Rigidity and glassy dynamics

in confluent biological tissues

Dapeng (Max) Bi
Syracuse University

Acknowledgements
Syracuse University

Xingbo Yang

Jen Schwarz

Cristina Marchetti

Lisa Manning

Harvard School of Public Health

Jae Hun Kim

Jin-Ah Park

Jeffrey Fredberg

Why study cell migration in dense tissues?

Wound healing assay

Tumor cell migration

youtube.com/watch?v=v9xq_GiRXeE
During development of embryo
Dieren'al*Adhesion*Hypothesis:*
Steinberg,*Science*1962

E-M. Schetz Thesis 2008

Caging and Viscoelasticity

a consequence of being crowded together
about 5000 cells

timescale ~ minutes

timescale ~ hours

log (distance2) (microns2)

3
2.5
2

10

1.5
1
0.5

1
log

1.5
(time) (mins)

10

Schetz et al
J. R. Soc. Interface
(2013)

Published on

colloidal experiments in the temperature-packing-fractio

A phase diagram for tissues?
parameter space. Their main conclusion is that these exper
ments are actually rather far from the critical regime of point
1/adhesion
In the past ten years, we have systematically investigate
systems of horizontally shaken
grains
in the vicinity o
Increasing
Adhesion
jamming.2427 For systems of rigid brass disks, we observed ver

Solid
Jammed

Cell
Motility
1/Density

Increasing
Temperature-packing
fraction Density
phase diagram for a give
Sadati et al, Differentiation
2013
conguration: at zero temperature, below jamming, there is always
Fredberg Group way to pack the particles without overlaps and the energy of th
Harvard School of Public system
Health is strictly zero. Above jamming, there is no packing withou
Fig. 1

However, for many tissues that are already confluent

(area density = 1), a density driven transition is not
possible.

What controls the behavior of confluent tissues?

How do confluent tissues transition from

fluid-like to solid like behavior?
1/adhesion

Solid
Jammed

Cell
Motility
1/Density

To study confluent tissues, we use the Vertex models for tissues monolayers
THE EUROPEAN
PHYSICAL JOURNAL E

modelling of cell packings in epithelia

C. R
oper2 , B. Aigouy2 , S. Eaton2 , and F. J
ulicher1,a

PHYSICAL REVIEW E 80, 011904 !2009"

HOEVAR
ANDothnitzerstrasse
P. ZIHERL
sics of Complex Systems, N
38, 01187 Dresden, Germany

r Cell Biology and Genetics, Pfotenhauerstrasse 108, 01307 Dresden, Germany

Developed about 15 years ago.

Good agreement with experimentally
heets of cells
are dynamically remodelled by cell division
80 and cell death
80 thatobserved
shapes.
e we describe the cell
shapes and packingscell
as networks
of polygons: stable and
urations obey force balance and are represented as local minima of a potential
60properties
60set of ground states,
the physical
of this vertex model, including the
mechanical
stable
Explain/predict
ological rearrangements. We furthermore discuss a quasistatic description of cell
o study the40
mechanics and dynamics of tissue remodelling
40during growth. The properties.
shapes
and
eir rearrangementscell
can account
for the morphology
of cell statistical
packings observed in
20
0
4 5 6 7 8 9

(b)

20
0
4 5 6 7 8 9

60
Drosophila
Hydra
Xenopus

50
40
Frequency [%]

(a)

Frequency [%]

Frequency [%]

Drosophila germband (late stage 8)

nd Received in final
form 7 October
2010
Drosophila
germband
(stage 6)
c aEDP
mber 2010
Sciences
/
Societ`
a
Italiana
di
Fisica
/
2010
aSpringer-Verlag
= 0.75
= 0.863

a = 0.75
a = 0.79
topological cell
proliferation
theory [10]

30
20

Frequency [%]

Frequency [%]

