B21a7e87 1

NanoDrop 2000c
Spectrophotometer
Both microvolume and cuvette
measurements, full spectral data
NanoDrop 8000
Spectrophotometer
Measure up to eight
1-2 L samples at a time
NanoDrop Lite
Spectrophotometer
Basic microvolume
measurements of puried
DNA, RNA and proteins
No matter the sample concentration or volume, scientists worldwide rely
on the patented Thermo Scientic
NanoDrop
technology to evaluate their

DNA, RNA, and protein samples. Expect reproducible quantity and purity
data from microvolume samples without having to perform dilutions or guess
which accessory you need. Data you can trust in seconds, backed by over
10,000 peer reviewed citations.
thermoso|ent|fo.oom/nanodrop
d
s
D
N
A

c
o
n
c
e
n
t
r
a
t
i
o
n

[
n
g
/
L
]
NanoDrop 2000c Performance Data
2000
1500
1000
500
0
1 2 3 4 5 6 7 8 9 10
Replicate

2
0
1
3

T
h
e
r
m
o

F
|
s
h
e
r

S
o
|
e
n
t
|
f

o

l
n
o
.

A
|
|

r
|
g
h
t
s

r
e
s
e
r
v
e
d
.

A
|
|

t
r
a
d
e
m
a
r
k
s

a
r
e

t
h
e

p
r
o
p
e
r
t
y

o
f

T
h
e
r
m
o

F
|
s
h
e
r

S
o
|
e
n
t
|
f

o

l
n
o
.

a
n
d

|
t
s

s
u
b
s
|
d
|
a
r
|
e
s
.

Discover More
Molecular Structures and Interactions
www.tainstruments.com
Nano ITC
Protein Protein Interactions
Pricrilize Drug Ccnciccle Tcrgel
Interactions
Vcliccle Ligcnc Bincing lc
Nucleic Acic
Cucnlify Lclh Fnlhclpy cnc Fnlrcpy
in Cne Tilrclicn
Nc lcLeling cr immcLilizclicn reuirec
Nano DSC
Protein Structural Domains and Stability
Fxcipienl lnfuence cn lcleculcr SlcLilily
SlcLilily cf Bicphcrmcceuliccls
Direcl lecsure cf lcleculcr
Thermccyncmics
Foreword
We are pleased to present this Best of reprint collection, which provides a chance
to refect on what has caught the attention of Biophysical Journal readers in 2013. This
collection includes a selection of thirteen of the most-accessed articles across a range of
topics. Article selection is primarily based on the number of requests for PDF and full-text
HTML versions of a given article. We acknowledge that no single measurement can truly
be indicative of the best research papers over a given period of time. This is especially
true when suffcient time has not necessarily passed to allow one to fully appreciate the
relative importance of a discovery. That said, we think it is still informative to look back at
the scientifc communitys interests in what has been published over the past year.
In this collection, you will see a range of the exciting topics, including cell biophysics,
motors and cytoskeleton, single-molecule microscopy, membrane biophysics, systems
biophysics, biomolecular structure, biophysical methods, and channel electrophysiology,
that have widely captured the attention and enthusiasm of our readers. They also represent
several of the different types of papers that BJ publishes: two Biophysical Reviews, two
Biophysical Letters, and nine regular articles.
We hope that you will enjoy reading this special collection and that you will visit http://www.
cell.com/biophysj/home to check out the latest fndings that we have had the privilege to
publish. To stay on top of what your colleagues have been reading over the past 30 days,
check out http://www.cell.com/biophysj/mostread. Also, be sure to visit http://www.cell.
com to fnd other high-quality papers published in the full collection of Cell Press journals.
Finally, we are grateful for the generosity of our sponsors, who helped make this reprint
collection possible.
Leslie M. Loew
Editor in Chief
For information for the Best of Series, please contact:
Jonathan Christison
Program Director, Best of Cell Press
e: jchristison@cell.com
p: 617-397-2893
t: @CellPressBiz
Visualize Beyond the 200nm Limit with Superresolution
ELYRA from Carl Zeiss incorporates up to two superresolution technologies, structured
illumination and PALM/dSTORM, for twice the resolution power of confocal or single
molecule resolution down to 20nm. Easily incorporate correlative studies and image
the exact same area with superresolution as well as electron microscopy.
www.zeiss.com/superresolution
Reviews
Biophysical Letters
Articles
Best of 2013
Modeling Stochastic Kinetics of Molecular Machines at
Multiple Levels: From Molecules to Modules
Nanoscale Distribution of Ryanodine Receptors and
Caveolin-3 in Mouse Ventricular Myocytes: Dilation of
T-Tubules near Junctions
Systems Biophysics of Gene Expression
Biophysical Model of Bacterial Cell Interactions with
Nanopatterned Cicada Wing Surfaces
Tilting and Wobble of Myosin V by High-Speed Single-
Molecule Polarized Fluorescence Microscopy
Temperature Dependence of the DNA Double Helix at the
Nanoscale: Structure, Elasticity, and Fluctuations
Lateral Membrane Diffusion Modulated by a Minimal
Actin Cortex
S3-S4 Linker Length Modulates the Relaxed State of a
Voltage-Gated Potassium Channel
Mechanism of Membrane Permeation Induced by
Synthetic -Hairpin Peptides
Distinct Stages of Stimulated FcRI Receptor Clustering
and Immobilization Are Identied through Superresolution
Imaging
Building KCNQ1/KCNE1 Channel Models and Probing
their Interactions by Molecular-Dynamics Simulations
Two-Photon Excitation STED Microscopy in Two Colors
in Acute Brain Slices
Kinetics and Energetics of Biomolecular Folding and
Binding
Debashish Chowdhury
Joseph Wong, David Baddeley, Eric A. Bushong, Zeyun Yu,
Mark H. Ellisman, Masahiko Hoshijima, and Christian Soeller
Jose M.G. Vilar and Leonor Saiz
Sergey Pogodin, Jafar Hasan, Vladimir A. Baulin, Hayden K.
Webb, Vi Khanh Truong, The Hong Phong Nguyen, Veselin
Boshkovikj, Christopher J. Fluke, Gregory S. Watson, Jolanta
A. Watson, Russell J. Crawford, and Elena P. Ivanova
John F. Beausang, Deborah Y. Shroder, Philip C. Nelson, and
Yale E. Goldman
Sam Meyer, Daniel Jost, Nikos Theodorakopoulos, Michel
Peyrard, Richard Lavery, and Ralf Everaers
Fabian Heinemann, Sven K. Vogel, and Petra Schwille
Michael F. Priest, Jrme J. Lacroix, Carlos A. Villalba-Galea,
and Francisco Bezanilla
Kshitij Gupta, Hyunbum Jang, Kevin Harlen, Anu Puri, Ruth
Nussinov, Joel P. Schneider, and Robert Blumenthal
Sarah A. Shelby, David Holowka, Barbara Baird, and
Sarah L. Veatch
Yu Xu, Yuhong Wang, Xuan-Yu Meng, Mei Zhang, Min Jiang,
Meng Cui, and Gea-Ny Tseng
Philipp Bethge, Ronan Chreau, Elena Avignone, Giovanni
Marsicano, and U. Valentin Ngerl
Christopher A. Pierse and Olga K. Dudko
Modeling Stochastic Kinetics of Molecular Machines at Multiple Levels:
From Molecules to Modules
Debashish Chowdhury*
Department of Physics, Indian Institute of Technology, Kanpur, India
ABSTRACT A molecular machine is either a single macromolecule or a macromolecular complex. In spite of the striking
supercial similarities between these natural nanomachines and their man-made macroscopic counterparts, there are crucial
differences. Molecular machines in a living cell operate stochastically in an isothermal environment far from thermodynamic
equilibrium. In this mini-reviewwe present a catalog of the molecular machines and an inventory of the essential toolbox for theo-
retically modeling these machines. The tool kits include 1), nonequilibrium statistical-physics techniques for modeling machines
and machine-driven processes; and 2), statistical-inference methods for reverse engineering a functional machine from the
empirical data. The cell is often likened to a microfactory in which the machineries are organized in modular fashion; each
module consists of strongly coupled multiple machines, but different modules interact weakly with each other. This microfactory
has its own automated supply chain and delivery system. Buoyed by the success achieved in modeling individual molecular
machines, we advocate integration of these models in the near future to develop models of functional modules. A system-level
description of the cell fromthe perspective of molecular machinery (the mechanome) is likely to emerge fromfurther integrations
that we envisage here.
INTRODUCTION
Some of the greatest thinkers of all time, ranging from
Aristotle to Descartes and Leibnitz, have been drawn by
some of the striking analogies between a living organism
and a man-made machine. However, the existence of molec-
ular machines in living bodies was rst speculated in the
17th century by Marcelo Malpighi (1). A molecular
machine is either a single protein or a macromolecular com-
plex (210). For its operation, a molecular machine needs an
energy input. It has an engine that transduces energy. In this
mini-review, we focus almost exclusively on specic types
of molecular machines, called molecular motors, whose
output is mechanical work (1121).
The cytoskeletal motor proteins drive motility and
contractility at the subcellular level. However, cell motility
and morphogenesis, which are also driven by these motors,
are beyond the scope of this mini-review. Many other
specialized enzymes involved in the manipulation, synthe-
sis, and degradation of macromolecules (e.g., proteins and
nucleic acid strands) can also be regarded as molecular
motors (4,22).
A cell has, at least supercially, similarities with a
microfactory (5) in which most of the crucial intracellular
functions require the coordination, cooperation, and compe-
tition of several machines that together form a functional
module (23). The interactions between different modules
are relatively weak, and the component machines of a single
functional module need not be contiguous in space.
Theoretical models are not only useful for systematic
analyses of the vast amounts of experimental data obtained
at different levels of spatiotemporal resolution, they can also
be used to predict new results and guide further experiments
(24). In a recent review published elsewhere (21), I have
presented a detailed overview of the results of multidisci-
plinary research on the kinetic models of molecular motors
and motor-driven processes, as well as models of a few
important modules. This mini-review is a nontechnical sum-
mary of that long review. Here I cite mostly books and
review articles; interested readers can refer to Chowdhury
(21) for technical details and a comprehensive bibliography.
CATALOG OF MOLECULAR MACHINES AND
FUELS
Microtubules (MTs) and lamentous actin (F-actin) are the
two types of cytoskeletal polar laments that also serve as
tracks for the cytoskeletal motor proteins (11). These
motors, which are listed in Table 1, function as intracellular
porters (25). These proteins carry intracellular cargoes (e.g.,
vesicles and organelles) over long distances by walking
along their respective tracks (Fig. 1). The porters power
their walk by hydrolyzing ATP, which is the most widely
used fuel for molecular machines.
In contrast, the motors listed in Table 2 slide one cytoskel-
etal polar lament with respect to another (15,20). The
sliding occurs when a slider motor that cross-links the two
laments tends to walk on both simultaneously, hydrolyzing
ATP (Fig. 1). Some sliders work in groups, and each group
detaches from the lament after every single stroke. These
are often referred to as rowers because of the obvious anal-
ogy to rowing, where the oars remain in contact with water
for a very brief period during each stroke (25). Sliders and
rowers drive contractility at cellular and subcellular levels.
Submitted August 23, 2012, and accepted for publication April 17, 2013.
*Correspondence: debch@iitk.ac.in
Editor: Brian Salzberg.
2013 by the Biophysical Society
0006-3495/13/06/2331/11 $2.00 http://dx.doi.org/10.1016/j.bpj.2013.04.042
Biophysical Journal Volume 104 June 2013 23312341 2331
Some cytoskeletal motor proteins are neither porters nor
rowers. Instead, they target MT tips and upon arrival start
to depolymerize the MT itself. For obvious reasons, these
are called depolymerases (26); their lament-chipping
activities are also powered by ATP hydrolysis. Two families
of these motors and their distinct characteristics are listed in
Table 3.
A polymerizing lamentous biopolymer can exert a push-
ing force against a light object (e.g., a membrane), thereby
mimicking a nanopiston (6) (Fig. 2). Similarly, a depolyme-
rizing MT can pull a light, ring-like object by inserting its
hook-like, outwardly curled, depolymerizing tip into the
ring (27) (Fig. 2). A few typical examples of such force
generators are listed in Table 4.
Some motors, while anchored on a xed support, can
import or export macromolecules across either the plasma
membrane or, in the case of eukaryotes, the internal mem-
branes of a cell. Packaging of the genome inside a prefabri-
cated empty viral capsid is facilitated by packaging motors,
located at the entrance of a viral capsid, that push the viral
genome into the capsid (9). These movers and packers are
among the strongest molecular motors discovered so far
the force generated can be as large as ~60 pN. Helicases,
which unwind double-stranded DNA, also share the dening
characteristics of molecular motors (13). Topoisomerases
are molecular machines that untangle DNA by passing
one DNA strand through a transient cut in another (4).
Polynucleotides (DNA and RNA) and polypeptides (pro-
teins) are linear polymers. The sequence of the monomeric
subunits of each of these polymers is dictated by that of the
corresponding template. A polymerizing machine selects
the successive monomers, as directed by the corresponding
template, and adds these step by step, thereby elongating the
product polymer. Depending on the nature of the template
and product polymers, specic machines for template-
directed polymerization are used (10); these are listed in
Table 5. Because the template also serves as the track for
the polymerizing machine, and movement of the machine
along the track is powered by input chemical energy, these
machines are also regarded as motors (4,22) (Fig. 3).
The cell also uses specialized machines for degradation of
the polymers. The exosome and proteasome are nanocages
a
b
FIGURE 1 Schematic representation of (a) porters and (b) sliders. In
panel a, a vesicular cargo (represented by the gray ellipse) is being hauled
by a two-headed kinesin (with the two heads denoted by black circles)
walking on an MT protolament that consists of periodic repetition of
a b tubulin dimers. In b, the sliding of the two MTs, which are antipar-
allel to each other, takes place when the heads of a four-headed kinesin (say,
Eg5) step toward the ends of the respective MTs, as indicated by the
curved arrows. The directions of the resulting movement of the MTs are
indicated by the straight arrows.
TABLE 2 Examples of rowers and sliders, and their locations
and biological functions
Motor family (Ref.) Filament Location and function
Myosin-II (79) F-actin actomyosin cross-bridge, muscle
contraction
F-actin stress ber, contraction of
nonmuscle cells
F-actin contractile ring, eukaryotic
cell division
Kinesin-5, kinesin-14 (80) MT cross-linked MT, mitotic spindle
size control
Axonemal dynein (81) MT axoneme, eukaryotic agellar
beating
Dynein (82) MT megakaryocytes, blood platelet
formation
TABLE 1 Examples of ATP-driven porters along with their
corresponding tracks
Motor family
(Ref.) Ancestor Track
Step
size
Location and
functions
Myosin-Vand
myosin-VI
(7375)
G-protein F-actin 36 nm melanophore
transport
Kinesin-1 (76,77) G-protein MT 8 nm transport in axons
and fungal hyphae
Cytoplasmic
dynein (77,78)
AAA proteins MT 8 nm transport in axons
and agella
The actual step size can be an integral multiple of the minimum value
quoted in this table.
TABLE 3 Examples of MT depolymerase motors and their
distinct features
Motor family
(Ref.)
Target
direction
Mode of reaching
the target
Purpose of
ATP hydrolysis
Kinesin-13 (26) either end
of MT
unbiased diffusion only for
depolymerization
Kinesin-8 (26) only end
of MT
directed walking for both walking and
depolymerization
Biophysical Journal 104(11) 23312341
2332 Chowdhury
into which RNA and proteins, respectively, are translocated
and shredded into smaller fragments (28).
Two extensively studied rotary motors are ATP synthase
(29) and the bacterial agellar motor (30); the former syn-
thesizes ATP from ADP, whereas the latter rotates the agel-
lum that propels swimming bacteria in bulk uid medium
(13). The ion-motive force (IMF) created by the concentra-
tion gradient of ionic charges across membranes drives most
of these rotary motors, exploiting the spatial arrangement of
some static charges and the trajectories of mobile charges
(Figs. 4 and 5).
SOME FUNDAMENTAL BIOPHYSICAL QUESTIONS
ABOUT MOLECULAR MACHINES
Operational mechanism of a single motor
Biophysical characteristic properties of a molecular machine
Directionality and processivity are two characteristic prop-
erties of a family of motors. Members of the same family
move in the same direction on a polar lamentous track.
Loosely speaking, the higher the processivity of a motor,
the longer is the distance it covers in a single run in between
its attachment to the track and its complete detachment from
it. The average fraction of the cycle time in which each head
remains attached to the track is called its duty ratio.
An external force that opposes the natural walk of a motor
is called a load force. The average velocity of a motor
decreases with increasing magnitude of the load force, even-
tually vanishing at a value that is called the stall force. The
force-velocity relation is one of the most fundamental char-
acteristics of a molecular motor. Two different motors with
identical average velocities may exhibit widely different
types of uctuations. Therefore, one can obtain a more
detailed characterization of the stochastic stepping kinetics
of a motor by examining the distribution of its dwell times
at successive spatial positions on its track.
Over the last two decades, it has become possible to
account for the above-mentioned characteristic properties
of a given family of motors in terms of the structure and
dynamics of its members. Model building has been assisted
by insights gained from experimental studies of these
motors, and in the next three sections we discuss the theoret-
ical and experimental approaches used in such studies.
Energy transduction: power stroke and Brownian ratchet
For a force-generating motor, efciency and power output are
good measures of its performance. In fact, efciency at
maximum power, rather than the maximum of the efciency
itself, is the main quantity of interest for all real motors whose
cycle time is nite (31). However, in contrast to the unique
denition of efciency of transduction of macroscopic
engines, several alternative denitions are possible in the
case of nanomachines operating against a viscous drag of
the surrounding aqueous medium. Different efciencies
characterize different aspects of energy transduction.
The power stroke of a molecular motor is bound to be
noisy because of the Brownian forces acting on it. Interest-
ingly, it is possible for a molecular motor to transduce
energy by a different mechanism, called Brownian ratchet
(32,33), that does not have any macroscopic counterpart.
Suppose the motor suffers unbiased forward and backward
displacements randomly because of the Brownian forces
acting on it. If input energy is utilized to operate a device
that prevents (or suppresses) backward movements without
affecting the forward displacements, then, on average, the
motor will move forward. The actual mechanism for a real
molecular motor may be a combination of the two idealized
extremes, namely, power stroke and Brownian ratchet.
The original actomyosin cross-bridge model of muscle
contraction, proposed by Andrew Huxley, was later inter-
preted as a Brownian ratchet. Other well-known examples
of (approximate) physical realizations of Brownian ratchet
include the movement of single-headed kinesin KIF1A
FIGURE 2 (a and b) Schematic representation of (a) a nanopiston and (b)
a nanohook. In panel a, a polymerizing MT pushes a membrane (repre-
sented by the gray plane) in the direction indicated by the arrow. In b,
the curved tips of a depolymerizing MT pull a ring in the direction indicated
by the arrow.
TABLE 4 Examples of nanopistons and nanohooks
Polymer (Ref.) Mode of force generation Function
MT (83) polymerization organizing cell interior
F-actin (6) polymerization cell motility
MT (27) depolymerization chromosome segregation
TABLE 5 Types of polymerizing machines, the templates that
dictate the sequence of monomers of the polymeric products,
and the corresponding function
Machine (Ref.) Template Product Function
DNA-dependent DNA
polymerase (84)
DNA DNA DNA replication
DNA-dependent RNA
polymerase (85,86)
DNA RNA transcription
RNA-dependent DNA
polymerase (87)
RNA DNA reverse transcription
RNA-dependent RNA
polymerase (88)
RNA RNA RNA replication
Ribosome (89) mRNA protein translation
Stochastic Kinetics of Molecular Machines 2333
along the MT track in vitro, messenger RNA (mRNA)
export from the nucleus, transcription by T7 RNA polymer-
ase, translocation of ribosome during translation, and rota-
tion of the F0-motor of mitochondria (see Chowdhury
(21) for references and details of the original works).
Speed versus delity: quality-control mechanism
Not all molecular motors are primarily force generators. For
example, for polymerases and ribosomes, the quality of
performance is best characterized in terms of the accuracy
of the template-directed polymerization that they carry out
(10). Such motors have to optimize the opposing demands
of speed and accuracy.
While selecting the monomeric subunits of the elongating
polymer, if the machine were to discriminate between the
correct and incorrect subunits based solely on the differ-
ences in the free energies of binding with the corresponding
substrates, the error rates would be orders of magnitude
higher than the observed values. Kinetic proofreading
(34,35) is one of the kinetic mechanisms that amplify the
achievable accuracy of template-directed polymerization.
This process involves dissipation of energy, a typical
example being the hydrolysis of one GTP molecule by a
ribosome for selection of the amino acid substrate during
translation.
The quality-control system of the machines for template-
directed polymerization is not restricted only to the stage of
subunit selection (10). Quality-control mechanisms have
been captured in kinetic models of transcription, translation,
and replication (21).
Coordination, cooperation, and competition
among several motors
Cooperative phenomena involving multiple molecular
motors are observed at several different spatiotemporal
scales in an organism. A packaged cargo (e.g., a vesicle)
is normally hauled by a team of motors. Often organelles
exhibit bidirectional movements on a track. The uctuation
in the direction of the resultant force experienced by the
cargo indicates the involvement of two teams of oppositely
directed motors (e.g., kinesins and dyneins) and may be a
manifestation of an ongoing tug-of-war between these two
teams (36). Given the force-velocity relation of the individ-
ual motors, how would the average velocity of a cargo
and the force required to stall it vary with the number of
motors (37)? How do these results change with the
decreasing processivity of the motors and increasing soft-
ness of the cargo (38)?
When many motors translocate simultaneously along the
same track, their collective movement resembles vehicular
trafc on a highway. The theoretically predicted possibility
of a molecular motor trafc jam is being pursued experi-
mentally. Although jamming of the cytoskeletal motors
would have important implications for cargo transport, jam-
ming of RNA polymerases and ribosomes would severely
affect gene expression.
Several complex machineries in a living cell may be
regarded as machines within a machine (39), typical exam-
ples being a replisome that replicates genomic DNA, and a
mitotic spindle that segregates postreplication eukaryotic
chromosomes (40). A replisome consists of a DNA-depen-
dent DNA polymerase and a few other accessory machines.
This is in contrast to genome-wide replication in eukaryotic
cells, which requires the coordination of many replisomes
that are spatially distributed instead of being segregated
contiguously. A mitotic spindle self-organizes by inte-
grating several different types of machines, which include
porters, sliders, chippers, nanopistons and nanohooks, etc.,
within a single assembly (41).
THEORETICAL MODELING OF MOLECULAR
MOTORS AT DIFFERENT SCALES
In the following subsections, we mention a few alternative
formalisms that model molecular motors and motor-driven
processes at different levels of spatiotemporal resolution.
We also explore the possible relations between them.
FIGURE 3 Schematic representation of template-
directed polymerization of a DNA chain using another
DNA strand as the template (i.e., DNA replication).
Only a short stretch of the partially replicated template,
excluding the site of initiation and termination of replica-
tion, is shown. Moreover, the strands are depicted as
straight tapes, ignoring their actual conformations.
FIGURE 4 Schematic representation of the turnstile model of rotary
motors. Upon entering from the bottom entrance, the mobile ions take a
ride on the rotor. The electrostatic interactions among the charges on the
stator and the rotor, as well as the rotational Brownian motion of the rotor,
result in the noisy directed rotation of the rotor. Each hitchhiking ion disem-
barks from the rotor after being transported up to a certain distance by the
rotor. Once the ion leaves, the rotor remains locked in its current position
and waits for the arrival of the next ion. The F0 part of the ATP synthase
exploits this mechanism (adapted from Chapter4 of Schliwa (13)).
2334 Chowdhury
Moreover, whenever possible, we mention a few alternative
formalisms at the same level of spatiotemporal resolution
and explain their relative advantages and disadvantages.
In general, to model molecular motors, one must make
four key choices: 1), the choice of the degrees of freedom,
or dynamical variables, consistent with the intended level
of spatiotemporal resolution; 2), the choice of the form of
the interactions between the variables; 3), the choice of
which dynamical equations to use depending on the nature
of the dynamical variables; and 4), the choice of which
solution methods to use to calculate the quantities of interest
under the given initial and/or boundary conditions. These
choices are normally guided by the questions intended to
be addressed by the model. Once these selections are
made, the aim of the theoretical model might be merely to
explain the fundamental principles underlying some generic
phenomena exhibited by all (or at least a class of) motors at
that level. However, it is equally (if not more) important to
understand the reasons for the observed diversities among
different motors in terms of the key structural and/or dynam-
ical features of those motors. Examples of both types of
models are discussed critically in my long review (21).
Fully atomistic and coarse-grained models
The relevant timescales for the kinetics of an entire molec-
ular motor are too long to be accessed by a molecular-
dynamics (MD) simulation of an atomistic model with the
currently available computational resources. However, MD
has been used successfully to study subprocesses of shorter
durations. Moreover, normal-mode analysis (NMA) of a
fully atomistic model, assisted by experimental insight,
helps to elucidate the collective conformational dynamics
of a molecular motor. Because of the macromolecular nature
of the motors, NMA of atomistic models are usually very
computer intensive.
For a coarse-grained description of a motor, a group of
atoms is represented by a single bead, and the beads are
assumed to interact with each other through appropriate
effective potentials. For example, one can develop an elastic
network model for a molecular motor by assigning a bead to
each amino acid and postulating that these beads are con-
nected by harmonic springs. The collective modes of the
coarse-grained model can be obtained by carrying out an
NMA (42). NMA results have been reported for coarse-
grained models of myosin, dynein, and ATP synthase,
among others (21).
However, in the absence of additional experimental infor-
mation, it is very difcult to identify unambiguously which
of the normal modes is functionally relevant for the given
motor. Obviously, NMA is not suitable for studying dynam-
ical processes that involve a low degree of collectivity. A
coarse-grained approach cannot resolve important chemical
details. For example, ATP binding, ATP hydrolysis, and
release of ADP and P
i
are captured indirectly by the making
and breaking of elastic links and relaxation of elastic strain
in the ATP-binding region of the coarse-grained model of
the motor. Therefore, a MD simulation of a fully atomistic
model of the ATP-binding site and its immediate surround-
ings, together with that of the ligands, would be more appro-
priate for elucidating the molecular mechanism of the
ATPase activity of a motor.
Stochastic mechanochemical model: wandering
on landscapes
In this section, we discuss an alternative approach that
involves far fewer dynamical variables than the methods
discussed above and is capable of capturing noncollective
dynamical processes. The key mathematical concepts at
the foundation of this formalism are summarized in Section
A of the Supporting Material.
The conformations of a molecular motor are described by
specifying the positions of all the constituent atoms in the
three-dimensional space. Since a motor is either a protein
or a macromolecular complex, it has a large number of
degrees of freedom. In the coarse-grained description that
we discuss in this subsection, only a few dynamical vari-
ables are treated explicitly; the remaining degrees of
freedom of the motor as well as those of the aqueous me-
dium are assumed to constitute a bath. The effects of the
bath enter into the equations of motion of the motor via
two terms that represent a viscous damping force and a
random (Brownian) force.
For chemically driven mechanical movements of a motor,
let us now divide the explicit dynamical variables into two
classes: mechanical variables and chemical variables. For
simplicity, let us discuss only a minimal version of such a
model. This version has only one mechanical variable, x,
which denotes the position of the center of mass of the
motor along its track, and a single chemical variable, x,
which accounts for the progress of the chemical reaction
that supplies the chemical input energy to the motor. The
FIGURE 5 Schematic representation of the turbine model of rotary
motors. The rotor is decorated with rows of charges that are tilted with
respect to the stator. Protons are constrained to move straight, hopping
from one binding site to the next in the stator. Thus, as the protons ow
down the stator along their concentration gradient, their interactions with
tilted rows of charges on the rotor rotate the rotor in a plane perpendicular
to the stator. The agellar motor of bacteria is believed to be a physical real-
ization of this mechanism (adapted from Chapter4 of Schliwa (13)).
free energy, Ux; x, which is also called the potential of
mean force, can be graphically represented as a landscape
(Fig. 6), where the height of the landscape at any point
with coordinates x; x represents the corresponding potential
Ux; x.
If the motor-binding sites on the track form a periodic
array, a cross-section of the landscape parallel to x (i.e.,
for x constant) must also be periodic; all the local minima,
which coincide with the location of the motor-binding sites
on the track, are equally deep (Fig. 7 a). However, within
each period it is, in general, not symmetric about the loca-
tion of the maximum. In contrast, a cross-section of this
landscape parallel to x (i.e., for x constant), which also
consists of local minima separated by potential barriers, is
tilted forward (Fig. 7 b) so that the bottom of the successive
minima is deeper by jDGj, which accounts for the lowering
of free energy caused, for example, by ATP hydrolysis.
The kinetics of the motor is mathematically described by
the Brownian motion of a particle in this time-independent
potential landscape. A coupling between the mechanical
and chemical cycles in this space gives rise to a chemically
driven mechanical motor. This formulation is useful for an
intuitive physical explanation of the coupled mechanochem-
ical kinetics of molecular motors (43).
Next, let us consider those special situations in which the
chemical states of the motor change in discrete jumps from
one long-lived state to another. We label the chemical states
by the integer index m 1%m%m. As an example, m 4
may be assigned to a motor powered by ATP hydrolysis to
denote the following four distinct chemical states: 1), ligand
free; 2), ATP bound; 3), ADP-P
i
bound; and 4), ADP bound.
During the period in which the motor dwells in the chemical
state n, its mechanical variable x can continue to change as
dictated by the corresponding time-independent potential
U
n
x. The transition from one chemical state to another is
so rapid that during such a transition, the mechanical vari-
ables remain frozen. No mixed mechanochemical transition
takes place in such special situations because of the clear
separation of the timescales of variation of the mechanical
and chemical variables. A purely chemical transition n/l
causes the corresponding change U
n
x/U
l
x of the
potential prole. A sequence of chemical transitions is
accompanied by the corresponding sequential change of
the potential prole U
m
x. The kinetics of the center of
mass of the motor is described by the Brownian motion
of a particle in the time-dependent potential U
m
x, where
m varies with time by discrete jumps.
Mechanochemical models based on explicit interactions
among the constituent parts of a motor have been developed
to elucidate the mechanisms of myosin-V, myosin-VI, and
dynein (21). The number of mechanical variables and their
nature depends on the molecular motor under consider-
ation. For example, an angle, rather than position, is the
appropriate mechanical variable in the mechanochemical
models of rotary motors such as F1-ATPase. Similarly,
while developing models for myosin-V and myosin-VI,
Lan and Sun (44) used two angular variables that were
dened with respect to the monomer body axis and the
F-actin plane.
For a track like MT, in the absence of any MT-associated
protein, the potential U
m
x is periodic, but because of the
intrinsic asymmetry of the a b-tubulin heterodimer,
FIGURE 6 Schematic representation of a landscape where the height
represents the potential of mean force (free energy) of a hypothetical
molecular motor that is described by a single mechanical variable and a
single chemical variable (adapted from Keller and Bustamante (43); cour-
tesy Ajeet K. Sharma).
a
b
FIGURE 7 (a and b) Cross sections of the landscape shown in Fig. 6
parallel to (a) the mechanical coordinate (i.e., for a constant value of the
chemical variable) and (b) the chemical variable (i.e., for a constant value
of the mechanical variable; courtesy Ajeet K. Sharma).
2336 Chowdhury
each period of this potential is expected to be spatially
asymmetric. Interestingly, on average, a temporal sequence
of such a periodic asymmetric potential can lead to
directed movement of the motor, albeit noisily, in spite of
the vanishing of spatially averaged force (32). This is one
of the possible mechanisms of a Brownian ratchet. Recti-
cation of the noise required for the Brownian ratchet
mechanism can also be achieved, for example, by the bind-
ing of a ligand that stabilizes conformations in the forward
direction (33).
Markov model: motor kinetics as a jump process
on a network of fully discrete mechanochemical
states
In this subsection we further simplify the continuum land-
scape-based scenario developed above to formulate a fully
discrete scheme (45,46). The main equations of these for-
malisms are summarized in Section B of the Supporting
Material.
Suppose PY DY is the probability of nding the sys-
tem between Y and Y DY on the continuous landscape.
Now, with each local minimum of the free-energy land-
scape we associate a discrete state. The probability P
i
of
nding the system in the i-th discrete state is obtained by
integrating the probability PY over the i-th zone, which
is the immediate surroundings of the local minimum
labeled by the discrete index i. Just like the continuum
formulation, the minimal model must have one mechanical
variable and a chemical variable, both of which are
discrete.
Let P
m
i; t be the probability of nding the motor at the
discrete position labeled by i and in the chemical state m at
time t. The time evolution of the system is described as
discrete jumps on the network of the discrete states gov-
erned by the master equations for P
m
i; t (which are also
referred to as stochastic rate equations).
The local minima in the free-energy landscape are sepa-
rated by low-energy passes, such that thermal uctuations
occasionally cause the system to leave the neighborhood
of one local minimum and arrive at that of a neighboring
one. Such wanderings on the free-energy landscape are
identied as transitions from one discrete state to another
in the fully discrete formulation. The corresponding rate
constants (i.e., the probabilities of transition per unit
time) can also be obtained from an analysis of the probabil-
ity uxes on the continuous landscape (43,44). Obviously,
the rate constants depend on the shape of the free-energy
landscape; the dependence of the rate constants on the
external force arise from that of the landscape shape on
the external force. The discrete state space of this formula-
tion can be regarded as a network. The mathematical prop-
erties of such networks are closely related to the formal
properties of electrical networks pioneered by Kirchoff
and others.
Markov modeling has been exploited successfully for
many motors, including myosin-V, kinesin-1, KIF1A,
RNA polymerase, DNA polymerase, and ribosome (21).
One distinct advantage of this approach is that often one
can solve the master equations to derive the main results
analytically. However, in contrast to the landscape-based
scenario, forces between different parts of the motor do
not appear explicitly through any interaction potential.
Consequently, the Markov models of motors are the most
suitable for interpreting a wide range of empirical data
without having to obtain the explicit form of the potentials
of interactions among the constituent parts or those among
the motor, ligand, and track.
INVERSE PROBLEM: FROM DATA TO MODEL BY
PROBABILISTIC REVERSE ENGINEERING
So far in this mini-review, we have discussed the forward
problem of modeling machine-driven kinetic processes.
The forward problem is solved by starting with an a priori
hypothesis (essentially, an educated guess) about the mech-
anochemical kinetics of the motor. An analytical or numer-
ical approach (or a combination thereof) yields data that can
be compared with the empirical data. In contrast, the inverse
problem is to infer the model from the empirical data; this
task is reminiscent of reverse engineering.
The likelihood P of a state trajectory is the conditional
probability for this trajectory, given the rate constants for
the interstate transitions. There are two alternative
approaches for estimating the rate constants from the likeli-
hood. Here we discuss these methods without mathematical
equations. The main steps of these formalisms are summa-
rized in Section C of the Supporting Material.
The maximum-likelihood (ML) approach is based on
estimating the set of rate constants that maximizes the likeli-
hood P. In other words, the estimated rate constants are
their most probable values.
In the Bayesian approach, both the rate constants and
empirical data are treated on an equal footing, i.e., both
are regarded as random variables. Starting from an
assumed a priori distribution of the rate constants (called
the prior, for obvious reasons), one estimates the a posteri-
ori probability distribution of the rate constants by utilizing
the distribution of the observed data. The choice of the
prior can be based on physical intuition or general argu-
ments based, for example, on symmetries. The choice of
the prior can be simple if some experience has been gained
from previous measurements. Often a uniform distribution
of the model parameter(s) can be assumed over its allowed
range if no additional information is available to bias its
choice.
For both the frequentist and Bayesian approaches out-
lined above, the state trajectories are required. However,
in practice, the actual sequence of states of the motor, gener-
ated by the underlying Markovian kinetics, is not directly
available for analysis. For example, the distinct chemical
states that correspond to the same mechanical state cannot
be distinguished in a purely mechanical measurement on a
single motor. The hidden Markov model is a powerful
tool for extracting information about hidden trajectories
from the trajectories that are visible in single-molecule
experiments.
Most often, one can infer multiple models, all at the
same level of spatiotemporal resolution, from a statistical
analysis of the same set of data. Therefore, it would be
desirable to follow Platts (47) principle of strong inference
(which is an extension of Chamberlins (48) method of
multiple working hypotheses). According to this principle,
the level of success of a model can be quantied by a
numerical score. The relative scores of the competing
models (and the corresponding underlying hypotheses)
would be a true reection of their merits. Very few attempts
have been made so far to rank the models of molecular
motors inferred from the data according to their relative
scores (21).
EXPERIMENTAL STUDIES OF MOLECULAR
MACHINES
The main emphasis of this mini-review is on theoretical
modeling of molecular machines. Nevertheless, the current
rapid progress in theory has been stimulated by experi-
mental observations, many of which were made possible
by the invention of novel techniques coupled with precision
technology. In this section, we present a micro-review of
some of these techniques and their use for studying molec-
ular machines.
X-ray crystallography, NMR, and electron microscopy
(EM) (49) are the standard techniques for experimentally
determining the structures of macromolecular systems,
including molecular machines. However, the main hurdle
in x-ray crystallography is the difculty of achieving
crystallization. Similarly, structure determination by NMR
becomes very difcult for most molecular machines
because these machines are usually large macromolecular
complexes. Therefore, for many of the molecular machines
for which x-ray crystallography has not been possible, high-
resolution atomic structures of some of their important
subunits have been obtained separately by x-ray crystallog-
raphy. In more recent times, a combination of these tech-
niques, particularly EM of the full machine and x-ray
structures of component subunits, have been found to be
more powerful than any of the individual techniques (50).
The results obtained by the above-mentioned techniques
are averaged over an ensemble of machines. Similarly,
rate constants for various chemical reactions involved in
the operation of a molecular machine are averaged over
an ensemble when extracted from the bulk sample. In
contrast, single-molecule techniques developed in recent
years avoid averaging over an ensemble (19,22,5153).
Some uorescence-based, single-molecule imaging tech-
niques (54) enable one to monitor the movement of
the motor or relative motion between different parts of
the motor. In some other single-molecule experiments, the
motor is manipulated by application of an external force
or torque; optical tweezers (52,55) and magnetic tweezers
(5658) are the most commonly used devices for this
purpose.
Single-molecule studies (51,52) have provided deep in-
sights into the mechanochemical kinetics of almost all types
of molecular machines (59), including cytoskeletal motors
(19), as well as machines for genomic processes (22,60),
such as helicase motors (45) and machines for template-
directed polymerization (e.g., RNA polymerases (46,61), re-
plisomes (62,63), and ribosomes (6466)). Single-molecule
techniques have also been used to study machines for pack-
aging of genome into viral capsids (67).
For a given motor, the data from single-molecule experi-
ments can also be used to obtain the dwell-time distribution
(DTD). The DTD can be utilized to extract a lower bound on
the number of kinetic states (68). The DTD is essentially a
distribution of rst-passage times (69), an interesting
concept in nonequilibrium statistical mechanics. The theo-
retically calculated DTDs for several motors are consistent
with the corresponding experimental data (21).
CONCLUSIONS AND OUTLOOK
Over the last two decades, investigators have been able to
produce a parts list of many motors by performing careful
experiments (18), and consequently have developed models
of single motors, at various levels of spatiotemporal resolu-
tion, by assembling those parts. However, dynamical
studies of fully atomistic models of even single motors
remain constrained by the current computational resources.
I have sketched several alternative strategies for modeling
molecular machines depending on the questions to be
addressed and the level of spatiotemporal organization.
Because of space limitations, I have not discussed any
individual examples in this mini-review; interested readers
are referred to my recent detailed review published else-
where (21).
The catalog of the machines presented here can be
divided into two parts: 1), machines for motility and
contractility (e.g., exporters and importers, packers and por-
ters, sliders and rowers); and 2), machines for genomic pro-
cesses (e.g., unwrappers, unzippers, and untanglers, and
synthesizers and degraders of DNA, RNA, and proteins;
Fig. 8). The operation of both groups of machines is regu-
lated by signaling and metabolic machineries (Fig. 8), which
I have not discussed here or in the longer review (21). How-
ever, one can incorporate the switching on and off of a
machine, as well as the supply of fuel and raw materials
for its operation, within the model of its operation by linking
it with signaling and metabolic machineries.
2338 Chowdhury
When modeling the full operation of a module, one
should incorporate all three stages of dynamic assembly:
1), assembly of the component devices at the right place
and the right time; 2), the successive steps of its main oper-
ation; and 3), disassembly, if it occurs, after completion of
its operation. Over the next decade, we can expect to see
major efforts to integrate coarse-grained models of motors
into dynamic modules. Next, further integration of weakly
interacting dynamic modules is likely to pave the way for
what may be called a system-level description (the mecha-
nome (70)) of an entire cell, at least in the interphase, as a
microfactory (5). Finally, modeling cellular homeostasis in
molecular detail will require integration of the multimodule
model of the cell interior with the external stressors at the
exterior-interior interface (71) (Fig. 8). The cell is the func-
tional unit of life. Therefore, this dream project, if it is ever
completed even for a minimal synthetic cell (72), will take
biophysicists closer to answering the eternal question: what
is life?
SUPPORTING MATERIAL
Additional analysis is available athttp://www.biophysj.org/biophysj/
supplemental/S0006-3495(13)00475-X.
I thank Alex Mogilner for suggesting the writing of this mini-review
and for valuable suggestions. I also thank all my collaborators and
students for enjoyable collaborations on molecular motors. I apologize
to all of the authors whose work could not be cited here due to space
limitations; I hope I have cited all the relevant works in my comprehen-
sive review (21).
This work was supported in part by the Indian Institute of Technology
Kanpur through a Dr. Jag Mohan Garg Chair professorship and by a
research grant from the Department of Biotechnology, India.
REFERENCES
1. Piccolino, M. 2000. Biological machines: from mills to molecules. Nat.
Rev. Mol. Cell Biol. 1:149153.
2. Mavroidis, C., A. Dubey, and M. L. Yarmush. 2004. Molecular
machines. Annu. Rev. Biomed. Eng. 6:363395.
3. Baumgaertner, A. 2005. Biomolecular machines. In Handbook of
Theoretical and Computational Nanotechnology. M. Rieth and
W. Schommers, editors. American Scientic Publishers, Stevenson
Ranch, CA. 189.
4. Cozzarelli, N. R., G. J. Cost, ., J. E. Stray. 2006. Giant proteins that
move DNA: bullies of the genomic playground. Nat. Rev. Mol. Cell
Biol. 7:580588.
5. Alberts, B. 1998. The cell as a collection of protein machines: prepar-
ing the next generation of molecular biologists. Cell. 92:291294.
6. Carlier, M. F., E. Helfer, ., F. Haraux. 2010. Living nanomachines. In
Nanoscience. P. Boisseau, P. Houdy, and M. Lahmani, editors.
Springer, New York.
7. Frank, J., editor. 2011. Molecular Machines in Biology: Workshop of
the Cell. Cambridge University Press, Cambridge.
8. Roux, B., editor. 2011. Molecular Machines.. World Scientic, River
Edge, NJ.
9. Rossmann, M. G., and V. B. Rao, editors. 2012. Viral Molecular
Machines.. Springer, New York.
10. Sharma, A. K., and D. Chowdhury. 2012. Template-directed bio-
polymerization: tape-copying Turing machines. Biophys. Rev. Lett.
7:141.
11. Howard, J. 2001. Mechanics of Motor Proteins and the Cytoskeleton.
Sinauer Associates, Sunderland, MA.
12. Vale, R. D., and R. A. Milligan. 2000. The way things move: looking
under the hood of molecular motor proteins. Science. 288:8895.
13. Schliwa, M., editor. 2003. Molecular Motors.. Wiley-VCH, New York.
14. Hackney, D. D., and F. Tanamoi. 2004. The Enzymes, Vol. 23. Energy
Coupling and Molecular Motors. Elsevier, New York.
15. Squire, J. M., and D. A. D. Parry. 2005. Fibrous Proteins: Muscle and
Molecular Motors. Elsevier, New York.
16. Kolomeisky, A. B., and M. E. Fisher. 2007. Molecular motors: a
theorists perspective. Annu. Rev. Phys. Chem. 58:675695.
17. Wang, H. 2008. Several issues in modeling molecular motors.
J. Comput. Theoret. Nanosci. 5:135.
18. Hwang, W., and M. J. Lang. 2009. Mechanical design of translocating
motor proteins. Cell Biochem. Biophys. 54:1122.
19. Veigel, C., and C. F. Schmidt. 2011. Moving into the cell: single-mole-
cule studies of molecular motors in complex environments. Nat. Rev.
Mol. Cell Biol. 12:163176.
20. Goldman, Y. E., and E. M. Ostap. 2012. Molecular Motors and
Motility, Vol. 4. Comprehensive Biophysics. Elsevier, New York.
21. Chowdhury, D. 2013. Stochastic mechano-chemical kinetics of molec-
ular motors: a multidisciplinary enterprise from a physicists perspec-
tive. Phys. Rep. 10.1016/j.physrep.2013.03.005 (Author-edited nal
version is available also at http://arxiv.org/abs/1207.6070).
22. Dulin, D., J. Lipfert, ., N. H. Dekker. 2013. Studying genomic pro-
cesses at the single-molecule level: introducing the tools and applica-
tions. Nat. Rev. Genet. 14:922.
23. Hartwell, L. H., J. J. Hopeld, ., A. W. Murray. 1999. From molecular
to modular cell biology. Nature. 402(6761, Suppl):C47C52.
24. Mogilner, A., R. Wollman, and W. F. Marshall. 2006. Quantitative
modeling in cell biology: what is it good for? Dev. Cell. 11:279287.
25. Leibler, S., and D. A. Huse. 1993. Porters versus rowers: a unied sto-
chastic model of motor proteins. J. Cell Biol. 121:13571368.
26. Howard, J., and A. A. Hyman. 2007. Microtubule polymerases and
depolymerases. Curr. Opin. Cell Biol. 19:3135.
FIGURE 8 Interactions of the different types of machines with each other
and with the signaling and metabolic machineries are indicated schemati-
cally by the two-headed arrows. The exchange of matter and energy
between the cellular machineries and the cell exterior across the cell mem-
brane is also indicated by two-headed double arrows.
27. McIntosh, J. R., V. Volkov, ., E. L. Grishchuk. 2010. Tubulin depoly-
merization may be an ancient biological motor. J. Cell Sci. 123:3425
3434.
28. Lorentzen, E., and E. Conti. 2006. The exosome and the proteasome:
nano-compartments for degradation. Cell. 125:651654.
29. Oster, G., and H. Wang. 2003. Rotary protein motors. Trends Cell Biol.
13:114121.
30. Sowa, Y., and R. M. Berry. 2008. Bacterial agellar motor. Q. Rev.
Biophys. 41:103132.
31. Seifert, U. 2012. Stochastic thermodynamics, uctuation theorems and
molecular machines. Rep. Prog. Phys. 75:126001.
32. Julicher, F., A. Ajdari, and J. Prost. 1997. Modeling molecular motors.
Rev. Mod. Phys. 69:12691281.
33. Howard, J. 2006. Protein power strokes. Curr. Biol. 16:R517R519.
34. Hopeld, J. J. 1974. Kinetic proofreading: a new mechanism for
reducing errors in biosynthetic processes requiring high specicity.
Proc. Natl. Acad. Sci. USA. 71:41354139.
35. Ninio, J. 1975. Kinetic amplication of enzyme discrimination.
Biochimie. 57:587595.
36. Welte, M. A., and S. P. Gross. 2008. Molecular motors: a trafc cop
within? HFSP J. 2:178182.
37. Berger, F., C. Keller, ., R. Lipowsky. 2011. Co-operative transport by
molecular motors. Biochem. Soc. Trans. 39:12111215.
38. Leduc, C., O. Campa`s, ., P. Bassereau. 2010. Mechanism of mem-
brane nanotube formation by molecular motors. Biochim. Biophys.
Acta. 1798:14181426.
39. Baker, T. A., and S. P. Bell. 1998. Polymerases and the replisome:
machines within machines. Cell. 92:295305.
40. McIntosh, J. R., M. I. Molodtsov, and F. I. Ataullakhanov. 2012.
Biophysics of mitosis. Q. Rev. Biophys. 45:147207.
41. Mogilner, A., and E. Craig. 2010. Towards a quantitative understanding
of mitotic spindle assembly and mechanics. J. Cell Sci. 123:3435
3445.
42. Dykeman, E. C., and O. F. Sankey. 2010. Normal mode analysis
and applications in biological physics. J. Phys. Condens. Matter.
22:423202.
43. Keller, D., and C. Bustamante. 2000. The mechanochemistry of molec-
ular motors. Biophys. J. 78:541556.
44. Lan, G., and S. X. Sun. 2012. Mechanochemical models of processive
molecular motors. Mol. Phys. 110:10171034.
45. Ha, T., A. G. Kozlov, and T. M. Lohman. 2012. Single-molecule views
of protein movement on single-stranded DNA. Annu. Rev. Biophys.
41:295319.
46. Bai, L., T. J. Santangelo, and M. D. Wang. 2006. Single-molecule anal-
ysis of RNA polymerase transcription. Annu. Rev. Biophys. Biomol.
Struct. 35:343360.
47. Platt, J. R. 1964. Strong inference: certain systematic methods of sci-
entic thinking may produce much more rapid progress than others.
Science. 146:347353.
48. Chamberlin, T. C. 1965. The method of multiple working hypotheses.
Science. 15:9296.
49. Frank, J. 2009. Single-particle reconstruction of biological macromol-
ecules in electron microscopy30 years. Q. Rev. Biophys. 42:139158.
50. Lander, G. C., H. R. Saibil, and E. Nogales. 2012. Go hybrid: EM, crys-
tallography, and beyond. Curr. Opin. Struct. Biol. 22:627635.
51. Greenleaf, W. J., M. T. Woodside, and S. M. Block. 2007. High-reso-
lution, single-molecule measurements of biomolecular motion. Annu.
Rev. Biophys. Biomol. Struct. 36:171190.
52. Moftt, J. R., Y. R. Chemla, ., C. Bustamante. 2008. Recent advances
in optical tweezers. Annu. Rev. Biochem. 77:205228.
53. Neuman, K. C., T. Lionnet, and J. F. Allemand. 2007. Single-molecule
micromanipulation techniques. Annu. Rev. Mater. Res. 37:3367.
54. Lord, S. J., H. L. Lee, and W. E. Moerner. 2010. Single-molecule spec-
troscopy and imaging of biomolecules in living cells. Anal. Chem.
82:21922203.
55. Lang, M. J., and S. M. Block. 2003. Resource Letter: LBOT-1: laser-
based optical tweezers. Am. J. Phys. 71:201215.
56. Hosu, B. G., J. Jacob, ., G. Forgacs. 2003. Magnetic tweezers for
intracellular applications. Rev. Sci. Instrum. 74:41584163.
57. Kim, K., and O. A. Saleh. 2009. A high-resolution magnetic tweezer
for single-molecule measurements. Nucleic Acids Res. 37:e136.
58. De Vlaminck, I., and C. Dekker. 2012. Recent advances in magnetic
tweezers. Annu. Rev. Biophys. 41:453472.
59. Peters, R. 2007. Single-molecule uorescence analysis of cellular
nanomachinery components. Annu. Rev. Biophys. Biomol. Struct.
36:371394.
60. Bustamante, C., W. Cheng, and Y. X. Mejia. 2011. Revisiting the cen-
tral dogma one molecule at a time. Cell. 144:480497 (Erratum in Cell.
2011. 145:160).
61. Herbert, K. M., W. J. Greenleaf, and S. M. Block. 2008. Single-mole-
cule studies of RNA polymerase: motoring along. Annu. Rev. Biochem.
77:149176.
62. van Oijen, A. M., and J. J. Loparo. 2010. Single-molecule studies of the
replisome. Annu. Rev. Biophys. 39:429448.
63. Manosas, M., T. Lionnet, ., V. Croquette. 2010. Studies of DNA-repli-
cation at the single molecule level using magnetic tweezers. Biol. Phys.
60:89122.
64. Aitken, C. E., A. Petrov, and J. D. Puglisi. 2010. Single ribosome
dynamics and the mechanism of translation. Annu. Rev. Biophys.
39:491513.
65. Uemura, S., and J. D. Puglisi. 2011. Real-time monitoring of single-
molecule translation. In Ribosomes. M. V. Rodnina, editor.. Springer,
New York. 295302.
66. Tinoco, I., Jr., and R. L. Gonzalez, Jr. 2011. Biological mechanisms,
one molecule at a time. Genes Dev. 25:12051231.
67. Roos, W. H., I. L. Ivanovska, ., G. J. L. Wuite. 2007. Viral capsids:
mechanical characteristics, genome packaging and delivery mecha-
nisms. Cell. Mol. Life Sci. 64:14841497.
68. Schnitzer, M. J., and S. M. Block. 1995. Statistical kinetics of proces-
sive enzymes. Cold Spring Harb. Symp. Quant. Biol. 60:793802.
69. Redner, S. 2001. A Guide to First-Passage Processes. Cambridge
University Press, Cambridge.
70. Lang, M. J. 2007. Lighting up the Mechanome. The Bridge. 34:1216.
71. Lipowsky, R. 2006. The physics of bio-systems: from molecules to net-
works. Biophys. Rev. Lett. 1:223230.
72. Forster, A. C., and G. M. Church. 2006. Towards synthesis of a minimal
cell. Mol. Syst. Biol. 2:45.
73. Hammer, J. A., 3rd, and J. R. Sellers. 2012. Walking to work: roles
for class V myosins as cargo transporters. Nat. Rev. Mol. Cell Biol.
13:1326.
74. Spudich, J. A., and S. Sivaramakrishnan. 2010. Myosin VI: an innova-
tive motor that challenged the swinging lever arm hypothesis. Nat. Rev.
75. Sweeney, H. L., and A. Houdusse. 2010. Myosin VI rewrites the rules
for myosin motors. Cell. 141:573582.
76. Block, S. M. 2007. Kinesin motor mechanics: binding, stepping,
tracking, gating, and limping. Biophys. J. 92:29862995.
77. Gennerich, A., and R. D. Vale. 2009. Walking the walk: how kinesin
and dynein coordinate their steps. Curr. Opin. Cell Biol. 21:5967.
78. Vallee, R. B., R. J. McKenney, and K. M. Ori-McKenney. 2012.
Multiple modes of cytoplasmic dynein regulation. Nat. Cell Biol.
14:224230.
79. Spudich, J. A. 2001. The myosin swinging cross-bridge model. Nat.
Rev. Mol. Cell Biol. 2:387392.
2340 Chowdhury
80. Peterman, E. J. G., and J. M. Scholey. 2009. Mitotic microtubule cross-
linkers: insights from mechanistic studies. Curr. Biol. 19:R1089
R1094.
81. Lindemann, C. B., and K. A. Lesich. 2010. Flagellar and ciliary
beating: the proven and the possible. J. Cell Sci. 123:519528.
82. Thon, J. N., and J. E. Italiano. 2010. Platelet formation. Semin.
Hematol. 47:220226.
83. Dogterom, M., J. Husson, ., C. Tischer. 2007. Microtubule forces
and organization. In Cell Motility. P. Lenz, editor. Springer, New
York. 93115.
84. Kornberg, A., and T. Baker. 1992. DNA Replication, 2nd ed. W.H.
Freeman and Co., New York.
85. Gelles, J., and R. Landick. 1998. RNA polymerase as a molecular
motor. Cell. 93:1316.
86. Buc, H., and T. Strick, editors. 2009. RNA Polymerases as Molecular
Motors. RSC Publishing, London.
87. Herschhorn, A., and A. Hizi. 2010. Retroviral reverse transcriptases.
Cell. Mol. Life Sci. 67:27172747.
88. Ort n, J., and F. Parra. 2006. Structure and function of RNA replication.
Annu. Rev. Microbiol. 60:305326.
89. Rodnina, M. V., W. Wintermeyer, and R. Green, editors. 2011.
Ribosomes. Springer, New York.
Systems Biophysics of Gene Expression
Jose M. G. Vilar
* and Leonor Saiz
Biophysics Unit (CSIC-UPV/EHU) and Department of Biochemistry and Molecular Biology, University of the Basque Country, Bilbao, Spain;
IKERBASQUE, Basque Foundation for Science, Bilbao, Spain; and

