
= −γκ n , = b (2ΔΓ κ + κ 3 − 4κK )n
Young’s moduli extracted from RT-DC measurements.2 Hence, δΓ δΓ 4
the Navier−Stokes equations in cell (i = 1) and fluid (i = 0) are
where K is the Gaussian curvature of the cell surface. The
⎛ ∂v ⎞ mathematical definition of the stretching shell energy Estretch is
ρi ⎜ i + (vi − vcell) ·∇vi⎟ = ∇·Si , in Ωi complicated and in general involves tangential surface tensors.41
⎝ ∂t ⎠ (1)
However, due to the spherical ground state of cells immersed in
∇·vi = 0, in Ωi (2) a fluid, it is possible to restrict computations to axisymmetric
scenarios which leads not only to an enormous reduction in
where vi, Si, ρi are the velocity fields, stresses, and densities in computational complexity but also to a simplified stretching
both phases, respectively. Note that the velocity fields are energy. The axisymmetric stretching energy can be expressed in
advected relative to the cell velocity, due to the comoving terms of the two principal stretches, that describe the relative
domains. Viscous and elastic bulk stresses are defined as change of a surface length, in lateral and rotational direction,
respectively, as illustrated in Figure 10 in the appendix,
S0 = −p0 I + η0(∇v0 + (∇v0)T ),
ds r
T
λ1 = , λ2 =
S1 = −p1 I + η1(∇v1 + (∇v1) ) + Selastic ds0 r0 (6)

where pi, ηi, and I are pressures, viscosities, and identity matrix, where s is the arc length, r the distance to the symmetry axis,
respectively. The elastic contribution in the cell is described by and s0 and r0 are the corresponding quantities at the same
two different models for incompressible elasticity. The Euler− material point in the undeformed state. The cortex stretch
Almansi strain is the constitutive law underlying the analytical elasticity can now be formulated as42,43
model in Mietke et al.2 Although it is geometrically nonlinear,
we will also refer to this model as linear elastic model, since it is Estretch = ∫Γ f (λ1, λ2)dA (7)
the standard model of linear elasticity theory with linear
relationships between the components of stress and strain. The with
second model we employ is the hyperelastic neo-Hookean ⎧ E2D ⎛ 1 2 1
⎪ ⎜ (λ − 1) + (λ 2 − 1)2 + ν(λ1 − 1)
⎪1 − ν ⎝2
model, which is known to be reasonably accurate when the 2 1
2
maximum strain is on the order of 100%.20 Accordingly, we ⎪ (λ − 1)⎟⎞
define the elastic stress ⎪ 2 ⎠
⎪
f=⎨ (thin shell)
⎧E E ⎪
⎪ (∇u1 + ∇u1T ) − ∇u1(∇u1T ) (linear) ⎪ E2D 2
⎪3 3 (λ1 + λ 22 + λ1−2λ 2−2 − 3)
Selastic =⎨ ⎪
⎪E ⎪ 6
⎪ (B − I) (neo‐Hookean) ⎪
⎩3 (3) ⎩ (neo‐Hookean shell) (8)