10
ganism from a fertilized
n and organize in multiDrosophila germband (late stage 8)
tant situationDrosophila
is the for- germband (stage 6)
0
eet-like two-dimensional
50
50
defines a surface that
= 0.87
the basal40
and thea api40 a = 0.78
lium exhibit an apical!a = 0.026
!a = 0.05
pical side,30
cell contacts
30
unctions: these junctions
ules and 20
are associated
20
d myosin in the adjacent Fig. 1. (a, left) Confocal microscope image of a developing
Drosophila wing (E-cadherin labelled with green fluorescent
junctions10
organizes the
10 (b, right) Epithelia in the
protein in a wing disc epithelium).
des cell rearrangements,
vertex model are represented as networks of polygons.
packings. 0
Remarkably,
0
epithelia (e.g.
polygon
0.5
0.9
0.7
0.5
0.9
0.7
using topological argu(d)
(c)
imaginal
disc,
which
consists
of
two
epithelial
layers.
Reduced
area a
Reduced areaThe
a
understanding the
menetwork of adherens junctions of an epithelium can be
ment at the cell scale deobserved experimentally, see fig. 1a. Its reorganization is
tissue mechanics.
governed by force balance and the biophysical properties
l tissues are dynamically
of cells and their adhesive contacts. A physical description
s topological changes of
of tissue organization and mechanics can thus be based on
for example, by cell dithe mechanics of the junctional network [3].
oundary rearrangement.
Tissue morphology can be described by vertex models,
rgo apoptosis and leave
which account for the geometry of the junctional network
over time scales of hours
using polygons. These polygons are characterized by the
m undergoes dramatic repositions of vertices and linear bonds connecting them,
esses. Such dynamic resee fig. 1b. Force balance in the junctional network is destudied experimentally
scribed using a potential or work function. Such vertex
system
is the ommatidia
develop- and simulated shapes. (A Upper) Mosaic ommatidia in which zero to four cone cells are mutant for the N-cadherin
son
of mutant
models are a coarse-grained representation of cell shape

FIG. 6. !Color online" Frequencies of polygon classes in the

sophila embryo germband at stage 6 !a" and late stage 8 !b":
erimental data !open columns, data from Ref. #13$" agree very
l with our theoretical predictions for a = 0.853 and a = 0.75, rectively !solid columns". From the experimental data #Figs. 2b
2c in Ref. #13$$ we also extracted the distribution of polygon

FIG. 7. !Color online" Structure of Drosophila wing disk, Hydra

tail epidermis, and Xenopus epidermis in terms of frequencies of
Nagai & Honda Phil. Mag. B vol. 81 (7) (2001)
polygon classes !data from Ref. #10$". The three sets of experimenHufnagel et al, PNAS vol. 104 (10) pp. 3835 (2007)
tal data agree nicely with the theory proposed in Ref. #10$. HowFarhadifar et al, Current Biology (2007)
ever, they are equally well described by our a = 0.79 tiling with the
Jlicher et al Phys. Rep. (2007)
polygon class frequencies virtually identical to that of Ref. #10$.
Hocevar & Ziherl PRE (2009)
The a = 0.75 tiling reproduces the observed frequency of quadrilatManning et al, PNAS (2010)
erals better.
Staple et al EPJE 33 (2) 117 (2010)
Katira et al Phys. Rev. Lett (2012)
Chiou et al PLOS Comp Bio 8 (5) e1002512 (2012)
areas andReview:
perimeters
barely et
affect
the topological
structure of
Fletcher
al Biophys.
J (2014)

the tiling reflected in the frequencies of polygon classes. We

e whether the truncated -catenin protein is also

Mechanical forces responsible for shape
generation
Cell-cell Adhesion
Adhesion molecules:
Cadherins, catenin, etc
Active cortical tension
Myosin II, actin
(experiments: Evans, Theory:
Joanny, Prost et al.)
Cortical Elasticity
Cytoskeletal networks

de Vries et al,
Development (2004)

Bulk effects
Bulk effects: fluid resists
volume change, cytoskeleton
resists shear

rec

Vertex models for tissues monolayers

A = single cell area
P = single cell perimeter

Mechanical energy for cell shape in confluent tissue:

Ecell = kA (A

A0 ) + k p P + P

2
3D incompressibility +2
A
0
p
0
resistance to height
fluctuations
Elasticity of acto-myosin

= k (A

A ) + k (P

P )

ring at cell periphery

Complete the square and

Non-dimensionalize:
Only two model
parameters:

Etissue =

X
i

p0 ) 2

(pi
r

Interfacial tension:
adhesion
cortical tension

+ (a

1)2

1. p0 = preferred perimeter: interfacial tension generated

by adhesion and cortical tension
2. r = inverse perimeter modulus: resistance to height
fluctuations normalized by perimeter contractility OR
ratio between bulk stiffness and interface stiffness

Migration in 2D happens via T-1 transitions

A T-1 transition

"T 1

E
Etot
/Ncells
tot/N
cells Etot
tot/Ncells
cells

2.602
2.602
2.602

"