Department of Biomedical Engineering, University of California at Davis,
Davis, California
ABSTRACT Gene expression is a process central to any form of life. It involves multiple temporal and functional scales that
extend from specic protein-DNA interactions to the coordinated regulation of multiple genes in response to intracellular
and extracellular changes. This diversity in scales poses fundamental challenges to the use of traditional approaches to fully
understand even the simplest gene expression systems. Recent advances in computational systems biophysics have
provided promising avenues to reliably integrate the molecular detail of biophysical process into the system behavior. Here,
we review recent advances in the description of gene regulation as a system of biophysical processes that extend from
specic protein-DNA interactions to the combinatorial assembly of nucleoprotein complexes. There is now basic mechanistic
understanding on how promoters controlled by multiple, local and distal, DNA binding sites for transcription factors can actively
control transcriptional noise, cell-to-cell variability, and other properties of gene regulation, including precision and exibility of
the transcriptional responses.
INTRODUCTION
The process that leads to functional RNA and protein mol-
ecules from the information encoded in genes is known
as gene expression. It starts with the binding of the RNA
polymerase (RNAP) to the promoter, continues with tran-
scription of the gene into RNA, and often concludes with
translation into protein (1). This simple description is just
the backbone of a much more complex set of events involv-
ing many processes that actively regulate, complement,
affect, and critically rene these three steps (2).
The complexity of gene expression is already evident at
the very early stages of the process. The RNAP rarely just
binds to DNA and starts transcription. There are molecules,
such as transcription factors (TFs), that enhance, stabilize,
hinder, and prevent the binding of the RNAP to the promoter
(3). This local layer of control is embedded in the underly-
ing dynamic organization of the genome, which determines
to a large extent the accessibility of the RNAP to the pro-
moter and to the information content of sets of genes that
are spatially in the same region (48). In addition, the
RNAP is not a simple molecule but a multisubunit complex
that, especially in eukaryotes, does not necessarily need to
come preassembled to the promoter region or to be ready
to start transcription upon binding (9,10). Along the way,
there are molecular mechanisms that affect RNA stability
and its information content, such as alternative splicing
and RNA editing (1). To close up the loop, proteins and
functional RNA are in charge of orchestrating all these pro-
cesses, thus regulating their own synthesis.
This short reviewfocuses on key, well-characterized guid-
ing principles that allow the description of gene expression
in terms of systems of biophysical processes and the applica-
tion of these principles to actual systems, exemplied by
the lac operon in prokaryotes and the retinoid X receptor
in eukaryotes, which are both amenable to concise informa-
tive mechanistic descriptions. The goal is to accurately cap-
ture the effects of molecular interactions across scales up to
the system behavior. To do so effectively, we will emphasize
approaches that are scalablenamely, approaches that can
be used with small and large systems, incorporate complex
phenomena such as DNA looping, employ as few free
parameters as possible, and the molecular parameters of
which can be inferred from the experimental data and
reused in modeling subsequent experiments.
There are two types of important situations that we will
not consider explicitly because of space limitations. One
type includes elementary mechanisms, such a cooperative
interactions, which are described in virtually any biochem-
istry and molecular biology textbook (1), and which are
applied to gene regulation exactly as described in the text-
books (1114). The other type includes complex situations
with missing key mechanistic information, such as eukary-
otic enhancers (15,16), which would extend the discussion
to cover many potential mechanisms that are compatible
with the observed experimental data. The most effective
avenue to modeling such complex problems so far has
been to supplement known biophysical mechanisms with
phenomenological rules and assumptions (17,18).
There are also many important aspects of gene expression
that we will not be able to reach, including the effects
of focused and dispersed transcription initiation (19),
transcription elongation regulation (20,21), transcriptional
trafc (22), and translation regulation by microRNAs (23),
to mention just a few. The general principles reviewed
here to a large extent also apply to those situations.
Submitted January 23, 2013, and accepted for publication April 12, 2013.
*Correspondence: j.vilar@ikerbasque.org or lsaiz@ucdavis.edu
Editor: Stanislav Shvartsman.
2574 Biophysical Journal Volume 104 June 2013 25742585
MODELING GENE EXPRESSION
Acommon starting point for most quantitative approaches to
gene expression is a description based on reactions among
molecular species (2430). This description considers that
there is a set of i different transcriptional states d
i
and that
for each of these states there is a given transcription rate G
i
that leads to mRNA, m. The simplest case with a single
state would be a constitutive promoter with a constant
transcription rate. The next step in complexity, a promoter
with two states, already includes the potential for regulation,
as for instance when a repressor turns off transcription
upon binding the promoter. In general, the transitions
between transcriptional states d
i
and d
j
with rates k
ij
depend
on the numbers of the different molecular species of the
system. For each mRNA molecule, proteins p are produced
at a rate U. Typically, mRNA and proteins are degraded
at rates g
m
and g
p
, respectively. These reactions can be
summarized as
d
i
!
k
ij
d
j
;
d
i
!
G
i
d
i
m;
m !
g
m
B;
m/
U
m p;
p !
g
p
B:
(1)
The advantage of using such an approach is that it allows
a direct connection of the description parameters with
biophysical properties such as free energies of binding and
DNA elastic properties.
THE DETERMINISTIC APPROACH
When uctuations are not relevant, either because they are
small or because they can be averaged out (31), the set
of expressions in Eq. 1 is usually written in terms of concen-
trations using traditional deterministic rate equations:
dP
i
dt

X
j
k
ji
P
j
k
ij
P
i
;
dm
dt

X
i
G
i
=V
c
P
i
g
m
m;
dp
dt
Um g
p
p:
(2)
Here, V
c
is the reaction volume; P
i
hd
i
i is the probability
of having the system in the transcriptional state i; [m] hmi/
V
c
is the mRNA concentration; and [p] hpi/V
c
is the
average protein concentration, with angular brackets h.i
representing averages.
The previous set of equations can be solved to obtain the
steady-state protein content concentration [p]
ss
as
p
ss
p
max
X
i
c
i
P
i
; (3)
where p
max
G
max
U/g
m
g
p
V
c
is the maximum concentra-
tion, G
max
is the maximum transcriptional activity, and
c
i
G
i
/G
max
is the normalized transcriptional activity.
This result is extremely important because, besides
collapsing the effects of many processes into a single param-
eter p
max
, it directly connects microscopic probabilities of
the transcriptional states with experimentally measurable
quantities.
The deterministic approach, also known as mean-eld
approach, has been very useful to study systems with large
numbers of molecules and negligible uctuations. In the
presence of uctuations of small numbers of molecules,
the average behavior of the system is still correctly
described by the set of expressions in Eq. 2. The applica-
bility of the deterministic approach, however, could break
down with the additional presence of nonlinear terms. The
reason is that the average of nonlinear terms cannot gener-
ally be expressed in terms of concentrations. For instance,
the kinetics of dimerization of a protein p would involve
the term hp
2
i, which is not equivalent to hpi
2
(V
c
[p])
2
.
In general, the validity of the deterministic approach should
be carefully assessed on a case-by-case basis, taking into
account that neither small numbers of molecules nor
nonlinear terms by themselves always prevent its applica-
bility, as illustrated by genetic nonlinear oscillators that
can function in the deterministic regime even with just a
few mRNA molecules per cell (32).
CONTROL OF GENE EXPRESSION
Control of gene expression is achieved through the depen-
dence of the probability of the transcriptional states on the
specic pattern of TFs that are assembled on DNA, as for
instance, binding of an activator and absence of a repressor
(2). In most instances, these interactions take place under
quasi-equilibriumconditions and statistical thermodynamics
can be used to express the probabilities of the states in terms
of standard free energies and concentrations of the different
regulatory molecules involved (3335). The validity of the
quasi-equilibrium assumption requires the binding kinetics,
which by itself would be a completely reversible reaction,
to be much faster than other cellular processes, such as cell
growth, that could affect the binding process.
The key quantity in the thermodynamic approach is the
statistical weight, or Boltzmann factor, which is dened
in terms of the free energy DG
i
of the state i as
Z
i
e
DG
i
=RT
. Its main feature is its proportionality to the
probability of the state i,
P
i

e
DG
i
=RT
Z
: (4)
Systems Biophysics of Gene Expression 2575
The normalization factor Z
P
i
Z
i
is known as the partition
function and the term RT is, as usual, the gas constant, R,
times the absolute temperature, T. This expression is partic-
ularly important because it encapsulates the dependence of
the probabilities on the different molecular concentrations
of regulatory molecules [p
j
] through
DG
i
DG
o
i

X
j
di; jRT ln
p
j
; (5)
where the terms d(i,j) correspond to the number of mole-
cules of the species j in the state i and DG
o
i
is the corre-
sponding standard free energy at 1 M concentration.
Therefore, if the free energies, or alternatively the probabil-
ities, of the different states are known for given values of the
concentrations of regulatory molecules, it is possible to
obtain the probabilities for any concentration using the pre-
vious two equations. These two equations can be combined
into
P
i

Y
j
p
j
di;j
e
DG
o
i
=RT
Z
; (6)
which has been a cornerstone in quantitative modeling of
gene expression since the beginning of the eld (30,34).
Its main advantage is that it only requires the values DG
o
i
for each transcriptional state, which has traditionally been
written in tabular form along with a description of the mo-
lecular conguration (34).
COMBINATORIAL COMPLEXITY
The main advantage of using a free energy value for each
transcriptional state may turn increasingly fast into a disad-
vantage when the number of components of the system
increases. The reason is that there are potentially as many
states as the number of possible ways of arranging the reg-
ulatory molecules on DNA, which grows exponentially with
the number of components. The resulting combinatorial ex-
plosion in the number of states makes the straightforward
application of Eq. 6 impracticable for systems with more
than just a handful of components.
Several general approaches have been developed to tackle
this exponentially large multiplicity in the number of states.
They involve a diversity of methodologies that range from
stochastic conguration sampling (36) to automatic genera-
tion of all the underlying equations (37). The complexity of
the general problem makes each of these approaches work
efciently only on a particular type of problem, be it confor-
mational changes, multi-site phosphorylation, or oligomeri-
zation (3842).
In the case of gene regulation, it has been possible to capi-
talize on the unambiguous modular structure that macromo-
lecular complexes typically have on DNA to capture this
complexity in simple terms (43). The key idea is to describe
the specic conguration, or state of the protein-DNA com-
plex, through a set of M state variables, denoted by s
(s
1
,.s
k
,.s
M
), which indicate whether a particular molecu-
lar component or conformation is present (s
k
1) or absent
(s
k
0) at a specic position within the complex (43). The
main advantage is that the free energy DG
i
DG(s) and
transcription rates G
i
G(s) for each state can be specied
as function of the state variables without explicitly enumer-
ating all the states.
THE LAC OPERON
The Escherichia coli lac operon is the genetic system that
regulates and produces the enzymes needed to metabolize
lactose (44,45). Besides opening the doors to the eld of
gene regulation, the lac operon has provided an example
of a sophisticated regulation mechanism where all the com-
ponents are known in great detail (4652).
The main player in the control of transcription is the tetra-
meric lac repressor. In the absence of allolactose, a deriva-
tive of lactose, the lac repressor can bind to the main
operator to prevent the RNAP from binding to the promoter
and transcribing the genes. Binding of allolactose to the
repressor substantially reduces its specic binding for the
operator and transcription is de-repressed. The effects of
the lac repressor on transcription are characterized as nega-
tive control. There is also positive control through the catab-
olite activator protein (CAP), which acts as an activator of
transcription when glucose is not present by stabilizing
the binding of the RNAP to the promoter.
This account of positive and negative control does not
offer the whole picture of the underlying complexity. There
are also two additional auxiliary operators that bind the
repressor without preventing transcription (Fig. 1 A). Early
on, they were considered just remnants of evolution because
they bind the repressor very weakly and because elimination
of either one of them has only minor effects in transcription.
It was later observed that the simultaneous elimination of
both auxiliary operators reduces the repression level by
~100 times (4648). The reason for this astonishing effect
is that the lac repressor can bind simultaneously two opera-
tors and loop the intervening DNA(Fig. 1 B). Thus, the main
operator and at least one auxiliary operator are needed to
form DNA loops that substantially increase the repressors
ability to bind the main operator. Without quantitative
approaches, however, it is difcult to fully grasp how such
weak auxiliary sites, as much as 300-times weaker in terms
of binding afnity than the main operator, can help the bind-
ing so much.
To illustrate the important effects of the presence of the
auxiliary operators, we examine in detail the case with
two operators, the main operator, O
m
, and an auxiliary oper-
ator, O
a
. The main operator is located at the position of O
1
and the auxiliary operator is located at the position of either
O
2
or O
3
(Fig. 1 A).
2576 Vilar and Saiz
The key piece of information that allowed capturing the
effects of DNA looping in quantitative detail was shown
to be the decomposition of the free energy of the looped pro-
teinDNA complex, DG
o
lc
, into different modular contri-
butions that take into account the binding to each operator
and the looping contribution (49). Explicitly, DG
o
lc

g
m
g
a
g
L
, where g
m
and g
a
are the standard free energy
of binding to O
m
and O
a
, respectively, and g
L
is the free en-
ergy of looping (Fig. 1 B).
This segmentation of the free energy allows for an
efcient representation of all the transcriptional states in
terms of state variables. These variables comprise s
m
and
s
a
, which indicate whether ( 1) or not ( 0) a repressor
is bound to the main and auxiliary operator, respectively,
and s
L
, which indicates whether ( 1) or not ( 0) DNA
forms the loop O
m
-O
a
.
The free energy of the system in terms of these three state
variables is given by
DGs g
m
RT lnns
m
g
a
RT lnns
a
g
L
RT lnns
m
s
a
s
L
N1 s
m
s
a
s
L
;
(7)
where [n] is the concentration of the lac repressor. The rst
two terms in the expression take into account the repressor
binding to O
m
or O
a
. The fourth term indicates that the pres-
ence of looping needs both operators occupied by a
repressor; otherwise, the free energy would be innite.
Finally, the rst three terms all together represent the bind-
ing of the repressor when the three state variables are equal
to 1, which indicates that a single repressor is bound simul-
taneously to O
a
and O
m
and that there is a looping contribu-
tion to the free energy.
The normalized transcriptional activity is expressed in
terms of state variables as
cs 1 s
m
c
a
s
a
1 s
a
: (8)
This expression species that there is no transcription when
the repressor is bound to O
m
. When O
m
is free, transcription
occurs at a maximum rate if O
a
is free, and at rate c
a
if the
repressor occupies O
a
.
The advantage of using state variables is that Eqs. 7 and
8 completely specify the transcriptional properties of the
lac operon. The steady-state protein production is computed
directly from [p]
ss
p
max
S
s
c(s)P(s), where the sum can be
performed by hand or automatically using software like
CplexA (53,54). The resulting repression level is given suc-
cinctly by
p
max
p
ss

e
gm=RT
n
e
ga=RT
n
ne
g
L
=RT
e
g
m
=RT
e
g
a
=RT
nc
a
: (9)
This expression is important because it connects macroscop-
ically measurable quantities, such as protein content in a
cell population, with microscopic binding parameters. The
value of p
max
can be obtained from measurements for strains
without repressor, which transcribe the lac genes at a
maximum rate, and it is customary to report just the ratio
p
max
/[p]
ss
, which is known as repression level. In the case
of the lac operon, all the parameters needed for modeling
can be inferred from the experimentally available data.
For instance, when the auxiliary operator is deleted, or
more precisely when it is mutated so that the binding is
very low (g
a
/N), the repression level reduces to
p
max
p
ss
1 ne
gm=RT
: (10)
From this expression and the data of experimental setups
that used the sequence of O
1
, O
2
, and O
3
as a main operator,
repressor concentration (nM)
r
e
p
r
e
s
s
i
o
n

l
e
v
e
l

X-O
1
-O
2
X-O
2
-O
2
X-O3
-O2
O
3
-O
1
-X
P
lacZYA
O
3
O
1
92bp 401bp
A C D E C
g
L
=8.30 g
L
=7.05
g
1X
= 7.65
g
1
= 13.1
g
2
= 11.6
g
3
= 9.45
g
a
g
m
g
L
B
FIGURE 1 Gene expression in the lac operon for different operator congurations. (A) The relative positions of the main operator O
1
and the auxiliary
operators O
2
and O
3
are shown (solid rectangles) on the solid line representing DNA. The binding site for CAP is also shown (shaded rectangle). (B) A
representation of the lac repressor is shown looping DNA (solid line) bound to the main and auxiliary operators (open rectangles). The contributions to
the standard free energy of the looped DNA-repressor state from binding the main and auxiliary operators and from looping DNA are indicated by g
m
,
g
a
, and g
L
, respectively. (CE) The repression level computed from Eq. 9 (lines) and experimentally measured by Oehler et al. (46) (symbols) are shown
for different combinations of operator replacements and deletions, including congurations with just the main operator, the O
1
-O
2
loop, and the O
3
-O
1
loop. The notation on each curve indicates the operator sequence at the positions of the operators O
3
-O
1
-O
2
. X corresponds to a complete deletion of
O
1
, O
2
, or O
3
; O
1X
corresponds to a partial deletion of O
1
. The values of g
1
, g
2
, g
3
, and g
1X
show the standard free energy of binding (in kcal/mol) to
the sequences of O
1
, O
2
, O
3
, and O
1X
, respectively. The value of g
L
used is shown (in kcal/mol) for each type of loop.
it is possible to obtain the free energy of binding for each
operator (Fig. 1 C), which can be reused in subsequent
modeling.
In the case of the O
1
-O
2
loop for different sequences of
the main operator, the only additional parameter needed to
accurately reproduce the experimental data is the free en-
ergy of looping g
L
(Fig. 1 D). In this case, binding of the
repressor to the auxiliary operator O
2
does not affect tran-
scription and c
a
1.
In the case of the O
3
-O
1
loop, binding of the repressor to
the auxiliary operator prevents CAP from activating tran-
scription and the transcription rate is reduced to c
a
0.03.
In this case as well, just a single additional parameter for
the free energy of looping g
L
is needed to reproduce most
of the experimental data (Fig. 1 E). It turns out, however,
that the deletion of O
1
in the strain labeled O
3
-O
1X
-X is not
complete and the site is still able to form the O
3
-O
1X
loop
even though its binding is reduced by 5.5 kcal/mol, a factor
10,000 in terms of binding afnity. This moderate decrease
in binding can be inferred from the position-weight matrix
score of the specic sequence of the incomplete deletion
(55). The computed repression level for the complete dele-
tion of O
1
, strain O
3
-X-X, is shown in Fig. 1 E as a discontin-
uous line that is clearly below the incomplete deletion.
THE RETINOID X RECEPTOR
Gene expression in eukaryotes is substantially more
involved than in prokaryotes (2,3,56). Just the core of the
eukaryotic transcriptional machinery itself involves a wide
variety of components with oscillatory patterns of macro-
molecular assembly and phosphorylation (9,57). In addition,
there are many additional layers of control that extend from
the accessibility and assembly of the transcriptional machin-
ery at the promoter to the intracellular transport and regula-
tion of mRNA and proteins. Despite all these differences, it
has been argued that there are many general principles that
apply to both prokaryotes and eukaryotes (2). We use the
retinoid X receptor (RXR) to illustrate how the main ideas
and methodology used in the lac operon can also be applied
to this complex eukaryotic system.
RXR is a nuclear receptor that is responsible for regu-
lating a large number of genes. It exerts its function by bind-
ing to DNA as homodimer, homotetramer, or obligatory
heterodimerization partner for other nuclear receptors (58).
Similarly to the lac operon, RXR can bind multiple sites
simultaneously as a tetramer by looping the intervening
DNA (Fig. 2 A). A distinct feature, however, is that in the
case of RXR, tetramers and dimers coexist in the cell and
their relative populations are regulated by the RXR cognate
ligands, which prevent the formation of tetramers besides
imparting RXR the ability to recruit coactivators of
transcription.
The rst step in the signaling cascade for sensing the
ligand concentration is regulation of the relative abundance
of the oligomerization states of the RXR, which include tet-
ramers, n
4
, dimers, n
2
, and nontetramerizing dimers, n
2
*.
The effects of the ligand are quantitated in general through
the modulator function f([l]) [n
2
*]/[n
2
], which describes
the partitioning into the tetramerizing and nontetramerizing
dimers by the ligand l. In this system, the canonical ligand is
the hormone 9cRA (9-cis-retinoic acid), a derivative of
Vitamin A, which binds each RXR monomeric subunit
independently of its oligomerization state (59) and prevents
dimers with their two subunits occupied from tetramerizing
(60). Therefore, considering [n
2
*] as the concentration of
dimers with two ligands bound and [n
2
] as the concen-
tration of dimers with one or zero ligand leads to
f l l
2
=K
2
lig
2K
lig
l, where K
lig
is the ligand-RXR
dissociation constant and [l] is the concentration of the
[9cRA] (nM)
ligand concentration (nM)
N
F
I
N
F
I
9
c
R
A

enhancer
coactivator
g
L
=7.75
g
L
=11.0 g
L
=9.60
Response R1 Response R2
enhancer
site
RXR
site 2
RXR
site 1
R
1
R
2
N
F
I
R
2
A
B
D
C
FIGURE 2 RXR-mediated transcriptional responses to 9cRA and atRA
ligands. (A) A prototypical arrangement of binding sites for RXR and an
enhancer element are shown (solid and shaded rectangles, respectively)
on a solid line representing DNA. (B) In response R1, an RXR tetramer
loops DNA (represented as a continuous line) to bring an enhancer close
to the promoter region. In response R2, an RXR dimer recruits a coactivator
to the promoter region. (C and D) The normalized fold induction for re-
sponses R1 and R2 (lines) was computed from Eqs. 15 with the experi-
mental values K
lig
8 nM for 9cRA (97) or K
lig
350 nM for atRA
(98), and K
td
4.4 nM (99). The value of the free energy of looping g
L
(shown in kcal/mol) depends on the specic promoter and cell line. (C)
Response to 9cRA for a promoter incorporating two RXR binding sites
with (left) and without (right) a distal enhancer, which considers responses
R1 and R2, respectively. Experimental gene expression data (symbols) was
obtained from Yasmin et al. (62). (D) Responses to 9cRA and atRA for pro-
moters without a distal enhancer (response R2). Experimental gene expres-
sion data (symbols) correspond to reporter plasmids ADH-CRBPII-LUC
(left) and TK-CRBPII-LUC (right) from Heyman et al. (70).
2578 Vilar and Saiz
ligand (61). This process determines dimer and tetramer
concentrations, which are related to each other through
[n
2
]
2
/[n
4
] K
td
, where K
td
is the tetramer-dimer dissocia-
tion constant.
Control of gene expression results from the dependence
of the transcriptional response on the type of oligomeric
species that are assembled on DNA (62). There are two
differentiated types of responses (Fig. 2 B): The rst type,
referred to as response R1, involves a tetramer that simulta-
neously binds two nonadjacent DNA sites. Upon binding,
the tetramer can bring a distal enhancer close to the pro-
moter region by looping DNA and control transcription. In
this case, dimers do not elicit transcriptional responses. In
general, promoting and preventing DNA looping has been
found to be a fundamental mechanism for controlling the
effects of distal enhancers (63,64). The second type, denoted
response R2, relies on differentiated recruitment abilities by
different oligomerization states. Specically, dimers can re-
cruit a coactivator by binding of a region that is secluded in
the tetramer (61).
The different congurations for binding of RXR to two
DNA sites are described by the state variables s
1t
and s
2t
that indicate whether ( 1) or not ( 0) a tetramer is bound
to site 1 and 2, respectively; s
L
that indicates whether ( 1)
or not ( 0) DNA forms the loop between these two sites;
and two additional state variables s
1d
and s
2d
that indicate
whether ( 1) or not ( 0) a dimer is bound to site 1 and
2, respectively.
The free energy of the system in terms of these state vari-
ables is given by
DGs g
1
RT lnn
4
s
1t
g
2
RT lnn
4
s
2t
g
L
RT lnn
4
s
1t
s
2t
s
L
N1 s
1t
s
2t
s
L
g
1
RT ln
n
2

s
1d
g
2
RT ln
n
2

s
2d
Ns
1t
s
1d
s
2t
s
2d
:
(11)
Here, g
1
and g
2
are the standard free energies of binding
to sites 1 and 2, respectively, which are assumed to be the
same for all three oligomeric species, and g
L
is the free
energy of looping. The rst four terms of this expression
are equivalent to those for the lac operon in Eq. 7 because
it is the same type of tetrameric binding to two sites.
The fth and sixth terms represent the binding of a dimeric
species to sites 1 and 2, respectively. The last term indicates
that dimers and tetramers cannot be bound simultaneously
to the same site by assigning an innite free energy to those
states.
The normalized transcriptional activities for responses
R1 and R2 are expressed in terms of state variables as
c
R1
s c
ref

1 c
ref
s
L
;
c
R2
s c
ref
c
d
s
1d
s
2d
c
dd
c
d
s
1d
s
2d
;
(12)
where c
ref
does not depend on the ligand concentration and
is the normalized basal activity of the promoter in absence
of any activation. The explicit forms of c
d
(1 (1
[l]/K
lig
)
2
)(1 c
ref
) and c
dd
(1 (1 [l]/K
lig
)
4
)(1
c
ref
) indicate that at least one of the ligand-binding sites
of one dimer and of a pair of dimers, respectively, needs
to be occupied by the ligand for the coactivator to be
recruited.
It is straightforward to obtain analytic expressions of the
transcriptional activity from Eqs. 11 and 12 using software
packages like CplexA (53), but it is more illustrative for
the purposes of this review to focus on the functional
regime, which guarantees that there is response to changes
in the ligand concentration.
The functional regime considers two properties. The rst
one is that the total RXR concentration is sufciently high
for it to signicantly bind DNA. The second one is that
the concentration of tetramers is low enough for them not
to completely saturate the binding. The reason is that for
typical values of g
L
, tetramers bind more strongly to two
DNA sites simultaneously than dimers do to a single DNA
site, as in the case of the lac operon (43,49,65). Under these
conditions the representative states, described by s (s
1t
,
s
2t
, s
L
, s
1d
, s
2d
), are those with a tetramer bound to the two
sites simultaneously, s (1,1,1,0,0), and with one dimer
bound to each of the two sites, s (0,0,0,1,1). The corre-
sponding statistical weights for these states, the only ones
needed in this case, are
Z
1;1;1;0;0
n
4
e
g
1
g
2
g
L
=RT
and
Z
0;0;0;1;1

n
2

2
e
g
1
g
2
=RT
;
respectively.
The key implication of this regime is that the steady-state
protein production, computed from
p
ss
p
max
X
s
csPs;
simplies in such a way that the transcriptional responses
are governed by the reduced expressions
p
ss
R1
p
max;R1
c
ref

1 c
ref
P
t
;
p
ss
R2
p
max;R2
c
ref

1
l
K
lig
1 c
ref
1 P
t
;
(13)
where
P
t

1
1 1 f l
2
e
g
L
=RT
K
td
(14)
is the probability of the state s (1,1,1,0,0).
The particular form of P
t
is exceptionally remarkable
because it imparts precision and exibility to the transcrip-
tional responsestwo properties that are the cornerstone of
natural gene expression systems but that have proved to
be highly elusive because of their seemingly antagonistic
character (66). Precision ensures that the transcriptional
response is consistently triggered at a given ligand concen-
tration irrespective of the particular total RXR concentra-
tion, which cancels out in the reduced equations that
govern the system behavior. Flexibility, on the other hand,
allows the precise triggering point to be altered both at the
individual promoter level through g
L
(67,68) and at a ge-
nomewide scale through f([l]) and K
td
.
To compare with the experimental data, the most conve-
nient approach is to use the normalized fold induction
(NFI), which is dened as NFI (FI 1)/(FI
max
1), where
FI is the fold induction and FI
max
is its maximum value. The
value of FI is obtained experimentally as the actual ex-
pression of a gene over its baseline expression and in math-
ematical terms as FI [p]
ss
/(p
max
c
ref
). In terms of the NFI,
the results do not depend on parameters related to the
baseline and maximum expression levels and it becomes
possible to effectively compare experiments on different
promoters and cell lines. The explicit form of the NFI for
responses R1 and R2 is
NFI
R1
P
t
;
NFI
R2

1
l
K
lig
1 P
t
;
(15)
respectively. Importantly, the only parameters needed to
characterize the shape of the response in the functional
regime are K
lig
and K
td
, which have been measured experi-
mentally, and g
L
, which can be inferred by adjusting its
value to reproduce the experimental data.
A fully predictive framework without free parameters has
been obtained with this approach because it collapses most
of the intracellular complexity into just one unknown
parameter g
L
. Therefore, once this parameter is known for
a particular experimental setup (specic cell type, cellular
conditions, and promoter), it can be used to predict other
responses just from thermodynamic principles.
One possibility is to use the value of g
L
inferred for one
type of response to predict the other one. There is experi-
mental data that tested in the same cell type and promoter
both types of transcriptional responses, one mediated by
an enhancer (response R1) and the other, by a coactivator
(response R2). The results of the model indicate that just a
single value of g
L
is needed to reproduce with high accuracy
the experimental data in both cases (Fig. 2 C).
Another possibility is to use the value of g
L
inferred for
one ligand to predict the response to other ligands. The
all-trans-retinoic acid (atRA) was tested early on as a poten-
tial candidate for the RXR cognate ligand, and it was
observed that binding was present but very weak (69,70).
The values of g
L
inferred for 9cRA responses can be used
to closely match the experimental transcription data in
response to atRA without any free parameter by simply
changing the value of the ligand-binding constant, K
lig
, to
the corresponding one for atRA (Fig. 2 D).
COMBINATORIAL ASSEMBLY OF
NUCLEOPROTEIN COMPLEXES
There are many situations in which the DNA loop is formed
not by a single protein, as in the lac operon and RXR, but by
a protein complex that is assembled on DNA as the loop
forms. The term combinatorial assembly is used because
there are many potential complexes that can arise from the
combinations of binding to multiple sites, even when just
a single TF is involved. An illustrative example is present
in the regulation of phage l. It has two operators located
2.4 kb away from one another and each operator contains
a tandem of three sites where phage l cI repressors can
bind as dimers. In this case, two dimers bound to an operator
can form an octamer with two dimers bound to another oper-
ator by looping the intervening DNA (43,71,72). Another
example is the interaction of TFs bound at distal enhancers
with the transcriptional complexes bound at the promoter
(63). To study this type of problem, it is crucial to properly
take into account that proteins bound to distal DNA regions
can interact with each other only if DNA looping is present.
Interactions mediated by DNA looping would lead to
terms with products of three or more state variables in the
free energy (43). An illustrative example is
g
L

X
i;j
e
i;j
s
U;i
s
D;j
s
L
;
where the state variables s
U,i
, s
D,j
, and s
L
indicate whether
( 1) or not ( 0) a protein is bound to site i at the upstream
DNA region, a protein is bound to site j at the downstream
DNA region, and DNA looping is present, respectively. The
quantities e
i,j
account for the interactions between proteins
bound at different DNA regions and g
L
is the free energy
of looping. The formation of the DNA loop would be ener-
getically favorable only when a sufcient number of interac-
tions can be achieved between the two DNA regions. In the
case of phage l, only octamers and dodecamers are able to
form the looped complex among the many possible combi-
nations of binding (43,71,72). In turn, the presence of DNA
looping can enhance DNA binding through the interactions
that can be established between the two DNA regions, which
can lead to highly cooperative phenomena in the formation
of the nucleoprotein complex (42).
STOCHASTIC KINETICS
The lac operon and RXR have been used so far in this re-
view to demonstrate how biophysical principles can be
2580 Vilar and Saiz
used to efciently capture the system behavior when noise
in the form of random uctuations is not relevant. The
very same principles can be extended to take into account
the inherent stochastic nature of the underlying processes
in a wide range of situations. An efcient avenue to do so
is to consider the dynamics of the macromolecular com-
plexes that control gene expression through the stochastic
dynamics of the state variables (43,73).
The dynamics of the macromolecular complex can be
described in terms of components that can change in a tran-
sition. For the widespread case in which only one compo-
nent can change at a given time (either the component i
gets into or out of the complex), one can dene on (k
i
on
)
and off (k
i
off
) rates for the association-like and dissocia-
tion-like rates, respectively, which in general depend on
the pretransition and posttransition states of the complex.
The explicit dynamics can be obtained by considering the
change in state variables as reactions given by
s
i
!
r
i
1 s
i
; with r
i
1 s
i
k
i
on
s s
i
k
i
off
s: (16)
These reactions change the variable s
i
to 1 when it is 0 and to
0 when it is 1, representing that the element gets into or out
of the complex. Typically, the on-rate does not depend as
strongly on the state of the complex as the off-rate. The
on-rate is essentially the rate of transferring the component
from solution to the complex. The off-rate, in contrast,
depends exponentially on the free energy change.
The principle of detailed balance (33) can be used to
obtain the off-rates from the on-rates:
k
i
off
s k
i
on
s
0
e
DGs
0
DGs=RT
: (17)
The remarkable property of this expression is that reactions
with known rates can be used to infer the rates of more com-
plex reactions from the equilibrium properties (for instance,
to infer dissociation rates for different binding sites from a
single association rate (43)). In general, the association
rate could also depend on the state of the complex and its
free energy, as for instance if the presence of a TF facilitates
the association of another TF. If this dependence is included
in the on-rate, Eq. 17 can also be applied straightforwardly
to obtain the off-rate.
The stochastic dynamics of the resulting networks of
reactions and transitions can then be obtained with kinetic
Monte Carlo simulations using well-established algorithms
(26,74,75).
NOISE AND FLUCTUATIONS IN THE LAC OPERON
Stochastic effects in the lac operon have been known to be
important since the late 1950s (76), predating the discovery
of gene regulation (45). The most salient example is the all-
or-none induction process (76), which was measured at the
single-cell level with a resolution of a few molecules of the
gene products per cell (77). This effect has its roots in
the amplication of the inherent stochastic uctuations of
transcription and translation processes (7881) close to the
boundary that separates the induced from noninduced
states of the lac operon (24,82).
The underlying molecular mechanisms and parameters
have been shown to shape transcriptional noise to a large
extent (43,49,8386). To illustrate these effects in the lac
operon, we discuss, rst, regulation through just the main
operator. The use of state variables leads to a single reaction
that describes both the binding and unbinding of the
repressor to the main operator:
s
m
!
rm
1 s
m
; with r
m
nk
a
1 s
m
s
m
e
gm=RT
n
:
(18)
Here, the on-rate is given by [n]k
a
, where k
a
is the associa-
tion rate constant, and the off-rate, k
a
e
g
m
=RT
, is obtained
from the detailed balance principle. The transcription rate
is described by
m!
G
S
m 1; with G
S
G
max
1 s
m
; (19)
and mRNA degradation, protein production, and protein
degradation are described by the stochastic counterpart of
the expressions in Eq. 1. The time courses of the number
of proteins produced from this promoter show relatively
small uctuations for the experimental values of the param-
eters (Fig. 3 A). The downside of having just a binding site
for regulation is that repression is relatively weak and
a substantial number of proteins are produced.
To increase repression, two simple alternatives exist. The
rst one is to consider a stronger site. For a site 50-times
stronger than the wild-type main operator, protein produc-
tion would be close to the value expected for the lac operon
with the three operators. In this case, the average protein
production is reduced ~50 times, as expected from the
deterministic theory, but uctuations increase dramatically
(Fig. 3 B). There are infrequent mRNA bursts that lead to
large protein amounts that decay in a few hours and long
periods of time without any protein at all. The second alter-
native is to include more repressors. For a repressor concen-
tration 50-times higher than in wild-type, the average
protein production is reduced ~50 times and the uctuations
remain relatively small (Fig. 3 C). In this case, mRNA pro-
duction happens in smaller quantities but more frequently.
The physiological downside is that the repressor production
would have to be 50-times higher than in wild-type and if
that happens for all the proteins of the cell, E. coli would
have to be 50-times more crowded.
A more efcient alternative to increase repression is to
use DNA looping, which has been chosen by evolution
not only in the lac operon but also in a large variety of sys-
tems. The computational approach in this case is slightly
more involved because it has to take into account that an
operator can be bound by a repressor in solution or by a
repressor bound to the other operator thus forming a DNA
loop (Fig. 1 B). For binding to the main operator, these
two processes are represented by
s
m
!
rm
1 s
m
; with r
m
nk
a
1 s
m
s
m
1 s
a
s
L
e
gm=RT
n
; (20)
fs
L
; s
m
g!
r
L;m
f1 s
L
; 1 s
m
g; with r
L;m
e
g
L
=RT
k
a
s
a
1 s
L
1 s
m
s
L
s
m
e
gmg
L
=RT
: (21)
For the binding to the auxiliary operator, the reactions have
the same representation except that the terms s
m
, s
a
, and g
m
are replaced by s
a
, s
m
, and g
a
, respectively.
The stochastic kinetics of the regulation through the O
1
-
O
2
loop shows a small average number of proteins with
low uctuations, thus behaving in a manner very similar
to a single operator with 50-times more repressor (Fig. 3
D). Therefore, DNA looping in this case allows the system
to achieve the same behavior as it would with 50-times more
repressors.
Intuitively, both looping and high repressor concentration
lead to lower noise than a single strong site because of the
characteristic timescales involved. In the strong site case,
there are long periods of time with maximum transcriptional
activity and long periods without any activity, which results
in the number of proteins uctuating strongly between high
and low values. In the cases of looping and high repressor
concentration, the off-rate of the repressor from the main
operator is 50-times larger than for the strong site and the
average on-rate increases accordingly to keep the same
repression level. Therefore, the switching between tran-
scriptional states is very fast and mRNA production is in
the form of short and frequent bursts. This lack of long
periods of time with either full or null production gives a
narrower distribution of the number of proteins. Explicitly,
the coefcient of variation of protein (mRNA) content
shown in Fig. 3 for the strong site, the high repressor con-
centration, and the DNA looping cases is 2.3 (12.9), 0.81
(4.8), and 0.95 (5.4), respectively.
DISCUSSION
Gene expression relies on intricate molecular mechanisms
to function in extraordinarily diverse intra- and extracellular
environments. Biophysical approaches have provided new
avenues to unravel how these different levels of molecular
complexity contribute to the observed behavior. The results
reviewed here show that the underlying complexity of bio-
logical systems is not just an accident of evolution but has
a functional role.
Explicitly, the lac operon exemplies how escalating
complexity from one to two operators introduces stronger
repression while preserving low transcriptional noise, which
is not possible with a stronger single binding site.
time (hr)

m
R
N
A

,
p
r
o
t
e
i
n
O
1
O
2
O
1
looping
50 x more repressor
50 x stronger site
single site
A
B
C
D
<prot>=319
<prot>=6.9
<prot>=5.9
<prot>=8.6
FIGURE 3 Transcriptional noise in the lac operon. (AC) Time courses
of the number of protein and mRNA (shown as negative values) produced
from a promoter with just the main operator described by Eqs. 18 and
19 and the stochastic implementation of the expressions in Eq. 1. The
values of the common parameters for all three panels are G
max
0.5 s
1
,
U 0.01 s
1
, g
m
3.3 10
3
s
1
, g
p
9.2 10
5
s
1
, and k
a

2.2 10
6
M
1
s
1
. The values of the remaining parameters are: (A)
g
m
13.1 kcal/mol and [n] 15 nM for wild-type O
m
and wild-type
repressor concentration; (B) g
m
13.1 RT ln 50 kcal/mol and [n]
15 nM for 50-times stronger O
m
and wild-type repressor concentration;
and (C) g
m
13.1 kcal/mol and [n] 750 nM for wild-type O
m
and
50-times more repressor. (D) Time courses of the number of protein and
mRNA (shown as negative values) produced from a promoter with the
operators O
1
and O
2
described by Eqs. 1921 and the stochastic implemen-
tation of the expressions in Eq. 1. The values of the parameters are the
same as in panel A with the addition of g
a
11.6 kcal/mol and g
L

8.30 kcal/mol. The illustrations to the right of panels A and D represent
respectively a lac repressor bound to the main operator and a lac repressor
bound simultaneously to the main and auxiliary operators looping the
intervening DNA.
2582 Vilar and Saiz
In the case of the RXR, the additional complexity
embedded in the control of its oligomeric state by the
cognate ligand and its ability to bind simultaneously single
and multiple DNA sites has been shown to impart precision
and exibility, two seemingly antagonistic properties, to the
sensing of cellular signals.
This type of regulated oligomerization has also been
observed explicitly in other transcription factors that can
bind multiple DNA sites simultaneously, such as the tumor
suppressor p53 (87), the nuclear factor kB (NF-kB) (88,89),
the signal transducers and activators of transcription
(STATs) (90), and the octamer-binding proteins (Oct)
(91,92). In these systems, the properties of self-assembly,
and the partitioning into low- and high-order oligomeric
species, are strongly regulated and modulated by several
types of signals, such as ligand binding (60), protein binding
(93,94), acetylation (95), and phosphorylation (92,96).
The combined presence of exibility and precision in the
control of gene expression, as explicitly shown for RXR,
allows a single TF to simultaneously regulate multiple genes
with promoter-tailored dose-response curves that consis-
tently maintain their diverse shapes for a broad range of
the TF concentration changes.
Thus, the complexity of multiple repeated distal DNA
binding sites both in prokaryotes and eukaryotes, far from
being just a remnant of evolution or a backup system as
often assumed, can confer fundamental properties that are
not present in simpler setups.
This work was supported by the El Ministerio de Econom a y Competitivi-
dad under grants No. FIS2009-10352 and No. FIS2012-38105 (to J.M.G.V.)
and the University of California at Davis (to L.S.).
REFERENCES
1. Alberts, B., A. Johnson, ., P. Walter. 2008. Molecular Biology of the
Cell. Garland Science, New York.
2. Ptashne, M., and A. Gann. 2002. Genes and Signals. Cold Spring
Harbor Laboratory Press, Cold Spring Harbor, NY.
3. Levine, M., and R. Tjian. 2003. Transcription regulation and animal
diversity. Nature. 424:147151.
4. Hermsen, R., P. R. ten Wolde, and S. Teichmann. 2008. Chance and ne-
cessity in chromosomal gene distributions. Trends Genet. 24:216219.
5. Teif, V. B., and K. Rippe. 2012. Calculating transcription factor binding
maps for chromatin. Brief. Bioinform. 13:187201.
6. Li, Q., G. Barkess, and H. Qian. 2006. Chromatin looping and the
probability of transcription. Trends Genet. 22:197202.
7. Fudenberg, G., and L. A. Mirny. 2012. Higher-order chromatin
structure: bridging physics and biology. Curr. Opin. Genet. Dev.
22:115124.
8. Naumova, N., and J. Dekker. 2010. Integrating one-dimensional and
three-dimensional maps of genomes. J. Cell Sci. 123:19791988.
9. Metivier, R., G. Penot, ., F. Gannon. 2003. Estrogen receptor-a
directs ordered, cyclical, and combinatorial recruitment of cofactors
on a natural target promoter. Cell. 115:751763.
10. Djordjevic, M., and R. Bundschuh. 2008. Formation of the open com-
plex by bacterial RNA polymerasea quantitative model. Biophys. J.
94:42334248.
11. Ay, A., and D. N. Arnosti. 2011. Mathematical modeling of gene
expression: a guide for the perplexed biologist. Crit. Rev. Biochem.
Mol. Biol. 46:137151.
12. de Jong, H. 2002. Modeling and simulation of genetic regulatory sys-
tems: a literature review. J. Comput. Biol. 9:67103.
13. Garcia, H. G., A. Sanchez, ., R. Phillips. 2010. Transcription by the
numbers redux: experiments and calculations that surprise. Trends
Cell Biol. 20:723733.
14. Karlebach, G., and R. Shamir. 2008. Modeling and analysis of gene
regulatory networks. Nat. Rev. Mol. Cell Biol. 9:770780.
15. Struf, P., M. Corado, ., S. Small. 2011. Combinatorial activation and
concentration-dependent repression of the Drosophila even skipped
stripe 37 enhancer. Development. 138:42914299.
16. Perry, M. W., A. N. Boettiger, and M. Levine. 2011. Multiple enhancers
ensure precision of gap gene-expression patterns in the Drosophila
embryo. Proc. Natl. Acad. Sci. USA. 108:1357013575.
17. Segal, E., T. Raveh-Sadka, ., U. Gaul. 2008. Predicting expression
patterns from regulatory sequence in Drosophila segmentation. Nature.
451:535540.
18. He, X., M. A. Samee, ., S. Sinha. 2010. Thermodynamics-based
models of transcriptional regulation by enhancers: the roles of syner-
gistic activation, cooperative binding and short-range repression.
PLOS Comput. Biol. 6:e1000935.
19. Juven-Gershon, T., and J. T. Kadonaga. 2010. Regulation of gene
expression via the core promoter and the basal transcriptional machin-
ery. Dev. Biol. 339:225229.
20. Boettiger, A. N., P. L. Ralph, and S. N. Evans. 2011. Transcriptional
regulation: effects of promoter proximal pausing on speed, synchrony
and reliability. PLOS Comput. Biol. 7:e1001136.
21. Levine, M. 2011. Paused RNA polymerase II as a developmental
checkpoint. Cell. 145:502511.
22. Klumpp, S., and T. Hwa. 2008. Stochasticity and trafc jams in the
transcription of ribosomal RNA: intriguing role of termination and
antitermination. Proc. Natl. Acad. Sci. USA. 105:1815918164.
23. Eulalio, A., E. Huntzinger, and E. Izaurralde. 2008. Getting to the root
of miRNA-mediated gene silencing. Cell. 132:914.
24. Vilar, J. M. G., C. C. Guet, and S. Leibler. 2003. Modeling network
dynamics: the lac operon, a case study. J. Cell Biol. 161:471476.
25. Paulsson, J. 2005. Models of stochastic gene expression. Phys. Life Rev.
2:157175.
26. McAdams, H. H., and A. Arkin. 1997. Stochastic mechanisms in gene
expression. Proc. Natl. Acad. Sci. USA. 94:814819.
27. Morelli, M. J., R. J. Allen, and P. R. Wolde. 2011. Effects of macromo-
lecular crowding on genetic networks. Biophys. J. 101:28822891.
28. Teif, V. B. 2010. Predicting gene-regulation functions: lessons from
temperate bacteriophages. Biophys. J. 98:12471256.
29. Cottrell, D., P. S. Swain, and P. F. Tupper. 2012. Stochastic branching-
diffusion models for gene expression. Proc. Natl. Acad. Sci. USA.
109:96999704.
30. Saiz, L. 2012. The physics of protein-DNA interaction networks in the
control of gene expression. J. Phys. Condens. Matter. 24:193102.
31. Gillespie, D. T. 2009. Deterministic limit of stochastic chemical
kinetics. J. Phys. Chem. B. 113:16401644.
32. Vilar, J. M. G., H. Y. Kueh, ., S. Leibler. 2002. Mechanisms of noise-
resistance in genetic oscillators. Proc. Natl. Acad. Sci. USA. 99:5988
5992.
33. Hill, T. L. 1960. An Introduction to Statistical Thermodynamics.
Addison-Wesley, Reading, MA.
34. Ackers, G. K., A. D. Johnson, and M. A. Shea. 1982. Quantitative
model for gene regulation by lphage repressor. Proc. Natl. Acad. Sci.
USA. 79:11291133.
35. Segal, E., and J. Widom. 2009. From DNA sequence to transcriptional
behavior: a quantitative approach. Nat. Rev. Genet. 10:443456.
36. Le Nove`re, N., and T. S. Shimizu. 2001. STOCHSIM: modeling of
stochastic biomolecular processes. Bioinformatics. 17:575576.
37. Hlavacek, W. S., J. R. Faeder, ., W. Fontana. 2006. Rules for
modeling signal-transduction systems. Sci. STKE. 2006:re6.
38. Borisov, N. M., N. I. Markevich, ., B. N. Kholodenko. 2006. Trading
the micro-world of combinatorial complexity for the macro-world of
protein interaction domains. Biosystems. 83:152166.
39. Bray, D., and S. Lay. 1997. Computer-based analysis of the binding
steps in protein complex formation. Proc. Natl. Acad. Sci. USA.
94:1349313498.
40. Ollivier, J. F., V. Shahrezaei, and P. S. Swain. 2010. Scalable rule-based
modeling of allosteric proteins and biochemical networks. PLOS
Comput. Biol. 6:e1000975.
41. Deeds, E. J., J. A. Bachman, and W. Fontana. 2012. Optimizing ring
assembly reveals the strength of weak interactions. Proc. Natl. Acad.
Sci. USA. 109:23482353.
42. Vilar, J. M. G., and L. Saiz. 2006. Multiprotein DNA looping. Phys.
Rev. Lett. 96:238103.
43. Saiz, L., and J. M. G. Vilar. 2006. Stochastic dynamics of macromolec-
ular-assembly networks. Mol. Syst. Biol. 2:2006.0024.
44. Muller-Hill, B. 1996. The Lac Operon: a Short History of a Genetic
Paradigm. Walter de Gruyter, Berlin, Germany.
45. Jacob, F., and J. Monod. 1961. Genetic regulatory mechanisms in the
synthesis of proteins. J. Mol. Biol. 3:318356.
46. Oehler, S., M. Amouyal, ., B. Muller-Hill. 1994. Quality and position
of the three lac operators of E. coli dene efciency of repression.
EMBO J. 13:33483355.
47. Oehler, S., E. R. Eismann, ., B. Muller-Hill. 1990. The three operators
of the lac operon cooperate in repression. EMBO J. 9:973979.
48. Mossing, M. C., and M. T. Record, Jr. 1986. Upstream operators
enhance repression of the lac promoter. Science. 233:889892.
49. Vilar, J. M. G., and S. Leibler. 2003. DNA looping and physical con-
straints on transcription regulation. J. Mol. Biol. 331:981989.
50. Saiz, L., and J. M. G. Vilar. 2008. Ab initio thermodynamic modeling
of distal multisite transcription regulation. Nucleic Acids Res. 36:
726731.
51. Kuhlman, T., Z. Zhang, ., T. Hwa. 2007. Combinatorial transcrip-
tional control of the lactose operon of Escherichia coli. Proc. Natl.
Acad. Sci. USA. 104:60436048.
52. Narang, A. 2007. Effect of DNA looping on the induction kinetics of
the lac operon. J. Theor. Biol. 247:695712.
53. Vilar, J. M. G., and L. Saiz. 2010. CPLEXA: a MATHEMATICA pack-
age to study macromolecular-assembly control of gene expression.
Bioinformatics. 26:20602061.
54. Vilar, J. M. G., and L. Saiz. 2010. CPLEXA. http://sourceforge.net/
projects/cplexa.
55. Vilar, J. M. G. 2010. Accurate prediction of gene expression by integra-
tion of DNA sequence statistics with detailed modeling of transcription
regulation. Biophys. J. 99:24082413.
56. Simicevic, J., and B. Deplancke. 2010. DNA-centered approaches to
characterize regulatory protein-DNA interaction complexes. Mol. Bio-
sys. 6:462468.
57. Lemaire, V., C. F. Lee, ., L. Glass. 2006. Sequential recruitment and
combinatorial assembling of multiprotein complexes in transcriptional
activation. Phys. Rev. Lett. 96:198102.
58. Altucci, L., and H. Gronemeyer. 2001. The promise of retinoids to ght
against cancer. Nat. Rev. Cancer. 1:181193.
59. Kersten, S., M. I. Dawson, ., N. Noy. 1996. Individual subunits of het-
erodimers comprised of retinoic acid and retinoid X receptors interact
with their ligands independently. Biochemistry. 35:38163824.
60. Chen, Z. P., J. Iyer, ., H. Gronemeyer. 1998. Ligand- and DNA-
induced dissociation of RXR tetramers. J. Mol. Biol. 275:5565.
61. Vilar, J. M. G., and L. Saiz. 2011. Control of gene expression by modu-
lated self-assembly. Nucleic Acids Res. 39:68546863.
62. Yasmin, R., K. T. Yeung, ., N. Noy. 2004. DNA-looping by RXR tet-
ramers permits transcriptional regulation at a distance. J. Mol. Biol.
343:327338.
63. Nolis, I. K., D. J. McKay, ., D. Thanos. 2009. Transcription factors
mediate long-range enhancer-promoter interactions. Proc. Natl.
Acad. Sci. USA. 106:2022220227.
64. Chopra, V. S., N. Kong, and M. Levine. 2012. Transcriptional repres-
sion via antilooping in the Drosophila embryo. Proc. Natl. Acad. Sci.
USA. 109:94609464.
65. Vilar, J. M. G., and L. Saiz. 2005. DNA looping in gene regulation:
from the assembly of macromolecular complexes to the control of tran-
scriptional noise. Curr. Opin. Genet. Dev. 15:136144.
66. Ptashne, M., and A. Gann. 1997. Transcriptional activation by recruit-
ment. Nature. 386:569577.
67. Saiz, L., J. M. Rubi, and J. M. G. Vilar. 2005. Inferring the in vivo loop-
ing properties of DNA. Proc. Natl. Acad. Sci. USA. 102:1764217645.
68. Jackson, P., I. Mastrangelo, ., J. Barrett. 1998. Synergistic transcrip-
tional activation of the MCK promoter by p53: tetramers link separated
DNA response elements by DNA looping. Oncogene. 16:283292.
69. Levin, A. A., L. J. Sturzenbecker, ., A. Lovey. 1992. 9-cis retinoic
acid stereoisomer binds and activates the nuclear receptor RXR a.
Nature. 355:359361.
70. Heyman, R. A., D. J. Mangelsdorf, ., C. Thaller. 1992. 9-cis retinoic
acid is a high afnity ligand for the retinoid X receptor. Cell. 68:
397406.
71. Ptashne, M. 2004. A Genetic Switch: Phage l Revisited. Cold Spring
Harbor Laboratory Press, Cold Spring Harbor, NY.
72. Revet, B., B. von Wilcken-Bergmann, ., B. Muller-Hill. 1999. Four
dimers of l repressor bound to two suitably spaced pairs of l operators
form octamers and DNA loops over large distances. Curr. Biol. 9:
151154.
73. Saiz, L., and J. M. G. Vilar. 2008. Protein-protein/DNA interaction net-
works: versatile macromolecular structures for the control of gene
expression. IET Syst. Biol. 2:247255.
74. Gillespie, D. T. 1976. General method for numerically simulating sto-
chastic time evolution of coupled chemical reactions. J. Comput. Phys.
22:403434.
75. Bortz, A. B., M. H. Kalos, and J. L. Lebowitz. 1975. New algorithm
for Monte Carlo simulation of Ising spin systems. J. Comput. Phys.
17:1018.
76. Novick, A., and M. Weiner. 1957. Enzyme induction as an all-or-none
phenomenon. Proc. Natl. Acad. Sci. USA. 43:553566.
77. Maloney, P. C., and B. Rotman. 1973. Distribution of suboptimally
induces -D-galactosidase in Escherichia coli. The enzyme content of
individual cells. J. Mol. Biol. 73:7791.
78. Golding, I., J. Paulsson, ., E. C. Cox. 2005. Real-time kinetics of gene
activity in individual bacteria. Cell. 123:10251036.
79. Paulsson, J. 2004. Summing up the noise in gene networks. Nature.
427:415418.
80. Elowitz, M. B., A. J. Levine, ., P. S. Swain. 2002. Stochastic gene
expression in a single cell. Science. 297:11831186.
81. Guet, C. C., L. Bruneaux, ., P. Cluzel. 2008. Minimally invasive
determination of mRNA concentration in single living bacteria.
Nucleic Acids Res. 36:e73.
82. Ozbudak, E. M., M. Thattai, ., A. van Oudenaarden. 2004.
Multistability in the lactose utilization network of Escherichia coli.
Nature. 427:737740.
83. Blake, W. J., M. Krn, ., J. J. Collins. 2003. Noise in eukaryotic gene
expression. Nature. 422:633637.
84. Coulon, A., O. Gandrillon, and G. Beslon. 2010. On the spontaneous
stochastic dynamics of a single gene: complexity of the molecular
interplay at the promoter. BMC Syst. Biol. 4:2.
85. Elf, J., G. W. Li, and X. S. Xie. 2007. Probing transcription factor
dynamics at the single-molecule level in a living cell. Science.
316:11911194.
2584 Vilar and Saiz
86. Hammar, P., P. Leroy, ., J. Elf. 2012. The lac repressor displays facil-
itated diffusion in living cells. Science. 336:15951598.
87. Wang, P., M. Reed, ., P. Tegtmeyer. 1994. p53 domains: structure,
oligomerization, and transformation. Mol. Cell. Biol. 14:51825191.
88. Phelps, C. B., L. L. Sengchanthalangsy, ., G. Ghosh. 2000.
Mechanism of kB DNA binding by Rel/NF-kB dimers. J. Biol.
Chem. 275:2439224399.
89. Sengchanthalangsy, L. L., S. Datta, ., G. Ghosh. 1999. Characteriza-
tion of the dimer interface of transcription factor NFkB p50 homo-
dimer. J. Mol. Biol. 289:10291040.
90. Zhang, X., and J. E. Darnell, Jr. 2001. Functional importance of
Stat3 tetramerization in activation of the a2-macroglobulin gene.
J. Biol. Chem. 276:3357633581.
91. Tomilin, A., A. Remenyi, ., H. R. Scholer. 2000. Synergism with the
coactivator OBF-1 (OCA-B, BOB-1) is mediated by a specic POU
dimer conguration. Cell. 103:853864.
92. Kang, J., M. Gemberling, ., D. Tantin. 2009. A general mechanism
for transcription regulation by Oct1 and Oct4 in response to genotoxic
and oxidative stress. Genes Dev. 23:208222.
93. van Dieck, J., M. R. Fernandez-Fernandez, ., A. R. Fersht. 2009.
Modulation of the oligomerization state of p53 by differential binding
of proteins of the S100 family to p53 monomers and tetramers. J. Biol.
Chem. 284:1380413811.
94. Hanson, S., E. Kim, and W. Deppert. 2005. Redox factor 1 (Ref-1) en-
hances specic DNA binding of p53 by promoting p53 tetramerization.
Oncogene. 24:16411647.
95. Li, A. G., L. G. Piluso, ., X. Liu. 2006. Mechanistic insights
into maintenance of high p53 acetylation by PTEN. Mol. Cell.
23:575587.
96. Wenta, N., H. Strauss, ., U. Vinkemeier. 2008. Tyrosine phosphoryla-
tion regulates the partitioning of STAT1 between different dimer con-
formations. Proc. Natl. Acad. Sci. USA. 105:92389243.
97. Bissonnette, R. P., T. Brunner, ., R. A. Heyman. 1995. 9-cis retinoic
acid inhibition of activation-induced apoptosis is mediated via regula-
tion of Fas ligand and requires retinoic acid receptor and retinoid
X receptor activation. Mol. Cell. Biol. 15:55765585.
98. Vuligonda, V., Y. Lin, and R. A. S. Chandraratna. 1996. Synthesis of
highly potent RXR-specic retinoids: the use of a cyclopropyl group
as a double bond isostere. Bioorg. Med. Chem. Lett. 6:213218.
99. Kersten, S., D. Kelleher, ., N. Noy. 1995. Retinoid X receptor a forms
tetramers in solution. Proc. Natl. Acad. Sci. USA. 92:86458649.
Nanoscale Distribution of Ryanodine Receptors and Caveolin-3 in Mouse
Ventricular Myocytes: Dilation of T-Tubules near Junctions
Joseph Wong,
6
David Baddeley,
k6
Eric A. Bushong,
Zeyun Yu,
{
Mark H. Ellisman,
Masahiko Hoshijima,
*
and Christian Soeller
***
Center for Research in Biological Systems and