2964 DOI: 10.1021/acsbiomaterials.6b00558

ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering
where E2D is the elastic surface modulus and Poisson’s ratio ν =
0.5 for an incompressible shell. The interfacial force resulting
from the shell stretching energy is derived in the Appendix.
If a square channel is employed in an RT-DC setup, the flow
can be mapped onto a cylindrical channel flow using the so-
called equivalent channel radius. This radius R is chosen such
that there is the same pressure drop along the cylindrical
channel and the square channel of width L, which leads to the
relation R = 0.547L.44 It has been shown that this methodology
allows accurate axisymmetric predictions of flow speed of
spherical objects in square channels as long as the cell occupies
less than 90% of the circular channel cross-section.2 The
corresponding surface stresses exerted by the fluid are
accurately predicted on the lateral sides of the object, but
overestimated at the diagonal sides of the object due to the
increased distance to the channel corners in a square channel. It
is currently not clear to which extent these additional stress
inhomogeneities impinge on the cell deformation. However,
the good agreement of the circular channel simulations with
experimental measurements of elastic beads suggests this is a
minor effect for moderate deformations and cell sizes.2
3. DISCRETIZATION
Two different numerical grids are used to represent Ω0 and Ω1,
respectively. Grid points of Ω1∪Γ are moved with the cell
velocity field v1. To advect the fluid domain the movement of
the boundary grid points along Γ is harmonically extended onto
Ω0.
The set of equations is split to subproblems on the different
domains, which can be solved successively. An implicit Euler
method is employed for the Navier−Stokes equations, an
explicit Euler method is used to evolve B. In the following, time
steps indices are denote by superscripts. The splitting scheme
in time step m+1 reads:
1. Solve the Navier−Stokes equations in the fluid (i = 0)
with boundary condition vm+1 0 = vm1 on Γ. The grid
velocity ṽ (see step (6)) is subtracted in the advection
operators in the same manner as the cell velocity vcell.
2. Update the left Cauchy-Green tensor B according to eq 5
by an explicit Euler time discretization.
3. Calculate λ1, λ2, κ, K and n from the position of boundary
grid points along Γ, see Hu et al.40 for details. This allows
δE tension m + 1
computation of the interfacial forces ( δΓ
δEstretch m + 1 δE m+1
( δΓ )
and δΓbend
, ( . ) al.1 In RT-DC experiments, camera snapshots are taken at the
4. Solve the Navier−Stokes equations in the cell (i = 1)
coupled to eqs 4 and 5 with boundary condition
S1m + 1n = (1 − ω)S1mn
⎡ δ Etension δ Estretch δ E bend ⎤
+ ω ⎢S 0 n − − −
⎣ δΓ δΓ δΓ ⎦
The relaxation constant is set to ω = 0.25 to remove the
stiffness of the coupling between flow and elasticity.
Advection terms are neglected in this step because Ω1 is
comoving (see next step).
5. Move every grid point of Ω1 ∪Γ with velocity v1−vcell.
6. Solve the following Laplace problem to harmonically
extend the interface movement into Ω0
Article
Δv ̃ = 0 in Ω 0 ,
ṽ = 0 on δ Ω 0 /Γ,
v ̃ = v1 − vcell on Γ
and move Ω1 including every grid point with velocity v.̃
All equations are discretized in the corresponding axisym-
metric formulation45 such that they can effectively be solved in
2D. The meshes are created by gmsh,46 the discretization with
the Finite Element method is implemented in the software
AMDiS.47
4. RESULTS
We simulate a typical RT-DC measurement of a single cell
flowing through a square channel of width L = 20 μm. To use
the cortex stretch elasticity model given above, the flow is
mapped onto an axisymmetric channel flow using the concept
of the equivalent channel radius (see Sec 2). Assuming
rotational symmetry of the cell deformation, axisymmetric
differential operators can be used to effectively restrict the
computations to two dimensions and largely reduce the
computational complexity of the simulations.
Hence, in the axisymmetric setting, a single spherical cell of
radius r is initially placed in the middle of a circular channel of
radius R = 0.547L. Because the channel moves with the cell
barycenter, it suffices to consider a small portion of the channel
length. We find that the cell deformation is not influenced by
the length of the computational channel for a channel length of
40 μm and larger. Accordingly, we specify Ω1 = {(x,y)|y ≥ 0, x2
+ y2 < r2}, Ω0 = ([0,40 μm] × [0,R])/Ω1.
No slip boundary conditions are imposed on the channel
wall. A pressure gradient pm is imposed between channel inlet
and outlet to drive the flow. Initially p0 = 2500 N/m2, over time
this pressure is adapted, pm+1 = pm/(0.9 + 0.1Q/(0.04 μl/s)), to
yield a defined flow rate Q = 0.04 μl/s as in RT-DC
experiments. The vertical velocity (in the y-direction) is set to
zero at the symmetry axis as well as at the channel inflow and
outflow. The densities is set to ρ0 = ρ1 = 1000 kg/m3.
The overall cell velocity is computed as vcell = ∫ Ω1v1dV/
∫ Ω11dV. The time step is chosen as dt = 0.2 μs (except for
simulations for the hyperelastic shell model with E2D ≤ 0.05 N/
m, where dt = 0.05 μs). The grid size is 0.2 μm around the cell
surface and 1 μm in the bulk phases. As end time for the
simulations we choose 500 μs except for the linear elastic bulk
model with E = 15 kPa (1200 μs) and for the hyperelastic bulk
) ,
model (2200 μs).
The viscosity of the solvent is η0 = 0.015 Pa·s, as in Otto et
end portion of the channel where the cells assume a stationary
shape. Because the velocity gradient inside the cell vanishes in
the stationary state, results are independent of the viscosity of
the cytosol. Hence, for simplicity and numerical stability we
m+1
⎥ on Γ choose η1 = η0, throughout this work. Time-dependent
simulations of channel entry and channel exit will depend on
η1. Such simulations may provide interesting additional
information about the cell state and are left to future work.
4.1. Comparison with Analytical Model. While the
numerical model can be used to combine bulk elasticity with
cortex elasticity and surface tension, we will focus in the
following on the two extreme cases that have also been
investigated in Mietke et al.:2 pure bulk elasticity and cortex
elasticity with surface tension. In the latter case, we fixed the
2965 DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

Figure 2. Time evolution of deformation and volume of a single simulated cell of initial radius r = 7.66 μm for different models. Left: The cell
deformation in the simulations assumes a stationary state after a few hundred microseconds. The four simulations displayed here are marked as black
dots in Figure 3 and Figure 4. Right: The relative volume change is determined by (V−V0)/V0 for volume V of the cell and initial volume V0. The
volume of the cell is very well conserved.

Figure 3. Isolines in area-deformation space for elastic solids (left) and elastic shells with a surface tension of γ = 0.1E2D (right). Lines mark the
analytical results,2 circles denote the results of the numerical linear elastic model. The two filled black circles in the top row indicate the solutions
corresponding to Figure 2. The bottom row shows a close-up of the respective top row diagram indicating good agreement between the models for
small deformations.

surface tension to the elastic modulus of the shell, γ = 0.1 E2D,
to stick to a single free cell parameter. The bending rigidity is
neglected for these first tests, cb = 0, and we will show in section
4.3 that it has no influence on cell deformation in a reasonable
parameter regime.
Cell deformation is quantified by the imaged two-dimen-
sional projection of the cell shape, as it would be seen by a
camera. Comparing the cell perimeter P with the perimeter of a
2966
circular cell of equal area A leads to the deformation measure
2 πA
d = 1 − P . Note that d = 0 for a circle and d > 0 for any
other shape. The time evolution of simulated cell deformation
is depicted in Figure 2 (left). All four models, i.e., linear bulk
elasticity, hyperelastic bulk material, linear shell elasticity and
hyperelastic shell elasticity reach a stationary state of
deformation within a few hundred microseconds. Note that
we do not simulate channel entry but start with an initially
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

Figure 4. Isolines in area-deformation space for elastic solids (left) and elastic shells with a surface tension of γ = 0.1E2D (right). Crosses mark the
results of the hyperelastic model, circles denote the results of the linear elastic model. The two filled black circles in the top row indicate the solutions
corresponding to Figure 2.