"initial
`initial

2.600
2.600
2.600
2.598
2.598
2.598

E11
E

2.596
2.596
2.596

2.590
2.590
2.590
-0.2
-0.2
-0.2
2.595
2.595
2.595
2.590
2.590
2.590
2.585
2.585
2.585
2.580
2.580
2.580
2.575
2.575
2.575

2.594
2.594
2.594
2.592
2.592
2.592

E22
E

-0.1
-0.1
-0.1

0.0
0.0
0.0

`f inal

`
E111
E

2.570
2.570
2.570

2.565
2.565
2.565
-0.3
-0.3
-0.3

-0.2
-0.2
-0.2

0.1
0.1
0.1

E333

E
E

-0.1
-0.1
-0.1

0.0
0.0
0.0

0.1
0.1
0.1

0.2
0.2
0.2

E
E222
E

0.2
0.2
0.2

0.3
0.3
0.3

L
DB et al, Soft Matter (2014)

Energy barrier statistics

y=
0

k
(k

1)!

k 1

kx

" ( "/ ")

10

The functional form

of distribution is
universal

10

Only the mean

10

energy barrier
depends on r and p0

10

x=

"/ "(r, p0 )

DB et al, Soft Matter (2014)

DB et al, submitted (2014) arXiv:1409.0593

Average energy barrier

height is controlled by p0
Etissue =

X
i

"

r=

r=

p0 ) 2

(pi
r

+ (a

1)2

0.5

r=2

r = 10

Strong cortical tension

p0

Strong cell-cell adhesion

Has features of a rigidity transition,

how to prove it is a true phase transition?

scaling function approach a common curve, f& "z# $

z#=! , so that precisely at " % "c , !!1 $ $#=! as $ ! 0
[24]. This is shown as the dashed line in both Figs. 2 and 3.
A similar scaling collapse of ! has been found in simulations [20] of a sheared Lennard-Jones glass, as a function
_ but only
of temperature and applied shear strain rate %,
above the glass transition, T > Tc . By comparing the goodCritical
in jamming:
ness of thescaling
scaling collapse
as parameters are varied, we
estimate &
theTeitel
accuracy
of the
critical exponents to be
Olsson
PRL
2007
roughly # % 1:7 & 0:2 and ! % 1:2 & 0:2.
That the crossover scaling exponent ! > 0, implies that
$ is a relevant variable in the renormalization group sense

that vanished in the thermodynamic limit. With these observations, we leave the question of criticality at finite $ to
future work.
The critical scaling found in Fig. 3 strongly suggests that
point J is indeed a true second-order phase transition and
thus implies that there ought to be a diverging correlation
length & at this point. Measurements of dynamic (time
dependent) susceptibilities have been used to argue for a
scaling
networks:
divergentCritical
length scale
in bothin
theFiber
thermally
driven glass
al Nature
Physics
transitionBroedersz
[25] and the et
density
driven jamming
transition
[17]. Here
we consider
equal time transverse
velocity
Das,
Quint the
& Schwarz
PLoS One
2012
correlation function in the shear driven steady state,

Examples of critical scaling

ARTICLES
104

2D

g"x# % hvy4"xi ; yi #vy "xi ' x; yi #i;

b
10
2D

(7)