Department of Medicine, University of California San Diego, La Jolla, California;

Department
of Physiology, School of Medical Sciences, University of Auckland, New Zealand;
{
Department of Computer Science, University of
Wisconsin-Milwaukee, Milwaukee, Wisconsin;
k
Department of Cell Biology, Yale University, New Haven, Connecticut; and **Biomedical
Physics, University of Exeter, United Kingdom
ABSTRACT We conducted super-resolution light microscopy (LM) imaging of the distribution of ryanodine receptors (RyRs)
and caveolin-3 (CAV3) in mouse ventricular myocytes. Quantitative analysis of data at the surface sarcolemma showed that
4.8% of RyR labeling colocalized with CAV3 whereas 3.5% of CAV3 was in areas with RyR labeling. These values increased
to 9.2 and 9.0%, respectively, in the interior of myocytes where CAV3 was widely expressed in the t-system but reduced in
regions associated with junctional couplings. Electron microscopic (EM) tomography independently showed only few couplings
with caveolae and little evidence for caveolar shapes on the t-system. Unexpectedly, both super-resolution LMand three-dimen-
sional EM data (including serial block-face scanning EM) revealed signicant increases in local t-system diameters in many
regions associated with junctions. We suggest that this regional specialization helps reduce ionic accumulation and depletion
in t-system lumen during excitation-contraction coupling to ensure effective local Ca
2
release. Our data demonstrate that
super-resolution LMand volume EMtechniques complementarily enhance information on subcellular structure at the nanoscale.
Received for publication 3 December 2012 and in nal form 25 February 2013.
6
Joseph Wong and David Baddeley equally contributed to this work.
*Correspondence: c.soeller@auckland.ac.nz or mhoshijima@ucsd.edu
The contraction of cardiac ventricular myocytes depends on
the rapid cell-wide transient increase in intracellular [Ca
2
]
upon depolarization of the cell-membrane potential. The car-
diac ryanodine receptor (RyR) (1), which is the intracellular
Ca
2
release channel in the sarcoplasmic reticulum (SR),
plays a central role in shaping Ca
2
transients. RyRs form
clusters of various sizes (2,3) with the majority located
within junctions between the SR and the surface membrane
and its cytoplasmic extension, the transverse tubular (t-) sys-
tem. It has been suggested that some RyR clusters are asso-
ciated with caveolae, a specialized signaling microdomain
of the surface membrane. Previous studies were complicated
by the limited resolution of optical imaging methods of
~250 nm, much larger than the nanometer scale of RyRs
and caveolae. Accordingly, these studies report varying
colocalization between RyRs and caveolin-3 (CAV3), a
caveolar marker also expressed in the t-system (4,5).
In this work, we investigated the relative distribution of
CAV3 and RyRs in mouse ventricular myocytes both in
the cytosol and near the cell surface with super-resolution
uorescence microscopy that achieves a resolution
approaching 30 nm. Our data revealed unexpected local
t-system swellings near junctional couplings, which was
supported by two different three-dimensional electron
microscopy (EM) modalities with <10-nm resolution: EM
tomography and serial block-face scanning EM (SBFSEM).
Super-resolution images of CAV3 and RyR labeling at
the surface sarcolemma of mouse myocytes showed little
overlap, suggesting that few RyRs were in couplings
with caveolae (Fig. 1 A, for detailed methods, see the Sup-
porting Material). Only ~4.8% of RyR labeling was asso-
ciated with CAV3 positive areas and ~3.5% of CAV3
associated with RyR positive areas (n 6 cells from three
animals, Fig. 1 B, see also Table S1 in the Supporting
Material), broadly consistent with previous data in rats
(6). To support this nding, EM tomography was applied
to mouse ventricular tissue that included a part of the
surface sarcolemma, to our knowledge for the rst time.
Segmentation of peripheral couplings (containing RyR
foot structures) and surface caveolae (~60 nm in diameter
and often interconnected) conrmed that the great majority
of peripheral couplings were in regions devoid of caveolae
(Fig. 1 C). A few junctional couplings containing feet were
between caveolae and subsarcolemmal SR (Fig. 1 D, see
also Fig. S1 and Movie S1 in the Supporting Material).
We conducted a similar analysis in the cytosol where
CAV3 expression occurs in the t-system (5) and RyRs
are abundant in dyadic junctions between the t-system
and SR terminal cisterns.
As shown in Fig. 2 A, the spatial distribution of CAV3 and
RyR clusters in super-resolution micrographs taken several
Editor: Leslie Loew.
http://dx.doi.org/10.1016/j.bpj.2013.02.059
L22 Biophysical Journal Volume 104 June 2013 L22L24
microns belowthe surface sarcolemma is consistent with this
view. The association of the two labels is slightly increased
(as compared to the surface), according to distance analysis
with 9% of CAV3 and 9.2% of RyR labeling associating
with each other (Fig. 2 B, n 6 cells from three animals).
The similarity of manually traced t-systemin EMtomograms
(Fig. 2 C) and super-resolved CAV3 labeling suggested that
CAV3 is widely distributed in the t-systemexcept for regions
where dyadic membrane junctions occur as CAV3 labeling
was much weaker in regions with strong RyR labeling. It
was notable that the t-system diameter appeared to increase
at regions of strong RyR labeling (Fig. 2 D), broadly consis-
tent with the behavior seen in tomograms (Fig. 2 C). This was
conrmed by a quantitative analysis of t-tubule diameters in
dyadic versus extradyadic regions on the basis of CAV3 and
RyR labeling, with full-width at quarter-maximum mean di-
ameters increasing from~150 nmdistal to dyads, to ~190 nm
(using CAV3 signal only) or ~280 nm (using CAV3 and RyR
signal) near dyads (Fig. 2, G and H, see also Methods in the
Supporting Material). The combined RyR and CAV3 signals
seemed to be a better representation of the entire t-system
lumen near junctions (see Fig. S2).
Taken together, super-resolution imaging and EM tomog-
raphystronglysupport the presence of local t-systemdilations
B
0 200 400 600 800
0.06
0.08
0.10
-200 0 200 400 600 800
0.00
0.02
0.04
0.06
0.08
0.10
Distance (nm)
RyR
CAV3
F
r
a
c
t
i
o
n

o
f

l
a
b
e
l
l
i
n
g
500 nm
A
200 nm
C
D
FIGURE 1 Colocalization of CAV3
and RyRs at the surface sarco-
lemma. (A) Super-resolution micro-
graph of the distribution of CAV3
(green) and RyRs (red) at the surface
of a mouse cardiac myocyte. (B)
Analysis of the association of
CAV3 with RyRs. The fraction of
RyR labeling within CAV3 positive
areas was ~4.8% (front data)
whereas ~3.5% of CAV3 was found
in RyR-positive membrane areas.
(C) Segmented EM tomogram containing a patch of surface sarcolemma (light blue) and associated caveolae (green) as well as
peripheral couplings (red). (D) Detailed view of a region with abundant caveolae. (Arrows) Couplings with caveolae.
250 nm
500 nm D E F
Distance (nm)
-200 0 200 400 600 800
0
0.05
0.10
0.15
0 200 400 600
800
0
0.05
0.10
0.15
B
RyR
CAV3
F
r
a
c
t
i
o
n

o
f

l
a
b
e
l
l
i
n
g
G
3000
0 200 400 600 800
Diameter (nm)
#

o
f

o
b
s
e
r
v
a
t
i
o
n
s
2000
1000
0
ex-dyad
(CAV3)
dyad
(CAV3)
dyad
(CAV3 &
RyR)
H
300
M
e
a
n

D
i
a
m
e
t
e
r
(
n
m
)
0
**
**
ex-dyad
(CAV3)
dyad
(CAV3)
dyad
(CAV3
& RyR)
500 nm
A C
500 nm
FIGURE 2 Distribution of CAV3 and RyRs in the cell interior. (A) Super-resolution micrograph of CAV3 (green) and RyR (red)
distribution at t-system. (Arrow) Direction of longitudinal cell axis. (B) Distance analysis of the CAV3 and RyR association (N 6 cells
per group). (C) Segmented EM tomogram of a similar region with three-dimensional mesh models of t-system membrane (green) and
dyadic couplings (red). (D) This image illustrates the tracing (white path) of t-tubules. The label distribution was extracted and
linearized along the path (E) to calculate a mask that shows the full width at quarter-maximum diameter along tubules, CAV3 (green)
and RyR (red) (F). (G) Histograms of local diameters extracted from traced t-tubules. (H) Mean diameters in junctional (dyad) and
nonjunctional (ex-dyad) regions. See main text and the Supporting Material for details. ** p < 0.01.
Biophysical Journal 104(11) L22L24
Biophysical Letters L23
in regions where the t-system opposes SR at dyads and such
t-system bulges are connected by narrower tubule segments.
Further support was provided by SBFSEM, another volume
EM technique to study larger cell volumes (albeit at the
expense of a slightly lower resolution). SBFSEM clearly
showed local t-system dilations were regularly involved in
the architecture of most (but not all) dyads (Fig. 3, see also
Fig. S3 and Movie S2), as also observed in full three-dimen-
sional super-resolution images (see Fig. S3 C).
Our data identify local dilations of the t-system associated
with dyads in mouse cardiac myocytes. Frequent tubule
distensions had been observed especially at the intersections
of transverse and axial tubules (7), and constrictions were
seen in rabbit myocytes although their relationship to dyads
was unknown (8). The increased local t-system lumen near
junctions may help reduce the predicted ionic accumulation/
depletion during excitation-contraction coupling (9). Alterna-
tively, it might simply be secondary to increasing local mem-
brane area and allowthe formation of large area junctions that
harbor many RyRs. In connection with this point, it would be
interesting toinvestigate the t-systemnear junctions inspecies
that have larger average tubule diameters (e.g., human and
rabbit (10)), or if this architecture changes in mouse heart fail-
ure models where t-tubule diameters are often increased.
Most peripheral couplings were in regions void of surface
caveolae, although a small number of RyR clusters were in
junctional couplings between subsarcolemmal SR and
caveolae as shown both by the low colocalization between
CAV3 and RyRs as well as direct evidence from EM tomog-
raphy. Similarly, a relatively small fraction of CAV3 colo-
calized with RyR clusters in the t-system although CAV3
was expressed widely in the t-system. A structural role of
CAV3 in the t-system is still uncleart-tubules in tomo-
gram data did not reveal distinct caveolae shapes on the
t-system membrane (see Fig. S4), although this might
change in pathology (11). In any case, the t-system exhibits
high curvature orthogonal to the tubule axis, which may be
supported by CAV3 oligomerization. In addition, the pres-
ence of CAV3 in the t-system may be important for regu-
lating other signaling systems (e.g., adrenergic signaling).
Finally, our data demonstrate that complementary data
from optical super-resolution and three-dimensional EM
images assists data interpretation and reliability. We suggest
that truly correlative optical and EM imaging approaches
should provide further information and improve our knowl-
edge of the basis of cardiac excitation-contraction coupling.
SUPPORTING MATERIAL
Methods, four gures, two movies, and movie legends are available at
http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)00432-3.
ACKNOWLEDGMENTS
We acknowledge support from the Marsden Fund, Lottery Health and Health
Research Council, New Zealand; National Institutes of Health grants No.
RR004050/GM103412 (to M.H.E.) and No. R15HL103497 (to Z.Y.); and
American Heart Association National Established Investigator Award
0840013N (to M.H.).
REFERENCES and FOOTNOTES
1. Franzini-Armstrong, C., and F. Protasi. 1997. Ryanodine receptors of
striated muscles: a complex channel capable of multiple interactions.
Physiol. Rev. 77:699729.
2. Baddeley, D., I. D. Jayasinghe, ., C. Soeller. 2009. Optical single-
channel resolution imaging of the ryanodine receptor distribution in
rat cardiac myocytes. Proc. Natl. Acad. Sci. USA. 106:2227522280.
3. Hayashi, T., M. E. Martone, ., M. Hoshijima. 2009. Three-dimen-
sional electron microscopy reveals new details of membrane systems
for Ca
2
signaling in the heart. J. Cell Sci. 122:10051013.
4. Scriven, D. R. L., A. Klimek, ., E. D. Moore. 2005. Caveolin-3 is
adjacent to a group of extradyadic ryanodine receptors. Biophys. J.
89:18931901.
5. Jayasinghe, I. D., M. B. Cannell, and C. Soeller. 2009. Organization of
ryanodine receptors, transverse tubules, and sodium-calcium
exchanger in rat myocytes. Biophys. J. 97:26642673.
6. Baddeley, D., D. Crossman, ., C. Soeller. 2011. 4D super-resolution
microscopy with conventional uorophores and single wavelength
excitation in optically thick cells and tissues. PLoS ONE. 6:e20645.
7. Forbes, M. S., L. A. Hawkey, and N. Sperelakis. 1984. The transverse-
axial tubular system (TATS) of mouse myocardium: its morphology in
the developing and adult animal. Am. J. Anat. 170:143162.
8. Savio-Galimberti, E., J. Frank, ., F. B. Sachse. 2008. Novel features
of the rabbit transverse tubular system revealed by quantitative analysis
of three-dimensional reconstructions from confocal images. Biophys. J.
95:20532062.
9. Pasek, M., J. Simurda, ., G. Christe. 2008. A model of the guinea-pig
ventricular cardiac myocyte incorporating a transverse-axial tubular
system. Prog. Biophys. Mol. Biol. 96:258280.
10. Jayasinghe, I., D. Crossman, ., M. Cannell. 2012. Comparison of the
organization of t-tubules, sarcoplasmic reticulum and ryanodine recep-
tors in rat and human ventricular myocardium. Clin. Exp. Pharmacol.
Physiol. 39:469476.
11. Hoshijima, M., T. Hayashi, ., J. Ross, Jr. 2011. Delta-sarcoglycan gene
therapyhalts progressionof cardiacdysfunction, improves respiratoryfail-
ure, and prolongs life in myopathic hamsters. Circ. Heart Fail. 4:8997.
A B
1 m
500 nm
FIGURE 3 Segmented SBFSEM data showing t-system dila-
tions near dyadic junctions. (A) The overview shows t-system
membranes (green) and jSR (red) in a mouse myocyte. (B,
enlarged inset from panel A) Thin connecting tubules (arrows)
and regular swellings in junctional regions at z-lines.
L24 Biophysical Letters
Kinetics and Energetics of Biomolecular Folding and Binding
Christopher A. Pierse and Olga K. Dudko*
Department of Physics and Center for Theoretical Biological Physics, University of California at San Diego, La Jolla, California
ABSTRACT The ability of biomolecules to fold and to bind to other molecules is fundamental to virtually every living
process. Advanced experimental techniques can now reveal how single biomolecules fold or bind against mechanical force,
with the force serving as both the regulator and the probe of folding and binding transitions. Here, we present analytical expres-
sions suitable for tting the major experimental outputs from such experiments to enable their analysis and interpretation. The
t yields the key determinants of the folding and binding processes: the intrinsic on-rate and the location and height of the acti-
vation barrier.
Received for publication 30 July 2013 and in nal form 23 September 2013.
*Correspondence: dudko@physics.ucsd.edu
Dynamic processes in living cells are regulated through
conformational changes in biomoleculestheir folding
into a particular shape or binding to selected partners. The
ability of biomolecules to fold and to bind enables them
to act as switches, assembly factors, pumps, or force- and
displacement-generating motors (1). Folding and binding
transitions are often hindered by a free energy barrier. Over-
coming the barrier requires energy-demanding rearrange-
ments such as displacing water from the sites of native
contacts and breaking nonnative electrostatic contacts, as
well as loss of congurational entropy. Once the barrier is
crossed, the folded and bound states are stabilized by
short-range interactions: hydrogen bonds, favorable hydro-
phobic effects, and electrostatic and van der Waals attrac-
tions (2).
Mechanistic information about folding and binding
processes is detailed in the folding and binding trajectories
of individual molecules: observing an ensemble of mole-
cules may obscure the inherent heterogeneity of these
processes. Single-molecule trajectories can be induced,
and monitored, by applying force to unfold/unbind a
molecule and then relaxing the force until folding or
binding is observed (35) (Fig. 1). Varying the force relax-
ation rate shifts the range of forces at which folding or
binding occurs, thus broadening the explorable spectrum
of molecular responses to force and revealing conforma-
tional changes that are otherwise too fast to detect. The
measured force-dependent kinetics elucidates the role of
force in physiological processes (6) and provides ways to
control the timescales, and even the fate, of these pro-
cesses. The force-dependent data also provides a route to
understanding folding and binding in the absence of
forceby extrapolating the data to zero force via a t to
a theory.
In this letter, we derive an analytical expression for the
distribution of transition forces, the major output of force-
relaxation experiments that probe folding and binding pro-
cesses. The expression extracts the key determinants of
these processes: the on-rate and activation barrier in the
absence of force. The theory is rst developed in the context
of biomolecular folding, and is then extended to cover the
binding of a ligand tethered to a receptor. In contrast to un-
folding and unbinding, the reverse processes of folding and
binding require a theory that accounts for the compliance of
the unfolded state, as well as the effect of the tether, to
recover the true kinetic parameters of the biomolecule of
interest.
In a force-relaxation experiment, an unfolded biomole-
cule or unbound ligand-receptor complex is subject to a
stretching force, which is decreased from the initial
value F
0
as the pulling device approaches the sample at
speed V until a folding or binding transition is observed
(Fig. 1) (35). Dene S(t) as the probability that the
molecule has not yet escaped from the unfolded (implied:
or unbound) state at time t. When escape is limited by
one dominant barrier, S(t) follows the rst-order rate
equation
_
Sth
dSt
dt
k
)
FtSt;
where k
)
(F(t)) is the on-rate at force F at time t.
Because, prior to the transition, the applied force decreases
monotonically with time, the distribution of transition
forces, p(F), is related to S(t) through pFdF
_
Stdt,
yielding
pF
k
)
F
_
FF
e
R
F
F
0
k)F
0
_
FF
0
dF
0
: (1)
Editor: Antoine van Oijen.
http://dx.doi.org/10.1016/j.bpj.2013.09.023
Biophysical Journal Volume 105 November 2013 L19L22 L19
Here
_
FFhdFt=dt<0 is the force relaxation rate. The
proper normalization of p(F) is readily conrmed by inte-
grating Eq. 1 from the initial force F
0
to negative innity,
the latter accounting for transitions that do not occur by
the end of the experiment. Note that the expression for the
distribution of folding/binding forces in Eq. 1 differs from
its analog for the unfolding process (7) by the limits of inte-
gration and a negative sign, reecting the property of a
relaxation experiment to decrease the survival probability
S(t) by decreasing the force. Converting the formal expres-
sion in Eq. 1 into a form suitable for tting experimental
data requires establishing functional forms for k
)
(F) and
_
FF and analytically solving the integral. These steps are
accomplished below.
The on-rate k
)
(F) is computed by treating the conforma-
tional dynamics of the molecule as a random walk on the
combined free energy prole G(x,t) G
0
(x) G
pull
(x,t)
along the molecular extension x. Here G
0
(x) is the intrinsic
molecular potential and G
pull
(x,t) is the potential of the pull-
ing device. When G(x,t) features a high barrier on the scale
of k
B
T (k
B
is the Boltzmann constant and T the temperature),
the dynamics can be treated as diffusive. The unfolded re-
gion of the intrinsic potential for a folding process, unlike
that for a barrierless process (8), can be captured by the
function
G
0
x DG
z
n
1 n
x
x
z
1
1n
DG
z
n
x
x
z
;
which has a sharp (if n 1/2, Fig. 2, inset) or smooth (if n
2/3) barrier of height DG
z
and location x
z
. The potential of a
pulling device of stiffness k
S
is G
pull
(x,t) k
S
/2(X
0
Vt x)
2
with an initial minimum at X
0
(corresponding to F
0
).
Applying Kramers formalism (9) to the combined potential
G(x,t), we establish the analytical form of the on-rate at
force F(t),
k
)
F k
0
1
k
S
k
U
F
1
n
1
2
1
nFx
z
DG
z
1
n
1
e
bDG
z
"
1
1
k
S
k
U
F
2n
1n
1
1
nFx
z
DG
z
1
n
#
;
where k
0
is the intrinsic on-rate, b h (k
B
T)
1
, and
k
U
F
n
1 n
2
DG
z
x
z2
1
nFx
z
DG
z
2
1
n
is the stiffness of the unfolded biomolecule under force F
(see the Supporting Material for details on all derivations).
The full nonlinear form of G
pull
(x,t) was necessary in
the derivation because, in contrast to the typically stiff
folded state, the unfolded state may be soft (to be exact,
1/2k
S
x
z2
(F) << k
B
T may not be satised) and thus easily
deformed by the pulling device. Because of this deforma-
tion, the folding transition faces an extra contribution (regu-
lated by the ratio k
S
/k
U
(F)) to the barrier height, typically
negligible for unfolding, that decreases the on-rate in addi-
tion to the applied force F.
The last piece required for Eq. 1, the loading rate
_
FF, is
computed as the time derivative of the force F(t) on the
unfolded molecule at its most probable extension at time t:
_
FF
k
S
V
1 k
S
=k
U
F
:
Finally, we realize that the integral in Eq. 1 can be solved
analytically exactly, both for n 1/2 and n 2/3, resulting
FIGURE 1 Schematic of the output from a force-relaxation
experiment. The applied force is continuously relaxed from the
initial value F
0
until the biomolecule folds or binds, as signied
by a sharp increase in the measured force. From multiple re-
peats of this experiment, distributions of the folding or binding
forces are collected (inset). Fitting the force distributions with
the derived analytical expression yields the key parameters
that determine the kinetics and energetics of folding or binding.
FIGURE 2 Contributions to the free energy prole for folding
(inset) and binding (main gure). The derived expression (Eq.
2) extracts the on-rate and the location and height of the activa-
tion barrier to folding. When applied to binding data, the expres-
sion extracts the parameters of the ligand-tether-receptor (LTR)
potential
~
G
0
(x); the proposed algorithm (Eqs. 3 and 4) removes
the contribution of the tether potential G
teth
(x) to recover the pa-
rameters of the intrinsic ligand-receptor (LR) potential G
0
(x).
in the analytical expression for the distribution of folding
forces:
pF
k
)
F
_
FF
k
)
F
bj
_
FFjx
z
1
k
S
k
U
F
n
n1
1
nFx
z
DG
z
1
1
n
:
(2)
Equation 2 can be readily applied to (normalized) histo-
grams from force-relaxation experiments to extract the pa-
rameters of the intrinsic kinetics and energetics of folding.
Being exact for n 1/2 and n 2/3, Eq. 2 is also an accurate
approximation for any n in the interval 1/2 < n < 2/3 as long
as k
S
(k
U
(F) (see Fig. S1 in the Supporting Material). For
simplicity, in Eq. 2 we have omitted the term containing F
0
as negligible if F
0
is large enough to prevent folding events.
The solution in Eq. 2 reveals properties of the distribution
of folding forces that distinguish it from its unfolding coun-
terpart (7):
1. The distribution has a positive skew (Fig. 3), as intui-
tively expected: the rare folding events occur at high
forces when the barrier is still high.
2. Increasing the relaxation speed shifts the distribution to
lower forces (Fig. 3): faster force relaxation leaves less
time for thermal uctuations to push the system over a
high barrier, causing transitions to occur later (i.e., at
lower forces), when the barrier is lower.
3. The stiffness k
S
and speed V enter Eq. 2 separately,
providing independent routes to control the range of
folding forces and thus enhance the robustness of a t.
The application of the above framework to binding exper-
iments on a ligand and receptor connected by a tether (3) in-
volves an additional stepdecoupling the effect of the
tetherto reconstruct the parameters of ligand-receptor
binding. Indeed, the parameters extracted from a t of
experimental histograms to Eq. 2 characterize the ligand-
tether-receptor (LTR) potential (
~
k
0
, ~x
z
, D
~
G
z
, n) (Fig. 2).
The parameters of the natural ligand-receptor (LR) potential
(k
0
, x
z
, DG
z
) can be recovered using three characteristics of
the tether: contour length L; persistence length p; and exten-
sion D of the tether along the direction of the force in the
LTR transition state. The values of L and p can be deter-
mined from the force-extension curve of the tether (10);
these dene the tether potential G
teth
(x) (Fig. 2). The value
of D can be found from an unbinding experiment (7) on
LTR and the geometry of the tether attachment points (see
Fig. S3). Approximating the region of the LR potential be-
tween the transition and unbound states as harmonic, with
no assumptions about the shape of the potential beyond x
z
,
the ligand-receptor barrier parameters are then
x
z

a 1
a 2
~x
z
; DG
z

a 1
2
2a 2
~x
z
F
teth
D ~x
z
; (3)
and the intrinsic unimolecular association rate is
k
0
z
~
k
0
bDG
z
3
2
bD
~
G
z
1
n
1
2
~x
z
x
z
2
e
bD
~
G
z
DG
z
: (4)
Here, the force value F
teth
D ~x
z
is extracted from the
force-extension curve of the tether at extension D ~x
z
and
a
2
D
~
G
z
G
teth
D G
teth
D ~x
z
~x
z
F
teth
D ~x
z
;
where G
teth
(x) is the wormlike-chain potential (see Eq. S13
in the Supporting Material). Equations 34 conrm that a
tether decreases the height and width of the barrier (see
Fig. 2), thus increasing the on-rate.
In Fig. 3, the developed analytical framework is applied
to folding and binding force histograms from Brownian dy-
namics simulations at parameters similar to those in the
analogous experimental and computational studies (3,5,11)
(for details on simulations and tting procedure, see the
Supporting Material). For the stringency of the test, the sim-
ulations account for the wormlike-chain nature of the mo-
lecular unfolded and LTR unbound states that is not
explicitly accounted for in the theory. With optimized
binning (12) of the histograms and a least-squares t, Eqs.
24 recover the on-rate, the location and the height of the
FIGURE 3 Force histograms from folding (left) and binding
(right) simulations at several values of the force-relaxation
speed (in nanometers per second, indicated at each histogram).
Fitting the histograms with the analytical expression in Eq. 2
(lines) recovers the on-rate and activation barrier for folding or
binding (Table 1).
TABLE 1 On-rate and the location and height of the activation
barrier from the t of simulated data to the theory in Eq. 2
Folding k
0
(s
1
) x
z
(nm) DG
z
(k
B
T) n
True 9.5 10
3
2.2 2.0
Fit 8 52 10
3
2.2 5 0.2 1.8 50.5 0.54
a
Binding (LTR)
~
k
0
(s
1
) ~x
z
(nm) D
~
G
z
(k
B
T) n
True 28 1.56 1.7
Fit 24 5 3 1.57 5 0.09 1.8 50.4 0.53
a
Binding (LR) k
0
(s
1
) x
z
(nm) DG
z
(k
B
T)
True 2.8 3.0 4.0
Fit 2.7 5 0.2 2.9 5 0.1 4.1 50.1
a
Fixed at value that minimized least-squares error.
Biophysical Letters L21
activation barrier, and the value of n that best captures how
the kinetics scale with force (Table 1). The accuracy of the
extracted parameters can be enhanced by ensuring that the
data sets possess:
1. Multiple relaxation speeds,
2. Folding/binding events at low forces, and
3. A large number of events at each speed.
SUPPORTING MATERIAL
Three gures, 29 equations and supplementary information are available at
http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)01071-0.
ACKNOWLEDGMENTS
We thank Robert Best, Brian Camley, Nathan Hudson, Jongseong Kim,
Timothy Springer, Debarati Chatterjee, Ashok Garai, and Yaojun Zhang
for valuable discussions.
This research was supported by National Science Foundation CAREER
grant No. MCB-0845099.
REFERENCES
1. Alberts, B., A. Johnson, ., P. Walter. 2007. Molecular Biology of the
Cell. Garland Science, New York.
2. Fersht, A. R. 1999. Structure and Mechanism in Protein Science. W.H.
Freeman, New York.
3. Kim, J., C. Z. Zhang, ., T. A. Springer. 2010. A mechanically stabi-
lized receptor-ligand ex-bond important in the vasculature. Nature.
466:992995.
4. Gebhardt, J. C. M., T. Bornschlogl, and M. Rief. 2010. Full distance-
resolved folding energy landscape of one single protein molecule.
Proc. Natl. Acad. Sci. USA. 107:20132018.
5. Yu, H., A. N. Gupta, ., M. T. Woodside. 2012. Energy landscape anal-
ysis of native folding of the prion protein yields the diffusion constant,
transition path time, and rates. Proc. Natl. Acad. Sci. USA. 109:14452
14457.
6. Bustamante, C., Y. R. Chemla, ., D. Izhaky. 2004. Mechanical pro-
cesses in biochemistry. Annu. Rev. Biochem. 73:705748.
7. Dudko, O. K., G. Hummer, and A. Szabo. 2006. Intrinsic rates and acti-
vation free energies from single-molecule pulling experiments. Phys.
Rev. Lett. 96:108101.
8. Tshiprut, Z., and M. Urbakh. 2009. Exploring hysteresis and energy
dissipation in single-molecule force spectroscopy. J. Chem. Phys.
130:084703.
9. Kramers, H. A. 1940. Brownian motion in a eld of force and the diffu-
sion model of chemical reactions. Physica. 7:284304.
10. Marko, J. F., and E. D. Siggia. 1995. Stretching DNA. Macromolecules.
28:87598770.
11. Best, R. B., and G. Hummer. 2008. Protein folding kinetics under force
from molecular simulation. J. Am. Chem. Soc. 130:37063707.
12. Doane, D. P. 1976. Aesthetic frequency classication. Am. Stat.
30:181183.
Biophysical Model of Bacterial Cell Interactions with Nanopatterned Cicada
Wing Surfaces
Sergey Pogodin,
Jafar Hasan,
Vladimir A. Baulin,
* Hayden K. Webb,
Vi Khanh Truong,
The Hong Phong Nguyen,
Veselin Boshkovikj,
Christopher J. Fluke,
{
Gregory S. Watson,
k
Jolanta A. Watson,
k
Russell J. Crawford,
and Elena P. Ivanova
Departament dEnginyeria Quimica, Universitat Rovira i Virgili, Tarragona, Spain;

Faculty of Life and Social Sciences and
{
Centre for
Astrophysics and Supercomputing, Swinburne University of Technology, Hawthorn, Victoria, Australia;

Institucio Catalana de Recerca i
Estudis Avancats, Barcelona, Spain; and
k
Centre for Biodiscovery and Molecular Development of Therapeutics, School of Marine and Tropical
Biology, James Cook University, Townsville, Queensland, Australia
ABSTRACT The nanopattern on the surface of Clanger cicada (Psaltoda claripennis) wings represents the rst example of
a new class of biomaterials that can kill bacteria on contact based solely on their physical surface structure. The wings provide
a model for the development of novel functional surfaces that possess an increased resistance to bacterial contamination and
infection. We propose a biophysical model of the interactions between bacterial cells and cicada wing surface structures, and
show that mechanical properties, in particular cell rigidity, are key factors in determining bacterial resistance/sensitivity to the
bactericidal nature of the wing surface. We conrmed this experimentally by decreasing the rigidity of surface-resistant strains
through microwave irradiation of the cells, which renders them susceptible to the wing effects. Our ndings demonstrate the
potential benets of incorporating cicada wing nanopatterns into the design of antibacterial nanomaterials.
INTRODUCTION
Several surfaces exist in nature that are capable of maintain-
ing a contaminant-free status despite the innate abundance of
potential contaminants in their surrounding environments
(15). The vast majority of these surfaces owe their self-
cleaning qualities to their superhydrophobic properties,
which in turn are largely due to their physical surface struc-
ture. Many animals (e.g., sharks (6,7), cicadae (8), butteries
(9), termites (10), mosquitos (11), and geckos (12)) and
plants (e.g., lotus (Nelumbo nucifera) (13,14) and cabbage
(Brassica oleracea) (15)) possess hierarchical surface
features that signicantly increase their hydrophobicity,
often to the point of becoming superhydrophobic (10,16).
A number of research groups have attempted to establish
a direct link between the self-cleaning and antibiofouling
properties of surfaces, i.e., the ability to prevent attachment
and accumulation of biological material (1720). More
recently, we demonstrated that superhydrophobic/self-clean-
ing surfaces are not necessarily inherently antibiofouling in
nature (8). Pseudomonas aeruginosa cells were found to be
capable of adhering relatively effectively onto the surface of
the wings of the Clanger cicada (Psaltoda claripennis);
however, those cells that were able to attach to the surface
were killed with extreme efciency by the wing surface
(8). We further demonstrated that cicada wings were efcient
at killing other Gram-negative bacteria (i.e., Branhamella
catarrhalis, Escherichia coli, and Pseudomonas uores-
cens), whereas Gram-positive bacteria (Bacillus subtilis,
Pseudococcus maritimus, and Staphylococcus aureus) re-
mained resistant (21). This result suggests that a common
mechanism underlies the observed phenomenon. Even
more signicantly, we also demonstrated that the bacteri-
cidal properties of the wings took the form of a mechanical
rupture of the bacteria arising from physical interactions
between the cells and the nanoscale wing surface structure.
To our knowledge, these cicada wings were the rst
described example of a surface that possesses biocidal
activity based solely on its physical surface structure (8).
The antibacterial properties of cicada wings have only
very recently been discovered, and hence there is still
much to be learned about the specic mechanisms that
lead to the observed bactericidal behavior (8,21). It is crit-
ical to obtain a greater fundamental understanding of these
mechanisms before any attempt can be made to apply these
structures in medical contexts. We developed a biophysical
model to provide insight into the interactions that take place
between the bacterial cells and the cicada wing surface
structures. Adsorption of the bacterial cell membrane on
the pattern of the cicada wing surface may lead to a drastic
increase of the total area, accompanied by stretching of the
membrane, which may in turn lead to irreversible membrane
rupture and death of bacteria. Previously, gold coating of
cicada wings was shown to signicantly alter the surface
properties while preserving both the topographical structure
and subsequently the bactericidal effect (8). This observa-
tion led to two research hypotheses that are the focus of
this work: 1), the mechanism is biophysical and no specic
biological interactions play a role; and 2), less rigid bacterial
membranes will be more affected by the bactericidal mech-
anism of the wings.
Submitted August 10, 2012, and accepted for publication December 31,
2012.
*Correspondence: vladimir.baulin@urv.cat or eivanova@swin.edu.au
Editor: Simon Scheuring.
Biophysical Journal Volume 104 February 2013 835840 835
MATERIALS AND METHODS
Sample preparation
Cicada (P. claripennis) specimens were collected from the greater
Brisbane parkland areas. It is known that the cell regions of the dorsal and
ventral sides of the wings possess a homogeneous nanopattern on their
surface (22). For consistency, all experiments were performed on the same
cell regions on the dorsal side of the forewing. Portions of the wings
(~0.5 cm 0.5 cm) were excised by a scalpel or scissors and attached
onto circular coverslips with adhesive tape. The wing samples were
then briey rinsed with MilliQ H
2
O (resistivity of 18.2 MU cm
1
;
Millipore, Billerica, MA) and nally blow-dried with nitrogen gas
(99.99% purity) (23).
Scanning electron microscopy
High-resolution scanning electron microscopy (SEM) images of cicada
wings with adhering bacteria were taken with the use of a eld-emission
scanning electron microscope (Supra 40 VP; Zeiss, Oberkochen, Germany)
at 3 kV under 35,000 and 42,000 magnication. Samples were coated
with thin gold lms using a Dynavac CS300 before they were viewed
with the microscope.
Atomic force microscopy
Atomic force microscopy (AFM) scans were performed with an Innova
microscope (Veeco/Bruker, Santa Barbara, CA) as described elsewhere
(8). Briey, scans were conducted using phosphorus-doped silicon probes
(MPP-31120-10; Veeco/Bruker) with a spring constant of 0.9 N/m, tips
with radius of curvature of 8 nm, and a resonance frequency of ~20 kHz
for surface imaging. Scanning was carried out in tapping mode per-
pendicular to the axis of the cantilever at 1 Hz.
Bacterial strains, growth, and sample preparation
Bacillus subtilis NCIMB 3610T, Planococcus maritimus KMM 3738, and
Staphylococcus aureus CIP 65.8T were used in this study. Bacterial strains
were obtained from the National Collection of Industrial, Food and Marine
Bacteria (NCIMB, Aberdeen, UK), the Collection of Marine Bacteria
(KMM, Russian Federation), and the Culture Collection of the Institute
Pasteur (CIP, France). Before each experiment, bacterial cultures were
refreshed from stocks on nutrient agar (Oxoid, UK) or marine agar (BD).
For cell attachment experiments, fresh bacterial suspensions were prepared
for each strain grown overnight at 37
C in 5 mL of nutrient broth (Oxoid) or

at 25
C in 5 mL of marine broth (Difco) with shaking (120 rpm). Bacterial

cells were collected at the logarithmic stage of growth and the suspensions
were adjusted to OD
600
0.3 as described elsewhere (8). The mounted
insect wings were immersed in 5 mL of the bacterial suspension and
incubated for 18 h.
Confocal laser scanning microscopy
Live and dead bacterial cells were visualized and differentiated using
a FluoView FV10i inverted confocal laser scanning microscopy (CLSM)
system (Olympus, Tokyo, Japan). Cells were stained using the LIVE/
DEAD BacLight Bacterial Viability Kit, L7012, which contains a mixture
of SYTO 9 and propidium iodide uorescent dyes (Molecular Probes/
Invitrogen, NY) according to the manufacturers protocol. SYTO 9
permeates all cells, binding to DNA and causing a green uorescence.
Propidium iodide only enters cells that have signicant membrane damage,
which is an indication of nonviability, and binds to nucleic acids with higher
afnity than SYTO 9.
Microwave experiments
Bacterial samples for microwave (MW) treatment were comprised of 2 mL
of cell suspensions (OD
600
0.1) that were transferred into a micro
Petri dish (35 mm i.d.; Greiner Bio-One, Frickenhausen, Germany). The
MW apparatus was a Lambda Technologies Vari-Wave model LT 1500
with the frequency xed at 18 GHz and other settings as described
elsewhere (24). The bulk temperature of the bacterial suspension during
exposure was controlled to remain below 40
C at all times. Each sample

was exposed to MW radiation for three consecutive exposures of 1 min
each, and the sample was allowed to cool back down to 20
C between expo-
sures. After treatment, the cell suspensions were incubated on insect wings
mounted on circular coverslips in the same manner employed for the
untreated cells.
RESULTS AND DISCUSSION
The surface structure of the wings of P. claripennis has been
extensively characterized by AFM and SEM imaging
techniques and described in earlier reports (8,2123). We
conrmed that the wing surfaces were covered by an array
of nanopillar structures arranged approximately hexago-
nally, spaced 170 nm apart from center to center (Fig. 1).
Each pillar was ~200 nm tall, with a conical shape and
a spherical cap 60 nm in diameter.
Pseudomonas aeruginosa ATCC 9027 bacterial cells in
contact with cicada wings are known to be deformed and
FIGURE 1 Cicada (P. claripennis) wing surface topography. (a)
Scanning electron micrograph of the surface of a cicada wing as viewed
from above (scale bar 200 nm). (b) Three-dimensional representation
of the surface architecture of a cicada wing, constructed from AFM scan
data and colored according to height. A three-dimensional animation of
the cicada wing surface is available at http://youtu.be/JDOEAUdqJGk.
836 Pogodin et al.
mechanically ruptured by the nanopattern on the surface of
the wing (8). Because the characteristic dimensions of the
nanopillars on the surface of the cicada wings (~100 nm)
are an order of magnitude larger than the thickness of the
bacterial membrane (~10 nm) (25), we can model the
membrane as a thin elastic layer and neglect the details of
the structure and composition of the layer. Similarly,
because the typical size of a bacterial cell (i.e., ~500
1000 nm) is at least several times larger than the spacing
between the nanopillars, we can also ignore the curvature
of the bacterial surface in the rst approximation and limit
our consideration to the adsorption of a planar piece of
a membrane onto an array of nanopillars. In our model,
the increase of the total area due to adsorption on the pillars
leads to nonuniform stretching due to a specic surface
pattern, which in turn leads to membrane rupture.
In such a macroscopic description, the bacterial outer layer
is characterized by the stretching modulus (k), the surface
density of the attraction sites on the relaxed layer (n
0
), and
the energy gain per adsorption site (). The microscopic
nature of the attraction forces between the layer and the
nanopillars is concealed into a single parameter , thus
providing a certain degree of universality. The stretching of
the layer due to the adsorption is described by the local rela-
tive stretching degree a(r) at point r. We assume that the
unperturbed membrane is characterized by the total area S
i
,
the initial stretching a
i
and initial uniform density n
0
of the
adsorption sites. The stretching due to contact with pillars
on the surface of the cicada wings leads to the redistribution
of the adsorption sites from n
0
to the local density n(r) n
0
/
(1 a(r)). Each site that is adsorbed on the nanopillar
surface contributes the energy gain ; therefore, the total
free energy gain due to the adsorption is given by
F
gain
Z
A
nrds
Z
A
n
0
ds
1 ar
; (1)
where ds is an element of the layer surface area, and the
integration is performed over the total contact area between
the layer and the nanopillar surface (A).
The energy gain due to adsorption on the nanopillars is
balanced by the free-energy loss associated with deforma-
tion of the membrane. The main contribution to the energy
loss, F
loss
, is due to local membrane stretching/compression,
which is proportional to (k/2)a
2
(r) for weak local deforma-
tions, ja(r)j 1. Thus, the integration over the total
adsorbed area of the layer (A) plus the total area of the layer
suspended between the nanopillars (B) gives
F
loss
Z
AB
k
2
a
2
r
nr
n
0
ds
Z
AB
k
2
a
2
r
ds
1 ar
: (2)
The local stretching a(r) is not a completely independent
variable. It relates the unperturbed area before adsorption
and the area stretched due to adsorption through the
following geometrical condition: the projection of unper-
turbed and stretched areas on the surface plane remains
constant. This condition can be taken into account in the
total free energy with the help of the Lagrange multiplier l:
F F
gain
F
loss
lk
0
@
Z
AB
ds
1 ar
S
0
1
A
; (3)
where S
0
is the initial area prior to contact with pillars. Mini-
mization of this expression with respect to a(r) yields the
local stretching of the layer in region A, where the
membrane interacts with the nanopillars, and region B,
where the membrane is suspended between pillars, leading
to the following condition:
1 ar
(
1 a
A

1 2l z
p
; region A
1 a
B

1 2l
p
; region B
(4)
where the dimensionless effective interaction parameter
z h n
0
/k is dened as the ratio between the attraction
of the layer to the nanopillar surface and the layer elasticity.
Equation 4 leads to an important general conclusion. In the
case in which adsorption is negative and hence z is posi-
tive, the stretching of the suspended region of the layer,
a
B
, is higher than the stretching of the adsorbed region of
the layer, a
A
. This means that the rupture point of the layer
will always be reached rst in region B. In other words, the
nanopillars do not pierce the membrane, but rather break the
membrane between the nanopillars. One might imagine
a scenario in which the nanopillars pierce the layer like an
array of needles; however, this would only be the case if
the diameter of the spherical caps were much smaller than
is actually the case, e.g., ~1 nm as opposed to the measured
60 nm.
Because a
A
and a
B
are constant for all points inside
regions A and B, respectively, we can simplify Eq. 3 by
converting the integrals into areas S
A
and S
B
of the corre-
sponding regions of the layer:
F
n
0
S
A
1 a
A
k
2
a
2
A
S
A
1 a
A
a
2
B
S
B
1 a
B
lk
S
A
1 a
A
S
B
1 a
B
S
i
1 a
i
:
(5)
Here we consider that the unperturbed membrane is
stretched up to the initial stretching degree, a
i
, and the total
initial area of unperturbed membrane is S
i
.
The geometry of the nanopattern on the surface of cicada
wings is described by four parameters (Fig. 2 a): the radius
of the cap on top of the pillar (R 30 nm), the pillar height
(h 200 nm), the pillar pitch (b 10
), and the average

distance between the pillars (d 170 nm). Assuming that
Model of Bacterial Cell Interaction with Cicada Wings 837
the membrane suspended between the pillars (region B)
remains horizontal with respect to the plane of the wing,
the membrane position can be characterized by a single
parameter: the vertical distance x from region B to the tip
of the nanopillars. It is also convenient to consider sepa-
rately two cases, in which region B is above (case I) and
below (case II) the junction point M between the spherical
cap and conical column of the nanopillar (Fig. 2 a). In
case I, it is more convenient to describe the position of the
layer by the angle q between the nanopillar vertical axis
and the contact point between the nanopillar and region B
of the layer. In case II, the most convenient parameter is
the vertical distance z between region B and the junction
point M. Assuming that the average initial area of the layer
per nanopillar is S
i
d
2
, the areas S
A
and S
B
are given by the
following expressions:
S
A
2pR
2
1 cos q
S
B
d
2
pR
2
sin
2
q
case I
S
A
2pR
2
1sin b
2pz
cos b
R cos b
z
2
tan b
S
B
d
2
pR cos b z tan b
2
9
>
=
>
;
case II
(6)
These expressions allow for numerical minimization of the
free energy, which gives the equilibrium position and the
equilibrium stretching of the membrane. Fig. 3 a shows
the calculated dependencies of the membrane stretchings,
a
A
(region A) and a
B
(region B), on the effective interaction
parameter, z, for different values of the initial degree of
stretching, a
i
. It was found that a
B
increases continuously
as z increases. This suggests that there is a critical value,
z
critical
, of the layer parameter, z, at which a
B
also reaches
a critical value and the membrane is ruptured.
The model suggests that the bactericidal mechanism is
biophysical and does not imply directly any specic biolog-
ical interactions. This is consistent with a previous experi-
ment in which cicada wing surfaces were coated with gold
(8). This technique preserves the geometry of the wings
but changes the surface properties. The result demonstrates
that such a pattern is lethal for P. aeruginosa cells, despite
the substantial difference in surface chemistry. To explore
the predictions of the proposed model and check the univer-
sality of the mechanism, we investigated the attachment
behavior of two species of Gram-positive cocci, Plano-
coccus maritimus and S. aureus, and the Gram-positive,
rod-shaped bacterium Bacillus subtilis on cicada wing
FIGURE 2 Biophysical model of the interactions between cicada
(P. claripennis) wing nanopillars and bacterial cells. (a) Schematic of
a bacterial outer layer adsorbing onto cicada wing nanopillars. The
adsorbed layer can be divided into two regions: region A (in contact with
the pillars) and region B (suspended between the pillars). Because region
A adsorbs and the surface area of the region (S
A
) increases, region B is
stretched and eventually ruptures. (be) Three-dimensional representation
of the modeled interactions between a rod-shaped cell and the wing surface.
As the cell comes into contact (b) and adsorbs onto the nanopillars (c), the
outer layer begins to rupture in the regions between the pillars (d) and
collapses onto the surface (e). Images be are screenshots from an anima-
tion of the mechanism available at http://youtu.be/KSdMYX4gqp8.
838 Pogodin et al.
surfaces (21). It is well documented that Gram-positive
bacteria are generally more rigid than their rod-shaped
counterparts, mostly due to the larger proportion of peptido-
glycan present in the cell wall (2527). Therefore, we per-
formed comparative attachment experiments to determine
whether Gram-positive cells respond in a similar manner
to the Gram-negative Pseudomonas aeruginosa. The results
of this experiment revealed that all three species were unaf-
fected by the nanopillar structures on the wing surface.
Scanning electron micrographs showed clearly that the cells
retained their characteristic morphologies, and CLSM
conrmed that the cells remained viable. According to the
model, the effective interaction parameter z is proportional
to the attraction between the bacterial layer and the wing
surface, and is inversely proportional to the layer rigidity.
Thus, more rigid cells require a stronger interaction with
the surface to sufciently stretch to the point of rupture.
This offers a possible explanation for the resistance of
B. subtilis, Planococcus maritimus, and S. aureus against
the action of the cicada wings, in that they possess increased
rigidity relative to Pseudomonas aeruginosa (2527).
If the presented model stands, one would expect that if the
cell rigidity is decreased and/or the initial stretching on the
membrane is sufciently decreased, the bacterial strains that
were previously resistant to the bactericidal action of the
wing surface could potentially become susceptible. MW
exposures under specied conditions were previously shown
to induce reversible poration in the membranes of bacteria
(28,29), allowing the release of some of the cellular
contents, decreasing internal turgor pressure, and releasing
some of the tension on the membrane. However, this tech-
nique itself is not lethal to the cells, and the pores in the
membranes self-seal after a few minutes. To test our theory,
we exposed cells of B. subtilis, Planococcus maritimus, and
S. aureus to MW radiation and then incubated them in the
presence of cicada wings. The morphology of irradiated
cells that came into contact with the wing surface was mark-
edly different from that of the nonirradiated cells (Fig. 4,
left). The MW-treated cells were considerably deformed
by the nanopillars, in a manner similar to that observed
for the untreated Pseudomonas aeruginosa (8), conrming
that the decrease in turgor pressure induced by the MW
treatment had rendered these cells susceptible to the
FIGURE 3 Modeled stretching dynamics of the outer layer of a bacterial
cell in contact with a cicada wing surface. (a) Stretching in region A (a
A
,
dashed lines) and region B (a
B
, solid lines) is plotted as a function of the
layer parameter z for layers under different degrees of initial stretching
(a
i
), denoted by color. (b) Stretching in a
A
and a
B
is plotted as a function
of the position of the layer relative to junction point M between the spher-
ical cap and conical base of the nanopillars. Both a
A
and a
B
are plotted for
different combinations of z and a
i
. The equilibrium position of the layer in
each case is marked with a dot.
FIGURE 4 Cell interactions of surface-resistant B. subtilis NCIMB
3610T, Planococcus maritimus KMM 3738, and S. aureus CIP 65.8T
strains after MW irradiation. All three strains were rendered susceptible
to the action of the wing surface by MW treatment. Typical scanning
electron micrographs (left) show substantial deformation of the cell
morphologies of all three species. A CLSM viability analysis (right) shows
that all cells were inactivated (shown in red).
Model of Bacterial Cell Interaction with Cicada Wings 839
bactericidal action of the wing surface. Subsequent viability
experiments conrmed that the cells were indeed inactivated
(Fig. 4, right). This is compelling evidence in support of the
proposed model, conrming that the primary factors that
determine the vulnerability of bacteria to the action of the
wing surface are the mechanical properties of the
membranes (i.e., the rigidity and initial stretching).
CONCLUSIONS
We developed a biophysical model of the interaction of
bacterial cells with superhydrophobic nanopillar structures
on the surface of cicada wings to provide a fundamental
understanding of the mechanisms behind the recently
discovered phenomenon of the bactericidal action of cicada
wings. As the bacterial cells adsorb onto the nanopillar
structures present on the wing surfaces, the cell membrane
stretches in the regions suspended between the pillars. If
the degree of stretching is sufcient, this will lead to cell
rupture. Due to their greater rigidity, Gram-positive cells
have a greater natural resistance to this effect than do
Gram-negative cells. However, by decreasing their internal
turgor pressure and hence their initial stretching and
(to some degree) rigidity through MW irradiation, one can
render these cells sensitive to the bactericidal mechanisms
of the wing surfaces. Designing bio/nanomaterials that
possess cicada-wing-like structures may be a promising
avenue of research for applications in which minimizing
bacterial contamination/infection is desirable.
This research was funded in part by the Advanced Manufacturing Co-
operative Research Centre.
REFERENCES
1. Marmur, A. 2004. The Lotus effect: superhydrophobicity and metasta-
bility. Langmuir. 20:35173519.
2. Su, Y., B. Ji, ., K. C. Hwang. 2010. Natures design of hierarchical
superhydrophobic surfaces of a water strider for low adhesion and
low-energy dissipation. Langmuir. 26:1892618937.
3. Bhushan, B., and Y. C. Jung. 2011. Natural and biomimetic articial
surfaces for superhydrophobicity, self-cleaning, low adhesion, and
drag reduction. Prog. Mater. Sci. 56:1108.
4. Guo, Z., W. Liu, and B.-L. Su. 2011. Superhydrophobic surfaces:
from natural to biomimetic to functional. J. Colloid Interface Sci.
353:335355.
5. Webb, H. K., J. Hasan, ., E. P. Ivanova. 2011. Nature inspired
structured surfaces for biomedical applications. Curr. Med. Chem.
18:33673375.
6. Bechert, D. W., M. Bruse, and W. Hage. 2000. Experiments with three-
dimensional riblets as an idealized model of shark skin. Exp. Fluids.
28:403412.
7. Chung, K. K., J. F. Schumacher, ., A. B. Brennan. 2007. Impact of
engineered surface microtopography on biolm formation of Staphylo-
coccus aureus. Biointerphases. 2:8994.
8. Ivanova, E. P., J. Hasan, ., R. J. Crawford. 2012. Natural bactericidal
surfaces: mechanical rupture of Pseudomonas aeruginosa cells by
cicada wings. Small. 8:24892494.
9. Fang, Y., G. Sun, ., L. Ren. 2007. Hydrophobicity mechanism of
non-smooth pattern on surface of buttery wing. Chin. Sci. Bull. 52:
711719.
10. Watson, G. S., B. W. Cribb, and J. A. Watson. 2010. How micro/nano-
architecture facilitates anti-wetting: an elegant hierarchical design on
the termite wing. ACS Nano. 4:129136.
11. Gao, X., X. Yan, ., L. Jiang. 2007. The dry-style antifogging proper-
ties of mosquito compound eyes and articial analogues prepared by
soft lithography. Adv. Mater. 19:22132217.
12. Mahdavi, A., L. Ferreira, ., J. M. Karp. 2008. A biodegradable and
biocompatible gecko-inspired tissue adhesive. Proc. Natl. Acad. Sci.
USA. 105:23072312.
13. Barthlott, W., and C. Neinhuis. 1997. Purity of the sacred lotus, or
escape from contamination in biological surfaces. Planta. 202:18.
14. Bhushan, B., Y. C. Jung, ., K. Koch. 2009. Lotus-like biomimetic
hierarchical structures developed by the self-assembly of tubular plant
waxes. Langmuir. 25:16591666.
15. Koch, K., and W. Barthlott. 2009. Superhydrophobic and superhydro-
philic plant surfaces: an inspiration for biomimetic materials. Philos.
Transact. A Math. Phys. Eng. Sci. 367:14871509.
16. Koch, K., B. Bhushan, ., W. Barthlott. 2009. Fabrication of articial
lotus leaves and signicance of hierarchical structure for superhydro-
phobicity and low adhesion. Soft Matter. 5:13861393.
17. Erbil, H. Y., A. L. Demirel, ., O. Mert. 2003. Transformation
of a simple plastic into a superhydrophobic surface. Science. 299:
13771380.
18. Alves, N. M., J. Shi, ., J. F. Mano. 2009. Bioinspired superhydropho-
bic poly(L-lactic acid) surfaces control bone marrow derived cells
adhesion and proliferation. J. Biomed. Mater. Res. A. 91:480488.
19. Nosonovsky, M., V. Hejazi, ., P. K. Rohatgi. 2011. Metal
matrix composites for sustainable lotus-effect surfaces. Langmuir.
27:1441914424.
20. Kirschner, C. M., and A. B. Brennan. 2012. Bio-inspired antifouling
strategies. Ann. Rev. Mater. Res. 42:8.18.19.
21. Hasan, J., H. K. Webb, ., E. P. Ivanova. 2012. Selective bactericidal
activity of nano-patterned superhydrophobic cicada Psaltoda claripen-
nis wing surfaces. Appl. Microbiol. Biotechnol. Dec 19. [Epub ahead
of print].
22. Hong, S.-H., J. Hwang, and H. Lee. 2009. Replication of cicada
wings nano-patterns by hot embossing and UV nanoimprinting.
Nanotechnology. 20:385303.
23. Zhang, G., J. Zhang, ., H. Shao. 2006. Cicada wings: a stamp from
nature for nanoimprint lithography. Small. 2:14401443.
24. Shamis, Y., A. Taube, Y. Shramkov, N. Mitik-Dineva, B. Vu, and E. P.
Ivanova. 2008. Development of a microwave treatment technique for
bacterial decontamination of raw meat. Int. J. Food Eng. 4:8.
25. Yao, X., J. Walter, ., T. J. Beveridge. 2002. Atomic force microscopy
and theoretical considerations of surface properties and turgor pres-
sures of bacteria. Colloids Surf. B Biointerfaces. 23:213230.
26. Whatmore, A. M., and R. H. Reed. 1990. Determination of turgor pres-
sure in Bacillus subtilis: a possible role for K in turgor regulation.
J. Gen. Microbiol. 136:25212526.
27. Arnoldi, M., M. Fritz, ., A. Boulbitch. 2000. Bacterial turgor pressure
can be measured by atomic force microscopy. Phys. Rev. E Stat. Phys.
Plasmas Fluids Relat. Interdiscip. Topics. 62(1 Pt B):10341044.
28. Shamis, Y., A. Taube, ., E. P. Ivanova. 2011. Specic electromagnetic
effects of microwave radiation on Escherichia coli. Appl. Environ.
Microbiol. 77:30173022.
29. Shamis, Y., S. Patel, ., E. P. Ivanova. 2009. A new sterilization tech-
nique of bovine pericardial biomaterial using microwave radiation.
Tissue Eng. Part C Methods. 15:445454.
840 Pogodin et al.
Tilting and Wobble of Myosin V by High-Speed Single-Molecule Polarized
Fluorescence Microscopy
John F. Beausang,
{
Deborah Y. Shroder,
Philip C. Nelson,
and Yale E. Goldman