Figure 5. Top row: Cell shapes for elastic solid and elastic shell model for various parameters. Analytical solution2 (blue circles), linear elastic model
(red dots), neo-Hookean model (black line). Depicted cell sizes are matched for better visibility. All models agree for small deformations, whereas
deviations from the analytical model become more prominent for larger deformations. Bottom row: The 3D illustration (left) shows a dent at the cell
rear that is hidden in the lateral camera view. Accordingly, computed cell cross-sections are convexified (right) for comparison with experiments, as
shown here for the linear elastic shell model.

spherical cell within the channel. Hence, the given times only
provide an indicator of the real time it takes for entering cells to
reach a stationary state. In general, we observe that, in our
simulations, the time until a stationary state is reached,
correlates with cell stiffness as larger elastic moduli lead to
earlier stationary states. Excellent volume conservation of
simulated cells is shown in Figure 2 (right).
Next, we compare stationary cell deformations of the
analytical linear elastic model2 with our numerical linear elastic
model. As a typical output of RT-DC measurements, cell
deformation, d, is plotted versus cell size (imaged cell area),
where each cell of the measured population creates a single data
points in a scatterplot.1 So-called isoelasticity lines based on the
analytical model have been introduced to to identify regions of
similar cell stiffness in such graphics.2 Hence, it is very
convenient to use these lines for a direct comparison of
analytical and numerical model in the following. Figure 3 shows
the isoelasticity lines for both, elastic solid and elastic shell. The
stationary deformations observed at the end time of Figure 2
corresponds to single points in this plot, highlighted by filled
black circles. The analytical model neglects nonlinear strain
2967
terms as well as the back-coupling of cell deformation on the
flow field and therefore only makes reliable predictions for
small deformations. Indeed, in the regime of small deformations
d < 0.02 we find good agreement with our numerical model.
For larger deformations and larger imaged cell areas (A ≳ 160
μm2) deviations between the analytical predictions and our
numerical model expectedly increase. This effect is even more
pronounced for pure shell elasticity with surface tension where
we find discrepancies between the analytical and the numerical
model already for d ≳ 0.005.
4.2. Hyperelastic Models. By including the back-coupling
of cell deformation on the flow field, our numerical model
naturally provides more accurate results for larger deformations.
Due to the higher strains in such cases we employ neo-
Hookean material laws for cell bulk and cortex. Such models
are widely used to describe incompressible biological tissues,
see section 1. In Figure 4, we compute isoelasticity lines for the
neo-Hookean bulk and shell model and compare them to
previous numerical results. For smaller deformations, d ≲ 0.02,
we find very good agreement between neo-Hookean and linear
elasticity. For larger deformations, neo-Hookean elasticity
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

Figure 6. Effect of bending stiffness and Reynolds number on deformation of a solid elastic cell. Left: The bending stiffness has no significant
influence below cb ≈ 1 × 104kbT. Right: A hundred-fold increase in Re changes the deformation by only ∼10%.

Figure 7. Comparison of RT-DC and AFM measurements of PAAm beads. Left: The sample of beads was measured with RT-DC at a flow rate of
0.08 μl/s. Right: The boxplot shows excellent agreement of elastic moduli predicted by RT-DC and measured by AFM.