where vy "xi ; yi # is the instantaneous velocity

f/ in the y^
direction, transverse 10
to2the direction of the average shear
102
Rigid
flow, for a particle at position
(xi , yi ). The average is over
Unjammed/Fluid
particle positions and time. In the inset to Fig. 4 we plot
g"x#=g"0# versus x for
100
100three different values
104 of " at fixed
EMT Sim. !4
3DWe see that
$ % 10 and number of particles N % 1024.
1
g"x#10decreases to negative values at a well-defined
mini102
2
2
2
10 before decaying
10
10 to zero as x increases. We define &
mum,
0
3
1
to be10the
position of this minimum.
That 10
g"&#
< 0 indicates
Floppy
Network
4
that 10
regions
separated by a distance & are
anticorrelated.
Jammed/Solid
4
4
102
10
10
We 10
can
5 thus interpret the sheared flow in the unjammed
state as6due to the rotation of correlated10
regions
of length &.
4
10
4
0 104 108
10
10of
Similar
behavior,
leading
to
a
similar
definition
&, has
6
106 4 2
10
2 100 102 of
4 10
6 108 1010
nonaffine
10 10 100 102 104 106previously
108 1010been found [26]
104 in
10correlations
10the
displacements of particles in a Lennard-Jones
p glass, in
p
FIG. 3 (color online). Plot of scaled inverse viscosity
response to small elastic distortions.
*Shear
stress
shear stress
z ) $=j" ! "c j! for the
!!1 =j" ! "c j# vs scaled
As with viscosity, we expect the correlation length
data of Fig. 2. We find an excellent
the scaling
Figure 3collapse
| Scalingto analysis
ofform
the mechanics and anomalous elasticity.
a,b, Scaling
of the shear modul
*Bending
stiffness
&"";
$#
to
obey
a
scaling
equation
similar
to
Eq.
(6). We
of Eq. (6) using values "c % 0:8415, # % 1:65, f and ! % 1:2.
d 1
scaling form G|1p| = G (|1p| ),consider
with G in
units
of
/`
and
the
bending
rigidity,
, like
in units of
here the inverse0 correlation length &!1 , which
The dashed line represents the large z asymptotic dependence,
1 black, 10 2 magenta, 10 3 cyan, 10 4 red, 10 5 p
filament bending
rigidities
(2.in units
of should
`20 : 10vanish
!!1
to those used
in Fig.
at the jamming transition, obeying the
$z#=! . Data point symbolsofcorrespond
and the simulations (b). The asymptotic form of the scaling function for low is shown as a dashed gr
three-dimensional fcc lattice is178001-3
shown as an inset in b. The EMT exponents are fCF = 1, = 2. In contra
G p f

f/

*Shear Modulus

G p f

*Inverse viscosity

A critical rigidity transition controlled by p0

Average energy barrier height obeys the scaling function

" = |p0

r "/|p0

p0 |

-
-
-
-

p0 < p0

p0 >

p0 |

p = 3.813 0.005
= 4.0 0.4
= 1.0 0.2
p0

z = r/|p0

p0 |

|p0

p0 |

Comparison to other rigidity transitions

Order
Parameter
Distance to
the critical
point

Exponents

DB et al, submitted (2014) arXiv:1409.0593

Rigidity Transition,
Confluent Tissues

Jamming
Transition (2D)

Cross-linked
Fiber Networks
(2D)

Energy barrier
height

Viscosity

Shear modulus

Preferred cell
perimeter

Packing
density

Bond occupation
probability

|p0

p0 |

J|

|p

pc |

=1

= 1.65

= 1.4

=4

= 1.2

=3

Jamming: Olsson & Teitel PRL 2007

Fiber networks: Broedersz et al Nature Phys 2011
Das, Quint & Schwarz 2012

Why is the transition at p0=3.813?

Preferred Perimeter
p0

Corresponding
Polygon (with a=1)

Number of sides
z

3.54

3.72

3.81

4.56

Isostatic conjecture

Number of degrees of freedom

2V

Number of constraint equations

3N
For a mechanical equilibrium solution to exist,
Number of Degrees of Freedom Number of Constraints

2V
Since

E+N =0
E = zN/2

zisostatic = 5

3N
Topological constraint for flat surface
Each edge is shared by two cells

p0

= Ppentagon 3.813

Vibrational modes give a clear picture of

linear mechanical response

N (!)

10

p0

p0 >
Appearance of Flu
id
-1 Soft Modes

10

Ri
gid

-2

p0 < p0

10

p
=
0
-3

10

10

-2

Debye Scaling

p0

N (!) ! d
or D(!) ! d

10

-1

10

10

A mean-field formulation
A cell can be described by a shape tensor

=
R

m=

R12
diagonalize
R22
!

A / det(R),
p0 ) 2
r

1
1

+ (a

0
2

P / tr(R)
E (r, p0 )m2 + (r, p0 )m4

1)2

R11
R21

Etissue =

X (pi

p0

A brief summary so far

Using energy barriers for local rearrangements
Rigid

Rigidity Transition
Fluid like

p0

p0

= 3.813

Next: What happens when cell motility is included

Does higher cell activity make the tissue more fluid-like?

How does the rigidity transition control the glass

transition?

Does this model exhibit signatures of glassy dynamics?