{
*
Department of Physics and Astronomy,

Pennsylvania Muscle Institute,

Graduate Group in Biochemistry and Molecular Biophysics, and
{
Department of Physiology, University of Pennsylvania, Philadelphia, Pennsylvania
ABSTRACT Myosin V is biomolecular motor with two actin-binding domains (heads) that take multiple steps along actin by
a hand-over-hand mechanism. We used high-speed polarized total internal reection uorescence (polTIRF) microscopy to
study the structural dynamics of single myosin V molecules that had been labeled with bifunctional rhodamine linked to one
of the calmodulins along the lever arm. With the use of time-correlated single-photon counting technology, the temporal reso-
lution of the polTIRF microscope was improved ~50-fold relative to earlier studies, and a maximum-likelihood, multitrace
change-point algorithm was used to objectively determine the times when structural changes occurred. Short-lived substeps
that displayed an abrupt increase in rotational mobility were detected during stepping, likely corresponding to random thermal
uctuations of the stepping head while it searched for its next actin-binding site. Thus, myosin V harnesses its uctuating envi-
ronment to extend its reach. Additional, less frequent angle changes, probably not directly associated with steps, were detected
in both leading and trailing heads. The high-speed polTIRF method and change-point analysis may be applicable to single-mole-
cule studies of other biological systems.
INTRODUCTION
Myosin V is a molecular motor that translocates along actin
laments in many cell types, transporting cargos toward the
barbed end of actin or tethering them at their destinations
(110). Defects in myosin V are associated with human
pigmentation, immunological, and neurological disorders
(1). Myosin V consists of two N-terminal globular ATP-
and actin-binding domains (heads), a lever arm comprising
a-helical light-chain-binding motifs and six tightly bound
calmodulins (CaMs) or CaM-like light chains per head,
a coiled-coil dimerization domain, and nally a C-terminal
cargo-binding tail that is usually absent in studies using re-
combinant proteins (2).
In eukaryotic cells, single molecules or very small groups
of myosin V molecules can transport cargos by producing
~50 mechanical steps without dissociating upon each diffu-
sional encounter with actin, a property termed processivity
(4,6,8). Many ensemble and single-molecule techniques
have been applied toward determining the chemomechani-
cal cycle of myosin V (48,1013), leading to a consensus
view of the mechanism in which a state with both heads
bound to actin, at a 36 nm separation, occupies much of
the ATPase cycle. The biochemical and structural states of
the two heads are coordinated, probably via intermolecular
strain, which minimizes backward stepping and dissociation
of the entire molecule (2). A 36 nm step is taken when the
trailing head dissociates, swings 72 nm forward (toward
the barbed end), and becomes the leading head. A rapid
forward-directed working stroke rotates the lever arm of
the attached head, accounting for most (2024 nm) of the
step (7). The remaining distance is covered by thermal uc-
tuations of the free head until it binds to an actin subunit
~36 nm ahead of the bound head. Evidence for this postu-
lated thermal-search period was obtained by high-speed
atomic force microscopy (AFM) (10) and by tracking gold
particles (14), microtubules (15), or actin (16) bound to
the lever arm. All of these experiments required investiga-
tors to attach a relatively large probe to the lever arm/or
slow down its motion by a nearby adherent surface.
Smaller motions and rotations of the lever arm have been
associated with ADP release from the trailing head, possibly
accounting for ~5 nm of stepping (7). Additional small steps
and lever arm rotations were detected when the leading-
head lever arm stroked before (17) or after (18) the trailing
head detached, and also apparently were independent of
productive stepping (10,19). The notion that the orientation
of the lever arm can change prior to stepping is supported by
cryoelectron microscopy (20,21), whereas combined single-
molecule uorescence localization and polarization experi-
ments support a straight-legged model for stepping (19).
The roles of these substeps (if any) in facilitating productive
translocation or in coordinating the two heads are not clear.
Single-molecule uorescence techniques (8,2224) are
powerful tools for detecting the position and orientation
changes of macromolecules with the use of relatively small
organic uorophore or quantum dot reporter probes. Polar-
ized total internal reection uorescence (polTIRF) micros-
copy measures the orientation and rotational motions of
dipolar uorophores by exploiting their preference for
absorbing light and emitting uorescence polarized along
their dipole axis. When the uorophore is bound to the
macromolecule in a known position and direction, as is
Submitted September 13, 2012, and accepted for publication January 28,
2013.
*Correspondence: goldmany@mail.med.upenn.edu
Editor: Christopher Berger.
Biophysical Journal Volume 104 March 2013 12631273 1263
the case with bifunctional rhodamine (BR) linked to a pair
of nearby cysteine residues engineered into CaM (25),
molecular rotations can be inferred from the probe angles
changes (12,23,26,27).
In addition to the three-dimensional (3D) orientation of
the uorophore and its linked protein domain, one can deter-
mine rotational mobility, including microsecond wobbling
motions that are not generally accessible in ensemble uo-
rescence anisotropy experiments (22,28,29). Microsecond-
timescale wobbling of a protein domain might be function-
ally important, such as in detecting the thermal search
period during myosin V stepping. Until now, the capability
of polTIRF to measure wobble has not been utilized for
dynamic measurements, except to verify stable probe
attachment. This is partly because the observation time
required to cycle through the range of input polarizations
that are necessary to measure the 3D orientation and wobble
(4080 ms) is longer than the expected lifetime of many
structural states, including myosin Vs postulated thermal
search.
In this work, we studied recombinant myosin V with BR
linked to one of the lever-arm CaM subunits to determine,
at high time resolution, the events that occurred while it
was taking translocation steps. We improved the time reso-
lution of the polTIRF microscope ~50-fold by incorpo-
rating a single-photon counting device and switching
the polarization of the uorescence excitation between
the requisite directions every 100 ms instead of every
510 ms as in most of our previous studies (12,23,26
28,30). Even when the 532 nm laser excitation was
made stronger, the signals were noisy due to the limited
number of photons collected during each 100 ms polariza-
tion time window. To help interpret the signals, we used
a maximum-likelihood, multitrace change-point algorithm
(31) that objectively identies changes in intensity (change
points), indicating structural changes with single-photon
precision and no binning of the raw data. We determined
the orientation of the BR and the extent of wobble during
the intervals between intensity changes, and found that the
changes often corresponded to structural motions of the
molecule. Short-lived substeps that displayed abrupt
increases in probe wobble were detected with the pattern
expected for hand-over-hand stepping, likely correspond-
ing to the random thermal uctuations of the detached
head during its search for the next actin-binding site. In
molecules showing these angular uctuations, they typi-
cally occurred when the trailing head stepped, as expected.
Additional, less frequent angle changes of the lever arm
were detected prior to stepping in both leading and
trailing positions. To our knowledge, this work represents
the rst use of single-molecule polTIRF microscopy
to report microsecond-timescale motions relevant to the
function of a biophysical macromolecule. This method
may be applicable to single-molecule studies of other bio-
logical systems.
Processive motility assay
Recombinant chicken myosin V with its full-length lever arm (amino acids
11099) and a FLAG afnity tag at its C-terminus was coexpressed with
CaM in SF9 cells (32). Chicken CaM with residues Pro-66 and Ala-73
mutated to cysteine was expressed in Escherichia coli (33), puried, and
labeled with BR (29,34) (a generous gift from Dr. J.E.T. Corrie). Myosin
V was labeled by exchanging endogenous wild-type chicken CaM
(WT-CaM) with exogenous BR-CaM at low stoichiometry (12,29). G-actin
was obtained from rabbit skeletal muscle (35). Alexa 647-labeled F-actin
was prepared from G-actin and Alexa-647 actin (Molecular Probes, Carls-
bad, CA) at a 5:1 ratio and 1 mM total actin subunit concentration and stabi-
lized with 1.1 mM phalloidin (Molecular Probes). In some experiments,
0.05 mM biotin-actin (Cytoskeleton, Denver, CO) was also incorporated
during polymerization.
PolTIRF experiments were performed as described previously (29,36).
Briey, all reagents were made in myosin buffer (M5; pH 7.6) containing
25 mM KCl, 20 mM HEPES, 5 mM MgCl
2
, and 1 mM EGTA, except for
the motility buffer (M5
) used for single-molecule myosin V motility

assays, which consisted of M5 buffer with 140 mM ATP, 100 mM dithio-
threitol (DTT), 100 mg/ml WT-CaM, and ~100 pM of BR-CaM-labeled
myosin V. For some experiments, 2,3-butanedione monoxime (BDM;
B0753; Sigma, St. Louis, MO), prepared fresh from powder, was included
in the nal motility buffer (nal concentration 50100 mM) and the pH was
readjusted as needed with KOH. Sometimes the M5
buffer included
10 mM phosphocreatine (P-7936; Sigma) and 0.3 mg/ml creatine phos-
phokinase (prepared daily from powder; C3755; Sigma), but no inuence
on myosin velocity or stepping kinetics was detected. To avoid BR
blinking, 100 mMDTTwas used as a reducing agent with no deoxygenating
system (23).
Myosin V processivity experiments were performed in a ow cell con-
sisting of a plasma-cleaned quartz slide (high purity fused silica grade
quartz, 212000-001; Quartz Scientic, Fairport Harbor, OH), spin coated
with 2 mg/ml poly(methyl methacrylate) (secondary standard grade,
37-003-7; Sigma-Aldrich, St. Louis, MO), and a glass coverslip (No. 1;
Fisher Scientic, Hampton, NH) held together by pieces of double-sided
tape (Cat. No. 665; Scotch). In most experiments, 60 nM N-ethylmaleimide
(NEM)-treated myosin II (37) was owed into the lane and incubated for
5 min, followed by a rapid ow of a 200 nM Alexa 647-actin laments.
The surface was passivated by a 5 min incubation with 1 mg/ml unlabeled
bovine serum albumin (BSA; A0281; Sigma Aldrich). Finally, the motility
buffer containing the myosin and ATP was owed into the chamber. Slight
variations in the protocol were also implemented, including substituting the
NEM myosin and Alexa-647-labeled actin with 1 mg/ml biotinylated
BSA (A8549; Sigma Aldrich), and incubation with 0.5 mg/ml streptavidin
(S-4762; Sigma Aldrich) and biotin-Alexa 647-actin laments, but no
signicant differences were detected.
High-speed polTIRF setup
We improved the time resolution of the polTIRF microscope (28,29,36)
by decreasing the time each laser polarization illuminated the sample
from the earlier 5 ms periods to 0.1 ms, and by storing the arrival time
and the polarization state of the illumination for each detected photon.
Two alternating beams from a 532 nm Nd:YAG laser (Fig. S1 in the Sup-
porting Material) were projected through a coupling prism at a glancing
incident angle, producing an evanescent eld at the quartz slide/aqueous
interface. Switching between the two beams (termed path 1 and path 2)
was achieved by computer-controlled voltages applied to a Pockels cell
(PC0) and a polarizing beam-splitting (PBS0) cube. Similarly, the polariza-
tion of the laser in each beam was switched between four different linear
polarizations (s, 90
; p, 0
; L, 45
; and R, 45
relative to the plane

dened by the incident and reected beams; Fig. S1) with an additional
1264 Beausang et al.
Pockels cell and a Berek compensator in each beam (PC1/BC1 and PC2/
BC2). Each polarization illuminated the sample for 0.1 ms in the sequence
s1, p1, p2, s2, R1, L1, L2, and R2 (letter indicates incident polarization,
number indicates path).
Fluorescence emission was collected by a Leica (Heerbrugg,
Switzerland) 100 1.2 NA water immersion lens, passed through a long-
pass blocking lter, and either imaged onto an intensied charge-coupled
device (CCD) camera (Cascade II; Photometrics, Tucson, AZ) or directed
through a polarizing beam splitter (PBS1) that resolved its x and y compo-
nents between two avalanche photodiodes (APDx and APDy). A time-
correlated single-photon counting PC adaptor board (TCSPC, SPC-130;
Becker and Hickl, Berlin, Germany) operating in FIFO mode, and a pulse
router (HRT-82; Becker and Hickl) were modied to record the excitation
polarization state according to a suggestion by Dr. Wolfgang Becker of
Becker and Hickl (details available upon request). Photon pulses during
the rst 3 ms of each polarization interval, while the Pockel cell voltage
was settling, were ignored. A 10 MHz pulse train from a digital delay
generator (DG645; Stanford Research Systems, Sunnyvale, CA) triggered
the timing of the TCSPC circuit and a 10 kHz pulse train, generated by
same DG645 unit, triggered digital counters that sequenced the high-
voltage ampliers driving the three Pockels cells. For visualization
purposes, single-photon counts from each of the 16 combinations of exci-
tation path and polarization and emission polarization were converted to
and plotted as polarized uorescence intensities (PFIs) by counting the
number of photons detected during successive constant time bins (here
0.1 ms). An analytical model of the expected PFIs was tted by maximum
likelihood to the 16 PFIs, or averages of the 16 PFIs over intervals between
sudden change points (see below) to determine the orientation of the probe
absorption and emission dipole (see Fig. 3 B, inset) (26,28,29); q is the axial
probe angle relative to the microscope optical axis, and f is its azimuth
around the optical axis. Using the orientation of the actin, known from
the initial uorescence video snapshot, q and f are converted to b and a,
the axial and azimuthal probe angles relative to the actin (26,28,29). The
probe wobbles due to protein motions and to faster local motions relative
to the CaM. On timescales slower than the uorescence lifetime (4 ns)
but faster than the measurement interval (0.1 ms), the extent of probe
motion is described as a cone of half-angle d that is centered on the orien-
tation (q, f). A control experiment in which the microscope stage was
moved stepwise successively in 100 nm increments, and with individual
BR-CaMs rmly adhered to the slide, showed that the measured probe
angles and wobble were not deected as the slide moved across the stage
area (~1.5 mm 1.5 mm) from which uorescence was captured by the
photodiodes.
Immediately before each polarization recording, two images of the eld
of candidate uorophores were recorded with the CCD camera, superim-
posed on an image of the Alexa-647-labeled actin laments and displayed
on a monitor. These images were used to estimate the position and direction
of the BR-CaM-labeled myosin V. A moving molecule was selected for
polarization analysis and identied by a mouse click. A custom-built Lab-
View program centered the molecule above the objective via a piezoelectric
stage, inserted a removable mirror that directed the uorescence emission
onto the APDs, and sent a trigger signal that reset the Pockels cell drivers
to the initial polarization setting and started the TCSPC. Photon counts
were recorded for 5 s, corresponding to 6250 cycles of the 16 polarization
channels. During recording of polarized uorescence, spatial information
from the uorophore was not available.
Multiple-channel change-point algorithm
In a previous study (31) we developed a multiple-channel change-point
(MCCP) algorithm tailored to single-photon polTIRF experiments. In this
algorithm, a likelihood function is used to compare two hypotheses: 1),
the hypothesis that there is a single, abrupt intensity change at the i
th
photon
in an interval containing N photons in time T; and 2), the null hypothesis
that there is no intensity change at the i
th
photon. Assuming that the uores-
cence emission is random, the only user-dened parameter is the proba-
bility of a false-positive change point, here chosen to be 5%, which
denes a threshold that the likelihood function must exceed for signi-
cance. The likelihood function, L
i
, is calculated for each photon as follows:
L
i

_
16
j 1
_
n
i;j
ln
_
n
i;j
N
j
_
_
N
j
n
i;j
_
ln
_
1 n
i;j
N
j
__
_
_
i ln
_
t
i
T
_
N iln
_
1 t
i
T
__
where n
i,j
is an N 16 dimensional matrix of the accumulated number of
photons i for each of the j PFIs (Fig. S7), t
i
indicates the arrival time of the
i
th
photon, and N
j
is the total number of photons in the j
th
PFI. A nonuniform
distribution of false positives across the interval is mitigated by applying
correction terms to the likelihood function and also including a small exclu-
sion region that prevents change points from being detected in the rst or
last 2.5% of the photons in the interval (31,38).
The algorithm is applied iteratively to the data until no additional change
points are detected. The likelihood surfaces for the change points at 0.46 s
and 0.49 s (points 5 and 6 correspond to the magenta and red lines in Fig. S3
C) exceed the threshold (black line) for 95% condence by over 10 log units
and correspond to abrupt changes in the PFIs (four of 16 are shown in
Fig. S3 A) and the kink in the accumulated photon trace (Fig. S3 B). The
likelihood surface of a third change point at ~0.4 s (point 4, blue line) is
broader and less pronounced with only a small change in the PFIs, but it
still exceeds the threshold for signicance by ~8 log units. The peaks of
the likelihood functions determine the most likely location of the change
points (vertical black dashed lines), and the relative sharpness of the likeli-
hood peaks determines the 95% condence interval (CI; gray-shaded
regions), which for change points 4, 5, and 6 are 1000, 226, and 248 photons
or 15, 3.6, and 4.5 ms, respectively.
Ambiguities in the dipole model for orientation
and wobble
The 16 combinations of eight time-multiplexed input polarizations and two
simultaneously measured emission polarizations used here, including
545
linear polarized excitations, enable unambiguous determination of

the probe angle within a hemisphere of orientation (26,29,39). In addition
to allowing a greater range of angles to be discerned, the 545
laser polar-
izations also resolve a more subtle ambiguity associated with d. When the
probe is completely free to rotate on the microsecond timescale, d should
be 90
(q and f are not dened at high d); however, when only s and p
(i.e., 0
and 90
) input laser polarizations are used (12) and the probe has
high wobble, an orientation in the laboratory frame of reference of
(q, f) (54.7
, 45
) and d z 50
is usually reported by the dipole

model because the corresponding eight PFIs are the same as in the case
when d 90
. In previous work, this artifact was not noticed because

periods of large probe wobble were not explicitly investigated.
Choice of hemisphere
For any given measurement of dipole orientation, either one of two vectors
((q, f) or (180
q, f 5180
)) may apply due to the twofold symmetry of

the probe dipole. The BR probe bound to the protein is not symmetrical,
however, due to its specic attachments to the two different Cys residues,
so it is meaningful to consider which of the dipole ends is being reported
by the angle and, importantly, to maintain the same end of the probe in
successive measurements on a given molecule. If the true rotational motions
of the molecule are smaller than 180
and within 90
of an average angle,
then all of the orientations for one end of the dipole will fall into one
hemisphere. A hemisphere that fullls this requirement is centered on the
Tilting and Wobble of Myosin V Lever Arm 1265
director axis determined fromthe measured orientations during a processive
run (see Supporting Material). Which end of the director axis to align with
the analysis hemisphere is still to be chosen, but for relative motions during
a run, it is irrelevant. The opposite hemisphere corresponds to the molecule
walking on the opposite side of the actin lament (27), an experimental
detail that is not known for any individual molecule. The end closest to
the initial direction of motion of the molecule, as measured from the
CCD images recorded prior to the polarization analysis, is chosen as the
hemisphere director. When (q, f) is rotated into the actin frame of reference
(b, a), the hemisphere does not change.
RESULTS
Rotational dynamics using high-speed polTIRF
microscopy and change-point analysis
We improved the maximum time resolution of standard
polarized uorescence microscopy (5 ms recording gates,
2040 ms cycle time (28,29)) by switching the excitation
laser polarizations every 0.1 ms and cataloging the arrival
time of each detected photon (Fig. S1). To obtain 2050
photons within the 0.1 ms gate time, we increased the laser
power from our standard 20 mW up to ~50 mW in a ~2000
mm
2
area at the sample. In many recordings the BR probe
photobleached before a myosin step occurred. The record-
ings used for analysis represented molecules that photo-
bleached after 12 s of high-intensity illumination, on
average, and after zero to six myosin stepping events (see
Materials and Methods). Fig. S2 shows the photon counts
recorded from the 16 combinations of input and detector
polarization binned at 0.1 ms intervals. As the molecule
walked processively along actin in the presence of 140
mM Mg.ATP, tilting of the BR (and myosin lever arm)
back and forth between two fairly well dened angles was
clearly evident from abrupt and simultaneous changes in
many of the polarized photon count rates, as previously
reported (12). All of the channels photobleached to the
background in a single step, consistent with signals from
a single uorophore.
To convert noisy polTIRF photon counting data into inter-
pretable intensity and angular time courses, a specialized
change point analysis (38) was developed (31). Instead of
binning the photons into intensity traces, the raw time-
stamped data are tted directly by a likelihood function
that compares the probability of a change in the photon
collection rate with that of the constant rate. By scanning
this likelihood function over all of the photons in an interval
from all of the 16 polarization channels, we can identify the
location and uncertainty of the most likely intensity change.
By iteratively applying this algorithm through the whole
trace, we can identify angle changes, bleaching events,
and any other large-magnitude changes in the individual
channels (dashed lines in Fig. 1) that are clearly seen by
eye. Smaller-magnitude but statistically robust changes are
also detected.
The uorescence intensity for each channel during the
interval is the average determined by summing the photons
and dividing by the duration between adjacent change
points. Each set of 16 intensities is used as input to
a maximum-likelihood tting algorithm that estimates the
dipole angles, (q, f); wobble, d; and brightness, k, of the
probe during each interval between change points (hori-
zontal lines in Fig. 2, BD). To interpret the angles in the
biological frame of reference, the orientation relative to
the microscope optical axis (q, f) is reexpressed in terms
of a frame referred to the long axis of the actin lament,
(b, a), where b is the polar angle with respect to the barbed
() end of the actin lament and a is the azimuthal angle
around the actin lament (12,29) (inset in Fig. 3 B). Due
to the dipole character of probe excitation and emission,
any one orientation
/
U (b, a) cannot be distinguished
from an equivalent orientation pointing in the opposite
direction
/
U
0
(180
-b, a 5 180
). Instead of imposing
a xed hemisphere for reporting the angles of all molecules
(26,39,40), we use an objective method to determine a favor-
able hemisphere for each molecule, based on the initial
direction of motility and the average direction of the mole-
cules probe orientations (i.e., the orientation of the director;
see Materials and Methods). To determine the CIs for the
measured angles, we calculate four additional estimates of
each intervals intensity based on the statistical uncertainty
of localizing the adjacent change points (see Materials and
Methods and Fig. S3 B).
Large changes in b and a (Fig. 2, B and C) correspond to
changes in the polarized count rates and indicate tilting
between two stable states with orientation (b, a) z (22
,
190
) and (81
, 140
) (Fig. S4). In Fig. 2, four major step-

ping events (; and 7) are observed at 0.17, 0.37, 0.46,
and 0.72 s as described previously (12). The angle changes
in b and a can be combined into the angle subtended by the
probe vector in the two orientations, z
i
cos
1
~
U
,
~
U
z
75
(where and indicate the state before and after step-

ping, respectively).
Importantly, d (microsecond wobble) transiently
increased during the recording on alternate steps when
b changed from low (trailing position) to high (leading
position) values, consistent with a labeled trailing head de-
taching from actin (; in Fig. 2 D), undergoing rapid rota-
tional motions and then rebinding to actin as the new lead
head with b z 90
(Fig. 2 E). The wobble parameter

remains below 50
during the steps when b changes

from high to low values (7). This is most easily inter-
preted as the nonstepping head undergoing a working
stroke while it is strongly bound to actin, as depicted in
Fig. 2 E (a cartoon showing the sequence of events with
the expected alternating periods of high wobble). A gallery
of angular time courses from additional molecules
(Fig. S5) demonstrates the same stepping pattern with
periods of high wobble when the probe switches from
a low to a high b angle, corresponding to the labeled
head detaching and wobbling while searching for its next
actin-binding site.
Binned data
For visualization purposes, the photon arrival events were
accumulated in bins to produce PFIs with time resolution
traded against photon shot noise (0.8 ms bins, gray/blue
curves in Fig. 1, A and G, and 8 ms bins in Fig. 1, BF).
Despite the high laser power, intensities binned at 8 ms
are relatively noisy, resulting in large uctuations in the
individual PFIs and in the estimated orientations. The vari-
ability of (b, a) during an otherwise constant dwell is larger
than might be expected from the variance in the intensity
traces, due to the highly nonlinear relationship between
the magnitude of the PFIs and the corresponding dipole
orientation. The wobble parameter d is also noisy (Fig. 2
D) but can be seen to increase in magnitude prior to the rst
and third tilting events when b switches from low to high
values (; in Fig. 2). During the nal 1/3 of the recording,
the probe is in a state of prolonged high-wobble (d>75
)
immediately prior to photobleaching, a characteristic that
was observed in many other molecules (Fig. S6).
Dynamics of periods of increased probe wobble
during steps
After manually screening for molecules with robust tilting
(i.e., strong anticorrelated changes in the PFIs), we detected
1224 tilting events in 306 molecules, 69 of which contained
a total of 84 substeps exhibiting increased wobble (inverted
solid triangles in Fig. 3 A) relative to the period just before
(solid triangles) and after (open triangles) each step. The
fraction of molecules with at least one such high-wobble
substep was ~23%, but the fraction of steps with a high-
wobble intermediate was only ~7%. One likely cause of
the low detection of these substeps is the limited number
of photons collected during short-duration events. Consis-
tent with this view, molecules recorded at higher laser power
contained more high-wobble substeps (~10%) than those re-
corded at low power (~4%). The substep durations are
distributed exponentially with an average duration of
~13 ms (solid black dots in Fig. 4 A). A fraction of missing
events is expected at or below the detection limit from the
exponential distribution (open black dots). Nevertheless,
some molecules with a very high signal/background ratio
and multiple myosin steps with strong polarization changes
did not exhibit detectable high-wobble substeps (Fig. S6 C).
Another way to test the origin of the substeps is to
collect the angles before and after the high-wobble period
A
B
C
D
E
F
G
FIGURE 1 High-time-resolution PFIs. (A) Sixteen PFIs are summed into
one total intensity with photons binned within each 0.8 ms cycle (blue and
gray) or averaged over 10 cycles (black) that is relatively constant before
photobleaching to background intensity at ~1.5 s. (B and CF) The total
intensity (B) and the 16 PFIs (CF) comprising the interval before photo-
bleaching (blue region in panel A) are shown expanded, with intensities
averaged over 10 cycles. The individual measured intensities depend on
the relative orientation between the uorophore dipole axis and the direc-
tion of the input laser polarizations p (blue), s (red), L (green), and R
(magenta), and the polarized uorescence detected on APDx (C and D)
and APDy (E and F) for beams 1 and 2, respectively (see Fig. S1 for details
on the experimental setup). (G) To illustrate the high time resolution of each
PFI, the photons collected during each 0.8 ms cycle of the laser polariza-
tions are shown for
p2
I
x
(blue) and
s2
I
x
(red) binned at 0.8 ms. The
MCCP algorithm gives the times of intensity change points (vertical black
dashed lines) along with a 95% CI (gray area). The average intensity
between neighboring change points (horizontal lines) is also shown in BG.
(solid red symbols and open red symbols in Fig. 3 B,
respectively). Considering all high-wobble substeps (98
events in 73 molecules), there is a strong tendency for low
b angles to precede the high wobble state (red points with
solid line) and high b angles to follow it. This provides
strong support for the sequence of events drawn in
Fig. 2, allowing the preceding orientation (low b) to be
identied with the trailing lever arm, the state following
the step (high b) with the leading lever arm and the
high-wobble period with the search. The distributions of
a before (solid green points in Fig. 3 A) and after (open
green points) stepping are uniform and the same before
and after stepping.
Previous work has shown that the myosin inhibitor buta-
nedione monoxime (BDM) slows the stepping rate of
myosin V by inhibiting the release of ADP (18) and also
slows the rebinding rate of the detached head (14). There-
fore, we performed motility experiments in the presence
of 50 or 100 mM BDM and ~2-fold higher MgATP concen-
tration to obtain comparable rates of stepping. On the
molecules with the most robust measurements of stepping,
we detected 43 substeps in 39 out of 123 molecules
A
B
C
D
E
FIGURE 2 Results of the dipole model for the 8.0 ms binned intensities (jagged lines) and the change-point intensities corresponding to the CIs (ve
horizontal lines), which are often very close to each other because the uncertainty in localizing the change points is small. (A) The tted intensity scale factor
k (gray) is similar to the total intensity (Fig. 1 B) and is relatively constant despite large changes in the underlying PFIs, the hallmark of tilts. (B) The polar
angle b (red) shows four large angle changes (denoted by ;and 7in panel D) that correspond to myosin stepping events at change points 1, 3, 5, and 8,
where b alternates between a low value of 22
and a high value of 81
. The magnitude of the total angular displacement z between change-point intervals is

represented by small crosses (5). (C) The azimuthal angle a (green) also changes with each step, alternating between 190
and 140
. (D) The extent of rapid

probe motion during the measurement (wobble) is represented by a cone with half-angle d (blue) and shows relatively low values during a dwell and increases
during steps from the low to high b states (between change points 1-2 and 5-6), but not after steps from the high to low b states (i.e., change points 3-4 and
7-8). Wobble at the end of the trace is also elevated, possibly indicating a laser-induced breakage of one of the bifunctional bonds on the probe, resulting in
a monofunctional attachment. (E) Cartoon depicting the BR-CaM-labeled myosin lever arm translocating along an actin lament (blue helix) that is attached
to the quartz slide (light gray) by NEMmyosin II (dark gray). The probe on the lever armis indicated by double-headed red arrows. Increases in probe wobble
(red disks) occur during every other step, consistent with the hand-over-hand mechanism and thermal search.
containing a total of 380 steps (11% of steps or 32% of
molecules contained high-wobble substeps). The detection
efciencies again depended on laser power, indicating
detection limited by photon counts (16% of steps and 34%
of molecules contained substeps at high laser powers,
compared with 6% of steps and 26% of molecules at low
laser power). Fitting single exponentials to histograms of
the substep durations in the absence (black) or presence
(gray, Fig. 4 A) of BDM led to estimates of the average sub-
step durations of 12.7 52 ms and 16.4 54 ms respectively
(595%CI). Varying the ATP concentration from1 to 20 mM
did not systematically affect the duration of the high-wobble
substep in the absence (black) or presence (gray, Fig. 4 B) of
BDM. These effects of BDM are smaller than expected from
results obtained by Dunn and Spudich (14) in gold-particle-
labeled myosin V, possibly because our molecules contained
only CaM in the lever arm, whereas the molecules used in
the previous study also included essential light chains, which
possibly affected the detached head rebinding rate.
Minor angle changes
In some molecules (52 events in 42 molecules), we detected
transient increases in wobble without corresponding main-
tained tilting motions of the myosin. Instead, the orienta-
tions before and after the high-wobble substep were
approximately the same (Fig. S6 D), consistent with the
labeled head detaching from actin and rebinding actin
without stepping (also known as a foot-stomp (10,19)).
The average duration of these events was 33 ms (n 22
events in 16 molecules). Nine of these molecules showed
this type of nonproductive high-wobble period in the
presumptive trailing head (b < 50
) and seven showed it

in the leading head (b > 50
). Neither the ATP concentra-

tion nor the presence of BDM appeared to inuence the
occurrence of these nonstep increases in wobble.
Smaller-magnitude angle changes also occurred between
the major steps. For the molecule in Fig. 2, the lever arm
A
B
FIGURE 3 Distributions of probe angle and wobble before and after
steps including a detectable high-wobble state (98 events in 73 molecules).
(A) The distributions of d (blue) during the intervals immediately before (:
with solid line) and after (6with dashed line) steps in which a high wobble
state was detected peak at lower values (~30
) compared with the distribu-

tion of d during the high-wobble state (;with dotted line), which peaks at
large values (~7080
). (B) Distributions of b (red) and a (green) during the

intervals immediately before (Cand -with solid lines) and after (Band
, with dashed lines) steps with detectable high-wobble states. Steps
with a high-wobble state allow trailing and leading lever arm states to be
identied unambiguously as the intervals before and after the step, respec-
tively. The inset illustrates the angle conventions for b, a, and d relative to
the actin lament.
,
0 mM BDM
100 mM BDM
0 20 40 60 80
0.0
30
25
20
15
10
0
5
0.2
0.4
0.6
0.8
Substep duration, ms
0 10 20 30 40
ATP, M
A
B
P
r
o
b
a
b
i
l
i
t
y
S
u
b
s
t
e
p

d
u
r
a
t
i
o
n
,

m
s
FIGURE 4 (A) Histogram of the duration of the high-wobble interval
between large orientation changes that likely represent detached heads.
Single exponential ts, excluding the rst bin (open symbol) in each distri-
bution, result in average lifetimes of 13 52 ms (n 99) and 16 54 ms
(n 53) in the absence (black) and presence (gray) of 50100 mM BDM.
(B) The duration of the high-wobble state for molecules in the absence
(black) and presence (gray) of BDM over a range of Mg.ATP concentra-
tions agrees within uncertainty (SE) with the mean duration (horizontal
dashed line), indicating that the kinetics of this state are independent of
Mg.ATP concentration. Experiments with BDM in the solution were per-
formed at a higher ATP concentration to compensate for the reduction
in velocity (18). The outlier at 2 mM Mg.ATP represents only six events de-
tected at this concentration.
appeared to undergo additional minor tilting at change
point 7 on the leading head and at points 4 and 912 on
the trailing head. The angular displacements (z) during
these motions were usually <10
but could be up to 45
.
Change points 1417 in Fig. 2 during the high-wobble
period before photobleaching represent statistically signi-
cant small changes in intensity, but they lack well-dened,
quantiable orientations.
The numbers of minor events detected by the change-
point algorithm before and after high-wobble substeps
were similar for the trailing and leading heads (solid and
open symbols in Fig. 5 A). For both the leading and trailing
heads, the majority of steps occurred with no additional
angle changes and the overall number of events averaged
~1.1 small change points per major step. Neither the magni-
tude (Fig. 5 B) nor the kinetics (data not shown) of these
nonstep angle changes differed appreciably between the
leading- and trailing-head states. By manually reviewing
these recordings, as well as those of molecules in which
no high-wobble states were detected during stepping (and
thus the leading and trailing states were less clear), we
conrmed that few molecules exhibited small, lasting angle
changes preceding their steps. Thus, at the ATP concentra-
tions used here (120 mM), the lever orientation is predom-
inantly constant prior to major rotations of the lever arm,
consistent with mainly straight-leg stepping (19).
Photophysical effects
Approximately 30% of the recordings contained an
extended period of high probe wobble immediately before
photobleaching (for example, the last ~0.5 s of Fig. 2 D
and Fig. S6). Previous studies reported occasional increases
in uorescence intensity or spectral shifting prior to photo-
bleaching (30,41). We also observed an increase in uores-
cence before photobleaching in some traces; however, the
increase in probe wobble was much more common. Photo-
bleaching times in this population were not different from
the normal traces, and the likelihood of obtaining a period
of high wobble before photobleaching did not depend on
laser intensity. The phenomenon is probably due to a photo-
physical artifact of laser light damage to the probe or probe-
labeled lever arm, based on several observations. Processes
that adversely affect actin binding or ATPase activity would
also be expected to produce a population of recordings with
a prolonged terminal state of constant angle, but molecules
without such sustained high-wobble episodes tilted back and
forth until they photobleached. Consequently, we speculate
that one of the bifunctional linkers in the probe may break
under the high laser illumination, resulting in a singly
attached, highly mobile probe that does not discriminate
between attached actomyosin states and detached ones.
Short bursts of multiple, very rapid polarization change
points were also detected infrequently in a subset of mole-
cules (Fig. S6). The timescale of these motions (~18 ms)
was much faster than myosin stepping and had no bias
toward the leading or trailing head. There were more of
these rapid events at higher laser powers, suggesting that
they may be another photophysical effect.
DISCUSSION
High-speed polTIRF detects periods of brief
detachment during myosin V stepping
In previous polTIRF studies of myosin V (26,27,39), the
time resolution was limited to 20 ms and the range of
discernible angles was often restricted to 1/8 of a sphere
due to symmetries that resulted from conning the input
laser polarizations to three orthogonal planes (12,28).
Here, laser polarizations aligned in four directions (45
,
0
, 45
, and 90
) relative to the plane dened by the input

and reected laser beams (the scattering plane) in both
beams decreased the ambiguity fourfold, resulting in a hemi-
sphere of unambiguously distinguishable orientations
(26,27,29) (Fig. S1). This scope of angles is the maximum
limit imposed by the intrinsic dipole symmetry of the probe.
Statistically valid change points in the photon collection
rates identied times of structural changes, such as tilting
A
B
FIGURE 5 Distributions of nonstep angle changes (n 73) obtained
using only molecules with steps exhibiting a high wobble state so that
the dwells could be unambiguously assigned to the trailing and leading
heads. (A) The distributions of the number of additional angle changes on
the trailing (n 59, solid line) and leading (n 69, dashed line) heads
are similar. (B) Histograms of the change in b angle during the dwell before
(solid symbols) and after (open symbols) large wobble events. Similar histo-
grams for Da are included in Fig. S8.
and increased wobble. An important assumption of the
change-point approach is that the molecule resides in
discrete structural states and switches abruptly between
these states, resulting in sudden changes in the polarized
uorescence photon count rates. This feature implies that
the rates are constant between change points, and that the
total number of counts divided by the interval between
change points is a good estimate of the uorescence
intensity. The 16 intensities were tted numerically by a
maximum-likelihood model of uorescence that depended
on the dipole orientation (b, a), and wobble, d (12,19,2729).
Wobble quanties the depolarization of the uorescence
signal on timescales slower than the uorescent lifetime
(~4 ns) but faster than the measurement timescale (~0.1 ms).
The expected microsecond rotational diffusion timescale
of a free myosin V head is thus quantied, for the rst time
to our knowledge, by this d parameter.
Increases in d during a switch in lever-arm orientation
were measured in many molecules (Fig. S5). The timing
of the structural changes derives purely from the uores-
cence intensities independently of any binning of data or
of any model predicting the relative intensities. The brief
increases in d are consistent with a highly mobile state of
the trailing head after it detaches from actin during the
thermal search for its next actin-binding site. This assign-
ment is supported by several of our results. First, the results
of tting the dipole model to the 8 ms binned data conrm
that the high-wobble substeps occur between steps (Fig. 3 A)
and are characterized by randomly distributed b and a,
consistent with microsecond tumbling of the detached
head. Second, in >90% of recordings where more than
one high-wobble substate was detected, they were observed
on alternating steps, as expected (for example, Fig. 2 E and
Fig. S5). Third, the average time for rebinding was indepen-
dent of ATP concentration (Fig. 4 B). Lastly, the ~13 ms
duration of these short intervals (Fig. 4 A) agrees well
with the period of high positional uctuations of gold-nano-
particle-labeled myosin V during steps along actin (14).
Experimental evidence for the single-head-attached
myosin intermediate was obtained by optical trapping
(17), uorescence microscopy (15,16), and high-speed
AFM (10), but measurement of the rebinding rate in these
studies was complicated by a bead, a large reporter mole-
cule, or surface interactions. The mean waiting time for re-
binding of the free head to actin was theoretically estimated
to be 0.11 ms from rst passage time calculations (7), the
processive run length (42), and Brownian dynamics simula-
tions (43). Both BR-CaM-labeled myosin V in the current
work (Fig. 4) and single-particle tracking of 40 nm gold-
particle-labeled CaM (14) indicated rebinding times of
1015 ms, even though the 40 nm gold particle probably
slowed diffusion of the head (the presence of a single rhoda-
mine uorophore (~ 600 Da) is not expected to have a signif-
icant inuence on the dynamics). The slower binding
compared with the expectation from diffusion-limited
models strongly suggests that a structural or biochemical
step, such as reversal of the working stroke (44,45), weak-
to-strong actomyosin binding (11), or P
i
release (13), and
not Brownian dynamics, limits the rebinding rate. The
increase in probe wobble during a step supports models of
hand-over-hand motion (8) with high rotational mobility
of the head as it steps from the trailing to the leading posi-
tion. Models in which the head remains in close proximity to
the actin lament and skips or slides along the lament
while reaching for the next strongly bound actin subunit
(47) are thus rendered less likely.
The polarization switching devices (Pockels cells),
photon detectors, and single-photon counting circuit are
all fast electronic elements with approximately microsecond
or faster response times. Consequently, the factor that limits
the time resolution is the number of detected photons from
the single probe (here ~50 photons/ms). The periods of
increased wobble during stepping of the labeled head last
~1020 ms, leading to ~5001000 total photons detected
on average from these intervals. These values are just above
the threshold required to detect change points with 95%
condence, which probably accounts for fewer high-wobble
substeps being detected (14%, compared with the expected
50% of all steps). The excitation intensity is not near satura-
tion of the absorption dipole, so higher excitation laser
powers would increase the numbers of detected photons
per time interval, but at the expense of shorter recording
times before photobleaching. Of the ~1000 total molecules
recorded at the laser intensities used, ~40% photobleached
before a step occurred, which limited the practical range
of input intensities.
Other factors may result in undercounting the expected
number of substeps. First, short-lived events that contain
enough photons to trigger detection of a change point but
not enough to be reliably t by the dipole model can be
missed because the tting typically underestimates the
wobble of low-photon states. Also, the number of stepping
events during a recording might be overestimated because
at low ATP and ADP concentrations, some large rotational
motions are not associated with stepping (10,19).
Distributions of orientation and wobble during
myosin V stepping
The distributions of probe orientations with each step were
characterized by two peaks in b at ~20
and ~85
(Fig. 2 A
and Fig. S4), which result in lever-arm orientations of 60
and 120
for the leading and trailing congurations when

the local orientation of the BR probe on the lever arm is
taken into account (12,27). In previous experiments in
which both position and orientation were measured (48),
the smaller magnitude peak in b
probe
was determined to be
the trailing head, in agreement with the relative orientation
between the probe axis and the myosin lever arm (27,49). In
the high-speed polTIRF experiments described here, this
assignment of angles to the two positions is additionally
supported by the observation that the high-wobble search
period is observed when b
probe
switches from smaller to
larger angle (Figs. 2 and 3 A, and Fig. S5), thereby conrm-
ing the trailing to leading positions, respectively, as sug-
gested by the earlier studies.
The fairly uniform a distributions (Fig. 2 B and
Fig. S4 B) are consistent with myosins binding at random
azimuthal angles around the actin lament. The magnitude
of the power stroke is estimated from angle changes of the
stepping head with Db, Da, and z equal to 68
5 7
,
2.4
5 4
, 83
5 4
, respectively (5SE). The small

magnitude of Da, leading to similar magnitudes of Db
and z, conrms that the power stroke is aligned closely
along the actin lament axis.
Minor angle changes
Previous single-molecule work on myosin V using optical
traps (7,17,18), electron micrographs (20,21), uorescence
(19), and atomic force microscopy (10) suggested that in
addition to the high-wobble search period during an active
step, there may be mechanistically important substeps
before or after a step. The base of the lever arm in the
leading head might tilt forward while the remainder of the
lever is restrained by its attachment to the rear lever arm,
causing a kink to form the so-called telemark conguration
(1921,50). In our data, nonstep change points were some-
times detected during the intervals between steps, but there
was no bias toward leading or trailing heads in either the
number of events (Fig. 5 A) or their magnitude (Fig. 5 B).
The occasional larger-magnitude angle changes that were
detected may be consistent with an alternate reaction
pathway (51). In our experiments, the most common cong-
uration observed while the molecule waited for ATP binding
was one in which both heads were bound to actin and both
lever arms were relatively straight. Upon ATP binding, the
stepping motion normally was rapid and proceeded without
a preceding tilting of either lever arm. These results apply to
the specic conditions optimized for detection of the high-
wobble state, i.e., a relatively narrow range of ATP concen-
trations, no added ADP, and a relatively brief recording
interval limited by photobleaching. High-speed AFM
images of myosin V also showed mainly straight lever
arms at micromolar ATP in the absence of ADP, but the
leading lever arms became kinked more often at higher
ADP concentrations (10). polTIRF measurements over
a broader range of nucleotide concentrations might identify a
role for such auxiliary rotations in the myosin molecule.
CONCLUSIONS
An upgraded polTIRF microscope was used to investigate
myosin V motility with high time resolution. Brief ~10
15 ms substeps, characterized by highly disordered lever-
arm orientations, were detected during stepping events and
likely correspond to the detached head searching by random
uctuations for its next actin-binding site. The majority of
myosin V molecules contained straight lever arms in both
leading and trailing positions (i.e., not bent or kinked),
and additional rotations of the lever arm were uncommon
on both the leading and trailing heads. Since biological
macromolecules generally function in an environment char-
acterized by signicant thermal uctuations, other impor-
tant enzymatic steps and short-lived intermediates might
be detected by rotational exibility that can be measured
at the single-molecule level using high-speed polTIRF
microscopy.
SUPPORTING MATERIAL
Additional analysis, eight supporting gures and their legends are available
at http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)00199-9.
We thank Drs. Yujie Sun and Martin Pring for helpful discussions, Xiaonan
Cui for help in labeling BR-CaM, and H. Lee Sweeney for providing the
myosin V.
This work was supported by National Institutes of Health grant R01
GM086352 (Y.E.G), and National Science Foundation grants DGE-02-
21664 (J.F.B.), EF-0928048 (P.C.N.), and DMR08-32802 (P.C.N. and
Y.E.G.). The authors thank the Aspen Center for Physics, National Science
Foundation Grant No. 1066293, for hospitality during various phases of
this work.
REFERENCES
1. Reck-Peterson, S. L., D. W. Provance, Jr., ., J. A. Mercer. 2000. Class
V myosins. Biochim. Biophys. Acta. 1496:3651.
2. Sweeney, H. L., and A. Houdusse. 2010. Structural and functional
insights into the myosin motor mechanism. Annu Rev Biophys. 39:
539557.
3. Sun, Y., and Y. E. Goldman. 2011. Lever-arm mechanics of processive
myosins. Biophys. J. 101:111.
4. Mehta, A. D., R. S. Rock, ., R. E. Cheney. 1999. Myosin-Vis a proc-
essive actin-based motor. Nature. 400:590593.
5. Rief, M., R. S. Rock, ., J. A. Spudich. 2000. Myosin-V stepping
kinetics: a molecular model for processivity. Proc. Natl. Acad. Sci.
USA. 97:94829486.
6. Sakamoto, T., I. Amitani, ., T. Ando. 2000. Direct observation of
processive movement by individual myosin V molecules. Biochem.
Biophys. Res. Commun. 272:586590.
7. Veigel, C., F. Wang, ., J. E. Molloy. 2002. The gated gait of the proc-
essive molecular motor, myosin V. Nat. Cell Biol. 4:5965.
8. Yildiz, A., J. N. Forkey, ., P. R. Selvin. 2003. Myosin V walks hand-
over-hand: single uorophore imaging with 1.5-nm localization.
Science. 300:20612065.
9. Schroeder, 3rd, H. W., C. Mitchell, ., Y. E. Goldman. 2010. Motor
number controls cargo switching at actin-microtubule intersections
in vitro. Curr. Biol. 20:687696.
10. Kodera, N., D. Yamamoto, ., T. Ando. 2010. Video imaging of
walking myosin V by high-speed atomic force microscopy. Nature.
468:7276.
11. De La Cruz, E. M., A. L. Wells, ., H. L. Sweeney. 1999. The kinetic
mechanism of myosin V. Proc. Natl. Acad. Sci. USA. 96:1372613731.
12. Forkey, J. N., M. E. Quinlan, ., Y. E. Goldman. 2003. Three-dimen-
sional structural dynamics of myosin V by single-molecule uores-
cence polarization. Nature. 422:399404.
13. Rosenfeld, S. S., and H. L. Sweeney. 2004. A model of myosin V proc-
essivity. J. Biol. Chem. 279:4010040111.
14. Dunn, A. R., and J. A. Spudich. 2007. Dynamics of the unbound head
during myosin V processive translocation. Nat. Struct. Mol. Biol.
14:246248.
15. Shiroguchi, K., and K. Kinosita, Jr. 2007. Myosin V walks by lever
action and Brownian motion. Science. 316:12081212.
16. Komori, Y., A. H. Iwane, and T. Yanagida. 2007. Myosin-V makes two
Brownian 90 degrees rotations per 36-nm step. Nat. Struct. Mol. Biol.
14:968973.
17. Cappello, G., P. Pierobon, ., J. Prost. 2007. Myosin V stepping mech-
anism. Proc. Natl. Acad. Sci. USA. 104:1532815333.
18. Uemura, S., H. Higuchi, ., S. Ishiwata. 2004. Mechanochemical
coupling of two substeps in a single myosin V motor. Nat. Struct.
Mol. Biol. 11:877883.
19. Syed, S., G. E. Snyder, ., Y. E. Goldman. 2006. Adaptability of
myosin V studied by simultaneous detection of position and orienta-
tion. EMBO J. 25:17951803.
20. Walker, M. L., S. A. Burgess, ., P. J. Knight. 2000. Two-headed
binding of a processive myosin to F-actin. Nature. 405:804807.
21. Oke, O. A., S. A. Burgess, ., J. Trinick. 2010. Inuence of lever struc-
ture on myosin 5a walking. Proc. Natl. Acad. Sci. USA. 107:2509
2514.
22. Forkey, J. N., M. E. Quinlan, and Y. E. Goldman. 2000. Protein struc-
tural dynamics by single-molecule uorescence polarization. Prog.
Biophys. Mol. Biol. 74:135.
23. Rosenberg, S. A., M. E. Quinlan, ., Y. E. Goldman. 2005. Rotational
motions of macro-molecules by single-molecule uorescence micros-
copy. Acc. Chem. Res. 38:583593.
24. Joo, C., H. Balci, ., T. Ha. 2008. Advances in single-molecule uores-
cence methods for molecular biology. Annu. Rev. Biochem. 77:5176.
25. Irving, M., T. St Claire Allen, ., Y. E. Goldman. 1995. Tilting of the
light-chain region of myosin during step length changes and active
force generation in skeletal muscle. Nature. 375:688691.
26. Beausang, J. F., H. W. Schroeder, 3rd, ., Y. E. Goldman. 2008. Twirl-
ing of actin by myosins II and Vobserved via polarized TIRF in a modi-
ed gliding assay. Biophys. J. 95:58205831.
27. Lewis, J. H., J. F. Beausang, ., Y. E. Goldman. 2012. The azimuthal
path of myosin Vand its dependence on lever-armlength. J. Gen. Phys-
iol. 139:101120.
28. Forkey, J. N., M. E. Quinlan, and Y. E. Goldman. 2005. Measurement
of single macromolecule orientation by total internal reection uores-
cence polarization microscopy. Biophys. J. 89:12611271.
29. Beausang, J. F., Y. Sun, ., Y. E. Goldman. 2008. Orientation and rota-
tional motions of single molecules by polarized total internal reection
uorescence microscopy. In Single Molecule Techniques. P. R. Selvin
and T. Ha, editors. Cold Spring Harbor Laboratory Press, Cold Spring
Harbor. 121148.
30. Quinlan, M. E., J. N. Forkey, and Y. E. Goldman. 2005. Orientation of
the myosin light chain region by single molecule total internal reec-
tion uorescence polarization microscopy. Biophys. J. 89:11321142.
31. Beausang, J. F., Y. E. Goldman, and P. C. Nelson. 2011. Changepoint
analysis for single-molecule polarized total internal reection uores-
cence microscopy experiments. Methods Enzymol. 487:431463.
32. Purcell, T. J., C. Morris, ., H. L. Sweeney. 2002. Role of the lever arm
in the processive stepping of myosin V. Proc. Natl. Acad. Sci. USA.
99:1415914164.
33. Putkey, J. A., G. R. Slaughter, and A. R. Means. 1985. Bacterial ex-
pression and characterization of proteins derived from the chicken
calmodulin cDNA and a calmodulin processed gene. J. Biol. Chem.
260:47044712.
34. Corrie, J. E., J. S. Craik, and V. R. Munasinghe. 1998. A homobifunc-
tional rhodamine for labeling proteins with dened orientations of a
uorophore. Bioconjug. Chem. 9:160167.
35. Pardee, J. D., and J. A. Spudich. 1982. Purication of muscle actin.
Methods Cell Biol. 24:271289.
36. Beausang, J. F. 2010. Single molecule investigations of DNA looping
using the tethered particle method and translocation by acto-myosin
using polarized total internal reection uorescence microscopy. PhD
dissertation, University of Pennsylvania, Philadelphia, PA.
37. Veigel, C., M. L. Bartoo, ., J. E. Molloy. 1998. The stiffness of rabbit
skeletal actomyosin cross-bridges determined with an optical tweezers
transducer. Biophys. J. 75:14241438.
38. Watkins, L. P., and H. Yang. 2005. Detection of intensity change points
in time-resolved single-molecule measurements. J. Phys. Chem. B.
109:617628.
39. Sun, Y., H. W. Schroeder, 3rd, ., Y. E. Goldman. 2007. Myosin VI
walks wiggly on actin with large and variable tilting. Mol. Cell.
28:954964.
40. Reifenberger, J. G., E. Toprak, ., P. R. Selvin. 2009. Myosin VI
undergoes a 180 degrees power stroke implying an uncoupling of the
front lever arm. Proc. Natl. Acad. Sci. USA. 106:1825518260.
41. Wazawa, T., Y. Ishii, ., T. Yanagida. 2000. Spectral uctuation of
a single uorophore conjugated to a protein molecule. Biophys. J.
78:15611569.
42. Smith, D. A. 2004. How processive is the myosin-V motor? J. Muscle
Res. Cell Motil. 25:215217.
43. Craig, E. M., and H. Linke. 2009. Mechanochemical model for myosin
V. Proc. Natl. Acad. Sci. USA. 106:1826118266.
44. Kinosita, Jr., K., K. Shiroguchi, ., H. Itoh. 2007. On the walking
mechanism of linear molecular motors. Adv. Exp. Med. Biol. 592:
369384.
45. Shiroguchi, K., H. F. Chin, ., K. Kinosita, Jr. 2011. Direct observation
of the myosin Va recovery stroke that contributes to unidirectional
stepping along actin. PLoS Biol. 9:e1001031.
46. Reference deleted in proof.
47. Okada, T., H. Tanaka, ., T. Yanagida. 2007. The diffusive search
mechanism of processive myosin class-V motor involves directional
steps along actin subunits. Biochem. Biophys. Res. Commun. 354:
379384.
48. Toprak, E., J. Enderlein, ., P. R. Selvin. 2006. Defocused orientation
and position imaging (DOPI) of myosin V. Proc. Natl. Acad. Sci. USA.
103:64956499.
49. Parker, D., Z. Bryant, and S. L. Delp. 2009. Coarse-grained structural
modeling of molecular motors using multibody dynamics. Cell Mol.
Bioeng. 2:366374.
50. Snyder, G. E., T. Sakamoto, ., P. R. Selvin. 2004. Nanometer locali-
zation of single green uorescent proteins: evidence that myosin V
walks hand-over-hand via telemark conguration. Biophys. J.
87:17761783.
51. Kad, N. M., K. M. Trybus, and D. M. Warshaw. 2008. Load and Pi
control ux through the branched kinetic cycle of myosin V. J. Biol.
Chem. 283:1747717484.
Temperature Dependence of the DNA Double Helix at the Nanoscale:
Structure, Elasticity, and Fluctuations
Sam Meyer,
Daniel Jost,
Nikos Theodorakopoulos,
{
Michel Peyrard,
Richard Lavery,
jj
* and Ralf Everaers
Universite de Lyon, Laboratoire de Physique, Ecole Normale Supe rieure de Lyon, Lyon, France;

Centre Blaise Pascal, Ecole Normale
Supe rieure de Lyon, Lyon, France;

Theoretical and Physical Chemistry Institute, National Hellenic Research Foundation, Athens, Greece;
{
Fachbereich Physik, Universita t Konstanz, Konstanz, Germany; and
jj
Bases Mole culaires et Structurales des Systemes Infectieux, University
Lyon I/Centre National de la Recherche Scientique, Institut de Biologie et Chimie des Prote ines, Lyon, France
ABSTRACT Biological organisms exist over a broad temperature range of 15
C to 120
C, where many molecular pro-

cesses involving DNA depend on the nanoscale properties of the double helix. Here, we present results of extensive molecular
dynamics simulations of DNA oligomers at different temperatures. We show that internal basepair conformations are strongly
temperature-dependent, particularly in the stretch and opening degrees of freedom whose harmonic uctuations can be consid-
ered the initial steps of the DNA melting pathway. The basepair step elasticity contains a weaker, but detectable, entropic contri-
bution in the roll, tilt, and rise degrees of freedom. To extend the validity of our results to the temperature interval beyond the
standard melting transition relevant to extremophiles, we estimate the effects of superhelical stress on the stability of the base-
pair steps, as computed fromthe Benhammodel. We predict that although the average twist decreases with temperature in vitro,
the stabilizing external torque in vivo results in an increase of ~1
/bp (or a superhelical density of Dsx 0:03) in the interval