mostly leads to lower deformations than the linear elastic
models.
Generic cell shapes are depicted in Figure 5. While linear
elastic and neo-Hookean model yield very similar shapes, the
analytical model shows larger deviations, in particular for more
deformed cells. In contrast to predictions of the analytical
model,2 the numerical model does not show any lateral grooves
for the elastic shell model. However, a dent occurs at the cell
rear for larger deformations with the elastic shell model. The
3D side view in Figure 5 (bottom left) depicts how the flatness
of the trailing edge observed in experiments (cf. inset Figure 9)
could in principle arise from an effective convexification of
indented cells due to the lateral imaging of cells in RT-DC. To
make our results comparable with RT-DC experiments, we
compute cell shape parameters, such as deformation and area,
from cell contours that are convexified at the cell rear,
throughout this work, see Figure 5 (bottom right).
4.3. Influence of Bending Rigidity and Finite Reynolds
Number. As pointed out above, the bending rigidity is believed
to have very low impact on the cell deformation. To verify this
hypothesis we conduct simulations with varying bending
rigidity in the range cb = 0−2.5 × 105 kbT. We use the linear
bulk elastic model with a fixed cell area of 115 μm2 and fixed
Young’s modulus, E = 2.29 kPa. The resulting cell deformations
2968
are shown in Figure 6 (left). The bending stiffness has no
significant effect below cb ≈ 1 × 104 kbT. Because typical values
for lipid bilayer membranes are cb = 25−100 kbT, this confirms
that bending stiffness can be neglected.
The Reynolds number Re = ρ0vcellR/η0 is defined as the ratio
of inertial forces to viscous forces and can be used to find a
scaling invariance between two different cases of fluid flow. For
the used parameter set, Re = 0.133, which is typically assumed
small enough to render the inertial term in eq 1 negligible.2 Our
numerical model allows for an analysis of the effects that a finite
Reynolds number has on cell deformations, which can be
particularly useful if future RT-DC devices may use larger flow
rates or smaller solvent viscosities.
To test the effect of larger Reynolds number, we use the
linear bulk elastic model with a fixed cell area of 115 μm2. The
Reynolds number is changed by variation of solvent density in
the range of ρ0 = 1 × 103 to 1 × 105 kg/m3, which yields an up
to 100-fold increase in Re. As shown in Figure 6 (right), we find
a linear dependence of cell deformation on Reynolds number. If
Re is increased by a factor of 10, the deformation increases
approximately by only 1%. This holds for small deformations
(E = 4.58 kPa) as well as for large deformations (E = 2.29 kPa).
Similar results are obtained with the other elastic models.
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering
Figure 8. Isoelasticity lines for the new deformation measure Λ1/Λ2 determined from the moment-of-area tensor.
We conclude that inertial forces can very well be neglected in
the original parameter setting and for a wide range of larger
Reynolds numbers. In this case, the Navier−Stokes eq 1
reduces to the Stokes equation, which is linear in the velocity.
As a consequence, given that the channel is long enough to
develop a Poisseuille flow profile at the inlet and outlet, the
quasi-stationary state of the system is determined by only three
nondimensional parameters: ER3/η0Q, E2DR2/η0Q, γR2/η0Q.
Hence, the universality of the isoelasticity lines found in the
analytical model2 carries over to the numerically computed
isoelasticity lines, i.e., a change of experimental parameters only
leads to rescaling of the involved elastic moduli without
changing the lines in the plot. To be more precise, a change of
(R, Q, η0) to (R′,Q′, η0′ ) can be accounted for by the replacing
the parameters (E,E2D,γ) in the graphic by (E R/R′, E2D, γ) ×
Q′η′0R2/(Qη0 R′2) and scaling the cell area axis by a factor of
(R′/R)2. This makes the isoelasticity lines a universal look-up
graphic that can easily be adapted to a given set of experimental
parameters. Plotting these lines together with data points from
RT-DC experiments allows identification of regions of similar
cell stiffness.2
4.4. Comparison with Experiments. We illustrate the
potential to correctly predict elastic moduli by comparison of
numerical results and RT-DC measurements. To this end, we
use beads of polyacrylamide (PAAm) that can be considered as
elastic solids. A sample of approximately 3000 beads was
measured by RT-DC, see Figure 7 (left). Note that an
experimental flow rate Q = 0.08 μl/s and viscosity η0 = 0.0079
Pa·s has been used which leads to a different scaling of the iso-
elasticity lines than before.
The numerical hyperelastic solid model was employed with a
bead size that corresponds to the mean of the experimental data
(area ≈144 μm2). The numerical values were interpolated by a
quadratic spline function providing a functional relationship
between Young’s modulus and bead deformation. From this
function, Young’s moduli were determined for beads of similar
area (144 μm2 ± 2 μm2). The resulting distribution of
predicted bead elasticity is depicted in Figure 7 (right). For
comparison a subset of beads was measured by atomic-force
microscopy (AFM). We find an excellent agreement between
RT-DC and AFM measurements of bead elasticity.
4.5. Determination of Multiple Cell Parameters. Cell
deformation as measured by the circularity of the cell, is very
sensitive to the cell boundary contour extracted from camera
images. Small deviations in the cell contour might occur by
2969
Article
processing the noisy images and can increase the measured cell
perimeter and lead to an overestimated deformation value.
Integral quantities are typically more robust to such
fluctuations. The moment-of-area tensor is such an integral
quantity that can be used to quantify cell deformation in the
following. For a two-dimensional cell image, the moment-of-
area tensor is given by
⎡ ⎤
⎢ ∫Ω (y − y0 )2 ∫Ω (x − x0)(y − y0 )⎥
T=⎢ ⎥
1 1
⎢ ⎥
⎢
⎣
∫Ω (x − x0)(y − y0 ) ∫Ω (x − x0) 2
⎥
⎦
1 1 (9)
where x0 = ∫ Ω1 x/|Ω1| and y0 = ∫ Ω1 y/|Ω1| are the coordinates
of the cell barycenter. The two eigenvalues Λ1, Λ2 of T are
related to the length of the two semiaxes of the cell. The
integration over the whole cell domain makes these values very
robust with respect to uncertainties in the cell boundary
contour. Note that, it is very easy to calculate Λ1, Λ2: If the cell
image is horizontally symmetric, the two eigenvalues are simply
the diagonal elements of T: Λ1 = ∫ Ω1 (y−y0)2, Λ2 = ∫ Ω1 (x−
x0)2. If the cell shape is given as a polygon of extracted contour
points, the moment-of-area tensor can be easily evaluated as
shown in the appendix section 7.
Elasticity lines are shown for the neo-Hookean bulk and shell
model for the ratio Λ1/Λ2 in Figure 8. For the solid elastic
model, we find no overlapping of lines for cells larger than 70
μm2, which confirms that this deformation measure is well-
suited to conclude cell elasticity for a given deformation. For
the elastic shell model, we find significant overlapping of
different isoelasticity lines, which makes it impossible to extract
a definite E2D for a given Λ1/Λ2.
However, the fact that Λ1/Λ2 appears so different for the
elastic shell and the elastic solid model can be very helpful to
discriminate between the effects of these two models and to
extract both parameters from a single cell. Although we
restricted our simulations so far to the two cases of either pure
bulk elasticity or pure cortex elasticity with surface tension, the
numerical model can be used to combine both effects. To
illustrate this, we conduct a parameter study where bulk and
cortex elasticity are varied simultaneously. The cell size and
surface tension are fixed to r0 = 7.27 μm, γ = 0.1E2D. The results
are plotted in a deformation space spanned by d and Λ1/Λ2 in
Figure 9. An RT-DC image of an embryonic mouse stem cell of
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering
Figure 9. Combined parameters study for r0 = 7.27 μm, γ = 0.1E2D.
The two deformation measures at the axis allow to discriminate
between effects from bulk and cortex elasticity. Different lines
correspond to different bulk moduli (in Pascal), whereas colors
indicate the modulus of the cortical shell (in N/m). The inset shows a
cell image acquired from an RT-DC measurement, as well as its
position in the shape space (black dot) illustrating how mechanical
properties can be determined for each cell.
the same size is included and corresponds to a single location in
this deformation space. Note that the flatness of the trailing
edge might indicate the presence of a hidden dent at the cell
rear (cf. Figure 5), which would also explain the rapid change in
surface curvature at the back corners.
By plotting lines for different bulk modulus E with colors
indicating the shell modulus E2D, it is possible to estimate both
elastic moduli of the experimental cells from nearby lines and
colors. In our example, the cell would have a bulk modulus of E
≈ 2.3 kPa and a shell modulus of E2D ≈ 500pN/μm. In
principle, all mechanical parameters could be determined from
shape data, which would require a more comprehensive
sampling of the space of mechanical parameters and cell
areas. This is not within the scope of this study, but currently
being addressed in a follow-up work.
5. CONCLUSION
In this paper, we presented numerical simulations of single cells
in a circular channel that can be used to extract elastic cell
parameters from RT-DC measurements. Therefore, we
describe the cell as a viscoelastic bulk material enclosed by a
thin cortical shell, subject to stretch elasticity, bending stiffness
and surface tension. The cell bulk as well as the cell cortex are
assumed as isotropic elastic materials, using either linear
elasticity or neo-Hookean hyperelasticity.
The discretization involves two separate Eulerian grids for
the cell and the fluid domain that overlap at the cell surface.
The full Navier−Stokes equations are solved within both
domains, the corresponding surface stresses arising from shear
and pressure forces enter as boundary conditions. The
equations of motion are coupled to additional equations for
the elastic stresses in the cell bulk and along the cell cortex by
an explicit operator splitting. Although the current simulations
are restricted to axisymmetric scenarios, the model is easily
extended to the fully three-dimensional case. The only change
would be the formulation of the membrane elasticity, which can
be included in 3D.41
We applied the model to typical RT-DC measurements and
thereby extend the first theoretical study on this topic that was
based on an analytical model, which neglected the interaction
between cell deformations and changes in the flow profile.2 We
2970
Article
used the concept of the equivalent channel radius2 to obtain
approximate cell deformations in a square channel. We could
confirm that bending stiffness of the cortex/membrane complex
has no effect on cell deformations for realistic values of bending
rigidity. Therefore, we focused on the other mechanical
elements of the cell and studied the influence of bulk elasticity
and shell elasticity separately. We found that once the cell is in
the channel, a stationary shape is reached within a few hundred
microseconds, which provides an indicator of the total time
between channel entry and stationarity. We compared the
stationary cell shapes with the results of Mietke et al.2 and
found good agreement for small deformations where the
analytical model is valid. For larger deformations (d ≲ 0.02)
and larger cell areas (A ≲ 160 μm2) the analytical model for
bulk elasticity appears to leave its range of validity probably due
to the missing back-coupling of cell deformation on the flow
field. This effect was found even more pronounced for pure
shell elasticity with surface tension where discrepancies
between analytical and numerical model already occurred for
d ≲ 0.005.
In an experimental study, with PAAm beads we illustrate the
potential of the numerical method to predict elastic moduli
from RT-DC measurements. We find an excellent agreement
between the numerically predicted Young’s modulus and AFM
measurements.
By including the back-coupling of cell deformation on the
flow field and keeping nonlinear strain terms, our numerical
model naturally provides accurate results for larger deforma-
tions. Because of the higher strains in such cases, we employed
neo-Hookean material laws for cell bulk and cortex and provide
isoelasticity lines that can be used for extraction of cell stiffness
of significantly deformed cells.
We further found that inertial effects play no role in the
original RT-DC parameter setting. This even holds for largely
increased Reynolds number, which confirms the theoretical
potential of RT-DC to work well with higher flow rates or less
viscous carrier fluids. As a consequence the scaling invariance
found in Mietke et al.2 is carried over to the nonlinear
numerical model, i.e., a change of experimental parameters only
leads to rescaling of the involved elastic moduli. Hence, the
numerically computed isoelasticity lines are universal and can
be scaled to account for any different flow rate, channel radius
and viscosity. Therefore, using the concept of the equivalent
channel radius, our computed isoelasticity lines can be overlaid
to data points of RT-DC measurements to quickly identify
regions of similar cell stiffness.
Further, we introduced two new measures of cell
deformation: the two eigenvalues of the moment-of-area
tensor. Being integral quantities, these measures are very
robust and well-qualified to characterize deformation of even
noisy cell images. We have shown that the quotient of these
eigenvectors has the potential to distinguish between
deformation effects stemming from cell cortex and cell bulk
elasticity. Although we restricted our simulations mostly to the
two cases of either pure bulk elasticity or pure cortex elasticity
with surface tension, the model can be used to combine all
these effects. We demonstrated this here by using two different
shape measures to simultaneously extract bulk elasticity and
cortex elasticity from an experimental cell image. We are
currently working on a combined study varying E, E2D and γ
simultaneously. The produced cell shapes will be fitted to the
three shape measures which should allow to extract the two
elastic cell moduli and the cortical tension at once, and hence to
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