From rigidity transition to glass transition

Self-propelled Voronoi-cells
(1) Calculate energy based on shapes of Voronoi tessellation
Etissue =

X (pi
i

p0 ) 2
r

+ (a

1)2

(2) Integrate equation of motion for points

d~ri
b
=
dt

@
Etissue + v0 n
i
@~ri

v0 = active cell motility force

n
i = cell polarization

(DB et al, in preparation) 2015

Po

n
o
i
t
a
z
i
r
la

(3) Cell polarization is random

n
i = (cos

i , sin

d i
= i
dt

i)

hi (t)j (t )i = 2Dr
Szab et al PRE 2006
Henkes, Fily & Marchetti PRE 2011
Li & Sun Biophys. J. 2014

ij

(t

t)

 ( )

Tissue dynamics movies

Cortical tension

Cell-cell adhesion

Mean-squared displacement

increasing p0

h r (t)i

-
-
-

Self-diffusivity is a good indicator of dynamics

2

h r(t) i
Ds = lim
t!1
4t

Quantifying dynamical
behavior

Fluid-like

Solid-lik
e

What about cell

shapes?

An order parameter based on the observed cell shape

q = hp/ ai
m.s.d boundary

q = hp/ ai

()

Fluid

q = 3.81

Solid

Phase diagram needs to be modified

Adhesion (p0)
1/adhesion

Solid
Jammed

Cell
Motility
1/Density

Experimental verification

Primary Bronchial Epithelial Cells

With Fredberg group, Harvard School of Public Health

Example of constant density

glass transition: Tissue remains
confluent as it matures, without
proliferation
Cilia
Nuclei
Cytoplasm

As tissue matures, it becomes

more jammed/glassy: the
maturation process is also a
model for injury repair.

Mature tissue develop cilia,

which help to trap mucus:
failure in maturation can result in
dysfunction in cilia formation

Park, Kim, DB, et al, Submitted (2014)

g.2 2
From Non-asthmatic
Non-asthma Patient
Non-asthma
G 6
GDay

100
m
100
m

100m
m
100

KK

Day
I I 14

LL

0.5
m/min
0.15 0.5
0.05 0.1
0.1 0.15
m/min
0 0 0.05
m/min
m/min

Day 14
Day 14

FF

Day 8

Day 10
CC

H 10
H
Day

Day 10
Day 10

EE

Day 6

B BDay 6

JJ

Day 6
Day 6

DD

Day 3

A ADay 3

From Asthmatic
Asthma Patient
Asthma

00

0.5m/min
m/min
0.150.5
0.05 0.1
0.1 0.15
0.05
m/min
m/min

Fig. 4

Can cell shapes predict mechanical rigidity?

A"

Non-asthma

B"

Asthma

4.2

4.8

4.1

10

15

Yes! q predicts the onset of glass

transition
No fitting parameters needed!

Jamming

qCell
=shape
hp/(p )ai

3.8

4.4
4.2

3.8

6
Day

Predicted
threshold

4.0

3.6

unjamming

q = hp/ ai

4.6

3.9

Fraction of cells that moved

more than 15% of their diameter
6

10

14

Day

Park, Kim, DB, et al, Submitted (2014)

Conclusions
Liquid-solid

transition in tissues is important for

biological processes.

Vertex

model for confluent tissue to predict a rigidity

transition independent of density.

This

rigidity transition controls a line of glass transition

points at finite cell motilities

We

predict an observable number (cell perimeter to

area ratio), that distinguishes between fluid/solid.
This

is precisely realized primary human cells from

non-asthmatic and asthmatic patients

T1

rosette

Karen Kasza PNAS 2014

2.0
1.0
0.0

WT EE AA sqh1

Cell sorting and segregation

40

time (min)

local (T1 process) and

which are mediated by
ltiple cell interfaces, reTo test whether RLC
relative contributions of
equency of T1 processes
in RLC-WT, contracting
in T1 processes (48

B
T1 process

time (min)

Abnormal cancer-like cells

embedded
in normal cells
er-Order
Rearrangements.

Rosette formation
in Drosophila during convergent
extension

Rosette

LC-WT and 16 min for

ith the expectation that
increase myosin activity
angement time was reboth faster contraction
1
C-AA and sqh display
accelerates rearrangearrangement is tuned by

rate (m/min)

leach fluorescence intensity

t1/2 compared with RLC-WT.

elongation, but RLC-EE displayed a 25% reductio

planar polarity (2.01 0.13) (Fig. 4 A, B, and E and
Ongoing
work
contrast,
RLC-AA
displayed an 83% reduction in p
polarity (1.23 0.03) (Fig. 4 A, B, and F and Fig.

20
0

WT EE AA sqh1

40
20
0

WT