0100
C. In the nal step, we show that the experimentally observed apparent bending persistence length of torsionally uncon-
strained DNA can be calculated from a hybrid model that accounts for the softening of the double helix and the presence of tran-
sient denaturation bubbles. Although the latter dominate the behavior close to the melting transition, the inclusion of helix
softening is important around standard physiological temperatures.
INTRODUCTION
DNA is the common substrate of genetic information in all
living organisms. The mechanical properties of the DNA
double helix play a crucial role in the molecular processes
related to the replication and the regulated transcription of
this information: examples include the tight wrapping of
DNA around histone (1) and histonelike (2) proteins in pro-
karyotes, and sequence recognition by other molecules such
as the TATA-box binding protein (3). With biological
organisms living at very different temperatures (so-called
extremophiles thrive over a temperature range of 15
C
to 120
C (4,5)), the question arises of how the properties

of DNA vary with temperature.
To discuss the basic ideas of this article regarding the
temperature dependence of the DNA double-helix elasticity
at different length scales, it is useful to briey consider
generic springlike degrees of freedom. In the simplest
example of a harmonic spring, the excitation free energy
has the form Fx 1=2 kx
2
, where k is the spring stiffness.
In a molecular mechanics force eld, this functional form
applies to bond lengths and suitable bond angles. Deforma-
tions lead to a purely energetic or enthalpic response. The
stiffness of such springs is independent of temperature,
kT k
h
, whereas the amplitude of the corresponding ther-
mal uctuations, hx
2
i k
B
T=k, is directly proportional to
the temperature. A rst complication arises if the mechani-
cal force eld is anharmonic. In this case, the linear
response of the system to external forces and torques
described by a harmonic approximation becomes tem-
perature-dependent. The opposite extreme from enthalpic
springs is entropic springs, with kT k
s
T and
hx
2
i k
B
=k
s
independent of temperature. Such springs
represent the behavior of soft matter at much larger scales.
The best-known examples are the entropic springs substitut-
ing random-walk-like polymer chains in the theory of rub-
ber elasticity (6). In this case, k
s
<0 as the entropy
decreases with the extension of the polymer chain. As a
result, a rubber band under a mechanical load contracts
when its temperature is raised. The nanoscale elasticity of
DNA combines all these complications. The double helix
has a well dened average shape and its local behavior is
not dominated by uctuations. However, since the descrip-
tion results from integrating out more microscopic degrees
of freedom, nanoscale force elds necessarily represent
temperature-dependent deformation free energies. In the
harmonic approximation and over a nite temperature
range, the temperature dependence can always be written
in the form kT k
h
Tk
s
and can be inferred from the
amplitude of thermal uctuations observed at different tem-
peratures, kT k
B
T=hx
2
i
T
. For k
s
% 0, the results can,
at least formally, be extrapolated to a temperature higher
than those used to calibrate the harmonic model. If, as turns
out to be the case, k
s
>0, the results cannot be extrapolated
beyond a spinodal temperature T
s
k
h
=k
s
, where the spring
Submitted July 24, 2013, and accepted for publication September 9, 2013.
*Correspondence: richard.lavery@ibcp.fr
Sam Meyers present address is LIRIS, CNRS UMR 5205, INRIA, INSA
Lyon, Universite de Lyon, F-69621 Lyon, France.
Editor: Michael Levitt.
1904 Biophysical Journal Volume 105 October 2013 19041914
ceases to resist extension and thus becomes mechanically
unstable. Consistency requires that the spinodal temperature
be much higher than the actual melting temperature, T
m
,
where the two DNA strands separate. In this case, the nano-
mechanical description of the double helix remains valid for
T
m
<T T
s
, even though this state now only represents a
metastable local free-energy minimum and no longer the
global free-energy minimum. If DNA only undergoes partial
melting, the mechanical properties of a given basepair step
can be described by two-state spring models, which account
for the different structures and elastic properties in the helix
and coil sections of the molecule (79). For the calculation
of suitable averages, the relative statistical weights of the
two states need to be inferred from sequence-dependent
models of DNA melting (1013). Interestingly, external me-
chanical forces and torques may change the relative stability
of the helix and coil states and hence modify the melting
temperature (14).
In this article, we address the temperature-dependent
properties of DNA over a wide span of length scales. We
start from an atomistic model (15), with which we investi-
gate DNA oligomers at different temperatures, thereby ex-
tending previous molecular dynamics (MD) results of the
ABC consortium (16,17). The central part of our work deals
with the nanoscale structure, elasticity, and stability of the
double helix. In particular, we propose and parameterize a
temperature-dependent generalization of the rigid-base
and rigid-basepair models of DNA (18). Although our sim-
ulations were performed for torsionally unconstrained
DNA, we estimate the effects of superhelical stress by intro-
ducing temperature-dependent torques estimated from the
Benham model (14). Biological control of such a torque
can serve to regulate spontaneous DNA opening at tran-
scription start sites (19). In particular, the introduction of
positive superhelical stress makes it possible to stabilize
the double helix in a temperature interval beyond the stan-
dard melting transition (14), which is comparable to the
conditions under which extremophiles exist. In the nal
step, we consider DNA on the wormlike chain level. We
coarse-grain the nanoscale model to determine the tempera-
ture-dependent persistence length of the double helix (20)
and reevaluate a recent kinked wormlike chain model (8)
to account for the presence of transient denaturation bubbles
(7) in the estimation of the apparent persistence length of
DNA (9).
The article is organized as follows. The Models and
Methods section describes the temperature-dependent
rigid-base and rigid-basepair models, the MD simulations,
the coupling with the Benham model, the coarse-graining
to the wormlike chain model, and the inclusion of transient
denaturation bubbles in the calculation of an apparent
bending persistence length. Details of the analysis of the
simulation data are described in the Supporting Material.
In the Results section, we present our ndings for the
entropic contribution to the DNA nanoscale elasticity.
This contribution is particularly strong for the internal base-
pair elasticity. At the basepair step level, the effect is weaker
but detectable, and it results in a softening of the large-scale
stiffness of the molecule. In the Discussion section, we
address the temperature dependence of internal uctuations
of the double helix, the path to the melting transition, the
effect of superhelical stress on the properties of the dou-
ble helix in vivo, and the large-scale bending rigidity
of DNA, including the temperature dependence of the
apparent persistence length over the whole experimentally
studied temperature interval. The article closes with a brief
conclusion.
MODELS AND METHODS
DNA elasticity in the rigid-base and rigid-basepair
models
We consider the uctuations of the DNA double helix at two
successive nanoscale levels, 1), inside the basepair (intra
parameters), and 2), between adjacent basepairs (step
parameters). In both cases, the conformation is described
by a six-dimensional vector, q q
1
; .; q
6
, corresponding
to the relative orientation and position of the relevant objects
(bases or basepairs), in conventional notation: buckle, pro-
pel, opening, shear, stretch, and stagger in the rst case,
and tilt, roll, twist, shift, slide, and rise in the second (18).
In the harmonic approximation, the Gibbs free energy
reduces to a quadratic form:
G
q; s; T

1
2
q q
0
s; T
t
ks; T
q q
0
s; T
; (1)
where T is the temperature, s is the DNA sequence, and we
have introduced the six-dimensional equilibrium conforma-
tion q
0
s; T and the 6 6 stiffness matrix ks; T. The
latter describes the couplings between the uctuations of
the different degrees of freedom and is proportional to the
inverse of the covariance matrix,
Cs; T k
B
T ks; T
1
: (2)
For the rigid-base(pair) model of DNA, previous studies
have extracted the sequence-dependent elastic parameters
at room temperature, T
0
300 K, from the analysis of
DNA and DNA-protein crystallographic structures (21),
from MD simulations of DNA oligomers (17,22), or from
combinations of both approaches (23). Note that for an
ensemble of crystal structures, the passage from the
observed covariance matrix to the stiffness (Eq. 2) relies
on the existence of an effective temperature governing the
ensemble of experimentally observed conformations
(22,23). This approach can therefore not be used to study
the actual temperature dependence of the mechanical prop-
erties of DNA. Instead, we extend previous MD simulations
of DNA oligomers (16,17) to a broad range of temperatures
Temperature Dependence of the DNA Double Helix 1905
to study the sequence and temperature dependence of DNA
structure and elasticity.
Inclusion of temperature effects
The knowledge of the elastic parameters at T
0
gives no
information on their temperature dependence around T
0
,
which depends on the relative importance of the enthalpic
and entropic contributions:
G
q; s; T
q; s; T
0
T S
q; s; T
0
: (3)
Here, H and S are quadratic forms:
H
q; s; T
0

1
2
q q
h
0
s
t
k
h
s
q q
h
0
s
(4)
S
q; s; T
0

1
2
q q
s
0
s
t
k
s
s
q q
s
0
s
: (5)
From Eq. 3, the temperature-dependent elastic parameters
ks; T and q
0
s; T of Eq. 1 can be written in terms of
enthalpic and entropic contributions. For the stiffness, the
relation is simply
ks; T k
h
s Tk
s
s k
0
s T T
0
k
s
s; (6)
where k
h
and k
s
are the enthalpic and entropic contributions
to the stiffness, and k
0
shks; T
0
is the stiffness at room
temperature.
Although the covariance matrix for the uctuations
(Eq. 2) takes the simple form C k
B
Tk
h
Tk
s
1
, the
equilibrium conformation q
0
s; T has a more complex
behavior, and in particular does not vary linearly with
temperature:
q
0
s; T
k
h
Tk
s
k
h
q
h
0
Tk
s
q
s
0
:
To facilitate the estimation of the temperature-dependent
contribution in the following numerical study, we consider
the linear expansion of the latter expression around T
0
:
q
0
s; Txq
0
0
s T T
0
q
0
0
s; (7)
where q
0
0
shq
0
s; T
0
is the equilibrium conformation at
room temperature, and q
0
0
q
h
0
; q
s
0
; k
h
; k
s
is the rst-order
coefcient.
The presence of two parameters in both Eqs. 6 and 7 sig-
nies that the temperature-dependent elastic model involves
twice as many parameters as the model at a single tempera-
ture, k
0
; q
0
0
. In the following sections, we estimate these
new parameters k
s
; q
0
0
from MD simulations of DNA olig-
omers. For different temperatures and sequences, q
0
s; T is
estimated from the mean value of the conformational distri-
bution, and ks; T is estimated by inverting the covariance
matrix (Eq. 2). Because Eqs. 6 and 7 have a linear temper-
ature dependence, the parameters can then be computed by
linear regression.
Simulations of DNA oligomers
The protocol for the MDsimulations was chosentobe as close
as possible to that used by the ABC consortium (16,17). The
oligomers used in the simulation were 18-mers built from
tetranucleotide repeats: an oligomer termed xyzw has the
sequence GCzw xyzw xyzw xyzw GC (uppercase letters
are basepairs conserved in all oligomers). Thus, the oligomer
AAAC has the sequence GCACAAACAAACAAACGC. To
eliminate possible end effects, we excluded the four terminal
basepairs at either end of the oligomers from our analysis.
We simulated four 18-mers of dsDNA (AAAC, AGAT,
GCGC, GGGG) at ve different temperatures (273 K,
283 K, 300 K, 325 K, and 350 K) for 50 ns each. This
data set contains all unique dinucleotide sequences and
includes the inuence of different anking sequences for a
limited sample (see Table S1 in the Supporting Material).
The two types of basepair (AT and GC) appear in nine
different trinucleotides and are used for studying intra
parameters. The 10 unique dinucleotides appear within sin-
gle tetranucleotides, except for AA, which appears in two
contexts (AAAC and CAAA, both in the oligomer
AAAC). Our data set is therefore less comprehensive than
the ABC study at room temperature but more varied than
the previous generation of MD simulations (22).
Each oligomer was constructed in the B-DNA conforma-
tion and simulated in Amber (24) for 50 ns in 150 mM
KCl. The parameters and protocol of the simulations can
be found in Lavery et al. (17). In particular, the DNA force
eld includes the parmbsc0 backbone parameters (15). We
assumed that these force elds remain valid for double-heli-
cal DNA in the temperature interval under consideration.
Water was modeled with the TIP4P/Ew model (25), which
was designed to reproduce the structural and dynamical prop-
erties of water in a broad temperature range (273350 K), in
contrast to the SPC/E model used in most ABC runs. The
results obtained with both solvents were compared at room
temperature and exhibited no signicant difference (17).
The temperature and pressure, P 1 atm, during the sim-
ulations were controlled with the Berendsen algorithm (26).
The simulations were stable at all temperatures with the
same time step, t 2 fs. Note that 350 K is beyond the
experimental melting temperature of short oligomers;
melting does not occur in the simulations, due to the limited
sampling time and/or to limitations in the force elds.
Analysis of the trajectory
DNA conformations were analyzed with the program
Curves (27), which uses commonly accepted conventions
for the denitions of the helical parameters (18,28). From
1906 Meyer et al.
the distributions of the basepair and basepair-step parame-
ters, we checked the validity of the harmonic approximation
of the free energy (Eq. 1) at all sampled temperatures (see
Figs. S2 and S3, and the detailed discussion in the Support-
ing Material). For all sequences, we then computed the
mean values and covariance matrices, which were inverted
to compute the stiffness matrices (Eq. 2). The entropic
and enthalpic parameters were estimated by linear regres-
sion of 1), each element of the 6 6 stiffness matrix (Eq.
6) and 2), each degree of freedom of the equilibrium values
(Eq. 7). We included a systematic estimation of statistical
errors for the computed quantities by the block-averaging
method (29). We used the weighted-least-squares tting
algorithm (30), which provides error estimates on the tted
model: the latter is therefore depicted by a shaded area
corresponding to one standard-error deviation rather than
a single line (see, for instance, Fig. 1 B). To discriminate
the cases where the effect of temperature can be reliably
estimated, we introduced a criterion based on the f-test
(30), a statistical test that compares the accuracy of the
two-parameter enthalpic entropic model with that of a
one-parameter enthalpic model. The different steps of this
analysis procedure, as well as its validation on articially
generated data, are described in more detail in section II
of the Supporting Material.
Effects of superhelical stress in biological DNA
We augment a thermodynamic model that efciently pre-
dicts the local opening properties of superhelical DNA
(19) by adding the temperature and basepair step depen-
dencies of torsional energetics, the computation of which
is included in this article. Here, we briey describe the
model, and we refer the reader to a study by Jost and
co-workers (19) for more details on the formalism.
Local opening of DNA basepairs can be described by
the Benham model, a thermodynamic model of DNA under
superhelical stress that couples the standard thermody-
namic description of basepairing with the torsional stress
energetics (31). In the limit of long sequences, this model,
dened in a superhelical density-imposed ensemble, is
equivalent to a similar model in the torque-imposed
ensemble where the constant applied torque can be
computed self-consistently for a given superhelical density
(19,32). In this Ising-like model, the energy of a DNA
conguration constrained by a torque, G, is given by
H H
ZB

X
i
G
2
2C
q
i

G
2
2K
i
Gw
i
1 q
i
; (8)
where H
ZB
is the Zimm-Bragg Hamiltonian describing the
denaturation of unconstrained DNA (13), q
i
1 if the base-
pair step i is open and 0 if it is closed, and C 3:1k
B
T is the
torsional stiffness of unpaired basepair steps. The sequence-
dependent torsional stiffness, K
i
, of double-stranded steps
and the natural helical twist, w
i
, given in Tables S9 and
S10, are temperature-dependent and have been computed
from Tables S5S8. Computations of equilibrium properties
for a givensequence (like the local openingprobabilities hq
i
i)
are then performed using the transfer-matrix method (19).
Coarse-graining the nanoscale elasticity to the
persistence length
Starting from the computed nanoscale stiffness parameters,
we neglect the inuence of the intrabasepair deformations
on the large-scale elasticity and consider only the step uc-
tuations. In an ideal B-DNA helix, only the two bending an-
gles (tilt, t, and roll, r) contribute to the large-scale bending
of the molecule:l
p
2b=ht t
0
2
i hr r
0
2
i, where
bx0:34 nm is the average rise. This relation is, however,
not true for the real molecule, where all six step degrees
of freedom contribute to this bending because of local dis-
tortions. We use a coarse-grain calculation (20) that takes
these effects into account to compute the persistence length
from the nanoscale parameters. The simpler tilt-roll relation
is used to compute error bars. Note that because the devia-
tions from the ideal B-DNA helix remain small, the discrep-
ancy between the two methods is typically <5% (20).
From the sequence-dependent parameters obtained from
the simulations, the sequence-neutral persistence length is
computed by averaging the mean rise and the angle uctu-
ations (covariance matrix) over the 16 possible dinucleo-
tides. The computed values are in the same order of
magnitude as the experimental values, but signicantly
lower (~40 nm instead of ~50 nm, or 120 instead of
150 bp). Similar deviations have already been observed by
Becker and Everaers (20), and it is unclear whether this
effect is a consequence of the MD-estimated microscopic
uctuations or the calculation method. We assumed that
this issue does not inuence the computed temperature
dependence and therefore rescaled them by their value at
278 K, following Theodorakopoulos and Peyrard (9).
Hybrid model
We construct a hybrid model by including the temperature
dependence of the double-helical stiffness, computed in
the previous paragraph, in a recently proposed model (9)
TABLE 1 Sequence-averaged standard deviations of the
different degrees of freedom
Buckle Propel Opening Shear Stretch Stagger
11.62
9.37
4.53
0.30 A
0.12 A
0.43 A
Tilt Roll Twist Shift Slide Rise

4.46
7.10
6.76
0.71 A
0.71 A
0.35 A
All values given are at 300 K, s

i

hq
i
q
0
i

2
i
T0
q
. These reference
values were used to express all covariance/stiffness elements in dimension-
less units.
that accounts for the reduction of the effective persistence
length of DNA by transient denaturation bubbles. We
describe here the main lines of the calculation; more detail
can be found in Theodorakopoulos and Peyrard (9).
We model DNA polymer elasticity in terms of a heteroge-
neous Kratky-Porod (KP) (33) chain of N segments of
length a with a congurational energy,
H
X
N
j 1
J
j;j1
~
R
j
,
~
R
j1
; (9)
N 1 basepairs (considered as point monomers for
simplicity), and contour length L Na.
~
R
j
is a unit vector
joining the jth to the j 1 th basepair, and the local stiff-
ness constants,
J
j; j1

1 q
j1
J q
j1
J
0
;
are weighted averages of hard (J) and soft (J
0
) couplings,
according to the probability q
j1
that the j 1 th basepair
is in the unbound (open) state. The local melting fractions,
fq
j
g, are computed in terms of the Peyrard-Bishop-Dauxois
model (12,34), using only the sequence information and
the molar ion concentration, c, with no further adjustable
parameters (35,36).
The average end-to-end distance can be numerically
computed for the heterogeneous KP model (9), because
the correlation functions factorize. It is then possible to
use it to extract an effective persistence length, l, from the
relationship <r
2
> 2lL 2l
2
1 e
l=L
; which is valid
in the continuum limit of the homogeneous KP chain
(known as the wormlike chain), and should be quantitatively
adequate as long as l[a.
The above scheme was used in the Theodorakopoulos
and Peyrard model (9), with c 0:004; J 6
10
12
erg; and J
0
0:14 10
12
erg, to compute the effec-
tive persistence length of the phage fragment studied in
Geggier et al. (37). In this work, we incorporated the tem-
perature dependence of the local stiffness constants arising
from harmonic uctuations, i.e.,
J/J
k
b
T
k
b
278K
; (10)
where k
b
T is the temperature-dependent double-helical
bending stiffness, as estimated from the MD simulations.
RESULTS
We simulated four 18-mers of DNA of different sequences
at temperatures of 273 K, 283 K, 300 K, 325 K, and
350 K at xed pressure. The atomistic trajectories were
analyzed with the conformational analysis software
Curves (27), which provided distributions of intra (inter-
nal basepair) and step (basepair step) parameters. Our data
set includes all unique mono- and dinucleotide sequences.
However, basepairs and basepair steps can be inuenced
by the anking sequences and in some cases (17) can lead
to bimodal parameter distributions. We observe such effects
(see Fig. S3) and deal with them by separately analyzing
each trinucleotide or tetranucleotide sequence fragment
(see Table S1). In this case, for the oligomers we studied,
A
AGA
shear
B
GGGG
roll
stiffness stiffness
covariance/T covariance/T
FIGURE 1 Linear tting of the stiffness (upper)
and corresponding temperature dependence of the
variance (lower) for two typical diagonal matrix el-
ements. (A) Intrabasepair parameter shear
(sequence AGA). (B) Step parameter roll
(sequence GGGG). In both cases, the data points
rule out a purely enthalpic model (dotted line).
The temperature dependence of the distribution is
compatible with a linear entropic contribution,
within the statistical errors of the simulations
(shaded area). This contribution is stronger within
the basepair (A) than for the step parameters (B).
Note that we plot the covariance divided by the
temperature, which is constant for a purely en-
thalpic phenomenon. All quantities are expressed
in reduced units, dened using the corresponding
standard deviation s
i
(see Table 1).
1908 Meyer et al.
the distributions can indeed be approximated as single
Gaussians, compatible with the harmonic approximation
of the free energy. The covariance matrices are inverted to
compute stiffness matrices, from which we estimate the
entropic contribution according to Eq. 6. We include a sys-
tematic estimation of the statistical errors for the matrix
elements, which are tted independently of each other
(see Models and Methods). Fig. 1 shows typical examples
of such ts for diagonal matrix elements of intra (Fig. 1
A) and step (Fig. 1 B) parameters, where the entropic contri-
bution is apparent.
Note that the different degrees of freedom cannot be
analyzed separately. The passage between the stiffness and
the covariance elements involves a matrix inversion, which
couples the 6 6 degrees of freedom and their respective
levels of statistical noise; an example of the complete tting
procedure is shown in Fig. S5. The good agreement between
the tted model and the data points in both representations
is therefore nontrivial and validates the approximations
used in our analysis. We now discuss separately the results
obtained for the intra and step parameters.
Internal basepair exibility
In this section, we describe the effect of temperature on in-
ternal basepair exibility. As explained previously, we esti-
mate the entropic contribution separately for each sequence
fragment (here trinucleotides) where the harmonic approxi-
mation is valid. Fig. 2 shows the temperature evolution of
the stretch and opening stiffnesses, averaged over the A-T
(Fig. 2 A) and C-G (Fig. 2 B) basepairs. Note that we use
reduced units, dividing all coordinates by their average stan-
dard deviations, s
i
, at T
0
300 K (Table 1).
We were not surprised to nd that the values depend
strongly on the type of basepair, reecting the different
number of hydrogen bonds (three for C-G and two for
A-T). The effect of temperature emerges in all cases as a
regular linear decrease of the stiffness. We measure the
strength of this entropic contribution by comparing it to
the enthalpic part (Eq. 6). For stretch (Fig. 2, upper), this
ratio is ~0.5 at room temperature, i.e., the stiffness has
already dropped to half its value at 0 K. For opening
(Fig. 2, lower), the effect is weaker (~0.3), and still more
so for the other degrees of freedom.
Interestingly, the inuence of base sequence is slightly
different for these two parameters. For stretch, although
the stiffness constant for the C-G basepair at T
0
300 K
is ~50% larger than that for A-T, the entropic stiffness
(slope) is of the same order: the relative entropic contribu-
tion is therefore stronger for A-T. For opening, on the other
hand, the stiffness of the C-G basepair at room temperature
is three times stronger than that of A-T, but here, the
entropic contribution is also approximately three times
higher, so that its relative weight is similar.
At this level, the error bars still partly reect the structural
heterogeneity due to the anking sequences. We therefore
look at the results obtained for the different trinucleotides,
including the remaining degrees of freedom. These results
are summarized in Fig. 3 A, which gives the relative entropic
contribution (as described in the previous paragraph) plotted
against the stiffness at room temperature. In this representa-
tion, the behavior noted previously for opening is reected
in the fact that the two groups of points are distant horizon-
tally (different stiffness at T
0
) but have approximately the
same values on the vertical axis.
We systematically determine which parameters present a
detectable entropic contribution by applying a quantitative
criterion, comparing the accuracy of the enthalpic
entropic model with that of a purely enthalpic one (see
Models and Methods). An entropic contribution is detected
for all sequences with respect to the opening angle and the
three translational parameters stretch, shear, and stagger.
Opening
Stretch
A
A-T stiffness
B
C-G stiffness
FIGURE 2 Temperature dependence of the stiff-
ness for the stretch-stretch (upper) and opening-
opening (lower) diagonal terms on the A-T (A)
and C-G (B) basepairs. The reduced units are
dened using the average standard deviation of
each parameter, s
i
, given in Table 1. Here, the error
bars partly reect the structural heterogeneity due
to the anking sequences. The effect of tempera-
ture is strongest in these two degrees of freedom:
at room temperature, the average stiffness has
decreased by ~45% and 30%, respectively, with
respect to the values at 0 K (enthalpic stiffness).
The error bars are small enough to discriminate the individ-
ual (trinucleotide) sequences, conrming the inuence of
anking sequences on basepairs. The contributions are
strongest for stretch and opening, as noted previously; for
shear and stagger, the relative entropic contribution is
~0.25 at room temperature. In contrast, the two other
angular parameters exhibit very little or no detectable
entropic contribution (for details, see Table S2).
We then look at the equilibrium values, i.e., Eq. 7. The
parameters for which temperature induces an average
displacement are generally also those for which the stiffness
changes; the average stretch increases, as does the average
opening, but mainly for G-C. The other parameters do not
exhibit a systematic effect.
Basepair step exibility: parameterization of a
T-dependent rigid-basepair model
We now focus on the step parameters. Our oligomers
contain the 10 unique dinucleotides, each located within a
single tetranucleotide except for AA, which is present in
two sequence contexts (see Table S1).
The stiffness values for the step parameters are less uni-
form than those for the intra parameters, as is the effect of
temperature. For most sequences, an entropic contribution
is found for the two bending angles, tilt and roll, and for
the translational-parameter rise. The relative weight of this
contribution, dened in the previous section as (k
s

300 K)=k
h
, spans the intervals 0.250.5, 0.150.35, and
0.250.4, respectively, depending on the sequence. As an
example, Fig. 1 B shows the roll-roll diagonal elements of
the GG dinucleotide: the entropic contribution clearly
emerges from the statistical errors but is indeed smaller
than for shear. More surprisingly, the twist and shift stiff-
nesses exhibit a very low sensitivity to temperature.
The temperature dependence of the equilibrium step pa-
rameters is a question of considerable interest: a modication
of the spontaneous basepair stacking would be an evolu-
tionary challenge to organisms living at high temperatures.
Such a modication is not observed in the data. The only
parameter where a clear and sequence-independent tendency
emerges out of statistical noise is rise, which typically in-
creases by 23% in the temperature interval considered.
The equilibrium twist angle decreases by 12
for about
half of the sequences (e.g., TA, GC, and GA) and remains
approximately constant for the other ones, except in the
case of CG, where it increases, yielding an average decrease
of ~0.5
. The relation between the living temperature of or-

ganisms and supercoiling is discussed in the next section.
These results showthat the entropic effect on basepair step
elasticity is detectable, but more limited than that on basepair
elasticity. Note that this conclusion may also be affected by
slower equilibration times, which would require a greater
computational effort to resolve. Our analysis provides, to
our knowledge, the rst parameterization of a temperature-
dependent rigid-basepair model of DNAwhere the entropic
contributions are sequence-dependent. Altogether, the model
includes 156 nonzero entropic parameters in addition to the
270 parameters required for describing the elasticity at a sin-
gle temperature. It is also the rst model, to our knowledge,
that systematically includes condence intervals for these
parameters, which can be used to estimate error bars for
the quantities computed at the coarse-grain level, as we
show in the next paragraph. The model parameters are given
in Tables S5 and S7. Note that for the AA dinucleotide,
treated in two different sequence contexts, we arbitrarily
chose the values computed for the CAAA data set, where
the mean values are closer to those reported in crystallo-
graphic structures (21), and where we had more data points.
DISCUSSION
Temperature dependence of soft vibration modes
Our results demonstrate that the basepair internal stiffness
contains an important entropic contribution. This contribu-
tion is strongest for the stretch and opening degrees of
A B
T=273K T=273K T=650K
Extrap.
T=650K
Extrapolation
FIGURE 3 Entropic contribution to internal
basepair elasticity: (A) Relative entropic contribu-
tion to the stiffness, compared to the enthalpic stiff-
ness, for different sequences. Note that because the
units are parameter-specic, the values of different
parameters should not be compared. Although the
stiffness of the A-T and C-G basepairs is very
different in the opening direction, the relative
entropic contribution is similar. (B) Illustration of
the sequence-averaged range of uctuations in
the different degrees of freedom for T 273 K
and extrapolated to T 650 K, i.e., close to the
destabilization of the basepair (see text) to empha-
size the temperature effect. Note that the colored
rectangles in B correspond to the colors of the
basepair parameters in A. In the directions most
sensitive to temperature, stretch, and opening, the
uctuations become considerable.
1910 Meyer et al.
freedom (Fig. 3 B). Although A-T and G-C basepairs differ
by their enthalpic stiffness constants, they show similar
ratios, k
s
=k
h
(see also below). In an experimental setting,
the corresponding temperature sensitivity of soft vibration
modes might be accessible via Raman spectroscopy or
neutron scattering.
Few studies have dealt with the temperature dependence
of these spectra. Among those that have, the only one (to our
knowledge) that focuses on the frequency shifts in the dou-
ble helix is that of Grimm and Rupprecht (38). The neutron-
scattering spectra in that study exhibit a mode at ~1.1 THz
(4.5 meV) at 193 K, which softens to ~0.85 THz at 300 K,
i.e., a relative softening of ~2025%, or, in our terms, a rela-
tive entropic weight at room temperature of ~0.45, consis-
tent with our results. These numbers must be treated with
some caution, however, since the data only allow a qualita-
tive estimation of this softening. Another problem is the
indexing of the associated mode with respect to the base
and basepair degrees of freedom. From the distributions in
our simulations, we estimate that the internal basepair vibra-
tions have frequencies between 0.6 THz (opening) and
2.4 THz (stretch); the latter value has been reported in
Raman spectra (39). The step frequencies are generally
lower, in the 0.30.6 THz range. This indexation thus
requires further experimental support, but if the observed
mode corresponds indeed to an intrabasepair eigenmode,
it is in qualitative agreement with the entropic contribution
observed in our simulations.
The path to the melting transition
Our harmonic model is the rst-order approximation of the
DNA double-helical free energy. For obvious reasons, these
uctuations have to be bounded for the model to be valid:
the eigenvalues of the stiffness matrix have to be positive.
Within our description of the temperature-dependent stiff-
ness, kT k
h
Tk
s
, the double helix is thus predicted
to become unstable at T
s
k
h
=k
s
, known in the thermody-
namics literature as the spinodal temperature (40). Depend-
ing on the sequence, this temperature is estimated to be
between 550 and 660 K (Table S4). As a comparison, we
note that the spinodal temperature of liquid water at atmo-
spheric pressure, T
s
x600 K (41), is comparable to the value
we estimate for the double helix.
These large values signify that our biologically relevant
range of temperature (273350 K) remains far below the
limit of (meta)stability of the double helix, thus justifying
the approximation of constant k
h
and k
s
values: the melting
transition occurs before these values change signicantly.
Note that the double helix might still exist as a metastable
state for temperatures T
m
<T<T
s
.
The eigenvector associated with the decomposition of the
basepair at T
s
reects the directions for which the uctua-
tions are most sensitive to temperature. Fig. 3 B shows
that the basepair is destabilized in the stretch direction,
with a simultaneous partial opening. Qualitatively, these
directions may also indicate the direction of the kinetic
pathway toward melting.
With our MD simulations, sampling only double-helical
states, we rely in the following on standard theories of
DNA melting (11,13) to provide information on the relative
stability of single-stranded DNA and the thermodynamics of
bubble formation.
DNA in vivo: torque control of the local opening
probability
How do our results apply to DNA in living cells, in partic-
ular for extremophiles living at temperatures close to or
above 85
C, the typical melting temperature of uncon-

strained DNA (42)? In these organisms, many proteins
bind to double-helical DNA (43), and the opening of a sig-
nicant fraction of the molecule would probably be lethal. It
has been suggested that the double helix could be stabilized
by over-twisting (14). Maintaining DNA under superhelical
stress by specic enzymes could thus be one of the strategies
used by extremophiles to keep their DNA closed even
beyond 85
C. In vivo, the double helix is therefore not

free, as in our simulations, but in a constrained torsional
state. Consequently, one may wonder if this constraint can
affect the elastic and structural properties of the DNA
double helix and therefore modify local physicochemical
features like the binding constants of transcription enzy-
matic complexes.
To account for this effect, we use a recently proposed
method that allows the efcient computation of DNA
melting properties under superhelical stress imposed by an
applied torque (19). Augmenting the Benham model to
incorporate the temperature and basepair step dependencies
of twisting energetics studied in this article (see Models and
Methods), we evaluate the suitable torque needed to stabilize
the double helix at high temperatures and combine this esti-
mation with our previous results to estimate the elastic prop-
erties of the biologically relevant constrained helix.
As a function of the temperature, we compute the torque
needed to maintain an open fraction of 1% for the
genomic sequence of the extremophilic bacteria Thermus
thermophilus. This fraction corresponds to the typical open-
ing probability of Escherichia coli under standard physiolog-
ical temperature and superhelical density Dsx 0:06 (19).
This propensity to open is mainly located at gene promoters
and transcription start sites (19), enhancing gene expression
by polymerase enzymes (44). The constraining torque in-
creases with temperature (see Fig. S6), and as noticed in
previous studies by Benham (14), the model predicts that
the stability of the double helix can be maintained only up
to a critical temperature, which depends on the sequence
(data not shown). In Fig. 4, we plot the different contributions
of the average basepair step stability. At high temperatures,
thermal destabilization of pairing and stacking (gray dashed
line) is prevented by the contribution of the average twist
(dotted line) in the free energy of the constrained helix, main-
taining an almost constant basepair-step opening penalty of
0.2 kcal/mol (black solid line). Above a certain temperature,
the stabilization is limited by the strong twisting stiffness of
double-stranded steps (dash-dotted line), and the double
helix starts to melt. For T. Thermophilus, we nd a critical
temperature of 106
C, which is 15
Chigher than the melting

temperature in the absence of superhelical constraints. Inter-
estingly, this value is close to the highest living temperatures
of extremophilic organisms, suggesting that the stability of
the double helix may indeed be the limiting condition for
their existence.
At this limit, the stabilizing torque is surprisingly low
2 k
B
T. Although the resulting twist excess is important
in 1% of open basepairs (~33
/bp, close to the average twist

of B-DNA; see Fig. S7), the action of the torque results only
in a weak local twist excess for double-stranded regions,
ranging from 0.8
/bp for stiff basepair steps to 2.5
/bp for
soft ones. These values should be compared to the typical
twist standard deviation at room temperature (~6
/bp). For
comparison, adding 1
to the spontaneous twist per basepair

increases the superhelical density, Ds, by ~0.03. At the
rigid-basepair level of description, the properties of the
constrained helix can therefore be described in the regime
of linear response, where the equilibrium conformation is
displaced but the elasticity remains identical to that of the
relaxed helix. The stiffness parameters extracted from our
simulations, with their limited temperature dependence,
are therefore relevant to biological DNA. Using these
parameters, we compute the mean conformation at the crit-
ical temperature by relaxing the sequence-neutral helix
under the suitable torque. The resulting twist modication
is shown in Fig. 4: the external torque goes against the small
spontaneous twist decrease and results in an average in-
crease of ~1
in the interval under consideration, between

0
C and the critical temperature.

Note that the computed values correspond to the base
content of a specic organism and neglect the inuence of
other physicochemical parameters (ionic strength), as well
as alternate structural transitions that may contribute to
releasing the superhelical stress, in particular the B-Z tran-
sition (45). These contributions may therefore shift the
computed critical temperature in an organism-dependent
way (for instance, we estimate that T
c
90
C for E. coli).
However, since they are (at rst order) purely enthalpic,
they should not modify the basic mechanism.
Altogether, our results show (14) that under constraints,
double-stranded DNA can exist above the typical DNA
melting temperature. Our simulations make it possible to
compute its elastic and structural properties, which remain
close to those of unconstrained DNA at room temperature.
These observations explain how similar molecular mecha-
nisms of DNA-protein associations can exist for organisms
living in the whole temperature range (43). Locally, the
melting proles may also be driven out of equilibrium by
in vivo active processes, in particular transcription (46),
which may induce a transient superhelical addition (or
removal) in the same range of values computed here (47).
Large-scale exibility of unconstrained DNA
As a nal point, we address the (apparent) large-scale
bending exibility of DNA. At the wormlike-chain level,
the details of the molecule are averaged out and the exibility
is described by the bending persistence length, l
p
T, i.e.,
the bending correlation length of the polymer, which is
related to the bending stiffness, k
b
T, by the relation
l
p
T b
k
b
T
k
B
T
; (11)
where b is the length of a rigid element of the polymer.
Within the rigid-basepair model, these parameters can be
derived from the nanoscale structure and elasticity via a sys-
tematic coarse-graining procedure (20). We note that the
three step parameters most inuential in terms of large-scale
bending are also the ones for which the effect of temperature
is strongest. Not surprisingly, our estimate for the bending
stiffness, k
b
T, of the DNA double helix exhibits a weak,
but noticeable, temperature dependence (Fig. 5, gray line;
shaded area indicates the estimated error bar).
FIGURE 4 Stabilization of the DNA double helix in vivo by superhelical
stress for temperatures between 273 K and 380 K, where extremophile or-
ganisms exist. Although unconstrained DNA (gray dashed line) is unstable
at high temperatures, the contribution of the average twist in the free energy
(dotted line) maintains the stability of the double helix (black solid line) for
an additional 15 K, with an almost constant opening penalty of ~0.2 kcal/
mol. Above a critical temperature T
c
z 378 K (dots), the contribution
from the twist stiffness of the double helix (dash-dotted line) prevents
further stabilization. The superhelical stress opposes the spontaneous
decrease of the average twist, and results in an ~1
/bp increase in the

considered interval.
1912 Meyer et al.
Recent measurements (37) have indeed ruled out a tem-
perature-independent stiffness, k
b
(Fig. 5, dotted horizontal
line). However, the data included in Fig. 5 show a temper-
ature effect, which is considerably stronger than expected
from our results for the DNA double helix. In addition
to softening the double helix, the exibility of DNA may
also increase due to the presence of transient denaturation
bubbles that generate local kinks (7,8). In a different
study, two of the authors of the article presented here
quantitatively implemented this idea and reproduced the
dramatic decrease of the apparent stiffness in the premelt-
ing stage (Fig. 5, last datapoint) (9). However, at lower
temperatures, the bubble formation is insufcient to repro-
duce the experimental slope if the stiffness of the double
helix is assumed to be independent of T (Fig. 5, dashed
line).
Here, we present the results from a hybrid model, which
accounts for the temperature dependence of the double-
helical stiffness, as well as for the effect of bubbles,
and which is in good agreement with the experimental
data (Fig. 5, solid line). At physiological temperatures, the
two effects contribute equally strongly. Closer to the melting
transition, the unharmonic effects dominate.
CONCLUSION
The nanoscale mechanical structure and elasticity of the
DNA double helix can be inferred from the analysis of crys-
tallographic data and MD simulations. To date, there is
almost no information on how these properties vary with
temperature. In this article, we present results from MD sim-
ulations of DNA oligomers at different temperatures. We
show that entropy plays a signicant role in double-helical
elasticity, both inside the basepair and, to a lesser extent,
between successive basepairs.
At the internal-basepair level, this entropic contribution is
particularly strong in the stretch and opening degrees of
freedom. With increasing temperature, it results in a signif-
icant broadening of the harmonic uctuations in these direc-
tions of the basepair plane, which can be considered the
initial step of the DNA melting pathway.
At the basepair step level, the entropic contribution is
weaker but detectable. It is strongest in the roll, tilt, and
rise degrees of freedom relevant to the large-scale bending
rigidity. We include the resulting temperature dependence
of the persistence length of the DNA double helix in a
description of kinking in transient denaturation bubbles.
Our predictions for the effective large-scale bending stiff-
ness agree with the experimentally measured values for
standard physiological conditions up to temperatures close
to the melting temperature of DNA.
It is remarkable that living organisms thrive over an even
broader temperature range. To describe DNA in vivo, we
have assumed that biological organisms control the stability
of the double helix by regulating superhelical stress. We
have estimated the required values from the Benham model
and included the effect in our description of the nanoscale
properties. Overall, we nd a remarkably small temperature
dependence of the structure and exibility of genomic DNA,
compatible with conservation of protein-DNA binding
mechanisms (43).
SUPPORTING MATERIAL
Ten tables, nine gures, references (48 and 49) and Supporting Methods are
available at http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)
01022-9.
S.M. thanks K. Zakrzewska for her help with the simulation software. We
acknowledge the Pole Scientique de Modelisation Numerique for
computing resources.
This work was supported by the Agence Nationale de la Recherche grants
FSCF (ANR-12-BSV5-0009-01) (to R.E.) and Chrome (ANR-12-BSV5-
0017-01) (to R.L.).
REFERENCES
1. Davey, C. A., D. F. Sargent, ., T. J. Richmond. 2002. Solvent medi-
ated interactions in the structure of the nucleosome core particle at
1.9 A
resolution. J. Mol. Biol. 319:10971113.

2. Dixon, N. E., and A. Kornberg. 1984. Protein HU in the enzymatic
replication of the chromosomal origin of Escherichia coli. Proc.
Natl. Acad. Sci. USA. 81:424428.
3. Kim, Y., J. H. Geiger, ., P. B. Sigler. 1993. Crystal structure of a yeast
TBP/TATA-box complex. Nature. 365:512520.
4. Tansey, M. R., and T. D. Brock. 1972. The upper temperature limit for
eukaryotic organisms. Proc. Natl. Acad. Sci. USA. 69:24262428.
5. Takai, K., K. Nakamura, ., K. Horikoshi. 2008. Cell proliferation at
122
C and isotopically heavy CH4 production by a hyperthermophilic