disentangle the mutual contributions of these parameters to cell ⎡ ⎛ df ⎞

shapes in RT-DC. With this, the presented model provides an dt Estretch = ∫Γ v·⎢⎢⎣(κ n − ∇Γ )⎜⎝f + dλ ⎟⎠
important stepping stone toward a quantitative, multivariate 1
analysis of cell mechanical properties at high-throughput with ⎛ df df ⎞ ⎤
many potential applications in medical diagnostics and biology. +⎜ λ2 − λ1⎟ r ⎥dA

■
⎝ dλ 2 dλ1 ⎠ ̅⎥⎦ (16)
APPENDIX
With the definition of the energy densities in eq 8, we obtain
6. Derivation of the Elastic Shell force
δE δEstretch ⎛ df ⎞ ⎛ df df ⎞
In this section, we derive the elastic surface force stretch . The = (κ n − ∇Γ )⎜f + ⎟+⎜ λ2 − λ1⎟ r
dλ1 ⎠ ̅
δΓ
variational derivative can be implicitly defined by the time δΓ ⎝ dλ1 ⎠ ⎝ dλ 2
derivative of the elastic surface energy (see also Figure 10) (17)
δEstretch ⎧ E2D
dt Estretch = ∫Γ δΓ
·vdA
(10) ⎪ 2
[(κ n − ∇Γ )(3λ12 + λ 22 + 4νλ1λ 2
⎪ 2(1 − ν )
⎪ + (1 + ν)(2 − 2λ 2 − 4λ1))
⎪
⎪ − 2 r ̅(λ12 − λ 22 + (1 + ν)(λ 2 − λ1))]
⎪
=⎨ (thin shell)
⎪
⎪ E2D 2 2 −2 −2
⎪ 3 [(κ n − ∇Γ )(3λ1 + 2λ 2 + λ1 λ 2 − 6)
⎪ 2 2
⎪ − r ̅(λ1 − λ 2 )]
⎪
⎩ (neo‐Hookean shell) (18)

Figure 10. Illustration of the two principal stretches of an In the case of a simple thin shell, we linearize the force to
axisymmetric thin shell. obtain
δEstretch E2D
= [(κ n − ∇Γ )(λ1 − 1 + νλ 2 − ν)
δΓ 1 − ν2
Now calculate the time derivative of the shell stretching energy − r ̅(1 − ν)(λ1 − λ 2)] (linear shell) (19)
given in eqs 7 and 8
df df 7. Calculation of Moment-of-Area Tensor from a Given Cell
dt Estretch = ∫Γ f ∇Γ ·v + dλ λ1̇ + dλ λ 2̇ dA
(11)
Contour
1 2
In experimental RT-DC data, the cell shape is usually described
where the overdot denotes the material derivative and ∇Γ· the by a set of contour points that determine the cell area by a
surface divergence operator. To calculate λ1̇ , λ 2̇ , we introduce polygon. In the following, we show how the moment-of-area
tensor can be calculated from these contour points without an
the distance r to the symmetry axis x, r(x , y , z) = y 2 + z 2 explicit representation of the cell interior.
and the vector r ̅ pointing away from the symmetry axis x, Let (x0,y0), ..., (xN−1,yN−1) be the coordinates of the
r(x,y,z)
̅ = (0,y,z)/(y2 + z2). Note that the magnitude of r ̅ scales considered contour points, where N denotes the total number.
like 1/r. From the definition of λ2 = r/r0, we conclude We introduce the following values:
λ 2̇ = r /̇ r0 = λ 2r /̇ r = λ 2 v· r ̅ (12)
• Area A of the cell polygon
N−1
To obtain λ1̇ we consider the stretch of a surface area element 1
A= ∑ (xiy − xjyi )
A/A0, which is related to the two principal stretches by A/A0 = i=0
2 j
λ1 λ2. From mass conservation, it is known that (A/A0)̇ = A/ j = i + 1mod N

A0∇Γ·v and hence • Integrals Ix = ∫ x, Iy = ∫ y over the polygon

(λ1λ 2̇ ) = λ1λ 2∇Γ ·v (13) N−1
1
Ix = ∑ (xi − yj − xjyi )(xi + xj)
λ1̇ = λ1(∇Γ ·v − v· r ̅) (14)
6 i=0
j = i + 1mod N

Substituting λ1̇ , λ 2̇ in the energy time derivative yields 1

N−1

⎛
Iy = ∑ (xi − yj − xjyi )(yi + yj )
df ⎞ ⎛ df df ⎞ 6
dt Estretch = ∫⎜f +
Γ⎝
⎟∇Γ ·v + ⎜
dλ1 ⎠ ⎝ dλ 2
λ2 − λ1⎟ r ·vdA
dλ1 ⎠ ̅
i=0
j = i + 1mod N