methanogen under high-pressure cultivation. Proc. Natl. Acad. Sci.
USA. 105:1094910954.
FIGURE 5 Persistence length of sequence-averaged DNA, multiplied by
temperature and rescaled by its value at 278 K. In this representation, a
purely enthalpic stiffness yields a temperature-independent value (dotted
line). The markers ;and -are the experimental data points from Geggier
et al. (37) obtained with two types of ligase enzymes. Gray solid line: MD-
derived values of the persistence length, with the estimated error bars
(shaded area), i.e., the contribution from the double-helical elasticity;
dashed line: contribution predicted from the denaturation bubbles (9); black
solid line: the hybrid model, including both contributions.
6. Rubinstein, M., and R. Colby. 2003. Polymer Physics (Chemistry).
Oxford University Press, Cary, NC.
7. Yan, J., and J. F. Marko. 2004. Localized single-stranded bubble mech-
anism for cyclization of short double helix DNA. Phys. Rev. Lett.
93:108108.
8. Wiggins, P. A., R. Phillips, and P. C. Nelson. 2005. Exact theory of
kinkable elastic polymers. Phys. Rev. E Stat. Nonlin. Soft Matter
Phys. 71:021909.
9. Theodorakopoulos, N., and M. Peyrard. 2012. Base pair openings and
temperature dependence of DNA exibility. Phys. Rev. Lett.
108:078104.
10. Zimm, B. H., and J. Bragg. 1959. Theory of the phase transition
between helix and random coil in polypeptide chains. J. Chem. Phys.
31:526535.
11. Poland, D., and H. A. Scheraga. 1966. Phase transitions in one dimen-
sion and the helix-coil transition in polyamino acids. J. Chem. Phys.
45:14561463.
12. Dauxois, T., M. Peyrard, and A. R. Bishop. 1993. Entropy-driven
DNA denaturation. Phys. Rev. E Stat. Phys. Plasmas Fluids Relat.
Interdiscip. Topics. 47:R44R47.
13. Jost, D., and R. Everaers. 2009. Genome wide application of DNA
melting analysis. J. Phys. Condens. Matter. 21:034108.
14. Benham, C. J. 1996. Theoretical analysis of the helix-coil transition in
positively superhelical DNA at high temperatures. Phys. Rev. E Stat.
Phys. Plasmas Fluids Relat. Interdiscip. Topics. 53:29842987.
15. Perez, A., I. Marchan, ., M. Orozco. 2007. Renement of the
AMBER force eld for nucleic acids: improving the description of
a/g conformers. Biophys. J. 92:38173829.
16. Beveridge, D. L., G. Barreiro, ., M. A. Young. 2004. Molecular
dynamics simulations of the 136 unique tetranucleotide sequences of
DNA oligonucleotides. I. Research design and results on d(CpG) steps.
Biophys. J. 87:37993813.
17. Lavery, R., K. Zakrzewska, ., J. Sponer. 2010. A systematic molecu-
lar dynamics study of nearest-neighbor effects on base pair and base
pair step conformations and uctuations in B-DNA. Nucleic Acids
Res. 38:299313.
18. Dickerson, R. E. 1989. Denitions and nomenclature of nucleic acid
structure components. Nucleic Acids Res. 17:17971803.
19. Jost, D., A. Zubair, and R. Everaers. 2011. Bubble statistics and posi-
tioning in superhelically stressed DNA. Phys. Rev. E Stat. Nonlin. Soft
Matter Phys. 84:031912.
20. Becker, N. B., and R. Everaers. 2007. From rigid base pairs to semiex-
ible polymers: coarse-graining DNA. Phys. Rev. E Stat. Nonlin. Soft
Matter Phys. 76:021923.
21. Olson, W. K., A. A. Gorin, ., V. B. Zhurkin. 1998. DNA sequence-
dependent deformability deduced from protein-DNA crystal com-
plexes. Proc. Natl. Acad. Sci. USA. 95:1116311168.
22. Lankas, F., J. Sponer, ., T. E. Cheatham, 3rd. 2003. DNA basepair
step deformability inferred from molecular dynamics simulations.
Biophys. J. 85:28722883.
23. Becker, N. B., L. Wolff, and R. Everaers. 2006. Indirect readout: detec-
tion of optimized subsequences and calculation of relative binding
afnities using different DNA elastic potentials. Nucleic Acids Res.
34:56385649.
24. Pearlman, D., D. Case, ., P. Kollman. 1995. Amber, a package of
computer programs for applying molecular mechanics, normal mode
analysis, molecular dynamics and free energy calculations to simulate
the structural and energetic properties of molecules. Comput. Phys.
Commun. 91:141.
25. Horn, H. W., W. C. Swope, ., T. Head-Gordon. 2004. Development
of an improved four-site water model for biomolecular simulations:
TIP4P-Ew. J. Chem. Phys. 120:96659678.
26. Berendsen, H., J. Postma, ., J. Haak. 1984. Molecular dynamics with
coupling to an external bath. J. Chem. Phys. 81:36843690.
27. Lavery, R., M. Moakher, ., K. Zakrzewska. 2009. Conformational
analysis of nucleic acids revisited: Curves. Nucleic Acids Res.
37:59175929.
28. Olson, W. K., M. Bansal, ., H. M. Berman. 2001. A standard refer-
ence frame for the description of nucleic acid base-pair geometry.
J. Mol. Biol. 313:229237.
29. Flyvbjerg, H., and H. Petersen. 1989. Error estimates on averages of
correlated data. J. Chem. Phys. 91:461466.
30. Press, W., S. Teukolsky, ., B. Flannery. 2007. Numerical Recipes: The
Art of Scientic Computing, 3rd ed. Cambridge University Press,
New York.
31. Fye, R. M., and C. J. Benham. 1999. Exact method for numerically
analyzing a model of local denaturation in superhelically stressed
DNA. Phys. Rev. E Stat. Phys. Plasmas Fluids Relat. Interdiscip.
Topics. 59:34083426.
32. Jost, D. 2013. Twist-DNA: computing base-pair and bubble opening
probabilities in genomic superhelical DNA. Bioinformatics. In press.
33. Kratky, O., and G. Porod. 1949. Rontgenuntersuchung geloster faden-
molekule. Recl. Trav. Chim. Pays-B. 68:11061122.
34. Peyrard, M., and A. R. Bishop. 1989. Statistical mechanics of a
nonlinear model for DNA denaturation. Phys. Rev. Lett. 62:27552758.
35. Theodorakopoulos, N. 2010. Melting of genomic DNA: predictive
modeling by nonlinear lattice dynamics. Phys. Rev. E Stat. Nonlin.
Soft Matter Phys. 82:021905.
36. Theodorakopoulos, N. 2011. Bubbles, clusters and denaturation in
genomic DNA: modeling, parametrization, efcient computation.
J. Nonlin. Math. Phys. 18:429447.
37. Geggier, S., A. Kotlyar, and A. Vologodskii. 2011. Temperature depen-
dence of DNA persistence length. Nucleic Acids Res. 39:14191426.
38. Grimm, H., and A. Rupprecht. 1997. Lowfrequency dynamics of DNA.
Phys. B Cond. Mat. 234236:183187.
39. Urabe, H., and Y. Tominaga. 1981. Low frequency raman spectra of
DNA. J. Phys. Soc. Jpn. 50:35433544.
40. Chandler, D. 1987. Introduction to Modern Statistical Mechanics.
Oxford University Press, Cary, NC.
41. Eberhart, J., and V. Pinks, II. 1985. The thermodynamic limit of super-
heat of water. J. Colloid Interface Sci. 107:574575.
42. Frank-Kamenetskii, F. 1971. Simplication of the empirical relation-
ship between melting temperature of DNA, its GC content and concen-
tration of sodium ions in solution. Biopolymers. 10:26232624.
43. Grayling, R. A., K. Sandman, and J. N. Reeve. 1996. DNA stability and
DNA binding proteins. Adv. Protein Chem. 48:437467.
44. Schneider, R., A. Travers, and G. Muskhelishvili. 2000. The expression
of the Escherichia coli s gene is strongly dependent on the superheli-
cal density of DNA. Mol. Microbiol. 38:167175.
45. Zhabinskaya, D., and C. J. Benham. 2012. Theoretical analysis of
competing conformational transitions in superhelical DNA. PLOS
Comput. Biol. 8:e1002484.
46. Liu, L. F., and J. C. Wang. 1987. Supercoiling of the DNA template
during transcription. Proc. Natl. Acad. Sci. USA. 84:70247027.
47. Tsao, Y.-P., H.-Y. Wu, and L. F. Liu. 1989. Transcription-driven super-
coiling of DNA: direct biochemical evidence from in vitro studies.
Cell. 56:111118.
48. Jones, E., T. Oliphant, and P. Peterson. 2001. Scipy: Open source
scientic tools for python. http://www.scipy.org/.
49. Hunter, J. 2007. Matplotlib: a 2d graphics environment. Comput. Sci.
Eng. 9:9095.
1914 Meyer et al.
Lateral Membrane Diffusion Modulated by a Minimal Actin Cortex
Fabian Heinemann,
6
Sven K. Vogel,
6
and Petra Schwille*
Max-Planck Institute of Biochemistry, Martinsried, Germany
ABSTRACT Diffusion of lipids and proteins within the cell membrane is essential for numerous membrane-dependent
processes including signaling and molecular interactions. It is assumed that the membrane-associated cytoskeleton modulates
lateral diffusion. Here, we use a minimal actin cortex to directly study proposed effects of an actin meshwork on the diffusion in
a well-dened system. The lateral diffusion of a lipid and a protein probe at varying densities of membrane-bound actin was char-
acterized by uorescence correlation spectroscopy (FCS). A clear correlation of actin density and reduction in mobility was
observed for both the lipid and the protein probe. At high actin densities, the effect on the protein probe was ~3.5-fold stronger
compared to the lipid. Moreover, addition of myosin laments, which contract the actin mesh, allowed switching between fast and
slowdiffusion in the minimal system. Spot variation FCS was in accordance with a model of fast microscopic diffusion and slower
macroscopic diffusion. Complementing Monte Carlo simulations support the analysis of the experimental FCS data. Our results
suggest a stronger interaction of the actin mesh with the larger protein probe compared to the lipid. This might point toward
a mechanism where cortical actin controls membrane diffusion in a strong size-dependent manner.
INTRODUCTION
A pivotal property of cell membranes is their uidity, which
allows rapid lateral diffusion of lipids and proteins neces-
sary for a continuous mixing of membrane components
and diffusion-limited biochemical interactions. In 1972,
the Singer-Nicholson uid mosaic model described the
plasma membrane as a rather simple two-dimensional
(2D) lipid matrix in a uid state with a mosaic of embedded
proteins (1). Nowadays, this description of the plasma
membrane has emerged to a more rened view (2), which
includes proposed lipid nanodomains (rafts) (3), protein
crowding (4), and interaction of the membrane with the
cytoskeleton (5,6). As a possible consequence of this
complexity, lateral diffusion coefcients in cell membranes
are typically 5 to 50 times reduced, compared to diffusion
coefcients determined in simple reconstituted membranes
(6). In addition, strong heterogeneity of diffusion coef-
cients, generically often described as anomalous diffusion,
can be observed for many species (7). However, the indi-
vidual contribution of these factors to the decrease and vari-
ation in lateral diffusion is not well understood.
Actin (and spectrin) laments bound to the inner cell
membrane surface constitute the membrane skeleton (8)
and have been suggested to inuence the diffusion of lipids
and proteins within the membrane (5,6,911). The actin
meshwork is also supposed to prevent micrometer-scale
phase separation, as indicated by recent experimental (12)
and theoretical data (13,14).
The membrane skeleton mesh subdivides the plasma
membrane into compartments. For two cell types the mesh
was quantitatively characterized by electron microscopy,
resulting in median mesh diameters of 50 and 200 nm (8).
Based on single particle tracking experiments on
membranes of living cells, it was reported that on small
spatial scales, corresponding to the membrane skeleton
mesh diameters, the diffusion was relatively rapid. On larger
scales, the diffusion was reported to be greatly reduced. This
led to the proposal of the hop-diffusion theory (6,8,1517).
In this model, diffusing species undergo fast diffusion (with
similar diffusion coefcients to simple uid-phase model
membrane systems) conned within the areas dened by
the actin mesh. Diffusing species could only occasionally
escape an area delimited by actin, when thermal uctuations
would create a sufciently large gap. As a consequence, the
macroscopic diffusion coefcient would be reduced. A
direct collision between the actin laments and protruding
headgroups of the diffusing molecule (i.e., a membrane
protein) or a collision with membrane-bound actin anchors
in the membrane plane were proposed as possible interac-
tion mechanisms. A further in vivo study suggested that
the actin mesh inuences the probe mobility in a strong
size-dependent manner (18). It was found that oligomers
of a membrane protein diffuse much slower than the respec-
tive monomers. This differs from the conventional weak size
dependency of membrane diffusion according to the Saff-
man and Delbruck model (19), which is valid for simple
homogeneous membranes and thus, does not consider
distortions such as interactions with the cytoskeleton. The
aspect of size-dependent diffusion may be very important
for signaling and cell polarization, where receptor proteins
often oligomerize or bind a ligand at the cell membrane.
For example, in an oligomerized (activated) state, receptors
would be trapped by the actin mesh, resulting in signal
localization on the cell membrane, whereas in a monomeric
(inactive) state, the receptors would diffuse rather quickly in
Submitted November 14, 2012, and accepted for publication February 19,
2013.
6
Fabian Heinemann and Sven K. Vogel contributed equally to this work.
*Correspondence: schwille@biochem.mpg.de
Editor: Lukas Tamm.
Biophysical Journal Volume 104 April 2013 14651475 1465
the membrane. However, it should be mentioned that due to
the difculty in observing the diffusion at the necessary high
temporal and spatial resolution, which typically requires
labeling of the molecules with colloidal gold, the evidence
from particle tracking for hop diffusion is still under debate
(17,20).
In a further in vivo study, the inuence of cortical actin on
the lateral diffusion was characterized by uorescence
correlation spectroscopy (FCS) with variable spot size
(10,21). Depending on the diffusing probe, a different diffu-
sion behavior was reported. For a transmembrane protein
labeled with green uorescent protein, the FCS data indi-
cated a faster diffusion on a small scale and slower diffusion
on a large scale in an actin-dependent manner corresponding
to hop-diffusion. For putative raft markers, a model of trap-
ping in regions of slow diffusion (i.e., lipid nanodomains)
was reported. It was also discussed that for some probes,
both mechanisms act simultaneously and may mask each
other. Due to this simultaneous presence of several factors
modulating the diffusion in vivo, it is challenging to charac-
terize the effect of the membrane skeleton on the lateral
diffusion of lipids and proteins independent of other factors.
In addition to the structural complexity of cellular
membranes, the cellular response to drugs modulating actin
is also typically complex and can result in undesired and
unknown side-effects.
Here, we used a minimal in vitro system of membrane-
bound actin to directly study the isolated effect of actin on
the lateral diffusion of a lipid and a membrane-binding
protein by FCS. The presence of membrane-linked actin
reduced the mobility of both probes in a concentration-
dependent manner. We found that this reduction in mobility
was much more pronounced for the larger protein, compared
to the lipid. Comparison with the mobility reduction ob-
tained by pure solvent viscosity supports the model that
actin reduces the probe mobility strongly depending on the
probe size. Spot variation FCS (sv-FCS) results were consis-
tent with fast microscopic diffusion inside the actin mesh
and slower macroscopic diffusion, with a stronger mesh
connement for the protein. Experiments using myosin-II
laments, which contract and condense the actin laments,
show the possibility to control the actin mesh to locally
tune probe diffusion. These results suggest a modulating
function of cortical actin in cells that might, for example,
locally change signaling properties of the cell membrane.
Monte Carlo simulations of diffusion in a partially reective
mesh complement our experimental data.
Actin preparation and labeling
Actin monomers (Molecular Probes/Life technologies, Paisley, UK) and
biotinylated actin monomers (tebu-bio/Cytoskeleton, Denver, CO) were
mixed in a 5:1 (actin: biotin-actin) ratio. Polymerization of the mixture
(39.6 mM) was triggered in F-Buffer containing 50 mM KCl, 2 mM
MgCl
2
, 1 mM DTT, 1 mM ATP, 10 mM Tris-HCl buffer (pH 7.5). The bio-
tinylated actin laments were stabilized with Alexa-Fluor 488 Phalloidin
(Molecular Probes). A nal concentration of 2 mM (refers to monomers)
Alexa-488-Phalloidin-labeled biotinylated actin laments was obtained.
Myosin preparation
Myosin was puried as previously described (22) from rabbit skeletal
muscle tissue. A classical motility assay where myosins bound to a nitrocel-
lulose-coated glass surface of a perfusion chamber (tebu-bio/Cytoskeleton)
propel actin laments was used to test the activity of myosin. Myolament
assembly was triggered in reaction buffer containing 50 mM KCl, 2 mM
MgCl
2
, 1 mM DTT, and 10 mM Tris-HCl (pH 7.5).
Free-standing membranes
Free-standing membranes were prepared as described by Heinemann and
Schwille (23) (for details see the Supporting Material). All experiments
using the free-standing membranes were performed at 23.5 5 1.5
C
(measured in the solution over the objective).
Minimal actin cortex (MAC) preparation
2 mg of neutravidin (Molecular Probes) dissolved in 200 ml reaction buffer
was addedtothe free-standingmembranes andincubatedat roomtemperature
for 10 min. The sample was washed several times with >2 ml reaction buffer
to remove unbound neutravidin. 50 ml of 2 mM (refers to monomers) Alexa-
488-phalloidin-labeled biotinylated actin laments were then added to the
lipid bilayer and incubated for 1 h. The sample was carefully washed with
~12 ml reaction buffer to remove unbound actin laments (see also (24)).
Correlation of mobility reduction and actin
density
Three consecutive FCS measurements were conducted at the center of the
membrane spots: First, after generation of the free-standing membrane,
second, after linking of neutravidin, and third, after linking of actin. The
measurements at the different states were performed after the respective
washing steps. Each measured spot was numbered (see Fig. 2 A) and changes
in diffusion were related to changes in the same spot. The density of actin at
each numbered spot was measured by acquisition of uorescence images
under standardized conditions (for details see the Supporting Material).
Point - FCS on membranes
The measurements were performed at the center of each membrane spot.
Before each measurement, the focus was moved vertically in steps of
200 nm to determine the position of maximal uorescence intensity. At
this position, the uorescence was recorded for 3060 s. The vertical
stability of the membrane was maintained by using a cast iron stage (JPK
manual precision stage, JPK Instruments, Berlin, Germany). After each
experiment, the vertical positioning was controlled, and experiments that
showed vertical drift were rejected.
Decrease in lateral mobility induced by solvent
viscosity
In a rst set of experiments, the viscosity of the buffer surrounding both
sides of the free-standing membranes was changed by preparing mixtures
of the reaction buffer and sucrose. The change in diffusion coefcient
was measured by line-scan FCS (LSFCS) (25). This method does not
1466 Heinemann et al.
rely on the shape of the focal volume and therefore avoids artifacts induced
by distortions of the focal volume due to changes in the refractive index
(26). The mass ratios 0%, 10%, 20%, and 30% were used. Corresponding
viscosity and refractive index values for water-sucrose mixtures were ob-
tained from (27). A line with the center on a free-standing membrane patch
with a length of 11.52 mm and 512 pixels was continuously scanned for
300 s and a constant scan speed of 768 ms/line on a LSM510 Meta setup
(for details see the Supporting Material). Photon arrival times were re-
corded in the photon mode of a hardware correlator (Flex 02-01D,
correlator.com, Bridgewater, VA). The spatio-temporal correlation was
computed in MATLAB(The MathWorks, Natick, MA) and tted to a model
function as described by Ries, Chiantia, and Schwille (25). The diffusion
coefcient D and the radial waist u were directly obtained from the t.
Supporting materials and methods
Details of the Free-Standing membrane preparation, calibration of spot
variation FCS, the tting of the autocorrelation data, the alignment and
calibration of the FCS setup, and the Monte Carlo simulations can be found
in the Supporting Material.
Free-standing MAC
To investigate the effect of a membrane-bound actin cyto-
skeleton on the lateral diffusion of lipids and proteins in
the membrane independent of other factors, we built
a minimal system with membrane-bound actin laments
mimicking an actin cell cortex (24,28). We aimed to use
a geometry of the membrane where both sides of the lipid
bilayer were surrounded by buffer solution to avoid fric-
tional coupling with a support as present in supported
membranes (29). Additionally, the possibility to remove
unbound proteins and to exchange buffer solutions by
washing with buffer was desired. These requirements were
met by a newly developed free-standing membrane system,
where membranes were suspended over holes with 2.5 mm
in diameter (23). Biotinylated actin laments were coupled
via neutravidin to both leaets of the suspended membrane,
which contains biotinylated lipids (24) (Fig. 1). The density
of the actin meshwork can be controlled by varying the
amount of biotinylated lipids present in the lipid bilayer
(24). By this procedure, we obtained a dense lamentous
actin meshwork associated with the membrane, which we
refer to as the MAC (Fig. 2 A, right image).
Correlation of actin density and mobility
To characterize the effect of actin on the lateral diffusion of
lipid and protein species, two different probes, the small
labeled lipid Atto647N-DOPE and the larger membrane
binding protein CtxB-Alexa647 were used. The density of
actin varied on the different membrane spots and was quan-
tied by acquisition of standardized confocal images of the
uorescently labeled actin, as shown in Fig. 2 A. This quan-
tication allowed testing for a potential correlation between
actin density and a change in probe mobility.
In Fig. 2 B, the change in the diffusion coefcient is
plotted versus the actin density. Each symbol corresponds
to a measurement on a single free-standing membrane
spot. The values were normalized to the diffusion coefcient
measured after the addition of neutravidin, i.e., directly
before the addition of actin (typically the addition of neutra-
vidin resulted in a slightly reduced mobility of up to 10% for
both probes compared to the membrane alone, data not
shown). For both probes, the graph shows a clear correlation
between reduced mobility and increasing actin density.
When comparing the normalized reductions in mobility at
high actin densities (regime II in Fig. 2 B), the maximal
effect was ~3.5-fold stronger for the protein CtxB-
Alexa647, compared to the lipid Atto647N-DOPE.
For a direct comparison of the results obtained using the
two probes, the changes in the nonnormalized diffusion
coefcients before and after actin coupling are helpful.
Before addition of actin, the diffusion coefcients were
D
lipid
9.9 50.6 mm
2
s
1
and D
CtxB
5.9 50.9 mm
2
s
1
(mean 5 standard deviation). The ratio of the two diffu-
sion coefcients corresponds to D
CtxB
/D
lipid
0.6.
Comparable ratios for molecules of similar difference in
size in a uid membrane have previously been reported
(30,31). It is important to mention that there is no general
theory that describes both, the diffusion of the very small
lipid and the larger protein, including the transition
between the two size regimes. The diffusion of small
molecules comparable to the size of the solvent is
described by the free area theory (32) showing an expo-
nential dependency on the size of the diffusant. For larger
molecules, the Saffman-Delbruck equation becomes valid,
showing a weaker logarithmic dependency of the lateral
diffusion coefcients on the radius of the diffusing species
(19). In the presence of the actin mesh, the diffusion of
FIGURE 1 Scheme of the MAC. The lamentous biotinylated (blue)
actin is coupled via neutravidin to the free-standing membrane (Egg PC)
containing biotinylated lipids (DSPE-PEG(2000)-Biotin). For a more
compact display the binding of actin is shown only to the upper leaet of
the membrane. In the experiments, actin presumably binds to both leaets,
because both membrane sides are accessible and contain biotinylated lipids.
Lateral Membrane Diffusion Modulated by a Minimal Actin Cortex 1467
both probes is slowed down to D
lipid
4.8 5 0.4 mm
2
s
1
and D
CtxB
0.7 5 0.1 mm
2
s
1
(high actin density
regime II in Fig. 2 B, mean 5 standard deviation). The
ratio decreases drastically from D
CtxB
/D
lipid
0.6 to
D
CtxB
/D
lipid
0.15, indicating that upon presence of the
actin mesh, the lateral diffusion of the larger protein is
more strongly affected compared to the lipid.
A probable cause for the stronger effect of actin on the
protein diffusion is a direct interaction of the part of the
probe that protrudes from the membrane plane with the actin
laments. In the case of CtxB, the possible interaction might
take place between the bulky protein and the actin. CtxB has
a height of ~3.5 nm (normal to the membrane plane) (33)
and an attached Alexa647 with a size of ~0.8 2.0 nm.
For the labeled lipid Atto647N-DOPE, the major interaction
with the actin could include the lipid headgroup and the
attached uorescent dye Atto647N with a size of 0.8
1.5 nm (structure from (34)). To compare these dimensions,
we estimated the distance of actin to the membrane. Neutra-
vidin has a diameter of ~5.0 nm and four biotin-binding sites
(35). Biotin at the biotinylated lipid is attached to a poly(eth-
ylene glycol) (PEG) spacer with a molecular mass of 2 kDa.
In the so-called mushroom conguration, which is predom-
inant at low PEG lipid concentrations, a size of ~2.8 nm is
expected (half sphere radius of the PEG) (36). Hence, we
roughly estimate an average distance of actin to the
membrane in the range of 38 nm.
Another possible source for the reduction in mobility is
the interaction with less mobile actin anchor points in the
membrane plane (in our case, biotinylated lipids linked to
actin), which might represent obstacles for the diffusion
(16,17,37). A variation of anchor points by reducing the
concentration of biotinylated actin monomers from 20% to
5% did not result in a difference in mobility reduction
(Fig. S2), indicating that this may not be the dominant
source of mobility reduction in our experiments. Neverthe-
less, the amount of 5% biotinylated actin monomers may
still be too high to completely rule out a role of the
anchoring lipids in mobility reduction. In addition, trans-
membrane proteins could be used as anchors in future exper-
iments as they would couple the actin meshwork to both
lipid leaets and possibly enhance the potential inner
membrane obstacle effect.
In summary, we assume that the major source of the
mobility reduction is a direct interaction of the actin la-
ments and components of the diffusing species, which
protrude from the membrane plane. This direct interaction
also explains the observed large difference in mobility
FIGURE 2 Correlation of the mean actin uorescence (actin density) and the decrease in lateral membrane diffusion. (A) Confocal uorescence images
of free-standing membranes containing Atto647N-DOPE (left) and Alexa-488-phalloidin-labeled actin laments (right) exhibiting a dense lamentous
meshwork associated with the membrane. Diffusion coefcients D
i
were determined by FCS in the center of numbered free-standing membrane
spots i. The corresponding actin density I
i
was determined from the average uorescence intensity in a circular area of the respective spot i. Scale
bars: 10 mm. (B) Relative change in diffusion coefcients plotted versus the mean actin intensity (measure of actin density) for the labeled lipid
(red circles) and the membrane binding protein (blue squares). Each point represents a pair of measurements of D
i
and I
i
, the solid line is an empirical
t with an asymptotic function. Changes in diffusion are shown after normalization to D
0
, the diffusion coefcient of the respective probe in the membrane
after neutravidin addition but before actin coupling. The gray dotted line separates regime I (left) and regime II (right). (C and D) Potential effects on the
shape of the FCS autocorrelation curves were investigated by classifying the data according to the actin density (I low density, II high density). FCS
curves were class-averaged (black) and tted with a model for 2D diffusion assuming single-component membrane diffusion (Eq. S2 or S3 in the Support-
ing Material, gray). At the timescales relevant for 2D diffusion, the theoretical models were a reasonable t to the experimental data for the lipid (C) and
the protein (D) independent of the actin density.
reduction for the two probes. For the bulky protein, it is
much more likely to collide with the actin laments,
compared to the small lipid probe.
Our results are in agreement with previous in vivo exper-
iments. Iino et al. (18) reported that the diffusion coefcient
of the membrane protein E-cadherin in different oligomeric
states decreases strongly with the size of the oligomers. It
was suggested that this effect is caused by the interaction
of the diffusing species with the membrane skeleton under-
neath the plasma membrane (6,18,37). For the larger oligo-
mers, the probability to collide with the actin mesh would
increase and result in a stronger connement in the mesh
and thus a stronger reduction of mobility.
Autocorrelation data correspond to single-
component diffusion
According to the hop diffusion model (6,17,8,37), species
such as membrane proteins or lipids diffuse relatively
rapidly inside the compartments formed by the membrane
skeleton but slow on a larger spatial scale, due to the barriers
formed by the membrane skeleton. In FCS, the simultaneous
presence of fast and slow diffusion can result in deviations
from one-component diffusion. In particular, at actin mesh
diameters comparable to the detection spot size, signicant
deviations from one-component diffusion may occur. In
theory, this was demonstrated by previous Monte Carlo
simulations (21) and further supported by the simulations
shown here (cf. the corresponding section below).
To test, whether we see these potential deviations from
single-component diffusion experimentally, we further
analyzed the autocorrelation data of the experiments shown
in Fig. 2. First, the experimental FCS data were classied ac-
cording to the respective actin density (see Fig. 2 B). In the
low actin density regime (I), the images showed a dense
meshwork with laments at the limit of optical resolution.
Hence, at this low density, the FCS detection spot size was
approximately comparable to the actin mesh diameter. For
higher densities, the appearance became more homoge-
neous, indicating a distance of the laments below the
diffraction limit. FCS results obtained at low (I) and high
actin densities (II) were normalized and averaged separately.
This procedure should allow for the detection of smaller
deviations from single-component diffusion, compared to
the consideration of single autocorrelation curves.
The averaged FCS curves were then tted with the single-
component FCS models (Eqs. S2 and S3 in the Supporting
Material) for the lipid and the protein, respectively (Fig. 2,
C and D). For both probes and at low and high actin densi-
ties, the single-component t models and the experimental
FCS data coincided at timescales >1 ms, which are relevant
for membrane diffusion. Thus, we do not directly observe
proposed deviations from one-component diffusion.
However, the Monte Carlo simulations presented in the
section below showed that for a rather weak interaction
between the actin and the diffusing species, deviations
from single-component diffusion are small, even in cases
of comparable observation spot size and actin mesh diam-
eter. Hence, from the constant spot FCS data alone, we
can neither conrm nor exclude the presence of fast micro-
scopic and slower macroscopic diffusion (hop diffusion) in
our system.
Decrease in lateral mobility induced by solvent
viscosity
Previous experimental data showed that solvent viscosity h
affects lateral membrane diffusion (3840), but so far, the
effect of the viscosity on the lateral diffusion of a protein
and a lipid has not been directly compared. Thus, we
changed the viscosity of the surrounding solvent and
measured the inuence on the diffusion of the lipid and
the protein. The aim of these experiments was to test
whether solvent viscosity could also induce similar mobility
differences for the two probes, as observed in the presence
of actin.
Increasing the solvent viscosity with sucrose also
increases the refractive index (27). Therefore, point-FCS
is not well suited for measuring lateral diffusion coefcients
in more viscous solutions, because for water immersion
objectives, a dened shape of the observation volume is
only guaranteed for solutions matching the refractive index
of water. Already slightly higher refractive indexes result in
an increased size and in a deformed shape of the observation
volume (26). To avoid these potential pitfalls, LSFCS was
used (Fig. 3 A), a technique belonging to the class of image
correlation spectroscopy, and therefore not depending on
a known size of the focal volume (25). In LSFCS, the diffu-
sion coefcient D and the effective 2D waist u are directly
obtained from a t. The effective waist indeed increases in
size with increasing viscosity (Fig. 3 B).
In Fig. 3 C, the absolute changes in lateral diffusion coef-
cient in the presence of sucrose from 0% to 30% (m/m) are
displayed. In agreement with our results from point FCS, the
absolute diffusion coefcient at the viscosity of water is
lower for the protein, compared to the lipid. With increasing
viscosity, the diffusion of both probes is reduced, due to the
increasing viscous drag. To compare the effect of solution
viscosity on the lipid and the protein, the normalized
changes are shown in Fig. 3 D. Comparing the normalized
changes of the lipid and the protein in diffusion clearly
reveals that the relative changes in mobility of both probes
are similar, and therefore in contrast to the strong difference
observed in the presence of the actin mesh (Fig. 2 B).
Conned diffusion in the actin mesh indicated by
spot variation FCS
In recent years, it was shown that FCS with a variable size of
the detection area is a useful tool to characterize diffusive
processes in membranes, where the diffusion coefcient
depends on the scale. In particular, Monte Carlo simulations
(21,41), an analytical treatment (42), and in vivo experi-
ments (10,34,43) demonstrated that sv-FCS can provide
additional information about the mode of lateral membrane
diffusion. In sv-FCS, autocorrelation curves are recorded at
different sizes of the focal spot u
2
. For free diffusion, a plot
of the diffusion time t
d
(u
2
) versus the focal area u
2
will
yield a linear relationship t
d
(u
2
) 1/(4D)u
2
where the
slope is 1/(4D) (compare Eq. S4 in the Supporting Material).
However, in many physiological cases, such as diffusion
conned by the membrane skeleton or trapping by microdo-
mains, the apparent diffusion coefcient D changes with the
observation scale. For diffusion conned by a mesh, rela-
tively fast diffusion D
micro
inside the single mesh cells is ex-
pected at small scales, whereas at larger scales, the slow
diffusion D
macro
<D
micro
in between different meshes domi-
nates. Even if this transition cannot be observed directly, due
to the diffraction limit of conventional confocal micro-
scopes, extrapolation of the recorded relationship t
d
(u
2
)
to u
2
0 allows to distinguish conned mesh diffusion
from trapping. For conned mesh diffusion, the extrapolated
intercept t
d0
t
d
(0) will be at t
d0
< 0, due to the intrinsic
transition to fast diffusion at small scales (21). For trapping,
in our experiments possibly induced by transient binding of
the diffusing species to the actin laments, an intercept
t
d0
> 0 is expected (21,41).
We used sv-FCS to characterize the predominant mode of
diffusion for the two probes. Fig. 4, A and B, shows the
measured diffusion times for the labeled lipid Atto647N-
DOPE and the protein CtxB-Alexa647 for the free
membrane (open symbols) and in the presence of actin (solid
symbols). In both cases, Eq. S2 or S3 (see the Supporting
Material) for the lipid and the protein, respectively, t the
autocorrelation data well at all spot sizes. For the free
membranes, the linear extrapolation to t
d0
intersects at the
origin (Atto647N-DOPE: t
d0
6 5 209 ms, CtxB-
Alexa647: t
d0
62 5 329 ms), as expected for free diffu-
sion with a constant diffusion coefcient independent on the
observation scale. In the presence of membrane linked actin,
the extrapolation intersects at slightly negative t
d0
values for
the lipid (Atto647N-DOPE: t
d0
501 5 241 ms) and
pronounced negative t
d0
values for the protein (CtxB-
Alexa647: t
d0
7880 5 2400 ms). This supports the
interpretation that a conned mesh-like diffusion is present
FIGURE 4 Spot variation FCS supports the view
that upon presence of the actin mesh a transition
from fast diffusion at a subdiffraction scale to
slow macroscopic diffusion occurs. (A) Before
actin linking, the diffusion time of the labeled
lipid changes linearly with the spot size (open
circles) and the extrapolation of the tted line
to zero spot size intersects at the origin
(t
d0
6 5 209 ms). This is in agreement with
a constant diffusion coefcient also at subdiffrac-
tion scales. In the presence of actin (solid circles),
the extrapolation to the origin intersects at a slightly
negative diffusion time (t
d0
501 5241 ms). This could indicate a transition from fast diffusion (similar to the diffusion without actin) at a subdiffraction
scale to slower diffusion on a larger scale (due to the presence of actin). (B) Furthermore, for the protein (open squares) the diffusion time in the membrane
without actin scales linearly with the spot size and the t intersects at the origin (t
d0
62 5 329 ms). In the presence of the actin mesh (solid squares)
the extrapolated diffusion time at zero spot size is strongly negative (t
d0
7880 5 2400 ms), consistent with a transition from faster to slow diffusion
at a subdiffraction scale, due to the presence of the actin mesh. All errors are standard deviations.
FIGURE 3 Effect of solution viscosity on the diffusion of the protein and
the lipid (in the absence of actin). The viscosity of the solution was varied
using different concentrations of sucrose. (A) Diffusion coefcients were
determined by LSFCS, because compared to point FCS this method is
less susceptible to distortions of the focal volume due to the use of a solution
with a refractive index deviating fromthat of water. Aline was continuously
scanned and the spatio-temporal correlation in a free-standing part
(indicated by the two strokes) was computed and tted with a model func-
tion for 2D diffusion. Scale bar 5 mm. (B) Increase in focal waist depending
on the viscosity of the buffer as determined by LSFCS. (C) Absolute
changes in the lateral diffusion D of the labeled lipid (circle) and the protein
(square). The mobility of both probes decreases with increasing viscosity.
(D) Plotting the relative changes in mobility shows that, in contrast to the
experiments in the presence of actin, the diffusion of both probes is affected
to a comparable extent. The diffusion coefcients were normalized to the
respective diffusion coefcient at 0% sucrose. All error bars are standard
deviations.
in both cases, and a transition from fast, free diffusion to
slower macroscopic diffusion occurs at subdiffraction
scales. In agreement with the results obtained with a constant
detection spot size, the more pronounced negative intersec-
tion for the protein indicates a stronger connement for the
protein by the actin mesh compared to the lipid.
Monte Carlo simulations of diffusion in a mesh
To examine the possible inuence of an actin mesh on the
experimental FCS autocorrelation data, we performed
Monte Carlo simulations of 2D diffusion in a meshwork,
as illustrated in Fig. 5. The simulated mesh was partially
reective to model the difference in the conning effect of
actin on different diffusing species. Previous simulations
of similar type by Wawrezinieck et al. (21) already showed
that different modes of diffusion can be distinguished by
sv-FCS.
However, the availability of FCS instrumentation capable
of sv-FCS is still limited and it is also insightful to study the
effect of diffusion in a meshwork monitored by conven-
tional FCS with a constant spot size. In particular, we aimed
to elucidate three issues in more detail: First, we wanted to
characterize the expected shape of the autocorrelation data
and previously described deviations from a one-component
model (21), depending on the mesh density in a quantitative
manner. Second, we wanted to test whether a variable
density of the simulated mesh could result in a comparable
mobility reduction with one-component diffusion at all
densities, as experimentally observed (Fig. 2). Third, we ad-
dressed the question of the impact on the mesh size distribu-
tion on the experimental results by comparing simulations
performed on heterogeneously distributed meshes (wide
distribution of mesh diameters) with simulations on homo-
geneous meshes (narrow distribution of mesh diameters),
as illustrated in Fig. S3.
In Fig. 6, A and B, two sets of simulated autocorrelation
curves for weaker connement (probability to escape the
mesh: p
jump
0.1) and stronger connement (p
jump
0.01)
are shown. Each set included different mesh diameters
a (dened as the square root of the mesh area). With
decreasing diameter of the mesh, the autocorrelation curves
were in both cases shifted toward longer diffusion times.
This effect was more pronounced for the stronger conning
FIGURE 5 Scheme of the Monte Carlo simulations. (A) The actin
membrane skeleton (gray) was modeled as a Voronoi mesh in an area of
10 mm edge length. Particles (not shown) performed a random walk and
could cross the boundaries only with a certain probability p
jump
and were
reected otherwise. Simulated uorescence traces were acquired in the
mesh with Gaussian-shaped detection spots with a waist of 250 nm
(circles). Subsequently, the corresponding autocorrelation functions were
computed. Each simulation was characterized by the average mesh size,
the mesh distribution (either a narrow, homogeneous mesh size distribution,
or a broad, heterogeneous mesh size distribution) and the connement
strength p
jump
. (B) Example trajectories illustrating the inuence of the
parameter p
jump
. At a low jump probability (bottom), the trajectory of the
diffusing particle was strongly affected by the mesh, resulting in a strong
connement. For a higher jump probability (top), the connement in the
mesh is weaker and the trajectory expands over a larger spatial scale.
Both trajectories represent a random walk over 5 s.
FIGURE 6 Simulated autocorrelation curves at varying mesh diameters
a and constant FCS detection spot area of u 250 nm. (A and B) Two
examples with difference in connement are shown. In both plots the auto-
correlation curves from left to right were simulated with decreasing mesh
diameter and by using the same set of Voronoi meshes. The leftmost curves
were obtained with a mesh larger than the FCS detection area, for the
middle curves mesh and detection area were similar, and the right curves
were simulated with a mesh smaller than the detection area (from left to
right: u/a 0, 0.5, 0.9, 1.1, 2.1, 3.1). In both cases a transition from fast
single-component microscopic diffusion (diffusion inside the mesh) over
an intermediate regime to slower macroscopic diffusion (diffusion on scales
larger than the mesh) occurs. (A) Autocorrelation curves simulated at
comparably weak connement (p
jump
0.1). With decreasing mesh diam-
eter the decay of the autocorrelation curve is shifted toward longer correla-
tion times. (B) Autocorrelation curves simulated at stronger connement
(p
jump
0.01). In this case, a clear deviation from single-component diffu-
sion at mesh diameters comparable to the detection spot size is evident, due
to the shape of the autocorrelation curves.
mesh (Fig. 6 B), because particles were on average in their
respective cell for a longer time before they could pass
a mesh-barrier. In the case of larger or smaller mesh diam-
eters, compared to the detection spot, the shape of the auto-
correlation functions resembled one-component diffusion.
However, for the simulations where the mesh had a size
comparable to the detection spot a z u, deviations from
one-component diffusion were visible. For the stronger
conning mesh (Fig. 6 B), these deviations were much
more pronounced, compared to the mesh with weaker
connement (Fig. 6 A).
To quantify the observed deviations from one-component
diffusion, the simulated autocorrelation curves were tted
with a one-component t, and the error, represented by the
standard deviation between t and simulated curve, was
calculated. In Fig. 7 A, the deviation from single-component
diffusion is plotted against the ratio of focal spot size and
mesh diameter. The dependency of the resulting autocorrela-
tion data on u/a, the spot size compared to the mesh size, is
visible. The calculation of the error, allowed comparing the
expected deviations from one-component diffusion for
different connement strengths. For large meshes compared
to the detection spot, u/a z0, the microscopic fast diffusion
in the mesh dominates. At comparable sizes, u/a z 1, the
microscopic diffusion will be simultaneously observed
with slower macroscopic diffusion in between adjacent
meshes. The stronger the connement, the more dominant
is the occurrence of two populations with a different diffusion
coefcient. Thus, the connement increases the observed
deviations from one-component diffusion. On the other
hand, for weak connement, these deviations are marginal.
Finally, if the mesh size becomes smaller than the detection
spot, u/a > 1, the macroscopic diffusion dominates and the
FCS data approach again a single-component model.
In the membrane skeleton of cells, the actin meshwork
does not have a single xed mesh size, but the mesh sizes
are rather broadly distributed (8). Therefore, when perform-
ing FCS measurements, it is conceivable that smaller and
larger meshes simultaneously affect the FCS measurement.
This could possibly result in a mixing of diffusing compo-
nents and affect the shape of the autocorrelation curves
compared to a homogeneous meshwork, where all meshes
have an almost identical size (if not explicitly stated the
homogeneous meshwork was used for the simulations). In
Fig. S4 the deviation of a single-component model is shown
in comparison for the homogeneous and the heterogeneous
meshwork. Surprisingly, only minor differences were ob-
tained. The maximal error of a one-component t was
slightly lower for the heterogeneous mesh, and the error
of the heterogeneous mesh was distributed over a broader
range of mesh sizes. Both results can be explained by the
averaging over a wider range of mesh dimensions for the
heterogeneous meshwork.
In our experiments with uorescently labeled actin la-
ments, the density of actin was measured from the average
uorescence intensity I in an area A around the FCS detec-
tion spot (Fig. 2 A). Under the assumption that quenching
and other processes, which reduce the observed uores-
cence can be neglected, the average intensity I will be
proportional to the sum of the individual lament lengths
l
i
in the detection area: I fA
1
P
l
i
. The proportionality
constant f is the average uorescence emission per lament
length. To compare the experimentally observed mobility
reduction in the presence of an increasing actin density
with the simulations, the mesh density r in the simulations
was quantied by r A
1
sim
P
l
i
, where A
sim
represented the
simulation area. The density r is directly proportional to the
experimentally determined actin density I.
Fig. 7 B shows the apparent diffusion coefcient plotted
versus the mesh density in the simulation. As expected, an
increase in density resulted in a decreasing mobility. The
result was practically identical for the homo- and the hetero-
geneously distributed mesh, as shown in Fig. S5. Initially, at
lowmesh densities, the decrease in mobility is rapid until the
density of the mesh is such that u/a z1 (corresponding to
a density of r z7 mm
1
). Above this density, the decrease
in mobility is less dramatic. For stronger connement, the
overall effect is generally stronger. However, most interest-
ingly, even for the simulations with weak connement,
which showed no deviation from single-component diffu-
sion, a distinct reduction in mobility could be obtained.
FIGURE 7 Analysis of the autocorrelation data from the Monte Carlo
simulations obtained by variation of the mesh diameter and connement
strength. (A) Error of a single-component diffusion model t to the autocor-
relation data from the simulations plotted versus the focal spot size u
divided by the mesh diameter a. The error was dened as the standard devi-
ation between the simulation result and a single component t. Different
shades of gray represent different strong connement. Fromu/a /0 (large
mesh) over u/a z 1 (mesh similar to detection spot) to u/a >> 1 (small
mesh) the transition from single-component diffusion over nonsingle-
component diffusion to single-component diffusion is evident. For strong
connement the deviations to single-component diffusion are strong,
when spot size and mesh diameter have similar dimensions. For weak
connement these deviations at u/a z1 are weak or negligible. (B) Change
in apparent diffusion coefcient D
app
plotted versus the mesh density (see
text for details). D
app
decreases with mesh density. Stronger connement
led to a faster decrease in diffusion. However, also for weak connement
(topmost two curves), which showed no deviation from single-component
diffusion in (A), a clear reduction in diffusion was obtained. D
app
was calcu-
lated from a one-component t to the simulation results. The data for (B)
represent the same data set as used for (A).
Qualitatively, the mesh size-dependent reduction in
mobility in the simulations (Fig. 7 B) and the experiments
(Fig. 2 B) agrees well. The experimental data show a fast
initial decrease in mobility and then a slower decay with
increasing actin density. Experimentally, at all actin densi-
ties, one-component autocorrelation curves were obtained.
If the simulations adequately describe the in vitro experi-
ments, they rule out the case of strong connement, because
in this case, clear deviations from one-component diffusion
should have been observed. The simulations match well to
the experimental data under the assumption of weak
connement, where one-component diffusion describes the
data adequately at all actin densities.
Switching the diffusive membrane state by
myosin II
An active regulation of the density of cortical actin is
a possible mechanism to tune diffusion in the membrane.
The actin density may be regulated in a myosin-dependent
manner during cytokinesis (44,45), and also by the interac-
tion with the manifold proteins modulating the actin mesh
during other cell cycle phases (46). To demonstrate that
tuning of the density of cortical actin is a possible way for
cells to control diffusion in the membrane, we added myol-
aments to membranes that were initially covered with actin,
and measured the change in mobility of the labeled lipid.
Before myosin addition and in presence of the actin mesh,
the mobility of the lipid was reduced as already observed
in the previously described experiments. After adding the
myolaments, the actin laments typically condensed in
a circular manner, exposing most of the free-standing
membrane patches as shown in Fig. 8. We assume that the
observed actin orientation along the support grid is induced
by the higher membrane friction with the support. However,
on a few free-standing patches, a concentration of actin was
observed (Fig. 8, right image, central spot). The diffusion
coefcients on the actin covered and the actin free spots
were modied accordingly. On the free spots, the diffusion
coefcients were close to the initial value before the addi-
tion of actin, whereas the areas covered with actin showed
a reduced mobility.
Thus, a variation of the actin density might be a mecha-
nism to efciently control diffusion in the plasma
membrane. A possible cellular function may be found in
the spatial control of signaling events (6,47). In combination
with a strong size-dependent lateral diffusion, as suggested
by our data and reported previously (18), receptors with
a bound cytoplasmic ligand or in an oligomeric state would
be localized.
CONCLUSION
Although lateral membrane diffusion is an essential process
for living cells, little is known about its regulation. The
simultaneous interplay of several factors modulating lateral
diffusion, including the interaction with the membrane
skeleton, complicates an in-depth understanding. To this
end, we developed a MAC to directly study the effect of
a membrane-bound actin mesh on the lateral diffusion.
The lateral diffusion of a lipid and a membrane binding
protein were both reduced in an actin density-dependent
manner, and the maximal reductions in diffusion were
much stronger for the protein, compared to the lipid. We
showed that this was a specic feature of membrane-bound
actin. This implies that membrane-bound cortical actin has
a complex inuence on the lateral diffusion, which depends
on the type and/or size of the diffusing species. By using
myolaments to locally change the actin density, our exper-
iments demonstrate a potential mechanism of how cells
might vary the actin mesh density to control lateral diffu-
sion. The proposed size dependency of lateral diffusion by
the actin mesh implies that a control of the cortical actin
density can help to localize cellular signals at certain areas
of the membrane.
SUPPORTING MATERIAL
Supporting analysis, equations, gures, and references (4854) are avail-
able at http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)
00260-9.
We thank Dr. Eugene P. Petrov (Max-Planck Institute of Biochemistry,
Martinsried, Germany), Dr. Jens Ehrig (Max-Planck Institute for the
science of light, Erlangen, Germany), and Le Mu for critical discussions
and helpful advice. We also thank Dr. Zdenek Petrasek (Max-Planck
Institute of Biochemistry, Martinsried, Germany) for discussion and helpful
comments during the manuscript preparation.
Financial support was provided by the Daimler und Benz foundation to
S.K.V. and F.H. (Grant 32-09/11), and the Gottfried Wilhelm Leibniz-
Program of the DFG to S.K.V. (SCHW716/8-1).
FIGURE 8 Addition of myosin II allows switching the diffusive state of
the membrane. The left image shows the uorescence of actin (labeled with
phalloidin-Alexa488) linked to membranes. Upon addition of myosin II, the
actin laments retract from most of the free-standing membrane patches,
but also condense on some spots as shown in the right image. The diffusive
behavior is modied accordingly. On the spots where actin was removed,
the diffusion is comparable to the diffusion in a membrane before actin
coupling, whereas the diffusion is reduced at spots where actin was concen-
trated. Both measurements are for the labeled lipid (Atto647N-DOPE).
Scale bars 5 mm.
REFERENCES
1. Singer, S. J., and G. L. Nicolson. 1972. The uid mosaic model of the
structure of cell membranes. Science. 175:720731.
2. Vereb, G., J. Szollosi, ., S. Damjanovich. 2003. Dynamic, yet struc-
tured: The cell membrane three decades after the Singer-Nicolson
model. Proc. Natl. Acad. Sci. USA. 100:80538058.
3. Simons, K., and M. J. Gerl. 2010. Revitalizing membrane rafts: new
tools and insights. Nat. Rev. Mol. Cell Biol. 11:688699.
4. Engelman, D. M. 2005. Membranes are more mosaic than uid.
Nature. 438:578580.
5. Sheetz, M. P., M. Schindler, and D. E. Koppel. 1980. Lateral mobility
of integral membrane proteins is increased in spherocytic erythrocytes.
Nature. 285:510511.
6. Kusumi, A., C. Nakada, ., T. Fujiwara. 2005. Paradigm shift of the
plasma membrane concept from the two-dimensional continuum uid
to the partitioned uid: high-speed single-molecule tracking of
membrane molecules. Annu. Rev. Biophys. Biomol. Struct. 34:351378.
7. Schwille, P., J. Korlach, and W. W. Webb. 1999. Fluorescence correla-
tion spectroscopy with single-molecule sensitivity on cell and model
membranes. Cytometry. 36:176182.
8. Morone, N., T. Fujiwara, ., A. Kusumi. 2006. Three-dimensional
reconstruction of the membrane skeleton at the plasma membrane
interface by electron tomography. J. Cell Biol. 174:851862.
9. Saxton, M. J. 1989. The spectrin network as a barrier to lateral diffusion
in erythrocytes. A percolation analysis. Biophys. J. 55:2128.
10. Lenne, P. F., L. Wawrezinieck, ., D. Marguet. 2006. Dynamic molec-
ular connement in the plasma membrane by microdomains and the
cytoskeleton meshwork. EMBO J. 25:32453256.
11. Andrews, N. L., K. A. Lidke, ., D. S. Lidke. 2008. Actin restricts
FcepsilonRI diffusion and facilitates antigen-induced receptor immobi-
lization. Nat. Cell Biol. 10:955963.
12. Baumgart, T., A. T. Hammond, ., W. W. Webb. 2007. Large-scale
uid/uid phase separation of proteins and lipids in giant plasma
membrane vesicles. Proc. Natl. Acad. Sci. USA. 104:31653170.
13. Machta, B. B., S. Papanikolaou, ., S. L. Veatch. 2010. A minimal
model of plasma membrane heterogeneity requires coupling cortical
actin to criticality. Arxiv preprint arXiv:1009.2095.
14. Ehrig, J., E. P. Petrov, and P. Schwille. 2011. Near-critical uctuations
and cytoskeleton-assisted phase separation lead to subdiffusion in cell
membranes. Biophys. J. 100:8089.
15. Sako, Y., and A. Kusumi. 1994. Compartmentalized structure of the
plasma membrane for receptor movements as revealed by a nano-
meter-level motion analysis. J. Cell Biol. 125:12511264.
16. Fujiwara, T., K. Ritchie, ., A. Kusumi. 2002. Phospholipids undergo
hop diffusion in compartmentalized cell membrane. J. Cell Biol.
157:10711081.
17. Umemura, Y. M., M. Vrljic, ., A. Kusumi. 2008. Both MHC class II
and its GPI-anchored form undergo hop diffusion as observed by
single-molecule tracking. Biophys. J. 95:435450.
18. Iino, R., I. Koyama, and A. Kusumi. 2001. Single molecule imaging of
green uorescent proteins in living cells: E-cadherin forms oligomers
on the free cell surface. Biophys. J. 80:26672677.
19. Saffman, P. G., and M. Delbruck. 1975. Brownian motion in biological
membranes. Proc. Natl. Acad. Sci. USA. 72:31113113.
20. Wieser, S., M. Moertelmaier, ., G. J. Schutz. 2007. (Un)conned
diffusion of CD59 in the plasma membrane determined by high-resolu-
tion single molecule microscopy. Biophys. J. 92:37193728.
21. Wawrezinieck, L., H. Rigneault, ., P. F. Lenne. 2005. Fluorescence
correlation spectroscopy diffusion laws to probe the submicron cell
membrane organization. Biophys. J. 89:40294042.
22. Smith, D., F. Ziebert, ., J. Kas. 2007. Molecular motor-induced insta-
bilities and cross linkers determine biopolymer organization.
Biophys. J. 93:44454452.
23. Heinemann, F., and P. Schwille. 2011. Preparation of micrometer-sized
free-standing membranes. ChemPhysChem. 12:25682571.
24. Vogel, S. K., Z. Petrasek, ., P. Schwille. 2013. Myosin motors frag-
ment and compact membrane-bound actin laments. Elife. 2:e00116.
25. Ries, J., S. Chiantia, and P. Schwille. 2009. Accurate determination of
membrane dynamics with line-scan FCS. Biophys. J. 96:19992008.
26. Muller, C. B., T. Eckert, ., W. Richtering. 2009. Dual-focus uores-
cence correlation spectroscopy: a robust tool for studying molecular
crowding. Soft Matter. 5:13581366.
27. Haynes, W. M., and D. R. Lide. 2012. CRC Handbook of Physics and
Chemistry. Taylor and Francis Group, London.
28. Vogel, S. K., and P. Schwille. 2012. Minimal systems to study
membrane-cytoskeleton interactions. Curr. Opin. Biotechnol.
23:758765.
29. Przybylo, M., J. Sykora, ., M. Hof. 2006. Lipid diffusion in giant uni-
lamellar vesicles is more than 2 times faster than in supported phospho-
lipid bilayers under identical conditions. Langmuir. 22:90969099.
30. Vaz, W. L. C., M. Criado, ., T. M. Jovin. 1982. Size dependence of the
translational diffusion of large integral membrane proteins in liquid-
crystalline phase lipid bilayers. A study using uorescence recovery
after photobleaching. Biochemistry. 21:56085612.
31. Lee, C. C., M. Revington, ., N. O. Petersen. 2003. The lateral diffu-
sion of selectively aggregated peptides in giant unilamellar vesicles.
Biophys. J. 84:17561764.
32. Vaz, W. L. C., R. M. Clegg, and D. Hallmann. 1985. Translational
diffusion of lipids in liquid crystalline phase phosphatidylcholine mul-
tibilayers. A comparison of experiment with theory. Biochemistry.
24:781786.
33. Zhang, R. G., D. L. Scott, ., E. M. Westbrook. 1995. The three-dimen-
sional crystal structure of cholera toxin. J. Mol. Biol. 251:563573.
34. Eggeling, C., C. Ringemann, ., S. W. Hell. 2009. Direct observation
of the nanoscale dynamics of membrane lipids in a living cell. Nature.
457:11591162.
35. Rosano, C., P. Arosio, and M. Bolognesi. 1999. The X-ray three-dimen-
sional structure of avidin. Biomol. Eng. 16:512.
36. Allen, C., N. Dos Santos, ., M. B. Bally. 2002. Controlling the phys-
ical behavior and biological performance of liposome formulations
through use of surface grafted poly(ethylene glycol). Biosci. Rep.
22:225250.
37. Murase, K., T. Fujiwara, ., A. Kusumi. 2004. Ultrane membrane
compartments for molecular diffusion as revealed by single molecule
techniques. Biophys. J. 86:40754093.
38. Vaz, W., J. Stumpel, ., M. De Rosa. 1987. Bounding uid viscosity
and translational diffusion in a uid lipid bilayer. Eur. Biophys. J.
15:111115.
39. Ollmann, M., A. Robitzki, ., H. J. Galla. 1988. Minor effects of bulk
viscosity on lipid translational diffusion measured by the excimer
formation technique. Eur. Biophys. J. 16:109112.
40. van den Bogaart, G., N. Hermans, ., B. Poolman. 2007. On the
decrease in lateral mobility of phospholipids by sugars. Biophys. J.
92:15981605.
41. Ruprecht, V., S. Wieser, ., G. J. Schutz. 2011. Spot variation uores-
cence correlation spectroscopy allows for superresolution chronoscopy
of connement times in membranes. Biophys. J. 100:28392845.
42. Destainville, N. 2008. Theory of uorescence correlation spectroscopy
at variable observation area for two-dimensional diffusion on a mesh-
grid. Soft Matter. 4:12881301.
43. Mueller, V., C. Ringemann, ., C. Eggeling. 2011. STED nanoscopy
reveals molecular details of cholesterol- and cytoskeleton-modulated
lipid interactions in living cells. Biophys. J. 101:16511660.
44. Robinson, D. N., and J. A. Spudich. 2004. Mechanics and regulation of
cytokinesis. Curr. Opin. Cell Biol. 16:182188.
45. Vale, R. D., J. A. Spudich, and E. R. Grifs. 2009. Dynamics of
myosin, microtubules, and Kinesin-6 at the cortex during cytokinesis
in Drosophila S2 cells. J. Cell Biol. 186:727738.
46. Pollard, T. D., and J. A. Cooper. 2009. Actin, a central player in cell
shape and movement. Science. 326:12081212.
47. Jaqaman, K., and S. Grinstein. 2012. Regulation from within: the cyto-
skeleton in transmembrane signaling. Trends Cell Biol. 22:515526.
48. Angelova, M., S. Soleau, ., P. Bothorel. 1992. Preparation of giant
vesicles by external AC electric elds. Kinetics and applications.
Trends in Colloid and Interface Science. VI:127131.
49. Abra`moff, M. D., P. J. Magalhaes, and S. J. Ram. 2004. Image process-
ing with ImageJ. Biophotonics International. 11:3642.
50. Hess, S. T., and W. W. Webb. 2002. Focal volume optics and experi-
mental artifacts in confocal uorescence correlation spectroscopy.
Biophys. J. 83:23002317.
51. Dertinger, T., V. Pacheco, ., J. Enderlein. 2007. Two-focus uores-
cence correlation spectroscopy: a new tool for accurate and absolute
diffusion measurements. ChemPhysChem. 8:433443.
52. Petrov, E., and P. Schwille. 2008. State of the art and novel trends in
uorescence correlation spectroscopy. Standardization and Quality
Assurance in Fluorescence Measurements II: Bioanalytical and
Biomedical Applications. 6:145197.
53. Heinemann, F., V. Betaneli, ., P. Schwille. 2012. Quantifying lipid
diffusion by uorescence correlation spectroscopy: a critical treatise.
Langmuir. 28:1339513404.
54. O
chsner, A., C. Henninger, and J. Gegner. 2003. Modellierung mikros-

kopischer Objektverteilungen mittels eines stochastischen Suchverfah-
rens. Austrian J. Stat. 32:297304.
S3-S4 Linker Length Modulates the Relaxed State of a Voltage-Gated
Potassium Channel
Michael F. Priest,
Je ro me J. Lacroix,
Carlos A. Villalba-Galea,
and Francisco Bezanilla
Committee on Neurobiology and

Department of Biochemistry and Molecular Biology, University of Chicago, Chicago, Illinois; and
Department of Physiology and Biophysics, Virginia Commonwealth University School of Medicine, Richmond, Virginia
ABSTRACT Voltage-sensing domains (VSDs) are membrane protein modules found in ion channels and enzymes that are
responsible for a large number of fundamental biological tasks, such as neuronal electrical activity. The VSDs switch from a
resting to an active conformation upon membrane depolarization, altering the activity of the protein in response to voltage
changes. Interestingly, numerous studies describe the existence of a third distinct state, called the relaxed state, also populated
at positive potentials. Although some physiological roles for the relaxed state have been suggested, little is known about the
molecular determinants responsible for the development and modulation of VSD relaxation. Several lines of evidence have
suggested that the linker (S3-S4 linker) between the third (S3) and fourth (S4) transmembrane segments of the VSD alters
the equilibrium between resting and active conformations. By measuring gating currents from the Shaker potassium channel,
we demonstrate here that shortening the S3-S4 linker stabilizes the relaxed state, whereas lengthening the linker or splitting
it and coinjecting two fragments of the channel have little effect. We propose that natural variations of the length of the
S3-S4 linker in various VSD-containing proteins may produce differential VSD relaxation in vivo.
INTRODUCTION
S4-based voltage sensors regulate voltage-dependent ion
channels and enzymes by switching between resting and
active conformations in response to changes in the
membrane potential. This conformational switch is driven
by gating charges, charged residues that directly sense the
membrane electrical eld focused on the hydrophobic core
of the voltage-sensing domain (VSD). The transition
between the resting and active states can be directly studied
by measuring gating currents, which are produced by the
movement of the gating charges in the electric eld
following a voltage change.
In addition to resting and active, the two classical VSD
conformations, virtually all VSD, when held for prolonged
periods of time at depolarized potentials, populate a sec-
ondary, more stable depolarized conformation that is struc-
turally distinct from the active state. This state was rst
associated with C-type inactivation in Kv channels (1), but
was later found to be intrinsic to the VSD (24) and named
the relaxed state. In contrast to the transition between the
resting and active states, the conformational change associ-
ated with the transition from the active to the relaxed state
(relaxation) does not produce detectable gating current.
Nevertheless, the VSD relaxation can be studied electro-
physiologically by measuring the deceleration of off-gating
currents when the membrane voltage returns to the resting
potential after a long-lasting depolarization (4). The VSD
transition from the relaxed state to the resting state (derelax-
ation) is slower than that from the active state to the resting
state (deactivation). Furthermore, measuring the return of
the gating charge (Q) from the relaxed to the resting state
(derelaxation) can result in a shift of the charge versus
voltage (Q-V) curve toward more negative voltages
compared to measuring the gating charge moved during a
transition from the resting to the activated state (activation)
(2). Although the concurrent slowing down of the off-gating
current may lead to an underestimation of the charge being
moved and an overestimation of this Q-V shift (4), this
apparent Q-V shift has constituted the hallmark of the
VSD relaxation for several decades.
Despite its ubiquity, the physiological relevance of the
VSD relaxed state has been challenged by the observations
that the entry into the relaxed state is generally very slow
compared to the time course of nerve impulses. However,
a recent study showed that in the Kv1.2 channel, VSD relax-
ation occurs faster than C-type inactivation (3). Addition-
ally, the slowing of VSD return in Kv1.2 from the relaxed
to the resting state appears to impact the pore closure
kinetics and as a result may help in controlling repetitive
ring in neurons (3). In the Kv11.1 channel (also known
as the human Ether-a`-go-go related Gene, or hERG), VSD
relaxation also controls pore closure kinetics, and occurs
in a timescale relevant to the cardiac action potential and
thus may be crucial to provide cardiac cells with enough
hyperpolarizing K
currents during the falling phase of

the cardiac action potential (5). Although the physiological
roles played by VSD relaxation are beginning to be
unveiled, the fundamental molecular mechanisms that
control VSD relaxation have remained unknown.
The S4 segment of the VSD contains most of the gating
charges (6,7); therefore, its movement in the eld generates
most of the gating current during the resting to active and
active to resting transitions. Several lines of evidence
Submitted June 11, 2013, and accepted for publication September 23, 2013.
*Correspondence: fbezanilla@uchicago.edu
Editor: Ian Forster.
2312 Biophysical Journal Volume 105 November 2013 23122322
suggest that the S3-S4 linker of the VSD plays a regulatory
role for this movement (8,9). The role of the S3-S4 linker is
of particular interest as its length varies widely between
different classes of voltage-gated ion channels, and couples
the S4 segment to the remainder of the VSD.
Although several studies have examined the effects of
deletions of the S3-S4 linker on Shaker ionic kinetics and
thermodynamics (1012), and in one case on Shaker gating
kinetics and thermodynamics and voltage sensor movement
(13), the role of the S3-S4 linker in modulating voltage-
gated potassium channels remains unclear.
In the hyperpolarization-activated cyclic-nucleotide-
gated channels, reducing the length of the S3-S4 linker shifts
the half-activation voltage (V
1/2
) of the steady-state ionic
current toward more depolarized potentials, while increasing
the length shifts the V
1/2
toward more hyperpolarized poten-
tials, compared to the wild-type (WT) channel (14). How-
ever, delayed-rectier voltage-gated potassium channels
have not shown this phenomenon. Instead, deletions of
amino acids from and insertions of amino acids into the
S3-S4 linker of the Shaker channel were seen to produce
small changes in ionic current activation and kinetics that
did not correlate with linker length (15). Additionally, re-
placing the S3-S4 linker of Shaker with the shorter linkers
from Shaw (Kv3 type) and Shab (Kv2 type) channels, pro-
duced slower activation and deactivation kinetics (15). How-
ever, for these substitutions no obvious trend was observed as
a function of the linker length. In contrast, shortening the
linker seven amino acids or less showed a clear regular
pattern of effects on the channel function; a helical structure
at the C-terminal end of the S3-S4 linker was inferred from a
periodic alteration of ionic activation kinetics of Shaker con-
structs with various deletions of the S3-S4 linker (12). How-
ever, the effect of such deletions on gating currents and
voltage sensor relaxation was not examined.
To investigate whether the S3-S4 linker controls VSD
relaxation, we generated several truncation mutants from a
nonconducting Shaker construct (16) with fast inactivation
removed (17) using the previously discovered periodicity
(12) to guide our systematic alterations of the S3-S4 linker.
Our results indicate that shortening the S3-S4 linker stabi-
lizes the relaxed state. The comparison of the putative
relaxed VSD structure from the Kv1.2 channel with a struc-
tural model of its resting state (1820) suggests that the
extracellular ends of the S3 and S4 segments move closer
together as the VSD enters the relaxed state. Thus, short-
ening the S3-S4 linker may stabilize the relaxed state by al-
lowing S3 and S4 to get closer to each other.
Construction of Shaker mutants
Deletions of the S3-S4 linker were made using site-directed mutagenesis by
polymerase chain reaction. Mutations were veried by sequencing. Gating
current constructs were produced on a W434F D 646 Shaker background
to remove ionic conduction (16) and fast inactivation (17), respectively. For
ionic current short linker constructs, residue W434F was mutated back to
tryptophan. For the split linker construct, the N-terminal construct was
terminated part way through the S3-S4 linker by inserting a stop codon;
for the C-terminal construct, the Shaker sequence was deleted from the
residue after the methionine start codon to the rst residue in the S3-S4
linker not contained in the N-terminal construct. DNA was linearized
with NotI (New England Biolabs, Ipswich, MA), and transcribed using
the mMESSAGE mMACHINE T7 kit (Life Technologies, Carlsbad, CA).
50 ng mRNA was injected into Xenopus oocytes and currents measured
25 days after injection; the two split linker constructs were coinjected in
a 1:1 ratio. Oocytes were kept at 16
C in a solution containing (in mM)