(15) • Coordinates xb, yb of the barycenter

Splitting ∇Γ·v into ∇Γ·(Pv) + κn·v with projection operator P Ix Iy
= I−n⊗n allows us to use integration by parts on the ∇Γ·(Pv) xb = , yb =
A A
term to obtain
2971 DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering
Now we define the point coordinates relative to the cell
barycenter, (x̃i, ỹi) = (xi−xb,yi−yb) for i = 0,..., N−1. The entries
of the moment-of-area tensor can now be calculated by
N−1
1
T11 = ∑ [xĩ yj̃ − xj̃ yi ̃ ]·[yi2̃ + yi ̃ yj̃ + yj2̃ ]
12 i=0
j = i + 1mod N
N−1
1
T22 = ∑ [xĩ yj̃ − xj̃ yi ̃ ]·[xĩ 2 + xĩ xj̃ + xj̃2]
12 i=0
j = i + 1mod N
N−1
1
T12 = T21 = ∑ [xĩ yj̃ − xj̃ yi ̃ ]·
24 i=0
j = i + 1mod N
[2xĩ yi ̃ + xĩ yj̃ + xj̃ yi ̃ + 2xj̃ yj̃ ].
Note that for horizontally symmetric cell shapes, the
eigenvalues of T are simply given by the diagonal elements:
Λ1 = T11, Λ2 = T22.
■
AUTHOR INFORMATION
Corresponding Author
*E-mail: sebastian.aland@yahoo.com.
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS
S.A. acknowledges support from the German Science
Foundation (grant AL 1705/3) and thanks the Isaac Newton
Institute for Mathematical Sciences for its hospitality during the
program Coupling Geometric PDEs with Physics for Cell
Morphology, Motility and Pattern Formation supported by
EPSRC Grant EP/K032208/1. Further, financial support
from the Sächsische Ministerium für Wissenschaft und Kunst
(TG70 grant to O.O. and J.G.) and the Bundesministerium für
Bildung und Forschung (ZIK grant to O.O.) and the Alexander
von Humboldt Foundation (Alexander von Humboldt
Professorship to J.G.) is gratefully acknowledged. Finally, we
thank the BIOTEC/CRTD Microstructure Facility (partly
funded by the State of Saxony and the European Fund for
Regional Development - EFRE) for the production of the
PAAm beads.
■
REFERENCES
(1) Otto, O.; et al. Real-time deformability cytometry: on-the-fly cell
mechanical phenotyping. Nat. Methods 2015, 12, 199−202.
(2) Mietke, A.; Otto, O.; Girardo, S.; Rosendahl, P.; Taubenberger,
A.; Golfier, S.; Ulbricht, E.; Aland, S.; Guck, J.; Fischer-Friedrich, E.
Extracting Cell Stiffness from Real-Time Deformability Cytometry:
Theory and Experiment. Biophys. J. 2015, 109, 2023−2036.
(3) Ekpenyong, A. E.; Whyte, G.; Chalut, K.; Pagliara, S.;
Lautenschläger, F.; Fiddler, C.; Paschke, S.; Keyser, U. F.; Chilvers,
E. R.; Guck, J. Viscoelastic properties of differentiating blood cells are
fate-and function-dependent. PLoS One 2012, 7, e45237.
(4) Lautenschläger, F.; Paschke, S.; Schinkinger, S.; Bruel, A.; Beil,
M.; Guck, J. The regulatory role of cell mechanics for migration of
differentiating myeloid cells. Proc. Natl. Acad. Sci. U. S. A. 2009, 106,
15696−15701.
(5) Tse, H. T. K.; Gossett, D. R.; Moon, Y. S.; Masaeli, M.; Sohsman,
M.; Ying, Y.; Mislick, K.; Adams, R. P.; Rao, J.; Di Carlo, D.
Quantitative diagnosis of malignant pleural effusions by single-cell
mechanophenotyping. Sci. Transl. Med. 2013, 5, 212ra163−212ra163.
2972
Article
(6) Suresh, S. Biomechanics and biophysics of cancer cells. Acta
Mater. 2007, 55, 3989−4014.
(7) Lincoln, B.; Erickson, H. M.; Schinkinger, S.; Wottawah, F.;
Mitchell, D.; Ulvick, S.; Bilby, C.; Guck, J. Deformability-based flow
cytometry. Cytometry 2004, 59A, 203−209.
(8) Guck, J.; Schinkinger, S.; Lincoln, B.; Wottawah, F.; Ebert, S.;
Romeyke, M.; Lenz, D.; Erickson, H. M.; Ananthakrishnan, R.;
Mitchell, D.; Käs, J.; Ulvick, S.; Bilby, C. Optical deformability as an
inherent cell marker for testing malignant transformation and
metastatic competence. Biophys. J. 2005, 88, 3689−3698.
(9) Remmerbach, T. W.; Wottawah, F.; Dietrich, J.; Lincoln, B.;
Wittekind, C.; Guck, J. Oral cancer diagnosis by mechanical
phenotyping. Cancer Res. 2009, 69, 1728−1732.
(10) Hur, S. C.; Tse, H. T. K.; Di Carlo, D. Sheathless inertial cell
ordering for extreme throughput flow cytometry. Lab Chip 2010, 10,
274−280.
(11) Cooke, B. M.; Mohandas, N.; Coppel, R. L. The malaria-
infected red blood cell: structural and functional changes. Advances in
parasitology 2001, 50, 1−86.
(12) Lee, G. Y.; Lim, C. T. Biomechanics approaches to studying
human diseases. Trends Biotechnol. 2007, 25, 111−118.
(13) Swaminathan, V.; Mythreye, K.; O’Brien, E. T.; Berchuck, A.;
Blobe, G. C.; Superfine, R. Mechanical stiffness grades metastatic
potential in patient tumor cells and in cancer cell lines. Cancer Res.
2011, 71, 5075−5080.
(14) Hou, H. W.; Bhagat, A. A. S.; Lin Chong, A. G.; Mao, P.; Wei
Tan, K. S.; Han, J.; Lim, C. T. Deformability based cell margin-
ationâĂ Ť a simple microfluidic design for malaria-infected erythrocyte
separation. Lab Chip 2010, 10, 2605−2613.
(15) Hou, H. W.; Li, Q.; Lee, G.; Kumar, A.; Ong, C.; Lim, C. T.
Deformability study of breast cancer cells using microfluidics. Biomed.
Microdevices 2009, 11, 557−564.
(16) Gossett, D. R.; Tse, H. T. K.; Lee, S. A.; Ying, Y.; Lindgren, A.
G.; Yang, O. O.; Rao, J.; Clark, A. T.; Di Carlo, D. Hydrodynamic
stretching of single cells for large population mechanical phenotyping.
Proc. Natl. Acad. Sci. U. S. A. 2012, 109, 7630−7635.
(17) Byun, S.; Son, S.; Amodei, D.; Cermak, N.; Shaw, J.; Kang, J. H.;
Hecht, V. C.; Winslow, M. M.; Jacks, T.; Mallick, P.; Manalis, S. R.
Characterizing deformability and surface friction of cancer cells. Proc.
Natl. Acad. Sci. U. S. A. 2013, 110, 7580−7585.
(18) Lange, J. R.; Steinwachs, J.; Kolb, T.; Lautscham, L. A.; Harder,
I.; Whyte, G.; Fabry, B. Microconstriction arrays for high-throughput
quantitative measurements of cell mechanical properties. Biophys. J.
2015, 109, 26−34.
(19) Wyss, H. M. Cell Mechanics: Combining Speed with Precision.
Biophys. J. 2015, 109, 1997−1998.
(20) Dao, M.; Lim, C. T.; Suresh, S. Mechanics of the human red
blood cell deformed by optical tweezers. J. Mech. Phys. Solids 2003, 51,
2259−2280.
(21) Eggleton, C. D.; Popel, A. S. Large deformation of red blood cell
ghosts in a simple shear flow. Phys. Fluids 1998, 10, 1834−1845.
(22) Zhou, E.; Lim, C.; Tan, K.; Quek, S. T. Finite element modeling
of the micropipette aspiration of malaria-infected red blood cells. Proc.
SPIE 2012, 763−767.
(23) Martins, P.; Natal Jorge, R.; Ferreira, A. A Comparative Study of
Several Material Models for Prediction of Hyperelastic Properties:
Application to Silicone-Rubber and Soft Tissues. Strain 2006, 42,
135−147.
(24) Miller, K. Method of testing very soft biological tissues in
compression. J. Biomech. 2005, 38, 153−158.
(25) Lim H. W., G.; Wortis, M.; Mukhopadhyay, R. Stomatocy-
teâĂ ŞdiscocyteâĂ Şechinocyte sequence of the human red blood cell:
Evidence for the bilayerâĂ Ş couple hypothesis from membrane
mechanics. Proc. Natl. Acad. Sci. U. S. A. 2002, 99, 16766−16769.
(26) Zhang, J.; Johnson, P. C.; Popel, A. S. An immersed boundary
lattice Boltzmann approach to simulate deformable liquid capsules and
its application to microscopic blood flows. Phys. Biol. 2007, 4, 285.
DOI: 10.1021/acsbiomaterials.6b00558
ACS Biomater. Sci. Eng. 2017, 3, 2962−2973
ACS Biomaterials Science & Engineering Article

(27) Devendran, D.; Peskin, C. S. An immersed boundary energy-

based method for incompressible viscoelasticity. J. Comput. Phys. 2012,
231, 4613−4642.
(28) Strychalski, W.; Copos, C. A.; Lewis, O. L.; Guy, R. D. A
poroelastic immersed boundary method with applications to cell
biology. J. Comput. Phys. 2015, 282, 77−97.
(29) Fai, T. G.; Griffith, B. E.; Mori, Y.; Peskin, C. S. Immersed
boundary method for variable viscosity and variable density problems
using fast constant-coefficient linear solvers I: Numerical method and
results. SIAM J. Sci. Comp. 2013, 35, B1132−B1161.
(30) Peskin, C. S. The immersed boundary method. Acta Numerica
2002, 11, 479−517.
(31) Borazjani, I. Fluid−structure interaction, immersed boundary-
finite element method simulations of bio-prosthetic heart valves.
Comp. Meth. Appl. Mech. Engrg. 2013, 257, 103−116.
(32) Dzwinel, W.; Boryczko, K.; Yuen, D. A. A discrete-particle
model of blood dynamics in capillary vessels. J. Colloid Interface Sci.
2003, 258, 163−173.
(33) Noguchi, H.; Gompper, G. Shape transitions of fluid vesicles
and red blood cells in capillary flows. Proc. Natl. Acad. Sci. U. S. A.
2005, 102, 14159−14164.
(34) Pozrikidis, C. Numerical simulation of the flow-induced
deformation of red blood cells. Ann. Biomed. Eng. 2003, 31, 1194−
1205.
(35) Sun, P.; Xu, J.; Zhang, L. Full Eulerian finite element method of
a phase field model for fluidâĂ Şstructure interaction problem. Comput.
Fluids 2014, 90, 1−8.
(36) Cottet, G.-H.; Maitre, E. A level set method for fluid-structure
interactions with immersed surfaces. Math. Models Methods Appl. Sci.
2006, 16, 415−438.
(37) Lan, H.; Khismatullin, D. B. A numerical study of the lateral
migration and deformation of drops and leukocytes in a rectangular
microchannel. Int. J. Multiphase Flow 2012, 47, 73−84.
(38) Turek, S.; Hron, J. Proposal for Numerical Benchmarking of Fluid-
Structure Interaction between an Elastic Object and Laminar Incompres-
sible Flow; Springer: New York, 2006.
(39) Salbreux, G.; Charras, G.; Paluch, E. Actin cortex mechanics and
cellular morphogenesis. Trends Cell Biol. 2012, 22, 536−545.
(40) Hu, W.-F.; Kim, Y.; Lai, M.-C. An immersed boundary method
for simulating the dynamics of three-dimensional axisymmetric vesicles
in Navier−Stokes flows. J. Comput. Phys. 2014, 257, 670−686.
(41) Barthes-Biesel, D.; Rallison, J. The time-dependent deformation
of a capsule freely suspended in a linear shear flow. J. Fluid Mech. 1981,
113, 251−267.
(42) Zhu, Y.; Luo, X.; Ogden, R. W. Nonlinear axisymmetric
deformations of an elastic tube under external pressure. European
Journal of Mechanics-A/Solids 2010, 29, 216−229.
(43) Höhn, S.; Honerkamp-Smith, A. R.; Haas, P. A.; Trong, P. K.;
Goldstein, R. E. Dynamics of a Volvox embryo turning itself inside out.
Phys. Rev. Lett. 2015, 114, 178101.
(44) Huebscher, R. Friction equivalents for round, square and
rectangular ducts. ASHVE Trans. (renamed ASHRAE Trans.) 1948, 54,
101−144.
(45) Tomé, M.; Grossi, L.; Castelo, a.; Cuminato, J.; McKee, S.;
Walters, K. Die-swell, splashing drop and a numerical technique for
solving the Oldroyd B model for axisymmetric free surface flows. J.
Non-Newtonian Fluid Mech. 2007, 141, 148−166.
(46) Geuzaine, C.; Remacle, J.-F. Gmsh: A 3-D finite element mesh
generator with built-in preand post-processing facilities. International
Journal for Numerical Methods in Engineering 2009, 79, 1309−1331.
(47) Witkowski, T.; Ling, S.; Praetorius, S.; Voigt, A. Software
concepts and numerical algorithms for a scalable adaptive parallel finite
element method. Advances in Computational Mathematics 2015, 41,
1145−1177.

2973 DOI: 10.1021/acsbiomaterials.6b00558

ACS Biomater. Sci. Eng. 2017, 3, 2962−2973