96 NaCl, 2 KCl, 1 MgCl
2
, 1.8 CaCl
2
, 10 HEPES, pH 7.4, supplemented
with penicillin 100 units/ml and streptomycin 0.1 mg/ml, or with
10 mg/L of gentamicin.
Gating and ionic current recordings
Recordings were performed at room temperature using the cut-open oocyte
voltage clamp technique (21). Gating currents were measured with external
recording solution containing (in mM), 120 N-methyl-D-glucamine-
methanesulfonic acid (NMG-MES), 10 HEPES, and 2 CaOH
2
and internal
recording solution containing 120 NMG-MES, 10 HEPES, and 2 EGTA.
Ionic currents were measured in external recording solution containing
(in mM) 118 NMG-MES, 2 KOH, 10 HEPES, and 2 CaOH
2
and internal
recording solution containing 120 KOH, 10 HEPES, and 2 EGTA. All
recording solutions were pH 7.4 to 7.5. Electrophysiological data were
ltered at 5 kHz and sampled at 1050 kHz. Agar bridges were lled
with 1 M NMG-MES. Current microelectrodes were lled with 3 M KCl
for gating currents and 3 M CsCl for ionic currents and had a resistance
of ~0.150.95 MU.
Derelaxation gating currents and ionic currents were not subtracted, but
were analogically compensated for capacitance; activation gating currents
were sometimes subtracted online using a standard P/4 or P/4 protocol
as appropriate with subtracting pulses from potentials in the saturated
portions of the Q-Vin addition to the capacitance compensation. Activation
gating currents were obtained upon pulsing from a hyperpolarized potential
to a depolarized potential. Deactivation and derelaxation gating currents
were obtained by pulsing from a depolarized potential to a hyperpolarized
potential. Derelaxation protocols and relaxation protocols were done by
pulsing to a position on the Q-V curve, termed the saturated depolarizing
potential, where essentially all gating charge had moved, as prior reports
had indicated this was necessary to enter the relaxed state (5). This
was 0 mV for all constructs except Sh D330347 and Sh D330350
(Fig. 1 A), which were pulsed to 20 mV.
Relaxation protocols were performed by pulsing to the saturated
depolarizing potential for various time durations and then to the membrane
potential at which weighted time constants (see data analysis below) t to
the decaying phase of the derelaxation gating currents were the largest, or,
more simply, to the membrane potential at which the derelaxation
gating currents were slowest. This membrane potential, at which the
slowing of kinetics of derelaxation currents was most dramatic, ranged
from 30 mV to 70 mV in our constructs. A Dagan (Minneapolis,
MN) CA-1B was used to voltage-clamp in the cut-open oocyte voltage
clamp conguration, and data was acquired using an Innovative Integration
(Simi Valley, CA) A4D4 conversion board as a front end of the Innovative
Integration digital signal processor-based SB6711 board that was under
software control by an in-house program (Gpatch).
Data analysis
Using an in-house analysis program (Analysis), time constants of activa-
tion, deactivation, and derelaxation were obtained by tting single, double,
S3-S4 Linker Modulates VSD Relaxation 2313
or triple exponentials to the decaying phase of activation, deactivation, and
derelaxation gating currents; gating currents included for analysis of dere-
laxation kinetics were taken from oocytes producing over 500 pC of gating
charge. Weighted time constants were calculated when more than one time
constant was necessary to t the curve using t (SA
i
t
i
)/(SA
i
), where A
i
is
the exponential coefcient, and t
i
is the time constant. For consistency, the
time constants of gating currents discussed are weighted time constants.
Time constants of C-type inactivation were similarly determined from
single exponential ts to ionic currents. Statistical differences were tested
using the unpaired students t-test, and statistical signicance was set at
P < 0.05. t-tests and exponential and Boltzmann ts to Q-V curves and
other plots were determined with Origin software.
RESULTS
Shortening the S3-S4 linker produces slower
derelaxation kinetics
To avoid alterations of gating current kinetics caused by
disruption of helical structures at the ends of the Shaker
S3-S4 linker, we used prior results (12) as a guide for our
deletions of the S3-S4 linker to produce various short linker
constructs (Fig. 1 A).
All these short linker constructs produced gating currents
(see example from a single construct, Fig. 1, B and C). This
was expected, as a Shaker construct with the full S3-S4
linker deleted was previously shown to produce voltage-
sensitive ionic currents (11). Deletions of the linker pro-
duced alterations in: i), the mid-point of the Q-V curve
measured during activation (activation V
1/2
) (Fig. 1 D)
and ii), the maximum activation kinetics (maximum activa-
tion Tau, or activation t
max
) obtained from the time constant
(Tau) versus V curve (Tau-V) of gating currents, and corre-
sponding to the time course of the slowest activation gating
current (Fig. 1 E). As the modications of the activation V
1/2
and t
max
did not correlate very strongly overall with S3-S4
linker length, we decided to examine whether S3-S4 linker
length altered the transition from the relaxed state to the
resting state, also known as derelaxation (Fig. 2 A). The
relaxed statea third state of the VSD that is reached after
an extended pulse to depolarized potentialsis notable for
slowing down the return kinetics of the VSD to its resting
state (5).
By pulsing to depolarizing potentials in the saturated
range of the Q-V curve for an extended period of time
(10 s) the voltage sensor enters its relaxed state (Fig. 2 A).
The derelaxation can be measured by analyzing the kinetics
of the decaying phase of the gating current obtained from a
subsequent hyperpolarizing pulse (Fig. 2 B) (4). The t
max
of
derelaxation obtained from the Tau-V analysis of each
construct (black trace in Fig. 2 B, black square in
Fig. 2 C) was plotted against S3-S4 linker length
(Fig. 2 D). As can be seen, there is a strong inverse correla-
tion between linker length and the t
max
of derelaxation
(adjusted R
2
of linear t to the derelaxation t
max
is 0.92).
Shortening the S3-S4 linker produces a more
rapid entry into the relaxed state
Given that shorter S3-S4 linkers slow down derelaxation, we
investigated whether shortening this linker also modulated
the entry into the relaxed state. To measure entry into
the relaxed state, Shaker constructs were prepulsed to a
A
B C
D E
FIGURE 1 Basic properties of short linker Shaker constructs. (A)
Sequence of the S3-S4 linker of constructs produced on a nonfast-inactivat-
ing, nonconducting Shaker background. Deleted residues are marked by as-
terisks. Linker length is determined using prior ndings (12) that suggest
the linker extends from residue 330 to 362 in the WT channel. (B) Example
of gating currents obtained from short linker construct Sh D330344 in
response to the activation protocol shown (prepulse to 120 mV followed
by 80 ms long, 10 mV steps from120 to 40 mV). The trace with the slow-
est kinetics is highlighted in red. (C) Tau-V (black) and normalized Q-V
(blue) of the Sh D330344 construct. The saturated depolarized membrane
potential, used in later derelaxation protocols, is marked by the green sym-
bol in the Q-V, at 0 mV for this construct. The slowest activation Tau, or
activation t
max
, is about 6 ms for this construct and is marked by the red
symbol in the Tau-V. (D) V
1/2
of Q-Vs obtained from activation protocol
plotted against linker length. N is between 3 and 9 for each construct; error
bars here and in all other gures show standard error. The V
1/2
of D341346
is more hyperpolarized than that of D345350 or D330335, which are
roughly equivalent. (E) t
max
obtained from the same protocol, plotted
against linker length. N is between 3 and 10 for each construct. D330
335 has much slower kinetics than D345350, which is comparable to
WT in kinetics, and also slower than D341346, which is the fastest.
2314 Priest et al.
depolarized potential where the Q-V is saturated (saturated
depolarizing potential, see Fig. 1, C and Fig. 3 A) for a
variable duration and then pulsed to the potential where
t
max
is reached (potential of derelaxation t
max
, see Fig. 2
C and Fig. 3 A). As the duration of the depolarizing pulse
increases, a larger fraction of the VSDs should enter the
relaxed state. Thus, as the duration of the depolarizing pulse
increases and more and more VSDs enter the relaxed state,
the derelaxation kinetics measured from the gating current
obtained in response to the subsequent repolarizing pulse
should slow more and more (for example traces, see Fig. 3
B). Thus, the duration of the depolarizing pulse required
to produce VSD slowing during derelaxation is a measure-
ment of the kinetics of VSD relaxation. In WT Shaker,
when relaxation kinetics are measured, the slowing is
biphasic, as is seen in a plotting of derelaxation t
max
versus
depolarizing pulse duration curves (Fig. 3 C, black). This
biphasic slowing suggests that two separate processes are
driving the slowing; it has been posited that the rst compo-
nent of the slow-down is driven by open-state stabilization,
whereas the second component of the slow-down is driven
by the VSD entering the relaxed state (4). Thus, in WT
Shaker, the rst component corresponds to the time course
of entry into the open-pore state, whereas the second
component corresponds to the time course of entry into
the relaxed-VSD state.
However, the derelaxation t
max
versus depolarizing pulse
duration curves of all examined short linker constructs were
well t by single exponentials. Although constructs with
linkers >20 amino acids in length could be approximately
t by double exponentials as well, as is seen for construct
D330335 (Fig. 3 C), constructs with shorter linkers typi-
cally could be well t by single exponentials. As a result,
it is not possible to discriminate the slowing produced by
open-state stabilization from the slowing produced by relax-
ation in the short linker constructs.
Our inability to distinguish the entry into the active state
from the entry into the relaxed state suggests two possible
mechanisms. First, shortening the S3-S4 linker may simply
abolish entry into the relaxed state. However, if entry into
the relaxed state was abolished, the slowing in gating
A
B C
D
FIGURE 3 Measurements of entrance into the relaxed state. (A) Protocol
for measuring entrance into the relaxed state. Depolarizing pulses of dura-
tions ranging from 1 ms to 10 s are followed by 700 ms hyperpolarizing
pulses to the membrane potential at which the derelaxation kinetics were
slowest (cf. 2 C). (B) Example traces of construct Sh D330350 in response
to depolarizing pulses of 10 s, 1 s, 8 ms, 3 ms, and 1 ms. (C) Derelaxation
t
max
versus depolarizing pulse duration for WT and example short linker
constructs (symbols) with single or double exponential curve ts as appro-
priate (continuous lines). Only the WT construct is not t by a single expo-
nential. (D) Time constants of exponential ts to relaxation curves as in
panel C versus linker length. All time constants shown came from exponen-
tial ts to relaxation curves with R
2
> 0.94. N is between 3 and 8 for each
construct. The relaxation process of Sh D330335 is faster than that of Sh
D341346. WT Sh is marked by a star.
A B
C D
FIGURE 2 Derelaxation properties of short linker Shaker constructs. (A)
Cartoon of voltage sensor states and transitions. Derelaxation measures the
entire transition from the relaxed state back to the resting state. (B) Example
of derelaxation gating currents obtained from short linker construct Sh
D330344 in response to the derelaxation protocol shown (prepulse
to 0 mV for 10 s, followed by 700 ms long, 10 mV steps from 0 mV
to 120 mV). The trace with the slowest kinetics is the bold black trace.
(C) The derelaxation Tau-V of Sh D330344. The derelaxation t
max
(~85 ms for this construct) is marked by the black symbol, as is the mem-
brane potential at which the maximum derelaxation kinetics are obtained
(40 mV for this construct). (D) Derelaxation t
max
, as described in A
and C, of various short linker constructs and the WT construct versus linker
length. N is between 3 and 9 for each construct.
kinetics as a function of the S3-S4 linker length observed in
Fig. 2 D would then have to originate 1), from a stabilization
of the active state or 2), from an increase of the energy
barrier between the active and resting states. If case one
were true, we would observe a correlation between linker
length and negative shifts in the activation Q-V curve, which
is not the case (Fig. 1 D). If case two were true, we would
observe a linear correlation between linker length and
activation gating kinetics, which does not appear to be the
case (Fig. 1 E).
Therefore, a different mechanism likely makes the time
course of the entrance to the active state indistinguishable
from the entrance to the relaxed state. The slowing down
of the repolarization gating current produced by the
entrance into the relaxed state may occur on roughly the
same timescale of the slowing down resulting from pore
opening. This could occur either from a slowing of the
activation kinetics of either gating or pore opening, or
from an acceleration of the transition from the active to
the relaxed state. The majority of our short linker con-
structs have linkers 10 amino acids or longer, and show
similarly rapid gating activation kinetics to WT Shaker,
in good agreement with previous ndings on ionic activa-
tion kinetics in short linker constructs (11). Thus, the
presence of only one component in the plots of the deacti-
vation/derelaxation kinetics as a function of the depolariz-
ing pulse duration is not likely due to either the abolition
of the relaxed state or a slowing of the activation transition
in the short linker constructs. Rather, this single component
comprising both the activation and the relaxation kinetics
stems from an acceleration of the entry of the VSDs into
the relaxed state.
The time constant of the second exponential in the double
exponential t of the t
max
versus depolarizing pulse duration
plot for the WT construct (black line, Fig. 3 C), then, corre-
sponds to the time course of VSD slowing produced by
relaxation. The single exponential t to the t
max
versus
depolarizing pulse duration plot for short linker constructs
(red, green, and blue lines, Fig. 3 C) is a composite of the
VSD slowing produced by relaxation with that produced
by pore opening. Thus, the time constants of these single
exponential ts measure the relaxation (and activation)
time course of the short linker constructs. By comparing
these time constants, we see that the relaxation process ac-
celerates in short linker constructs compared to WT,
although, as opposed to the deceleration of derelaxation
observed, this acceleration of relaxation does not correlate
strongly with linker length (Fig. 3 D). The membrane poten-
tial difference used in our protocol, i.e., the difference be-
tween the variable duration saturated depolarizing
potential and the subsequent hyperpolarizing pulse to the
potential of derelaxation t
max
, could be a driver of relaxa-
tion kinetics. However, the membrane potential difference
does not account for the acceleration of the relaxation Tau
that we observe in constructs with shorter S3-S4 linkers;
for example, this difference is 50 mV for the WT
construct, 60 mV for Sh D330335, and 40 mV for Sh
D330341 and Sh D330350. One possible explanation is
that shortening the S3-S4 linker might accelerate the transi-
tion from the active to the relaxed state; further investigation
of this possibility may be valuable.
It has been proposed that the slowing of the voltage sensor
is linked to C-type inactivation (1,22,23). However, it is not
immediately clear how altering S3-S4 linker length would
alter C-type inactivation. If C-type inactivation were linked
to voltage sensor slowing, it should correspond to entrance
to the relaxed state. Therefore, C-type inactivation should
occur more rapidly in our short linker constructs compared
to WT Shaker. However, when we measured the time con-
stant of C-type inactivation from ionic currents during
long depolarizing pulses (15 s) to the saturated depolarizing
potential, we found that this was not the case (Fig. 4, A
and B). Similarly, recovery from C-type inactivation
should follow the time course of derelaxation, and our short
linker constructs should have slower recoveries. C-type
FIGURE 4 Relaxation is separable from C-type inactivation (A) C-type
inactivation was recorded from conducting (W434W) WT, D330347,
and D330350 constructs by pulsing from a 90 mV resting potential to
the saturated depolarizing potential for 15 s and tting with an exponential.
Representative traces are shown here. (B) When kinetic ts are performed
on the C-type inactivation of these ionic currents, the C-type inactivation
does not change signicantly. N 5, 4, and 5, respectively. This is in
contrast to the faster relaxation kinetics displayed by the gating currents
of these short linker constructs. (C) Similar to B, but the fast component
of the recovery from C-type inactivation for these constructs. Recovery
from C-type inactivation was recorded by pulsing to the saturated depola-
rizing potential for 15 s, recovering to a 90 mV resting potential for
durations ranging from 3 ms to 3 s, and measuring the ionic current
obtained upon a subsequent pulse to the saturated depolarizing potential.
As opposed to the slower derelaxation kinetics displayed by the gating
kinetics, the short linker constructs displayed faster recoveries from
C-type inactivation. N 5, 2, and 4, respectively. (D) As in C, but the
slow component of C-type inactivation recovery.
2316 Priest et al.
inactivation recovery was measured by the classical re-
covery from inactivation protocol (24). A 15 s depolarizing
pulse to induce C-type inactivation was followed by a vari-
able duration hyperpolarizing pulse to recover from inacti-
vation and then followed by a test pulse to the saturated
depolarizing potential. The recovery of ionic current during
the test pulse as a function of the duration of the hyperpola-
rizing pulse was t to an exponential to obtain the time con-
stant of recovery from C-type inactivation. We found that
the recovery from C-type inactivation does not correspond
to derelaxation for either the fast (Fig. 4 C) or slow
(Fig. 4 D) component of C-type inactivation recovery.
Thus, the effects of S3-S4 linker length on the relaxation
of the voltage sensor are not likely mediated by C-type inac-
tivation following open-state stabilization.
The W434F background mutation used to record gating
currents complicates our comparison of C-type inactivation
to relaxation, as this mutation produces ultrafast or perma-
nent C-type inactivation (25). Despite this limitation, our
results that C-type inactivation appears independent from
relaxation are in good agreement with recent ndings that
alterations of gating charge movement in response to pro-
longed depolarizations do not correlate with C-type inacti-
vation in WT Kv channels (3) nor with slow inactivation
in Nav channels (26).
Splitting the S3-S4 linker does not abolish gating
currents
Shortening the S3-S4 linker speeds up transitions into the
relaxed state and slows down transitions out of the relaxed
state. We hypothesized that reducing the restraint placed
by the coupling of the S1 to S3 segments of the voltage
sensor domain to the S4 voltage sensor via the S3-S4 linker
would speed transitions out of the relaxed state. An exami-
nation of Shaker type channels indicated that the length of
the S3-S4 linker in Drosophila was near the maximum
found in nature. Additionally, prior reports had indicated
that coinjection of two Shaker construct subunits, one con-
sisting of the portion from the N-terminal to part way
through the S3-S4 linker and the other consisting of the
remainder of the S3-S4 terminal through the C-terminus,
could result in ionic currents (27). Therefore, we made a
similar split construct on a nonconducting Shaker back-
ground (Fig. 5 A). Injection of either of these two constructs
individually produced no observable gating currents (Fig. 5,
B and C). However, coinjection of these constructs produced
robust gating currents (Fig. 5 D).
Interestingly, splitting the S3-S4 linker had no signicant
effect on derelaxation or activation kinetics (Fig. 5 E). As
the derelaxation kinetics did not differ between the split
linker and the WT construct, we examined whether the
leftward Q-V shift seen between the activation Q-V and
the derelaxation Q-V was also similar for these two con-
structs. This leftward shift between the activation Q-V and
the Q-V obtained upon gating currents returning to the
resting state following a prolonged depolarization was the
original descriptor of the relaxed state (1,2). Subsequent
work showed that in the WT Shaker, this Q-V shift was quite
small (4). Indeed, the magnitude of the leftward Q-V shift in
the split linker construct was ~13 mV, and was not signi-
cantly different from the WT construct (Fig. 5 F). However,
the activation Q-V was shifted slightly, but signicantly, to
more depolarized potentials (Fig. 5 F), an effect in good
agreement with ionic current recordings from a voltage-
gated potassium channel with an enzymatically split linker
(10). The simplest explanation of these ndings is that the
split linker construct slightly reduces the stability of the
active state, but does not alter the stability of the relaxed
state.
Lengthening the S3-S4 linker does not alter
derelaxation kinetics
With the unexpected nding that a Shaker construct with a
S3-S4 linker of theoretically innite length from the
perspective of the S4 had no effect on derelaxation, we
FIGURE 5 Gating currents and properties of a split linker construct. (A)
Sequence of the two portions of the split linker construct, compared to the
WT background construct. The asterisk denotes a stop codon. (B and C)
Injection of the N-terminal portion (B) or C-terminal portion (C) of the
Shaker channel does not produce gating currents. (D) Coinjection of the
N-terminal and C-terminal portions of the Shaker construct produces gating
currents in response to an activation protocol with subtraction. (E) t
max
of
WT (black) and split linker (white) gating currents produced by derelaxa-
tion and activation protocols. N from left to right: 9, 5, 7, 5. F. V
1/2
of acti-
vation and (F) V
1/2
shift upon derelaxation. N from left to right: 4, 3, 4, 3.
examined whether this phenomenon was also found in
Shaker constructs with elongations of the S3-S4 linker.
Amino acids were inserted just after residue 341 (Fig. 6
A); this site was chosen both based on our own ndings
and on the basis that previous insertions at this site had pro-
duced functional constructs with minimal alterations to
properties of ionic currents (15). Corroborating the ndings
from our split linker construct, lengthening the S3-S4 linker
did not result in any consistent changes in activation t
max
(Fig. 6 B) or in derelaxation t
max
(Fig. 6 C).
The amino-acid composition of the S3-S4 linker
inuences derelaxation kinetics
Three constructs, D345350, D341346, and D330335
(Fig. 1 A) with 25 amino acid long linkers produced by
deletions of different portions of the linker show striking
differences in derelaxation kinetics (Fig. 7 A, shown in
brown, red, and gray, respectively). Similarly, two con-
structs with 31 amino acid length linkers, WT Shaker and
a construct with the EEED sequence near the S3 side of
the linker neutralized to QQQN, show variable derelaxation
kinetics (Fig. 7 A, WT in black, neutralized mutant in blue).
Finally, two constructs with 38 amino acid linkers (Fig. 6 A,
Sh7G and ShTEV) show slightly different derelaxation
kinetics (Fig. 7 A, Sh7G in dark green, ShTEV in light
green). Hence, although S3-S4 linker length clearly plays
a crucial role in modulating the thermodynamic and kinetic
properties of the relaxed state, other molecular mechanisms,
including the primary sequence of the linker, may alter the
kinetics of entry into and exit out of the relaxed state.
For example, constructs with deletions from residue 330
to 337 or less produced unexpectedly slow gating that
became extremely slow in constructs containing one or
more of the negatively charged amino acids present in
the N-terminus of the S3-S4 linker (Fig. 7 B, black circles;
Fig. 7 A, example Sh D330335). Disrupting this region
had produced abnormal ionic current behavior in previous
examinations of the S3-S4 linker (12), raising the possi-
bility that there was something unique about these con-
structs that was producing abnormal gating currents apart
from the length of the S3-S4 linker. To test this, we
made two additional short linker constructs, D341346
and D345350, with the same linker length as the construct
D330335, which has the most extreme derelaxation
kinetics, but with deletions taken from the central region
of the linker. The derelaxation t
max
measured from these
constructs, with deletions not taken from the N-terminal
region of the S3-S4 linker, followed the general trend of
shorter linker lengths derelaxing more slowly (Fig. 7 B,
in red, D341346 has the faster derelaxation kinetics of
the two).
The aberrantly slow derelaxation observed in constructs
D330335 and D330334 could also originate from altering
the position of negatively charged residues that are normally
at the top of the S3 segment (Fig. 1 A). Neutralizing the
endogenous charges found at the N-terminus of the S3-S4
linker, by mutating the EEED sequence to QQQN, slows
derelaxation (Fig. 7 A, neutralized construct in blue com-
pared to WT in black). Thus, the alteration of charge may
underlie some of the slowed derelaxation kinetics observed
in our short linker constructs.
A
B C
FIGURE 6 Kinetic properties of longer linker constructs. (A) Sequences
of longer linker constructs produced by the insertion of glycines or other
amino acid sequences at residue 341. (B) t
max
produced by activation pro-
tocol for the longer linker constructs, shown with WT and split. ShTEV
has slower activation kinetics than Sh7G. The t
max
of the split linker
construct is shown beyond the x axis break. (C) As in B, but showing der-
elaxation time constants. Once again, ShTEV has slower kinetics than
Sh7G. Extending the linker appears to have no consistent effect on either
activation or derelaxation kinetics. N is between 3 and 9 for each construct.
FIGURE 7 Primary structure of N-terminal region of S3-S4 linker inu-
ences derelaxation. (A) Shaker constructs with the same linker length can
display different derelaxation kinetics. Shown are the derelaxation t
max
of seven constructs. Two extended linker constructs with 38 amino acid
length linkers (for sequences, see Fig. 6 A) are shown in dark green
(Sh7G) and light green (ShTEV). WT Shaker (black) and a neutralized
mutant (333336 EEED mutated to 333336 QQQN) (blue) both have 31
amino acid linkers. Three short linker constructs with 25 amino acid linker
lengths are also shown, Sh D345350 (brown), Sh D341346 (red), and Sh
D330335 (gray) (for sequences, see Fig. 1 A). (B) Short linker constructs in
which deletions begin at the N-terminal of the S3-S4 linker and delete fewer
than nine residues (black circles) do not show typical derelaxation kinetics
seen in previously displayed Shaker constructs (gray squares). However,
similar sized deletions from other regions of the linker do show expected
derelaxation kinetics (red squares). N is between 5 and 8 for constructs in
black or red.
2318 Priest et al.
DISCUSSION
Mechanism of S3-S4 linking
We provide evidence that shortening the S3-S4 linker in
Shaker channels may increase the rate at which voltage sen-
sors enter the relaxed state, and that this linker shortening
decreases the rate of leaving that state. This is, to our knowl-
edge, the rst demonstration of a systematic effect of the
length of the S3-S4 linker on the gating current of
delayed-rectier voltage-gated potassium channels. Addi-
tionally, it provides a molecular mechanism for modifying
the properties of the relaxed state of the voltage sensor.
For the S3-S4 linker to alter the kinetic properties of
voltage sensor movement in a systematic way, the S4
segment must move in relation to the S3 segment. Indeed,
recent cross-linking experiments (28,29), omega gating
pore current discoveries (30), and spectroscopic experi-
ments (31) demonstrate that S4 moves relative to S3.
We summarize the effect of S3-S4 linker length on the
free energy and movement of the voltage sensor in a
simplied energy diagram (Fig. 8 A). Upon depolarizations,
a shorter link (dark and light gray), between the S3 and S4
segments facilitates the transition of the VSD to its thermo-
dynamically more stable relaxed state, whereas a longer
linker (black) gives more degrees of freedom before reach-
ing that relaxed state. This could occur through two possible
mechanisms that are not mutually exclusive. One is that the
shorter linkers lower the energy barrier between the active
and the relaxed state. In this case, the more constrained
degrees of freedom produced by the shorter linkers lead to
less energy being lost to peripheral displacement of the
voltage sensor moving to degenerate states, rather than a
direct displacement along the route traveled from the active
to the relaxed state. A similar phenomenon has been previ-
ously modeled to describe the gating charge displacement of
the voltage sensor (32); the key difference here is that no
gating displacement that results in gating current occurs
during the transition from the active to the relaxed state. A
second possibility is that the active state in many of these
constructs is less stable than in the WT construct, which is
supported by the depolarized shift in activation Q-V
1/2
(Fig. 1 D) for constructs with linkers longer than seven
amino acids. In this case, the lower degrees of freedom pro-
duced by the shorter linkers lowers the entropy of the active
state, destabilizing it compared to the long WT linker, and
speeding the transition from the active to the relaxed state
in the process.
The transition from the relaxed state to the resting state,
on the other hand, does not appear to be primarily con-
strained by the degrees of freedom allowed by the linker.
Instead, there seems to be a mechanical constraint produced
by a shortened linker that prevents the voltage sensor from
moving quickly from the relaxed state to the resting state.
This constraint by the linker can be thought of as a dog
(the voltage sensor) being held by a leash (the S3-S4 linker)
FIGURE 8 Model and relevance of S3-S4 linker length and the relaxed
state. (A) We provide evidence that shortening the S3-S4 linker in Shaker
K
channels increases the stability of voltage sensors in the relaxed state.

The model shows long, WT S3-S4 linker (black), intermediate linker
(dark gray), and short linker (light gray). During derelaxation, the shorter
linker acts as a mechanical constraint, slowing the voltage sensor as it
moves from the relaxed to the resting state, where the distance between
S3 and S4 is larger. Additionally, shortening the linker may increase the
rate at which voltage sensors enter the relaxed state. During relaxation,
the shorter linker may reduce the degrees of freedom available to the
voltage sensor, speeding the transition from the resting to the relaxed state.
As this could be occurring due to destabilization of the active state, reduc-
tion of the barrier between the active and relaxed state, or both, these
changes are denoted using dashed lines. (B) The leftward Q-V shift
obtained upon derelaxation of short linker Shaker constructs. N is between
3 and 8. (C) Sequence homology of the S3-S4 linker of Shaker and other
voltage-gated ionic channels. Shaded in light gray are apparently conserved
negatively charged residues in the N-terminal portion of the linker. (D and
E) The activation Q-V
1/2
from voltage-gated potassium channels (4,3742)
is shown to have no obvious correlation with linker length, similar to what
was found with mutant Shaker constructs in Fig. 1, D. Similar to B, the left-
ward Q-V shift obtained upon derelaxation of voltage-gated ionic channels
taken from the literature (4,37,4345). The linker length of Nav is the
average from the four domains of the sodium channel from Doryteuthis
opalescens; the linker length of Cav1.2 is the average of the S3-S4 linker
length from the rst three domains from Cavia porcellus, as the
fourth likely undergoes splicing of an undetermined amount in this model
organism.
as it traverses from the relaxed to the resting state. With a
long enough linker, the voltage sensor is no longer con-
strained, and behaves as it would with an even longer, or
split, linker, which is what we observe.
Shorter linkers, therefore, show a slower exit from the
relaxed state and may show a faster entry into the relaxed
state. In addition to altering kinetics, the shorter linkers pro-
duce an increase in the leftward Q-V shift produced during
derelaxation (Fig. 8 B). Thus, shorter S3-S4 linkers appear
to stabilize the relaxed state (Fig. 8 A).
A key prediction of these explanations of the effect of
linker length on transitions between the resting and relaxed
states is that the S3 and S4 helices are closer to each other in
the relaxed state than in the resting state. By comparing a
recent model of the resting state of a Kv channel VSD
(20) with putative relaxed states of the VSD obtained
from Kv channel crystal structures (18,19), the distance
between the top of S3 and the top of S4 is ~911 A
in the
relaxed state compared to ~17 A
in the resting state. This

observation appears to be consistent with recent molecular
dynamics simulations of voltage sensor motion from depo-
larized to hyperpolarized conformations (33).
Finally, although the composition of residues in the linker
can alter the kinetics of ionic activation (15) and of gating
activation and derelaxation (Fig. 7 A), these changes remain
poorly understood and difcult to predict. One example of
this is the function, if any, of the three prolines present in
the S3-S4 linker, spaced every three amino acids. Replacing
these prolines with alanines produced only small changes in
ionic activation energies and kinetics (15) and deletion or
disruption of this proline motif, in this study and others
(11,12), caused no consistent changes to channel function.
Together, these ndings suggest that the middle portion of
the S3-S4 linker, from residue 338 to around residue 353,
may be exible. Thus, although we cannot rule out that
our S3-S4 linker deletions, which typically occurs every
three amino acids, are following the helicity of the S3-S4
linker, this possible helicity seems unlikely to explain all
the results shown here. Rather, it appears that derelaxation,
and to a lesser extent relaxation, are largely dependent on
linker length, provided that the structured regions at the
ends of the linker are not disrupted.
Split linker constructs and membrane insertion
The ability of a split Shaker construct to successfully pro-
duce gating currents is unsurprising. Recently, it has been
demonstrated that a synthetic S4 from Shaker spontaneously
inserts into a membrane (34). Additionally, the S4 segment
of Kv1.3 integrates into the membrane on its own following
translation (35). Finally, the S3-S4 linker is not required for
VSD function, as its complete deletion does not abolish
voltage-activation of Shaker channels (8,11).
Some split linker constructs with fewer amino acids at the
top of S4 were made, but did not become functional (data
not shown). The simplest explanation of this nding is
that these constructs lacked a necessary, previously un-
known signal sequence that properly orients the N-terminal
of S4 to the endoplasmic reticulum lumen. These functional
and nonfunctional split linker constructs may provide a
useful tool for dissecting what is required for successful
membrane protein translation, insertion, and trafcking. A
last point of interest concerning the split linker and mem-
brane insertion is that S3 does not insert into the membrane
spontaneously (35), and likely requires S4 to integrate it
after translation (36). It has been suggested that the S3-S4
linker may serve to increase peptide integration into the
membrane (37). Indeed, gating currents produced by
coinjection of the split linker constructs, or of short linker
constructs of two to three amino acids in length, were typi-
cally smaller than those produced by injection of other short
linker or WT constructs in the same batch of oocytes (data
not shown), suggesting that membrane integration of the
constructs may be improved by the presence of the S3-S4
linker.
Extremely short S3-S4 linkers slow both
activation and derelaxation kinetics
In addition to reduced expression, short linker constructs
with linker lengths of two or three amino acids had mark-
edly slowed activation kinetics as well (Fig. 1 E). Thus, in
addition to a stabilized relaxed state, these constructs likely
require high activation energy to transition from the resting
to the active state. One possible cause of these high energy
barriers is the severity of these truncations, which, accord-
ing to other reasonable sequence alignments, are deleting
not only the entirety of the S3-S4 linker, but also residues
within S3 and S4 (9,18). Under such constraints, it is not
surprising that a larger energy barrier is encountered when
moving the voltage sensor in these constructs.
Potential physiological roles of S3-S4 linker
length and relaxation
Our nding that lengthening the S3-S4 linker speeds up der-
elaxation kinetics in Shaker may extend to other voltage-
gated ion channels. Voltage-gated potassium channels
have very diverse S3-S4 linkers, both in terms of amino
acid sequence and length (Fig. 8 C). By comparing our re-
sults to those obtained from a literature search of prior
gating current studies of voltage-gated potassium channels,
we can examine the plausibility of these ndings extending
beyond Shaker mutants to more physiologically relevant
contexts. In Shaker, V
1/2
s of activation Q-V curves showed
no correlation with linker length overall (Fig. 1 D); simi-
larly, V
1/2
s of Q-Vs taken from the literature (3,3843)
showed no correlation between activation V
1/2
and length
of the S3-S4 linker (Fig. 8 D). Finally, it is of interest that
the unique region of the N-terminal portion of the S3-S4
2320 Priest et al.
linker, suggested by the extreme derelaxation kinetics in
some of our constructs (Fig. 7 B), corresponds quite well
to a region of sequence homology between the S3-S4 linkers
of Shaker and other Kv1 family channels (Fig. 8 C, high-
lighted in gray).
Examples of derelaxation kinetics in the literature are
extremely scarce, due to its novel description (4). However,
derelaxation can also be measured as a leftward shift in the
Q-V obtained from the relaxed state when compared to an
activation Q-V (2). Although this measure is prone to
methodical biases resulting from short integration times
(4), it can still be used as a rough measure of derelaxation
in voltage-gated ionic channels. In this way, a trend can
be appreciated between the S3-S4 length and the Q-V shifts
of other voltage-gated ion channels available in the litera-
ture (Fig. 8 E) (3,38,4446). This trend ts with the leftward
Q-V shift observed upon derelaxation in our short linker
Shaker constructs (Fig. 8 B). This suggests that the effects
on the relaxed state described in Shaker may extend to other
voltage-gated ion channels.
Other factors affecting VSD relaxation
Our proposed direct effect of the S3-S4 linker on the
motion of the voltage sensor to inuence the relaxed state
is in good accordance with growing evidence that the
relaxed state is intrinsic to the voltage sensor. Indeed,
although pore coupling has been implicated as a require-
ment for entrance to the relaxed state (23), recent works
showed that the isolated VSD from the Ciona intestinalis
voltage-sensitive phosphatase (Ci-VSP) displays relaxa-
tion, albeit faster than the full-length Ci-VSP (3). Thus,
the presence of the catalytic (pore or phosphatase) domain
seems able to inuence the energetics underlying VSD
transitions toward and from the relaxed state, but is not
necessary to trigger them. Other recent ndings have
implicated open-state stabilization via intracellular cations
(47) and closed-state stabilization by phosphatidylinositol-
4,5-bisphosphate (48,49). Hence, modulations of VSD
conformations by intracellular cations and membranal
phosphoinositides may also affect relaxation/derelaxation
kinetics in vivo.
In summary, our results show the importance of the length
of the S3-S4 linker in stabilizing the relaxed state. Although
lengthening or genetically splitting the linker has minimal
effects on gating currents, short S3-S4 linkers appear to
provide a mechanical constraint on the voltage sensor as it
derelaxes to the resting state. The relationship between
S3-S4 linker length and the relaxed state that is presented
here may underlie important physiological processes
controlled by a wide range of voltage-gated ion channels.
This work was supported by the National Institutes of Health grant R01
GM030376. M.F.P. was supported by the Pritzker Fellowship in Neurosci-
ence, a Howard Hughes Medical Institute Med into Grad fellowship, and
the National Institutes of Health grants T32 GM7839 and F31 NS081954.
REFERENCES
1. Olcese, R., R. Latorre, ., E. Stefani. 1997. Correlation between charge
movement and ionic current during slow inactivation in Shaker K
channels. J. Gen. Physiol. 110:579589.
2. Villalba-Galea, C. A., W. Sandtner, ., F. Bezanilla. 2008. S4-based
voltage sensors have three major conformations. Proc. Natl. Acad.
Sci. USA. 105:1760017607.
3. Labro, A. J., J. J. Lacroix, ., F. Bezanilla. 2012. Molecular mechanism
for depolarization-induced modulation of Kv channel closure. J. Gen.
Physiol. 140:481493.
4. Lacroix, J. J., A. J. Labro, and F. Bezanilla. 2011. Properties of deac-
tivation gating currents in Shaker channels. Biophys. J. 100:L28L30.
5. Tan, P. S., M. D. Perry, ., A. P. Hill. 2012. Voltage-sensing domain
mode shift is coupled to the activation gate by the N-terminal tail of
hERG channels. J. Gen. Physiol. 140:293306.
6. Seoh, S. A., D. Sigg, ., F. Bezanilla. 1996. Voltage-sensing residues
in the S2 and S4 segments of the Shaker K channel. Neuron.
16:11591167.
7. Aggarwal, S. K., and R. MacKinnon. 1996. Contribution of the S4
segment to gating charge in the Shaker K channel. Neuron.
16:11691177.
8. Xu, Y., Y. Ramu, and Z. Lu. 2010. A Shaker K channel with a
miniature engineered voltage sensor. Cell. 142:580589.
9. Xu, Y., Y. Ramu, ., Z. Lu. 2013. Energetic role of the paddle motif in
voltage gating of Shaker K() channels. Nat. Struct. Mol. Biol.
20:574581.
10. Sand, R., N. Sharmin, C. Morgan, and W. J. Gallin. 2013. Fine-tuning
of voltage sensitivity of the Kv1.2 potassium channel by inter-helix
loop dynamics. J Biol Chem. 288:96869695.
11. Gonzalez, C., E. Rosenman, ., R. Latorre. 2000. Modulation of the
Shaker K() channel gating kinetics by the S3-S4 linker. J. Gen.
Physiol. 115:193208.
12. Gonzalez, C., E. Rosenman, ., R. Latorre. 2001. Periodic perturba-
tions in Shaker K channel gating kinetics by deletions in the S3-S4
linker. Proc. Natl. Acad. Sci. USA. 98:96179623.
13. Srensen, J. B., A. Cha, ., F. Bezanilla. 2000. Deletion of the S3-S4
linker in the Shaker potassium channel reveals two quenching groups
near the outside of S4. J. Gen. Physiol. 115:209222.
14. Tsang, S. Y., H. Lesso, and R. A. Li. 2004. Dissecting the structural and
functional roles of the S3-S4 linker of pacemaker (hyperpolarization-
activated cyclic nucleotide-modulated) channels by systematic length
alterations. J. Biol. Chem. 279:4375243759.
15. Mathur, R., J. Zheng, ., F. J. Sigworth. 1997. Role of the S3-S4 linker
in Shaker potassium channel activation. J. Gen. Physiol. 109:191199.
16. Perozo, E., R. MacKinnon, ., E. Stefani. 1993. Gating currents from a
nonconducting mutant reveal open-closed conformations in Shaker K
channels. Neuron. 11:353358.
17. Hoshi, T., W. N. Zagotta, and R. W. Aldrich. 1990. Biophysical and
molecular mechanisms of Shaker potassium channel inactivation.
Science. 250:533538.
18. Long, S. B., X. Tao, ., R. MacKinnon. 2007. Atomic structure of a
voltage-dependent K channel in a lipid membrane-like environment.
Nature. 450:376382.
19. Chen, X., Q. Wang, ., J. Ma. 2010. Structure of the full-length Shaker
potassium channel Kv1.2 by normal-mode-based X-ray crystallo-
graphic renement. Proc. Natl. Acad. Sci. USA. 107:1135211357.
20. Vargas, E., F. Bezanilla, and B. Roux. 2011. In search of a consensus
model of the resting state of a voltage-sensing domain. Neuron.
72:713720.
21. Stefani, E., and F. Bezanilla. 1998. Cut-open oocyte voltage-clamp
technique. Methods Enzymol. 293:300318.
22. Shirokov, R. 2011. Whats in gating currents? Going beyond
the voltage sensor movement. Biophys. J. 101:512514, discussion
515516.
23. Haddad, G. A., and R. Blunck. 2011. Mode shift of the voltage sensors
in Shaker K channels is caused by energetic coupling to the pore
domain. J. Gen. Physiol. 137:455472.
24. Hodgkin, A. L., and A. F. Huxley. 1952. The dual effect of membrane
potential on sodiumconductance in the giant axon of Loligo. J. Physiol.
116:497506.
25. Yang, Y., Y. Yan, and F. J. Sigworth. 1997. How does the W434F
mutation block current in Shaker potassium channels? J. Gen. Physiol.
109:779789.
26. Gamal El-Din, T. M., G. Q. Martinez, ., W. A. Catterall. 2013. A
gating charge interaction required for late slow inactivation of the
bacterial sodium channel NavAb. J. Gen. Physiol. 142:181190.
27. Naranjo, D., L. Kolmakova-Partensky, and C. Miller. 1997. Expression
of split Shaker K channels in Xenopus oocytes. 1997 Biophysical
Society Meeting Abstracts. Biophysical Journal, Supplement, A11,
Abstract, M-AM-G4.
28. Broomand, A., R. Mannikko, ., F. Elinder. 2003. Molecular
movement of the voltage sensor in a K channel. J. Gen. Physiol.
122:741748.
29. Henrion, U., J. Renhorn, ., F. Elinder. 2012. Tracking a complete
voltage-sensor cycle with metal-ion bridges. Proc. Natl. Acad. Sci.
USA. 109:85528557.
30. Gamal El-Din, T. M., H. Heldstab, ., N. G. Greeff. 2010. Double gaps
along Shaker S4 demonstrate omega currents at three different closed
states. Channels (Austin). 4:93100.
31. Posson, D. J., and P. R. Selvin. 2008. Extent of voltage sensor move-
ment during gating of Shaker K channels. Neuron. 59:98109.
32. Sigg, D., and F. Bezanilla. 1997. Total charge movement per channel.
The relation between gating charge displacement and the voltage sensi-
tivity of activation. J. Gen. Physiol. 109:2739.
33. Jensen, M. O., V. Jogini, ., D. E. Shaw. 2012. Mechanism of voltage
gating in potassium channels. Science. 336:229233.
34. Tiriveedhi, V., M. Miller, ., M. Li. 2012. Autonomous transmembrane
segment S4 of the voltage sensor domain partitions into the lipid mem-
brane. Biochim. Biophys. Acta. 1818:16981705.
35. Tu, L., J. Wang, ., C. Deutsch. 2000. Transmembrane biogenesis of
Kv1.3. Biochemistry. 39:824836.
36. Sato, Y., M. Sakaguchi, ., N. Uozumi. 2002. Integration of Shaker-
type K channel, KAT1, into the endoplasmic reticulum membrane:
synergistic insertion of voltage-sensing segments, S3-S4, and indepen-
dent insertion of pore-forming segments, S5-P-S6. Proc. Natl. Acad.
Sci. USA. 99:6065.
37. Tu, L. W., and C. Deutsch. 2010. A folding zone in the ribosomal exit
tunnel for Kv1.3 helix formation. J. Mol. Biol. 396:13461360.
38. Piper, D. R., A. Varghese, ., M. Tristani-Firouzi. 2003. Gating cur-
rents associated with intramembrane charge displacement in HERG
potassium channels. Proc. Natl. Acad. Sci. USA. 100:1053410539.
39. Dougherty, K., and M. Covarrubias. 2006. A dipeptidyl aminopepti-
dase-like protein remodels gating charge dynamics in Kv4.2 channels.
J. Gen. Physiol. 128:745753.
40. Dougherty, K., L. Tu, ., M. Covarrubias. 2009. The dipeptidyl-
aminopeptidase-like protein 6 is an integral voltage sensor-interacting
beta-subunit of neuronal K(V)4.2 channels. Channels (Austin).
3:122128.
41. Zhang, S., S. J. Kehl, and D. Fedida. 2001. Modulation of Kv1.5 potas-
sium channel gating by extracellular zinc. Biophys. J. 81:125136.
42. Taglialatela, M., and E. Stefani. 1993. Gating currents of the cloned
delayed-rectier K channel DRK1. Proc. Natl. Acad. Sci. USA.
90:47584762.
43. Shieh, C. C., K. G. Klemic, and G. E. Kirsch. 1997. Role of transmem-
brane segment S5 on gating of voltage-dependent Kchannels. J. Gen.
Physiol. 109:767778.
44. Kuzmenkin, A., F. Bezanilla, and A. M. Correa. 2004. Gating of the
bacterial sodium channel, NaChBac: voltage-dependent charge move-
ment and gating currents. J. Gen. Physiol. 124:349356.
45. Bezanilla, F., R. E. Taylor, and J. M. Fernandez. 1982. Distribution and
kinetics of membrane dielectric polarization. 1. Long-term inactivation
of gating currents. J. Gen. Physiol. 79:2140.
46. Shirokov, R., R. Levis, ., E. R os. 1992. Two classes of gating current
from L-type Ca channels in guinea pig ventricular myocytes. J. Gen.
Physiol. 99:863895.
47. Goodchild, S. J., H. Xu, ., D. Fedida. 2012. Basis for allosteric open-
state stabilization of voltage-gated potassium channels by intracellular
cations. J. Gen. Physiol. 140:495511.
48. Abderemane-Ali, F., Z. Es-Salah-Lamoureux, ., G. Loussouarn.
2012. Dual effect of phosphatidylinositol (4,5)-bisphosphate PIP(2)
on Shaker K() channels. J. Biol. Chem. 287:3615836167.
49. Rodriguez-Menchaca, A. A., S. K. Adney, ., D. E. Logothetis. 2012.
PIP2 controls voltage-sensor movement and pore opening of Kv
channels through the S4-S5 linker. Proc. Natl. Acad. Sci. USA.
109:E2399E2408.
2322 Priest et al.
Mechanism of Membrane Permeation Induced by Synthetic b-Hairpin
Peptides
Kshitij Gupta,
Hyunbum Jang,
Kevin Harlen,
Anu Puri,
Ruth Nussinov,
Joel P. Schneider,
and Robert Blumenthal
Basic Research Laboratory, Center for Cancer Research,

Basic Science Program, SAIC-Frederick, and

Peptide Design and Materials
Section, Chemical Biology Laboratory, National Cancer Institute, National Institutes of Health, Frederick, Maryland
ABSTRACT We have investigated the membrane destabilizing properties of synthetic amphiphilic cationic peptides, MAX1
and MAX35, which have the propensity to formb-hairpin structures under certain conditions, and a control non-b-hairpin-forming
peptide MAX8V16E. All three peptides bind to liposomes containing a mixture of zwitterionic POPC and negatively charged
POPS lipids as determined by Zeta potential measurements. Circular dichroism measurements indicated folding of MAX1
and MAX35 in the presence of the POPC/POPS liposomes, whereas no such folding was observed with MAX8V16E. There
was no binding or folding of these peptides to liposomes containing only POPC. MAX1 and MAX35 induced release of contents
from negatively charged liposomes, whereas MAX8V16E failed to promote solute release under identical conditions. Thus,
MAX1 and MAX35 bind to, and fold at the surface of negatively charged liposomes adopting a lytic conformation. We ruled
out leaky fusion as a mechanism of release by including 2 mol % PEG-PE in the liposomes, which inhibits aggregation/fusion
but not folding of MAX or MAX-induced leakage. Using a concentration-dependent quenching probe (calcein), we determined
that MAX-induced leakage of liposome contents was an all-or-none process. At MAX1 concentrations, which cause release
of ~50% of the liposomes that contain small (R
h
<1.5 nm) markers, only ~15% of those liposomes release a uorescent dextran
of 40 kDa. A multimeric model of the pore is presented based on these results. Atomistic molecular dynamics simulations show
that barrels consisting of 10 b-hairpin MAX1 and MAX35 peptides are relatively more stable than MAX8V16E barrels in the
bilayer, suggesting that barrels of this size are responsible for the peptides lytic action.
INTRODUCTION
Antimicrobial peptides (AMPs) are a unique and diverse
group of molecules, which target and disrupt the function
of cell membranes (1). These peptides are capable of killing
bacteria (2), fungi (3), and cancer cells (46). We designed a
class of hydrogel materials from self-assembling peptides
whose surfaces display antibacterial activity (7). Interest-
ingly, the primary sequence and amphiphilic secondary
structure of these self-assembling peptides resemble clas-
sical AMPs. However, distinct from AMPs is their ability
to self-assemble into macroscopic gels. Our peptides,
initially unordered when dissolved in aqueous solution,
can be triggered to fold into an amphiphilic b-hairpin
conformation that rapidly self-assembles to form a network
of brils that constitute the formation of a hydrogel. For
example, the peptide MAX1 is composed of N- and C-ter-
minal b-strands containing alternating hydrophobic (valine)
and hydrophilic (lysine) residues (Table 1). A central four-
residue sequence (-V
D
PPT-) connects the two strands. At
neutral pH and solution of low ionic strength, electrostatic
repulsion between protonated lysine side chains keeps the
peptide unfolded. Increasing the ionic strength of a neutral
solution with NaCl to 150 mM (8) or simply increasing
the pH to 9 reduces the positive charge (9,10), allowing
the peptide to fold into a facially amphiphilic b-hairpin
that rapidly self-assembles. We have reported earlier that
the surface of the resulting macroscopic gel confers antibac-
terial activity. As will be shown, we have found that nega-
tively charged surfaces, such as that provided by model
liposomes can also induce the folding of MAX1. Here,
electrostatic interactions between the positively charged
lysine side chains and negatively charged lipid headgroups
drive the folding of the peptide into an amphiphilic hairpin.
When the peptides fold at the surface of liposome mem-
brane it is unable to undergo gelation. Rather, the folded
conformer acts to destabilize the membrane resulting in
liposome leakage. Herein, we use a combination of biophys-
ical techniques and peptide design to probe the mechanism
of action. The inuence of peptide hydrophobic content is
assessed via the study of MAX35, a derivative of MAX1
that contains four isoleucine residues in place of four
original valine side chains. Finally, a control peptide,
MAX8V16E, was prepared to study the importance of
membrane-induced folding with respect to the peptides
membrane-disrupting action. MAX8V16E contains two
sequential glutamate residues that render it incapable of
folding at the surface of the membrane.
We used unilamellar vesicles (liposomes) with an average
diameter of 100300 nm consisting of a neutral phos-
pholipid, 1-palmitoyl-2-oleoyl-sn-glycero-phosphocholine
(POPC)
#
, and a negatively charged phospholipid, 1-palmi-
toyl-2-oleoyl-sn-glycero-3-phosphor-L-serine (POPS), at
Submitted August 6, 2013, and accepted for publication September 26,
2013.
*Correspondence: blumenthalr@mail.nih.gov
Editor: Huey Huang.
Biophysical Journal Volume 105 November 2013 20932103 2093
various mole ratios. Circular dichroism (CD) spectroscopy
was employed to assess the b-hairpin forming propensity
of these peptides as a function of negative charge on the
liposomes. Release of liposome contents as a result of inter-
action with MAX peptides was monitored using different
molecular weight markers encapsulated into the liposomes.
Our data show that addition of MAX1 and MAX35 to
liposomes containing a sufcient negative charge (>10%
PS) results in a rapid release of small molecules in a dose
and negative surface charge-dependent manner. At MAX1
concentrations, which cause release of ~50% of the
liposomes that contain small (R
h
<1.5 nm) markers, only
~15% of those liposomes release a uorescent dextran of
40 kDa. We surmise that the 15% value represents the
formation of very large (diameter >10 nm) pores and the
remaining liposomes release their contents via pores that
can discriminate between molecules of different sizes.
Atomistic molecular dynamics (MD) simulations are
consistent with the notion that these peptides form b-barrels
in the lipid bilayer, which are relatively stable and allow
permeation of molecules with dened size limitations.
Phospholipids POPC, POPS, and 1,2-distearoyl-sn-glycero-3-phosphoetha-
nolamine-N-[methoxy(polyethyleneglycol)-2000] (PEG-PE) were pur-
chased from Avanti Polar Lipids (Alabaster, AL). Tb (III) Chloride
Hexahydrate (TbCl
3
$6H
2
O), 2,6-Pyridinedicarboxylic acid (Dipicolinic
acid), octyl b-D-glucopyranoside (OG), and other reagents were purchased
from Sigma-Aldrich (St. Louis, MO). SephadexG-50 and Sepharose CL-6B
were purchased from GE Healthcare Biosciences AB (Uppsala, Sweden).
Texas red dextran 40,000 (TRD-40 k) was purchased from Invitrogen.
Appropriately side-chain protected Fmoc-amino acids were purchased
from Novabiochem (Billerica, MA). 2-(6-Chloro-1H-benzotriazole-1-yl)-
1,1,3,3-tetramethyl-aminium hexauorophosphate (HCTU) was purchased
from Peptides International (Louisville, KY).
3
H-inulin was purchased
from Perkin Elmer (Boston, MA). Buffer 1, 10 mM TES, 100 mM NaCl,
1 mM EDTA (pH 7.4); Buffer 2, 10 mM TES, 100 mM NaCl (pH 7.4);
Stock solutions: Tb/DPA, 7.5 mM TbCl3, 75 mM Na citrate, 75 mM Na
dipicolinate, 10 mM TES (pH 7.4). Inulin solution was prepared in buffer
1 (pH 6.6) at a concentration of 1% weight by volume by incubation for
2 h at 55
C followed by cooling at room temperature. Any insoluble mate-

rial was removed by centrifugation and the clear supernatant was used
further. Calcein solution was prepared at a nal concentration of 100 mM
(pH 7.4).
Peptides were prepared on PL-Rink Resin in 0.25 mM quantities via
automated Fmoc peptide synthesis employing an ABI 433Apeptide synthe-
sizer (Applied Biosystems, Snoqualmie, WA) and HCTU activation. The
resulting dry resin-bound peptides were cleaved and side-chain-deprotected
using a TFA: thioanisole: ethanedithiol: anisole (90:5:3:2) cocktail. Crude
peptides were puried by RP-HPLC (Agilent technologies, Santa Clara,
CA) using preparative Grade C18 peptide/protein column (Vydac, Hespe-
ria, CA). Peptides were puried using a linear gradient from solvent A
(0.1% TFA in water) to solvent B (90% Acetonitrile, 9.9% water, 0.1%
TFA). The following gradients were used for each peptide: MAX1 15%
to 100% B over 159 min; MAX35 18% to 100% B over 170 min;
MAX8V16E 16% to 100% B over 175 min. MAX1 electrospray ioniza-
tion-mass spectrometry (ESI-MS, Shimadzu, Columbia, MD): 2231.4
[(MH)
calculated 2229.96]; MAX35 ESI-MS: 2287.4 [(MH)
calcu-
lated 2286.1]; MAX8V16E ESI-MS: 2262.2 [(MH)
calculated
2260.88]. More detailed methods including liposome preparation, size
and Zeta potential measurements, CD studies, solute release assays, and
atomistic MD simulations are available in the Supporting Material, Docu-
ment S1.
RESULTS
Structural changes for MAX peptides upon
binding to negatively charged liposomes
For our studies on MAX peptide-liposome interactions we
used POPC (zwitterionic) and/or POPS (negatively charged)
as the prototype lipids. We characterized these liposomes
with respect to their size and surface charge (Table 1). We
performed a one way analysis of variance of the liposome
size and Zeta potentials listed in Table 1 (see Document
S1 in the Supporting Material). The measured parameters
with peptide were tested against the liposomes without the
peptide (controls). This analysis shows no signicant
changes in size of liposomes (POPC alone or POPC/
POPS) in the presence or absence of MAX peptides. There-
fore, we conclude that the binding of MAX peptides to
POPC and POPS/POPC liposomes does not affect their
size distribution. These data rule out MAX-induced aggre-
gation or fusion of these liposomes. There were no signi-
cant changes in Zeta potentials of POPC liposomes in the
presence of the MAX peptides and of POPS/POPC lipo-
somes in the presence of MAX8VE16. By contrast, Zeta
potentials of POPS/POPC liposomes were signicantly
different in the presence of MAX1 and MAX35 indicating
binding of these peptides to these liposomes. Because bind-
ing and folding of MAX1 and MAX35 peptides are electro-
statically driven, there are much higher probabilities for
TABLE 1 Physical properties of liposomes
Liposome formulation Physical characteristic No peptide MAX35 MAX1 MAX8V16E
POPC Diameter (nm) 97 519 112 53 104 5 10 110 5 7
Zeta potential (mV) 7.5 52.0 3.0 52.0 3.7 5 1.5 5.7 5 1.2
POPC/POPS (1:1) Diameter (nm) 79 5 1 106 514 109.0 524 96.0 516
Zeta potential (mV) 35.4 5 2.2 19.7
a
51.6 23.9
a
52.3 30.4 5 2.4
Sequence MAX1: VKVKVKVKV
D
PPTKVKVKVKV-NH
2
MAX35: IKVKIKVKV
D
PPTKIKVKIKV-NH
2
MAX8V16E: VKVKVKVKV
D
PPTKVEEKVKV-NH
2
Diameter and Zeta potential of liposome with or without MAX peptides. The sequences of the MAX peptides are shown in the bottom row.
a
Values were signicantly different from no peptide control according to an analysis of variance test, p % 0.05, n 2.
2094 Gupta et al.
these peptides to bind anionic lipid liposome compared to
zwitter-ionic POPC vesicles.
CD spectroscopy shows that MAX1 and MAX35 fold at
the surface of negatively charged liposomes. Fig. 1 A shows
that when MAX1 is dissolved in buffer alone, it remains
unfolded, but folds when negatively charged POPS-contain-
ing liposomes are present as evident by the increase in nega-
tive mean residue ellipticity at ~218 nm. In contrast, when
neutral POPC liposomes are present, the peptide remains
unfolded. The data suggest that MAX1 binds to and folds
at the surface of negatively charged liposomes and that the
driving force for folding is electrostatically driven. Presum-
ably, the lysine side chains of the peptide can form salt
bridge pairs with the POPS headgroups to drive the folding
event, which leads to the formation of an amphiphilic
hairpin. When the peptide initially folds at the surface of
the liposome, the lysine-rich face of the hairpin is engaged
with the lipid headgroup region and its hydrophobic face is
exposed to bulk water. Solvating this hydrophobic surface
area with water is an energetically unfavorable proposition.
In response, the peptide partitions into the membrane to
release ordered water and solvate its hydrophobic face
with lipids hydrophobic core. Fig. 1 B shows that
MAX35, the more hydrophobic peptide behaves in a similar
fashion. However, Fig. 1 C shows that the control peptide,
MAX8V16E is unable to fold at the surface of negatively
charged liposomes. Here, the anionic glutamates, through
charge repulsion with the negatively charged liposome,
disfavors binding (see Table 1) and folding. The CD data
taken together indicate that both MAX1 and MAX35 fold
at the surface of negatively charged liposomes, adopting
amphiphilic hairpin conformations. The exact mechanism
by which the peptide partitions into the membrane after
the initial binding and folding events is not known. But as
will be shown, once partitioned into the membrane, the
folded amphiphiles can induce liposomal leakage.
b-hairpin peptides induced membrane
permeabilization
To monitor the permeabilization of liposomes upon interac-
tion with MAX peptides, we have employed a uorescence-
based Tb
3
/DPA assay (11). Tb
3
/DPA chelates, which are
~10,000 times more uorescent than free Tb
3
, were encap-
sulated into liposomes. Upon the release of liposome
contents into a medium containing EDTA, uorescence
decreases due to the dissociation of the Tb
3
/DPA com-
plexes. Maximum release was calculated by adding OG
to the medium. Fig. 2 shows the kinetics of leakage of
POPC/POPS liposomes upon addition of MAX1 and
MAX35. Strikingly, the addition of MAX1 (Fig. 2 A) and
MAX35 (Fig. 2 B) resulted in a rapid release of liposome
contents reaching a steady state within ~3 s.
The CD and leakage data taken together indicate that
formation of b-structure is required for the permeabilization
of the liposome membrane and that the interaction of posi-
tively charged peptide with negatively charged lipids is
necessary but not sufcient. Various mechanisms have
been invoked to account for the mode of action of lytic pep-
tides on membrane permeabilization. One such mechanism
posits that the peptide induces aggregation/fusion of lipo-
somes that is accompanied by leakage of liposome contents
(leaky fusion) (12). However, aggregation/fusion of lipo-
somes is prevented by inclusion of pegylated lipid into
liposomes (12). CD spectra indicate that folding of the
MAX peptides was identical in our liposome formulations
prepared with or without inclusion of 2 mol % PEG-PE
(data not shown). Fig. 3 shows similar leakage for pegylated
(right panels) versus non-pegylated (left panels) liposomes
ruling out leaky fusion as a mechanism for MAX peptide-
induced membrane permeabilization.
Even though sufcient peptide is available at L/P 24 to
form at least one pore (see below) in all the liposomes, the
FIGURE 1 CD spectra of MAX peptides in buffer 1 (B), POPC
liposomes (,), and POPC/POPS (1:1) liposomes (-) for (A) MAX1 (B)
MAX35, and (C) MAX8V16E. All samples contain 50 mM peptide and a
lipid/peptide ratio of 50:1.
Membrane Permeation Induced by Synthetic b-Hairpin Peptides 2095
extent of leakage only reached ~50% (Fig. 2, A and B). We
therefore examined whether adding more peptide would
cause additional leakage. Indeed, at POPS/POPC ratios
R50%, more extensive leakage is realized (Fig. 3). How-
ever, at 10 mol % POPS the extent of leakage did not go
beyond ~30% even at low L/P ratios. These data indicate
that a threshold of POPS in the liposome is required for
membrane permeabilization.
Fig. 4 shows dose-response curves for the extent of
leakage induced by MAX1 (Fig. 4 A) and MAX35
(Fig. 4 B) peptides, respectively. Overall, the activities of
the two peptides do not appear to be very different except
for higher activity of MAX35 at 10 mol % POPS (see above)
in the liposomes. MAX activities in liposomes containing 50
and 75 mol % POPS seem to be higher than in 98% POPS
liposomes (Fig. 3), whereas in liposomes containing
10 mol % POPS the maximal release even at high peptide
concentrations remains in the range of 1540%. Fitting
the curves to a Langmuir isotherm (see legend) yields
maximal extents of release and the peptide concentrations
for half-maximal activity. In Fig. 4 C we show the maximal
extents of leakage for MAX1, MAX35, and MAX8V16E as
a function of mol % POPS in the liposomes. Whereas the
leakage activities of MAX1 and MAX35 are similar at
POPS >25 mol %, it appears that at 10 mol % POPS in
the liposomes, MAX35 induces higher leakage activity.
This may be the result of greater hydrophobicity of
MAX35 resulting in a higher probability of pore formation.
Interestingly, a decrease is seen in the extent of leakage from
liposomes containing very high POPS levels suggesting that
initial peptide binding is not sufcient for membrane disrup-
tion. A similar observation has been reported by Ramamoor-
thy and co-workers (13) who showed that the antimicrobial
peptide pexiganan cannot disrupt purely anionic lipid bila-
yers like POPG or POPS as they bind strongly to the bilayer
surface. MAX8V16E had no signicant binding activity
(Table 1) and did not induce leakage from liposomes at all
POPS concentrations tested.
Release mechanisms
Liposomes containing differentially sized markers were
incubated with MAX1 and fractionated through a sizing
column to separate liposome-containing macromolecules
from the free macromolecules. Fig. 5 A shows that MAX1
induced ~33% leakage of inulin (mw. ~5000; R
h
~1.39 nm) from POPC/POPS liposomes. To test whether
MAX peptide-induced leakage may result from formation
of pores that can discriminate between molecules of
different sizes, we encapsulated both calcein (mw.
~622.55; R
h
~0.74 nm) and Texas red dextran 40,000
(TRD-40 k, mw. ~40,000; R
h
~5 nm) into the same lipo-
somes and then examined the differential release of both
molecules induced by MAX1. Under conditions that cause
release of ~50% of the liposomes that contain calcein,
only ~15% of those liposomes released the TRD-40 k
(Fig. 5 B). The data indicate that under these experimental
conditions 15% of the liposomes release their contents as
a result of very large (diameter >10 nm) pore formation,
whereas the remaining liposomes contain pores that allow
permeation of molecules with dened size limitations.
The question is whether leakage via the (remaining) pores
is graded or all or none. In other words when we measure
50% leakage do all vesicles leak 50% of their contents or
do 50% of the vesicles leak all of their contents and the re-
maining vesicles remain intact. In the former case the pores
will open for a very short amount of time, insufcient to
allow complete depletion of vesicle contents. We followed
the procedure developed by Weinstein et al. (14) to make
this determination. Because these molecules exhibit a con-
centration-dependent quenching mechanism, quenching ra-
tios (Q) following MAX-induced leakage and repassing the
vesicles through a column should indicate the appropriate
mechanism. In the case of all or none the recovered vesicles
should exhibit the same quenching ratios as the original
vesicles, whereas in the graded case the quenching ratios
should change. Fig. 6 clearly shows that the recovered ves-
icles closely follow the original curve indicating an all-or-
none mechanism. The data suggest the formation of a very
FIGURE 2 Kinetics of leakage of Tb/DPA from POPC/POPS liposomes
containing varying % of POPS viz; 10% (;), 25% (-), 50% (C) upon
addition of (A) MAX1 and (B) MAX35 peptides at a lipid/peptide ratio
of 24:1. To see this gure in color, go online.
2096 Gupta et al.
stable pore allowing enough time for the peptide-challenged
vesicles to deplete their contents.
Atomistic MD simulations
To gain more insight into possible pore-forming properties
of MAX peptides, we performed atomistic MD simulations
using the CHARMM program (15). Using the coordinates of
an idealized b-hairpin, we constructed annular b-sheets in
the barrel (Fig. 7 A) and channel (Fig. 7 B) topologies. To
construct the barrel structure, the b-hairpin was inclined
~37
relative to the pore axis and then a 10-fold rotational

symmetry operation was performed with respect to the
pore axis, creating a decameric MAX barrel (16). To
construct the channel structure with the conventional
b-strands arrangement, 10 b-hairpins were inserted without
inclination, generating a decameric MAX channel (17,18).
We modeled MAX barrels/channels with the b-sheet struc-
ture. For MAX barrels, our simulations employed the b-bar-
rel morphology by mimicking naturally occurring b-barrels
observed in transmembrane proteins that are found
frequently in the outer membranes of bacteria, mitochon-
dria, and chloroplasts. The b-barrel motif is a large b-sheet
composed of an even number of b-strands. Some known
structures of b-barrel membrane proteins have b-strands
ranging in size from 8 to 22 (19,20). Our modeled 10-mer
MAX barrels contain 20 b-strands enclosing the solvated
pore. This number is also in the range of the number of
b-strands for natural b-barrels ranging from 8 to 22, which
can form a b-barrel motif. For MAX channels, we modeled
the b-sheet channel morphology for an extension of the
b-barrel morphology. The decameric MAX b-sheet chan-
nels also contain 20 b-strands enclosing the solvated pore.
Our previous simulations for Ab channels, which contain
the same b-sheet channel morphology, indicate that
different numbers of Ab monomers could produce channels
with different outer and pore dimensions (2125). For the
b-sheet channel morphology, we found that Ab channels
obtained a preferred size in the range of 1624 b-strands lin-
ing the pores (23,24). This range was also found to hold for
other toxic b-sheet channels; K3 (a fragment of b
2
-micro-
globulin) channels with 24 b-strands (26) and protegrin-1
(PG-1) channels with 1620 b-strands (17,18). In both bar-
rel/channel topologies, the b-hairpin arrangements give rise
to two potential b-sheet motifs; turn-next-to-tail b-hairpins
in NCCN packing mode and turn-next-to-turn b-hairpins
in NCNC packing mode (27), where N and C denote the
N- and C-terminal strands, respectively (Fig. 7). In both
cases, the b-hairpins form antiparallel b-sheets, positioning
the positively charged Lys side chains into the central pore
(Fig. S1 in the Supporting Material). For PG-1 b-hairpin,
intermolecular packing and alignment in the ordered
b-hairpin oligomerization have been determined (2830).
These experiments suggested some possible b-hairpin pack-
ings with antiparallel and parallel intermolecular b-strand
arrangements. However, for the MAX b-hairpin, no experi-
mental data for the peptide arrangement is currently avail-
able. In our simulation, we modeled MAX b-sheet barrels/
channels with the antiparallel b-hairpin arrangement. This
ensures that all charged Lys side chains point toward the
solvated pore. However, in the parallel MAX b-hairpin
arrangement, half of the peptides in the barrels/channels
FIGURE 3 Sequential addition of MAX pep-
tides to the same liposomal formulation (nonpegy-
lated and pegylated 2 mol % PEG-PE) of POPC/
POPS cause additional Tb/DPA leakage. Each
liposome formulation contains a different % of
POPS. MAX1 and MAX35 were added at a con-
centration of 0.25 mM successively to each lipo-
somal formulation after a certain interval of time
resulting in Tb/DPA leakage at each addition.
Each successive increase in the prole shows Tb/
DPA leakage corresponding to the peptide addi-
tion. The last addition corresponds to the addition
of detergent OG to obtain maximal leakage.
unfavorably point their charged Lys side chains toward the
lipid hydrophobic core. We constructed 12 independent
initial conformations of MAX barrels and channels for
explicit solvent MD simulations with a mixed zwitterion/
anionic lipid bilayer composed of POPC/POPS (mole ratio
3:1).
The MAX barrels/channels are preassembled initially as
an annular shape with 10 b-hairpins. During the simulations,
the initial annular conformation is further optimized in the
MAX1 and MAX35 barrels preserving the stable b-barrel
topology through the formation of the inter- and intramolec-
ular backbone hydrogen bonds (H-bonds) (Fig. 8), although
the control barrel composed of MAX8V16E and all chan-
nels (composed of MAX1, MAX35, and MAX8V16E) grad-
ually dismiss the annular shape via relaxation of the lipid
bilayer yielding discontinuities in the b-sheet network
(Fig. S2). All MAX barrels/channels increase their outer
diameter, 12 nm from the starting points, due to the charge
repulsion between the Lys side chains in the solvated pore.
FIGURE 4 (A) Dose response curve for the extent of Tb/DPA leakage
induced by the addition of MAX1 peptide to liposomes of POPC/POPS/
PEG-PE. Each liposome formulation contains a different % of POPS.
(B) Dose response curve for the extent of Tb/DPA leakage induced by
the addition of MAX35 peptide to liposomes of POPC/POPS/PEG-PE.
Each liposome formulation contains a different % of POPS. The data
were tted to a Langmuir isotherm f a*x/(bx), where f is extent of
release, a represents maximal release, and b is the concentration of peptide
at half-maximal release. (C) The maximal extents of release are plotted
against %POPS in the liposome. Blue bars: MAX1; brown bars: MAX35;
green bars: MAX8V16E.
FIGURE 5 Release of macromolecules of different hydrodynamic radii
(R
h
). Release from POPC/POPS/PEG-PE (1:1:0.02, mole ratio) liposomes
was determined upon addition of MAX1 peptide at L/P 24. (A) % Release
of
3
H-inulin (mw. ~5000; R
h
~1.39 nm), blue line denotes untreated lipo-
somes and red line denotes liposomes treated with MAX1. (B) Coencapsu-
lated TRD-40 k (mw. ~40,000; R
h
~5 nm) and calcein (mw. ~622.55; R
h
~0.74 nm). The uorescence intensity was measured at l
ex
490 nm
and l
em
517 nm (calcein) for untreated (blue line) and MAX1 treated
(red line), respectively, and at l
ex
590 nm and l
em
615 nm (TRD-
40 k) for untreated (green line) and MAX1 treated (purple line), respec-
tively. The peaks at higher fraction numbers represent released molecules.
The % release was calculated from the formula; 100 [total counts (
3
H or
uorescence) in the released fractions]/[the total counts in all fractions].
The average release calculated from two separate experiments was 51 5
2, 33.3 58.3, and 15.5 51.5% for calcein, inulin, and TRD-40 k, respec-
tively.
2098 Gupta et al.
Both MAX1 and MAX35 barrels preserve the solvated pore
with pore diameter ~2.0 nm, wide enough for conducting
ions or small molecules as calculated by the HOLE program
(31). Although the b-sheets are partially broken in the
MAX1 and MAX35 channels, they also retain the solvated
pore. However, the MAX8V16E barrels/channels generally
collapse the pore due to unstable b-hairpins in the lipid
bilayer (Fig. S3).
The interaction of individual MAX b-hairpins with
surrounding solvent including lipids, water, and ions is inho-
mogeneous. Well-balanced b-hairpin interactions with
solvent could support stable barrel/channel conformation
in the complex lipid environment. We calculated the interac-
tion energy for each MAX b-hairpin with lipids, water, and
anions (Cl
). The b-hairpin interaction energy was then

averaged over time and the number of peptides in the bar-
rel/channel. The percentage of peptide-solvent interactions
for the three components, i.e., lipid, water, and anion, repre-
sents the relative strength of peptide interactions made with
the surrounding environment (Fig. S4). For the MAX1 and
MAX35 barrels, the lipid and water interactions are rela-
tively weak, whereas anion interaction is relatively strong.
In our simulations, the initial locations of ions were in the
bulk with jzj > 3 nm from the bilayer center. No ions
were initially placed in the water pore. However, during
the course of preequilibration, these ions rapidly migrated
into the water pore. Especially, the anion, Cl
penetrated
into the pore and screened the charged Lys-Lys side-chains
interactions. After the initial transient t > 30 ns, we found
that the average number of anions in the pore is ~40 Cl
.
Three peaks in the probability distribution curve for Cl
denote the anion locations in between four Lys side chains

lined along the pore axis, indicating that anions effectively
screen the positively charged Lys side chains in the solvated
pore (Fig. 9). The MAX1 and MAX35 channels exhibit
similar interaction patterns as the barrel motif, suggesting
that the channel topology is in the metastable state toward
more stable b-barrel conformation. In the starting conforma-
tions of MAX1 and MAX35 channels, b-hairpins are in-
serted conventionally without any tilt with respect to the
membrane normal. However, we observed that the b-strand
tilt angle increases, close to the values of inclined b-hairpins
in the MAX barrel topology during the simulations
(Fig. S5). The MAX8V16E is unstable in the lipid bilayer
and partitions back into solution.
DISCUSSION
Peptide-induced leakage of various molecules through lipo-
some membranes studied here involves binding of the pep-
tide to the membrane surface, followed by conformational
changes, peptide aggregation, insertion, and formation of
a pore (1). We have been able to follow these events using
FIGURE 6 MAX1-induced relief of self-quenching of calcein. The
quenching ratio (Q) in each POPC/POPS/PEG-PE (1:1:0.02, mole ratio)
liposome preparation that encapsulated calcein at different concentrations
was determined by measuring uorescence at l
ex
490 nm and l
em

517 nm before and after adding 1% OG. Black diamonds: Q of the original
liposomes; White diamonds: calculated Q for an all or none release mech-
anism; Dark triangles: calculated Q for a graded mechanism; Crosses:
experimentally determined Q after treatment of liposomes with MAX1 at
L/P 24 and removing the released calcein by passage through a Sepharose
CL-6B column.
FIGURE 7 Topological diagrams for the MAX1 b-hairpins in the anti-
parallel b-sheet arrangement for the MAX1 (A) barrel and (B) channel in
the NCCN and NCNC packing modes. Blue and white beads represent
the positively charged (Lys) and hydrophobic (Val,
D
Pro, and Pro) residues,
respectively, and the polar residue (Thr) is denoted by green beads. Dotted
lines indicate the backbone hydrogen bond (H-bond). The rst and the last
residues for each monomer are indicated by the residue number. The car-
toons representing the barrels/channels are in lateral view.
a variety of biophysical techniques. The peptides MAX1
and MAX35 bind to negatively charged membranes as
shown by Zeta potential measurements but not to liposomes
composed of phospholipids containing only zwitterionic
headgroups. Therefore, peptide binding occurs through
electrostatic interactions between positively charged amino
acids on the peptide and negatively charged headgroups on
phospholipids. Although in this study we have focused on
phosphatidylserine as the negatively charged lipid because
of its presumed overexpression in cancer cells (5,32), incor-
poration of alternative negatively charged lipids such as
phosphatidylglycerol showed similar leakage kinetics
(Fig. S6).
MAX1 and MAX35 are capable of folding on POPC/
POPS membranes as shown by CD measurements. By
contrast MAX8V16E does not fold at the surface of those
membranes presumably due to the presence of the two
sequential glutamate residues. In contrast to MAX8V16E,
both MAX1 and MAX35 induce rapid content release
when added to POPS-containing liposomes. We ruled out
leaky fusion (12) as a mechanism of release by incorpo-
rating lipid-anchored PEG (PEG-PE) into the liposome
membrane, which inhibits aggregation/fusion. Such leaky
fusion can be demonstrated in the case of Ca
2
-mediated
fusion of POPS liposomes (Fig. S7).
At high peptide/lipid ratios release of contents of all lipo-
somes can be achieved, but at lower ratios only a given
percentage of vesicles release their contents in an all-or-
none fashion. Because there are nine positively charged res-
idues on MAX1 and MAX35, nine POPS molecules on the
liposome surface would constitute a binding site that sat-
ises all the charge-charge interactions between lipids and
peptide. POPS:POPC (1:1) liposomes would then have
1553 binding sites on their surface and sufcient peptide
is available to bind to these sites (see Document S2 in the
Supporting Material for calculations). Therefore, under
these conditions a large excess of peptide is available to
form a 10 peptide-containing pore that should release small
molecules in all liposomes. The apparent contradiction
between high amounts of peptide required to produce
FIGURE 8 Averaged pore structures calculated
by the HOLE program (31) embedded in the aver-
aged barrel conformations during the simulations
for the MAX1 and MAX35 barrels in the NCCN
and NCNC packing modes. In the top view,
MAX barrels are shown in the surface representa-
tion with the color representations: hydrophobic
residues are shown in white, positively charged
residues are shown in blue, and polar residues are
shown in green. In the angle and lateral views of
the pore structure, barrel structures are shown
with the ribbon representation. For the pore struc-
tures in the surface representation, red denotes
pore radius of r < 0.8 nm, green denotes pore
radius in the range, 0.8 nm % r % 1.2 nm, and
blue denotes pore radius of r > 1.2 nm.
2100 Gupta et al.
quantitative leakage and low amounts of peptide to produce
a pore can be reconciled by considering studies showing that
certain AMPs initially bind to the membrane surface
causing an increase in outer monolayer tension that allows
peptide insertion to form a pore (33,34). Accordingly, suf-
cient binding of MAX peptides to POPS-containing
liposomes is required to produce the outer monolayer
expansion required for MAX peptide insertion. For instance,
at a peptide/lipid ratio of 1/24 only about half of the popu-
lation of POPS:POPC (1:1) liposomes release calcein and
~15% of that population release TRD-40 k (Fig. 5). From
these experiments we surmise that in 50% of the liposomes
insufcient outer monolayer tension is produced to allow
peptide insertion and no leakage is observed. We have enter-
tained the possibility that the leakage of TRD-40 k is due to
the formation of very large (>10 nm) lipidic or toroidal
pores (33,34). However, based on the theoretical analysis
and MD simulations presented here we do favor a barrel
stave model in the case of the MAX peptides. Final resolu-
tion of the pore structure will require solid-state NMR (35)
or x-ray diffraction (36).
Although no signicant differences in the leakage-
inducing capabilities between MAX1 and MAX35 could
be detected at POPS R25%, at 10% POPS, MAX35 clearly
induced greater leakage than MAX1 (Fig. 2). Presumably, in
a signicant portion of the 10% POPS containing liposome
population, an insufcient amount of surface POPS is
available to provide the underlying matrix required for
peptide binding and outer monolayer expansion. However,
at a given outer monolayer tension the more hydrophobic
MAX35 appears to have a larger probability of insertion
than MAX1. The POPS distribution on the liposome surface
and aggregation of peptides required to form a pore are
complex issues still under investigation. Because liposome
contents are emptied out within ~10 ms (see Document S2
in the Supporting Material for calculations) following pore
formation the rate-limiting steps in the overall processes
are presumably aggregation/membrane expansion/insertion.
Previous studies on AMP PG-1 have also shown that
membrane perturbation activity was dependent on their
b-hairpin forming nature (17,18,27,37,38). PG-1 is a small
b-hairpin peptide consisting of 18 amino acids with a high
content of positively charged Arg residue, and two disulde
S-S bonds between Cys side chains that constrain the hairpin
conformation (39,40). PG-1 is able to alter the permeability
of bilayers, forming b-sheet channels in the membrane
(30,35,41,42), and electrical recordings show multiple sin-
gle channel conductances (18). PG-1 and MAX peptides
are very similar: i), Both share the same b-hairpin secondary
structure. ii), They are highly positively charged. iii),
Conformational b-hairpin constraints depend on their
sequences, i.e., disulde S-S bonds versus tetrapeptide
turn sequence (-V
D
PPT-). However, the charged side chains
of the peptides play a critical role in the formation of chan-
nels in the membrane. In the PG-1 peptide, positively
charged side chains are located at the b-hairpins turn and
termini. Thus, when assembled in a channel, both ends of
the cylindrical channel are charged. This bifurcated charge
distribution in the PG-1 channel aligns the channel in the
lipid bilayer. In contrast, our modeled MAX barrels/chan-
nels contain their positively charged side chains in the inte-
rior of a solvated pore. The asymmetric charge distribution
across the b-sheet plane induces strong repulsive forces at
FIGURE 9 Three-dimensional density maps of Na
(yellow surface) and

Cl
(cyan surface) for the MAX1 barrels in the (A) NCCN and (B) NCNC
packing modes. In the maps, the averaged barrel structures are shown as
cartoons in gray. Probability distribution functions for Na
(orange line),
Cl
(blue line), and water (gray line) as a function of the pore axis are
shown.
the clustered Lys side chains. This suggests that small MAX
barrels/channels may not be stable due to closely packed
charged side chains and larger barrels, greater than the 10-
mer barrel in our simulations, can be populated.
CONCLUSIONS
CD, liposome release, and sizing experiments, as well as
modeling suggest a mechanism of membrane permeation
consistent with surface-induced peptide folding followed
by outer monolayer expansion, peptide insertion, and barrel
formation. CD shows that MAX1 and MAX35 are capable
of folding at the surface of negatively charged liposomes
forming amphiphilic hairpins. The lysine-rich face of the
hairpin interacts with the negatively charged headgroup
region of the liposome and it is this interaction that drives
the folding event. Once folded at the liposome surface, the
hydrophobic face of the hairpin would be largely exposed
to bulk water, an energetically unfavorable proposition. In
response, the hairpin partitions into the membrane to release
ordered water and solubilize its hydrophobic side chains in
the lipid. Modeling and release studies indicate that once
associated with the liposome, the hairpin rapidly assembles
to form barrel-like structures with pore diameters of at least
3 nm. The data also indicate that these barrels induce
leakage in an all-or-none fashion where the liposomes that
do undergo disruption rapidly leak all of their content. It
should be noted that the mechanism leading to liposomal
disruption undercover in this study is distinct from the
mechanism of antibacterial action enjoyed by MAX1 and
MAX35 in their gelled state. Once self-assembled into the
hydrogels brillar network, there is no free peptide avail-
able for barrel formation. Thus, MAX1 and MAX35 are
capable of disrupting membranes in both their self-assem-
bled, gel-state and their freely soluble state by distinct
mechanisms.
SUPPORTING MATERIAL
Seven gures, references (4355) and supplemental information are avail-
able at http://www.biophysj.org/biophysj/supplemental/S0006-3495(13)
01088-6.
This project has been funded in whole or in part with Federal funds from the
Frederick National Laboratory for Cancer Research, National Institutes of
Health, under contract HHSN261200800001E. This research was supported
[in part] by the Intramural Research Program of NIH, Frederick National
Lab, Center for Cancer Research. All simulations had been performed using
the high-performance computational facilities of the Biowulf PC/Linux
cluster at the National Institutes of Health, Bethesda, MD (http://biowulf.
nih.gov).
REFERENCES
1. Shai, Y. 2002. Mode of action of membrane active antimicrobial
peptides. Biopolymers. 66:236248.
2. Shai, Y., A. Makovitzky, and D. Avrahami. 2006. Host defense peptides
and lipopeptides: modes of action and potential candidates for the treat-
ment of bacterial and fungal infections. Curr. Protein Pept. Sci.
7:479486.
3. van der Weerden, N. L., M. R. Bleackley, and M. A. Anderson. 2013.
Properties and mechanisms of action of naturally occurring antifungal
peptides. Cell. Mol. Life Sci. 70:35453570. http://dx.doi.org/10.1007/
s00018000130126000011.
4. Mader, J. S., and D. W. Hoskin. 2006. Cationic antimicrobial peptides
as novel cytotoxic agents for cancer treatment. Expert Opin. Investig.
Drugs. 15:933946.
5. Sinthuvanich, C., A. S. Veiga, ., J. P. Schneider. 2012. Anticancer
b-hairpin peptides: membrane-induced folding triggers activity.
J. Am. Chem. Soc. 134:62106217.
6. Hoskin, D. W., and A. Ramamoorthy. 2008. Studies on anticancer
activities of antimicrobial peptides. Biochim. Biophys. Acta.
1778:357375.
7. Salick, D. A., J. K. Kretsinger, ., J. P. Schneider. 2007. Inherent anti-
bacterial activity of a peptide-based beta-hairpin hydrogel. J. Am.
Chem. Soc. 129:1479314799.
8. Ozbas, B., J. Kretsinger, ., D. J. Pochan. 2004. Salt-triggered peptide
folding and consequent self-assembly into hydrogels with tunable
modulus. Macromolecules. 37:73317337.
9. Rajagopal, K., M. S. Lamm, ., J. P. Schneider. 2009. Tuning the pH
responsiveness of beta-hairpin peptide folding, self-assembly, and
hydrogel material formation. Biomacromolecules. 10:26192625.
10. Schneider, J. P., D. J. Pochan, ., J. Kretsinger. 2002. Responsive
hydrogels from the intramolecular folding and self-assembly of a
designed peptide. J. Am. Chem. Soc. 124:1503015037.
11. Wilschut, J., N. Duzgunes x, ., D. Papahadjopoulos. 1980. Studies on
the mechanism of membrane fusion: kinetics of calcium ion induced
fusion of phosphatidylserine vesicles followed by a new assay for
mixing of aqueous vesicle contents. Biochemistry. 19:60116021.
12. Yang, S. T., E. Zaitseva, ., K. Melikov. 2010. Cell-penetrating peptide
induces leaky fusion of liposomes containing late endosome-specic
anionic lipid. Biophys. J. 99:25252533.
13. Lee, D. K., J. R. Brender, ., A. Ramamoorthy. 2013. Lipid composi-
tion-dependent membrane fragmentation and pore-forming mecha-
nisms of membrane disruption by pexiganan (MSI-78). Biochemistry.
52:32543263.
14. Weinstein, J. N., R. D. Klausner, ., R. Blumenthal. 1981. Phase tran-
sition release, a new approach to the interaction of proteins with lipid
vesicles. Application to lipoproteins. Biochim. Biophys. Acta.
647:270284.
15. Brooks, B. R., R. E. Bruccoleri, ., M. Karplus. 1983. Charmm - a
program for macromolecular energy, minimization, and dynamics
calculations. J. Comput. Chem. 4:187217.
16. Jang, H., F. T. Arce, ., R. Nussinov. 2010. b-Barrel topology of
Alzheimers b-amyloid ion channels. J. Mol. Biol. 404:917934.
17. Jang, H., B. Ma, ., R. Nussinov. 2008. Models of toxic b-sheet chan-
nels of protegrin-1 suggest a common subunit organization motif
shared with toxic alzheimer b-amyloid ion channels. Biophys. J.
95:46314642.
18. Capone, R., M. Mustata, ., R. Lal. 2010. Antimicrobial protegrin-1
forms ion channels: molecular dynamic simulation, atomic force
microscopy, and electrical conductance studies. Biophys. J. 98:2644
2652.
19. Schulz, G. E. 2002. The structure of bacterial outer membrane proteins.
Biochim. Biophys. Acta. 1565:308317.
20. Sansom, M. S., and I. D. Kerr. 1995. Transbilayer pores formed by
beta-barrels: molecular modeling of pore structures and properties.
Biophys. J. 69:13341343.
21. Jang, H., J. Zheng, and R. Nussinov. 2007. Models of b-amyloid ion
channels in the membrane suggest that channel formation in the bilayer
is a dynamic process. Biophys. J. 93:19381949.
2102 Gupta et al.
22. Jang, H., J. Zheng, ., R. Nussinov. 2008. New structures help the
modeling of toxic amyloidbeta ion channels. Trends Biochem. Sci.
33:91100.
23. Jang, H., F. T. Arce, ., R. Nussinov. 2009. Misfolded amyloid ion
channels present mobile b-sheet subunits in contrast to conventional
ion channels. Biophys. J. 97:30293037.
24. Jang, H., F. Teran Arce, ., R. Nussinov. 2010. Structural convergence
among diverse, toxic b-sheet ion channels. J. Phys. Chem. B.
114:94459451.
25. Jang, H., F. T. Arce, ., R. Lal. 2010. Truncated b-amyloid peptide
channels provide an alternative mechanism for Alzheimers Disease
and Down syndrome. Proc. Natl. Acad. Sci. USA. 107:65386543.
26. Mustata, M., R. Capone, ., R. Nussinov. 2009. K3 fragment of
amyloidogenic b(2)-microglobulin forms ion channels: implication
for dialysis related amyloidosis. J. Am. Chem. Soc. 131:1493814945.
27. Jang, H., B. Ma, and R. Nussinov. 2007. Conformational study of the
protegrin-1 (PG-1) dimer interaction with lipid bilayers and its effect.
BMC Struct. Biol. 7:21.
28. Buffy, J. J., A. J. Waring, and M. Hong. 2005. Determination of peptide
oligomerization in lipid bilayers using 19F spin diffusion NMR. J. Am.
Chem. Soc. 127:44774483.
29. Mani, R., M. Tang, ., M. Hong. 2006. Membrane-bound dimer struc-
ture of a b-hairpin antimicrobial peptide from rotational-echo double-
resonance solid-state NMR. Biochemistry. 45:83418349.
30. Tang, M., A. J. Waring, and M. Hong. 2007. Phosphate-mediated argi-
nine insertion into lipid membranes and pore formation by a cationic
membrane peptide from solid-state NMR. J. Am. Chem. Soc.
129:1143811446.
31. Smart, O. S., J. M. Goodfellow, and B. A. Wallace. 1993. The pore
dimensions of gramicidin A. Biophys. J. 65:24552460.
32. Kenis, H., and C. Reutelingsperger. 2009. Targeting phosphatidylserine
in anti-cancer therapy. Curr. Pharm. Des. 15:27192723.
33. Lee, M. T., W. C. Hung, ., H. W. Huang. 2008. Mechanism and
kinetics of pore formation in membranes by water-soluble amphipathic
peptides. Proc. Natl. Acad. Sci. USA. 105:50875092.
34. Tamba, Y., H. Ariyama, ., M. Yamazaki. 2010. Kinetic pathway of
antimicrobial peptide magainin 2-induced pore formation in lipid
membranes. J. Phys. Chem. B. 114:1201812026.
35. Mani, R., S. D. Cady, ., M. Hong. 2006. Membrane-dependent olig-
omeric structure and pore formation of a b-hairpin antimicrobial pep-
tide in lipid bilayers from solid-state NMR. Proc. Natl. Acad. Sci. USA.
103:1624216247.
36. Lee, M. T., T. L. Sun, ., H. W. Huang. 2013. Process of inducing pores
in membranes by melittin. Proc. Natl. Acad. Sci. USA. 110:14243
14248.
37. Jang, H., B. Ma, ., R. Nussinov. 2006. Interaction of protegrin-1 with
lipid bilayers: membrane thinning effect. Biophys. J. 91:28482859.
38. Jang, H., F. T. Arce, ., R. Lal. 2011. Antimicrobial protegrin-1 forms
amyloid-like brils with rapid kinetics suggesting a functional link.
Biophys. J. 100:17751783.
39. Fahrner, R. L., T. Dieckmann, ., J. Feigon. 1996. Solution structure of
protegrin-1, a broad-spectrum antimicrobial peptide from porcine
leukocytes. Chem. Biol. 3:543550.
40. Miyasaki, K. T., and R. I. Lehrer. 1998. b-sheet antibiotic peptides as
potential dental therapeutics. Int. J. Antimicrob. Agents. 9:269280.
41. Sokolov, Y., T. Mirzabekov, ., B. L. Kagan. 1999. Membrane channel
formation by antimicrobial protegrins. Biochim. Biophys. Acta.
1420:2329.
42. Gottler, L. M., R. de la Salud Bea, ., E. N. Marsh. 2008. Using uo-
rous amino acids to probe the effects of changing hydrophobicity on the
physical and biological properties of the beta-hairpin antimicrobial
peptide protegrin-1. Biochemistry. 47:92439250.
43. Szoka, Jr., F., and D. Papahadjopoulos. 1980. Comparative properties
and methods of preparation of lipid vesicles (liposomes). Annu. Rev.
Biophys. Bioeng. 9:467508.
44. Duzgunes x, N., J. Wilschut, ., D. Papahadjopoulos. 1983. Physico-
chemical characterization of large unilamellar phospholipid vesicles
prepared by reverse-phase evaporation. Biochim. Biophys. Acta.
732:289299.
45. Ames, B. N., and D. T. Dubin. 1960. The role of polyamines in the
neutralization of bacteriophage deoxyribonucleic acid. J. Biol. Chem.
235:769775.
46. Gupta, K., V. P. Singh, ., S. Maiti. 2009. Nanoparticles of cationic
chimeric peptide and sodium polyacrylate exhibit striking antinocicep-
tion activity at lower dose. J. Control. Release. 134:4754.
47. Duzgunes x, N., H. Faneca, and M. C. Lima. 2010. Methods to monitor
liposome fusion, permeability, and interaction with cells. Methods Mol.
Biol. 606:209232.
48. Kucerka, N., S. Tristram-Nagle, and J. F. Nagle. 2005. Structure of
fully hydrated uid phase lipid bilayers with monounsaturated chains.
J. Membr. Biol. 208:193202.
49. Mukhopadhyay, P., L. Monticelli, and D. P. Tieleman. 2004. Molecular
dynamics simulation of a palmitoyl-oleoyl phosphatidylserine bilayer
with Na counterions and NaCl. Biophys. J. 86:16011609.
50. Klauda, J. B., R. M. Venable, ., R. W. Pastor. 2010. Update of the
CHARMM all-atom additive force eld for lipids: validation on six
lipid types. J. Phys. Chem. B. 114:78307843.
51. Durell, S. R., B. R. Brooks, and A. Bennaim. 1994. Solvent-induced
forces between two hydrophilic groups. J. Phys. Chem. 98:21982202.
52. Mackerell, Jr., A. D., M. Feig, and C. L. Brooks, 3rd. 2004. Extending
the treatment of backbone energetics in protein force elds: limitations
of gas-phase quantum mechanics in reproducing protein conforma-
tional distributions in molecular dynamics simulations. J. Comput.
Chem. 25:14001415.
53. Connelly, L., H. Jang, ., R. Lal. 2012. Atomic force microscopy and
MD simulations reveal pore-like structures of all-D-enantiomer of
Alzheimers b-amyloid peptide: relevance to the ion channel mecha-
nism of AD pathology. J. Phys. Chem. B. 116:17281735.
54. Capone, R., H. Jang, ., R. Lal. 2012. All-d-enantiomer of b-amyloid
peptide forms ion channels in lipid bilayers. J. Chem. Theory Comput.
8:11431152.
55. Phillips, J. C., R. Braun, ., K. Schulten. 2005. Scalable molecular
dynamics with NAMD. J. Comput. Chem. 26:17811802.
We dont
compromise
on quality.
We understand that respect is not a given,
its earned. Our standards are high, but you
wouldnt have it any other way. Thats why you
choose to publish with us.
When you choose Cell Press you get the
attention you deserve.
For more information visit: www.cell.com/values
Distinct Stages of Stimulated FcRI Receptor Clustering and
Immobilization Are Identied through Superresolution Imaging
Sarah A. Shelby,
David Holowka,
Barbara Baird,
and Sarah L. Veatch
Department of Chemistry and Chemical Biology, and Field of Biophysics, Cornell University, Ithaca, NY; and

Department of Biophysics,
University of Michigan, Ann Arbor, MI
ABSTRACT Recent advances in uorescence localization microscopy have made it possible to image chemically xed and
living cells at 20 nm lateral resolution. We apply this methodology to simultaneously record receptor organization and dynamics
on the ventral surface of live RBL-2H3 mast cells undergoing antigen-mediated signaling. Cross-linking of IgE bound to FcRI by
multivalent antigen initiates mast cell activation, which leads to inammatory responses physiologically. We quantify receptor
organization and dynamics as cells are stimulated at room temperature (22
C). Within 2 min of antigen addition, receptor diffu-

sion coefcients decrease by an order of magnitude, and single-particle trajectories are conned. Within 5 min of antigen addi-
tion, receptors organize into clusters containing ~100 receptors with average radii of ~70 nm. By comparing simultaneous
measurements of clustering and mobility, we determine that there are two distinct stages of receptor clustering. In the rst stage,
which precedes stimulated Ca
2
mobilization, receptors slow dramatically but are not tightly clustered. In the second stage,
receptors are tightly packed and conned. We nd that stimulation-dependent changes in both receptor clustering and mobility
can be reversed by displacing multivalent antigen with monovalent ligands, and that these changes can be modulated through
enrichment or reduction in cellular cholesterol levels.
INTRODUCTION
Mast cell activation results in secretion of chemical media-
tors of inammation from intracellular granules as part of
the adaptive immune response, which is responsible for
the symptoms of allergy. The rst steps of this process occur
at the plasma membrane, where antigen-specic immuno-
globulin E (IgE) bound to its receptor, FcRI, is cross-linked
by soluble multivalent antigen (1,2). Cross-linking of IgE-
FcRI complexes causes signal initiation by Lyn kinase
phosphorylation of immunoreceptor tyrosine-based activa-
tion motifs on FcRI b and g
2
subunits. The resulting
tyrosine phosphorylation leads to Ca
2
mobilization and
cellular degranulation (3). Oligomerization has readily
observable effects on the spatial distribution and diffusion
behavior of IgE-FcRI. Receptors are uniformly distributed
and mobile on the membrane before activation. Stimulation
with antigen causes clustering of IgE-FcRI into punctate
aggregates on the cell surface and a marked decrease in
receptor mobility (4,5).
Cross-linked IgE-FcRI puncta can be visualized with
conventional uorescence microscopy. However, quantita-
tive measurements of receptor cluster formation require
subdiffraction-limited spatial resolution, comparable to the
dimensions of clusters. Previous work has used a variety
of experimental approaches to characterize the dynamic,
antigen-induced, nanoscale reorganization of FcRI to
understand this initiation step in signaling. Receptor distri-
butions at high spatial resolution have been studied with
electron microscopy using immunogold labeling of IgE-
FcRI (68). In addition, dynamics of IgE-FcRI during
the time course of activation have been captured by uores-
cence photobleaching recovery (FPR) (5,9,10) and uores-
cence correlation spectroscopy (FCS) (11). Single-particle
tracking (SPT) of individual receptors labeled with uores-
cent-dye- or quantum-dot-conjugated IgE has characterized
the motion of single proteins (1214). To date, these ap-
proaches have either achieved high-resolution spatial mea-
surements in chemically xed systems or have measured
receptor mobility in live cells using nonimaging methods
or imaging without nanoscale spatial resolution.
Advances in uorescence microscopy now enable subdif-
fraction imaging using photoconvertible uorescent dyes.
Superresolution techniques, including (direct) stochastic
optical reconstruction microscopy (STORM/dSTORM)
(15,16) and (uorescence) photoactivation localization
microscopy (PALM/FPALM) (17,18), have been used in
xed cells to quantify membrane protein distributions and
clustering in other systems (1924). Superresolution tech-
niques in live cells capture high-resolution maps of protein
distributions (25,26) in addition to diffusion information
from single-molecule trajectories (2729). Further, the use
of localization microscopy for single-particle tracking
methods provides improved number statistics compared to
traditional single-particle tracking because the photoconver-
sion process allows for sampling of an ensemble of recep-
tors over the course of a single live-cell measurement.
This work applies superresolution uorescence localization
microscopy, exploiting its capabilities for both high-resolu-
tion imaging and single-molecule recording of receptor
diffusion. With this technique, we monitor the kinetics of
clustering and mobility changes of IgE receptors in live
Submitted June 7, 2013, and accepted for publication September 9, 2013.
*Correspondence: sveatch@umich.edu
Editor: Anne Kenworthy.
Biophysical Journal Volume 105 November 2013 23432354 2343
rat basophil leukemia (RBL)-2H3 mast cells undergoing a
stimulated immune response. We do this both by quanti-
fying average properties of receptors and by examining
the behavior of single molecules. In addition, we explore
how receptor mobility and diffusion are altered by perturba-
tions, including reversal of receptor cross-linking with
monovalent hapten and modulation of the cholesterol con-
tent of cell membranes.
Redistribution of IgE-FcRI upon stimulation in
live cells
Through superresolution imaging of living cells, we simulta-
neously observe nanometer-scale receptor organization and
dynamics in real time. Fig. 1 A shows a representative live
RBLcell that is imaged before and after the addition of multi-
valent antigen at room temperature under buffer conditions
that both support superresolution imaging and preserve
downstream functional responses (Fig. S1 in the Supporting
Material). We chose to image at room temperature rather
than at 37
C because key signaling stages occur at the lower

temperature, although at a slower rate. These include
receptor phosphorylation, Ca
2
mobilization, and endocy-
tosis (11,30). Cells were sensitized by incubation with IgE
antibodies specic for dinitrophenyl (DNP) and stimulated
with the multivalent antigen DNP-bovine serum albumin
(DNP-BSA). IgE-FcRI complexes were uorescently
labeled by sensitizing with IgE directly conjugated to an
Alexa Fluor 647 (AF647) and imaged as described in Mate-
rials and Methods in the Supporting Material. Live cell im-
ages are produced by following single-molecule trajectories
in raw images, then reconstructing time-averaged images
using only the rst localized position in each trajectory.
Each reconstructed image in Fig. 1 A is compiled from
2000 raw image frames acquired over 68 s of imaging time
at 31 frames/s. The relatively short imaging time produces
a reconstructed image that is inherently undersampled; only
a fraction (estimated to be between 30%and 60%) of individ-
ual IgE proteins are represented in each image. Despite this
limitation, images clearly indicate that receptors are nearly
randomly organized in unstimulated cells and become more
clustered in response to cross-linking by multivalent antigen.
We utilize the spatial pair correlation function as a function
of radius, g(r), to quantify clustering of IgE-FcRI complexes
in reconstructed images. Pair autocorrelation functions mea-
sure the normalized probability of nding a second localized
uorophore at a given distance, r, from the average localized
uorophore. These functions are tabulated as described pre-
viously (31) and summarized in Materials and Methods in the
Supporting Material. For the resulting curves, a value of 1 in-
dicates that receptors are randomly organized. Values of >1
indicate that receptors are clustered, and the range in r over
which g(r) > 1 is a measure of cluster size. The g(r) curves
shown in Fig. 1 B were tabulated from images reconstructed
using 500 frames of raw image data acquired over 16 s. In
agreement with visual observations, autocorrelation func-
tions generated from time-resolved images show that recep-
tors are nearly randomly distributed before antigen addition,
with g(r) ~ 1 at all radii, and become dramatically more
densely clustered after stimulation. Correlation functions
measured in live cells are in good quantitative agreement
with those observed in cells chemically xed at specic
FIGURE 1 Quantitative superresolution localization microscopy imag-
ing of IgE-FcRI redistribution after antigen addition in live cells. (A)
Reconstructed superresolution uorescence localization images of an
AF647-IgE-labeled living cell at various times in the stimulation sequence,
where antigen (DNP-BSA, 1 mg/mL) is added at 0 min. Each image is re-
constructed from 68 s of acquired data, as described in Materials and
Methods (see Supporting Material). A movie showing complete time-lapse
imaging of this cell is supplied in the Supporting Material. (Insets) Magni-
ed images of the square regions outlined in black. (B) Autocorrelation
functions, g(r), are calculated from reconstructed single-molecule centers
acquired over 16 s, as described in Materials and Methods in the Supporting
Material (solid symbols) and are t to single exponentials (Eq. 1; solid
lines). (Inset) The correlation function from data recorded 3 min before an-
tigen stimulation on an expanded scale. (C) Correlation function parameters
from 11 live-cell experiments, distinguished by different colors: the corre-
lation length, x (upper), the correlation amplitude, A (middle), and the
average number of correlated proteins, N (lower). Solid black lines indicate
averages over 11 cells, and error bars represent the mean 5SE. Fit param-
eters extracted from two-color xed-cell experiments are reproduced from
Fig. S2, D and E and plotted for comparison in C as open black diamonds.
(Inset) Average x for time points between 3 and 15 min after antigen addi-
tion on an expanded y-axis scale. To see this gure in color, go online.
2344 Shelby et al.
time points after stimulation (Fig. S2). Although recon-
structed images of live cells are undersampled compared to
xed-cell images, as long as undersampling is random, its
effects alone will not change the correlation function beyond
decreasing the signal/noise ratio (31).
Measured autocorrelation functions are t to a single
exponential to extract information on average cluster size
and density according to the equation
g
Fit
r 1 Aexpr=x (1)
for r > 20 nm, where A is the amplitude of correlations,
which is proportional to the increased density of receptors
in clusters, and x is the correlation length, which is approx-
imately the average cluster radius. The average number of
correlated proteins (N), or the number of correlated proteins
within the average cluster, is the summation of the measured
g(r) over r times the average surface density of receptors,
dened by the equation
N r
ave
S2p rDrgr 1; (2)
where we assume that the overall average surface density of
receptors (r
ave
) is 200/mm
2
(31,32). When curves are well t
to the single-exponential form given in Eq. 1 in the limit of
small Dr, this sum over values of r from zero to innity can
also be written as N r
ave
4pAx
2
. In practice, we evaluate
Eq. 2 for radii between 0 and 300 nm, with Dr 15 nm.
This quantitative analysis, averaged over 11 cells, and a
summary of extracted t parameters is shown in Fig. 1 C.
We observe dramatic redistribution of receptors into clusters
after addition of multivalent antigen with weak, long-range
correlations in unstimulated cells, and strong, shorter-range
correlations after antigen stimulation, consistent with
previous reports in live cells (33). In unstimulated cells,
we observe correlations with very low values of A and N
and large values of x. This is further illustrated by the corre-
lation function for unstimulated cells plotted in the inset of
Fig. 1 B. x extends to ~200 nm in unstimulated live cells,
whereas we observed x z 80 nm in chemically xed cells
(Fig. S2). The larger x observed in live-cell images could
arise from overcounting single molecules that are lost by
our tracking algorithm, lateral motion of any correlated
structures observed during data collection, or, possibly,
the fact that live cells were imaged at room temperature
whereas chemically xed cells were incubated at 37
C.
We observe time-dependent increases in A and N during
the rst 5 min after antigen addition. After this time, the cor-
relation amplitude, A, remains constant, the average number
of correlated proteins, N, continues to increase at a slower
rate, and the correlation length, x, slowly increases (Fig. 1
C upper, inset). The average x decreases within 3 min of an-
tigen addition to ~70 nm, in good agreement with x in stim-
ulated xed cells (Fig. 1 C, diamonds). The continuous
decrease in x soon after antigen addition likely indicates
the increasing presence of small and dense clusters in a
background of larger more diffuse structure, as suggested
by the image reconstructed from data acquired 1 min after
antigen addition in Fig. 1 A, although we do not attempt
to resolve two distinct components in g(r).
Our choice to quantify single-color live-cell images using
autocorrelations relies on the assumption that live-cell
superresolution images are not greatly affected by artifacts
associated with overcounting single receptors. We expect
this to be the case because individual uorescently labeled
receptors will typically diffuse over distances much larger
than correlated structures with dimensions of several hun-
dred nanometers or less during the time that a uorescent
molecule labeling an individual receptor remains in the
dark state. This assumption may not be valid in stimulated
cells, where receptors may become conned within densely
cross-linked clusters, and the same receptor may be counted
multiple times within a single cluster. If there is a contri-
bution to correlation functions from overcounting, our
reported results would lead to an overestimate of A. How-
ever, the values of Awe observe from autocorrelation func-
tions in live cells are systematically lower than values of A
extracted from cross-correlation functions in xed cells
(Fig. 1 C, black diamonds), which are not affected by over-
counting artifacts. This supports the assumption that over-
counting does not affect correlation functions throughout
the time course of imaging.
Our results using chemically xed cells (Fig. S2) are
consistent with and complementary to our live cell measure-
ments in several regards. Agreement of A and N between
xed and live cells at early stimulation time points indicates
that clustering occurs to approximately the same extent,
although at a somewhat slower rate, in cells stimulated at
room temperature compared to those stimulated at 37
C.
Further, results from xed-cell experiments demonstrate
that receptors are not clustered before stimulation. This
result is important in the context of our live-cell experi-
ments, because small and highly mobile clusters that diffuse
much farther than their size over a typical image acquisition
time would not be detected by our live-cell quantication
methods.
Mobility of IgE-FcRI in live-cell measurements
The majority of individual uorophores remain in a uores-
cent state for multiple sequential frames, and we track these
probes to form trajectories from localizations of the protein
in time and space (see Materials and Methods in the Sup-
porting Material). Visual inspection of trajectories obtained
from 16 s of acquired data in unstimulated and stimulated
cells suggests that IgE-FcRI diffusion is relatively uncon-
strained before stimulation and that mobility decreases
signicantly after antigen addition (Fig. 2 A). Trajectories
are quantied by tabulating the mean-square displacement
(MSD) as a function of time interval (t). Several represen-
tative curves are calculated by averaging MSD(t) values
STORM Identies FcRI Signaling Stages 2345
over all trajectories acquired within 16 s, as shown in Fig. 2
B. The magnitude of the average MSD(t) decreases after
stimulation, indicating reduced receptor mobility. The
representative data shown in Fig. 2, A and B, are acquired
from the same cell shown in Fig. 1, A and B.
In most cases, we nd that the slopes of MSD(t) are not
linear, as expected for free diffusion, but instead are de-
ected to lower values at long t, indicating that receptors
are conned. We quantify both diffusion and connement
of IgE-FcRI complexes as a function of stimulation time
by tting MSD(t) to obtain both short- and long-time diffu-
sion coefcients, D
S
and D
L
, respectively, which are ob-
tained by tting distinct time ranges of MSD(t) curves to
extract linear slopes. D
S
is obtained by tting the equation:
MSDt
24
4D
S
t
24
C
S
; (3A)
where t
2-4
indicates the second, third, and fourth time inter-
vals of the MSD(t) curve, typically corresponding to
roughly 50100 ms, and C
S
is the y-intercept of the t that
accounts for the nite localization precision of the single-
molecule data. D
L
is obtained by tting the MSD curve at
time intervals between 250 and 500 ms to the analogous
equation,
MSDt
250500ms
4D
L
t
250500ms
C
L
: (3B)
Best-t lines to Eqs. 3A and 3B, whose slopes are propor-
tional to the values of D
S
and D
L
, are shown for the repre-
sentative MSD(t) curves in Fig. 2 B. In addition, we
quantify connement by taking a ratio of these values,
D
S
/D
L
. In the examples shown in Fig. 2 B, D
S
/D
L
> 1, indi-
cating that receptors are conned, and this ratio increases
after antigen addition.
D
S
, D
L,
and their ratio for IgE-FcRI in cells undergoing
signaling responses are shown in Fig. 2, CE, for the same
11 cells characterized in Fig. 1. Both D
S
and D
L
dramati-
cally decrease within 5 min of antigen addition. After
5 min of stimulation, D
S
decreases from 0.1 to 0.02 mm
2
/s
and D
L
decreases from 0.075 to 0.01 mm
2
/s. We also observe
changes in connement with stimulation time, as measured
by D
S
/D
L
, which rapidly increases after stimulation before
decreasing slightly at later stimulation time points.
Our measured values of D
S
versus stimulation time are
in agreement with similar diffusion-coefcient parameters
measured previously using SPT (10,12,13,33), FCS (11),
and FPR (5,9), ranging from 0.03 to 0.26 mm
2
/s before
stimulation and from 0.01 to 0.16 mm
2
/s after stimulation
with a multivalent antigen. Previous measurements of
receptor diffusion using SPT approaches similar to ours
(10,12,13,33) agree best with our observed values, where
here reported values range from 0.07 to 0.1 mm
2
/s before
stimulation and from 0.01 to 0.05 mm
2
/s after stimulation.
Our measurements of receptor connement are also consis-
tent with the observation of restricted or compartmentalized
diffusion in previous SPT studies of FcRI (10,12,33).
In one of these past studies, the diffusion compartments of
IgE-FcRI were reported to shift to smaller sizes upon anti-
gen addition, accompanied by a decrease in the diffusion
coefcient for movement between compartments (33).
That result is consistent with the antigen-induced increase
in receptor connement measured in this study.
Correlating receptor mobility with receptor
clustering
The data presented in Figs. 1 and 2 allow for direct compar-
ison between changes in IgE-FcRI receptor diffusion
versus spatial distribution. Taken together, the results indi-
cate that IgE-FcRI receptor complexes have decreased
FIGURE 2 Antigen stimulation leads to slower and more conned diffu-
sion of IgE-FcRI receptor complexes. (A) Single-molecule trajectories of
IgE-FcRI complexes on the surface of cells under TIRF illumination before
(Ag) and after (Ag) stimulation with 1 mg/mL DNP-BSA for 5 min.
Tracks shown are accumulated over 1 min, only tracks observed for ve or
more frames (0.16 s) are displayed, and coloring from blue to red indicates
the relative time at which a single probe was observed within the 1 min
time frame. (B) MSD curves are generated by averaging over all tracks
observed within a 500 frame (16 s) time period at the times during antigen
stimulation indicated, as described in Materials and Methods in the Support-
ing Material. MSD curves are t to Eqs. 3A and 3B to extract the short- and
long-time diffusioncoefcients D
S
andD
L
, respectively. (Cand D) Summary
of D
S
(C) and D
L
(D) extracted from MSD curves tabulated from single-
molecule trajectories acquired over 500 frames (~20 s and variable from
cell to cell) for 11 distinct cells. Error bars represent the mean 5 SE of
the 11 live-cell experiments. (E) Connement as a function of stimulation
time, as measured by D
S
/D
L
from the same 11 live-cell experiments. Error
bars represent the mean 5SE. To see this gure in color, go online.
2346 Shelby et al.
mobility (D
S
and D
L
) and increased connement (D
S
/D
L
)
that plateaus within 12 min after antigen addition. In
contrast, the density of receptor clusters (A) increases
more slowly, with the amplitude of correlations plateauing
after ~5 min (Fig. 1 C).
To explore more directly the relationship between cluster
properties and receptor mobility, we plot in Fig. 3 A the
average short-time receptor diffusion coefcient, D
S
, versus
the average number of correlated proteins, N, for the stimu-
lation time course averaged from 11 live-cell experiments
(average D
S
and N as a function of time are shown indepen-
dently in Figs. 2 C and 1 C, respectively). Interestingly, this
representation suggests two distinct regimes of receptor
mobility and clustering. In the rst regime, D
S
decreases
dramatically without a large corresponding change in N.
In the second regime, receptors become increasingly clus-
tered, without a large corresponding decrease in D
S
. The
crossover between regimes occurs for N between 20 and
30 and for D
S
between 0.035 and 0.02 mm
2
/s, which corre-
spond to stimulation times between 1 and 1:45 min, respec-
tively, after antigen addition.
Interestingly, the beginning of the crossover between the
two regimes shown in Fig. 3 A roughly coincides with the
onset of Ca
2
signaling in RBL-2H3 cells imaged using
the Ca
2
-sensitive dye Fluo-4 under nearly identical stimu-
lation conditions (Fig. 3, B and C). Fluo-4 is loaded into
sensitized RBL cells, and the uorescence intensity is moni-
tored across a eld of several hundred cells as a function of
stimulation time, as described in Materials and Methods
(see Supporting Material). Soon after antigen addition,
Fluo-4 intensity averaged over the population of cells begins
to rapidly increase, with the bulk of the increase coming
between 1 and 4 min (Fig. 3 B). There is a large cell-to-
cell heterogeneity in the timing of the onset of the Ca
2
response, as indicated in the cumulative distribution shown
in Fig. 3 C, with some cells initiating a response ~1 min after
antigen addition under these imaging conditions. Approxi-
mately 50% of cells have experienced a Ca
2
response
within 2:15 min. Keeping in mind this large heterogeneity
and the fact that we sample only a limited number of single
cells in superresolution experiments, we use the Ca
2
mobi-
lization measurement as a rough indicator of the commence-
ment of cellular signaling.
The results reported in Fig. 3 indicate that the initial,
rapid decrease in D
S
of IgE-FcRI complexes is a
consequence of interactions that precede Ca
2
mobilization,
whereas the accumulation of receptors into densely packed
clusters represents receptors after the onset of Ca
2
mobili-
zation. This suggests that early signaling events leading to
the Ca
2
response do not require that receptors be densely
clustered or fully immobilized. This interpretation of our
results is consistent with observations from previous studies
that small IgE-FcRI clusters that retain mobility can elicit a
degranulation response (13), and that there is a high level of
receptor tyrosine phosphorylation within the rst few
minutes of antigen stimulation, both at 37
and 15
C (34).
In the latter study (34), receptor tyrosine phosphorylation
at 15
C occurs on a timescale similar to that of the Ca

2
response that we measure at room temperature, as we would
expect, since tyrosine phosphorylation precedes Ca
2
mobi-
lization in the IgE receptor signaling cascade. The same
study also demonstrated that exposure of DNP haptens on
the surface of DNP-BSA is transient, and that antigen bind-
ing is dominated by cross-linking of receptors after a minute
of exposure to antigen (34). This is also consistent with the
idea that the immobilization and Ca
2
responses we observe
at early stimulation times occur concurrently with the for-
mation of small clusters, and that receptors become more
heavily cross-linked at later times.
Because diffusion of cross-linked IgE-FcRI decreases
rapidly, without a corresponding large increase in N, it
likely occurs as a result of IgE-FcRI coupling to down-
stream signaling partners. This could be due to receptor as-
sociation with other freely diffusing membrane-anchored
FIGURE 3 Average receptor diffusion displays two different phases of
dependence on the number of proteins in the average cluster. (A) Average
D
S
(as in Fig. 2 C) is shown as a function of average N from the same
live-cell experiments (as in Fig. 1 C). Each point corresponds to values
of D
s
and N at a given time before (diamonds) and after (circles) stimulation
averaged over the 11 cells imaged, and data from individual cells are binned
every 15 s to facilitate averaging. Time after the addition of antigen is
indicated by the color bar. Antigen (1 mg/ml) was added after the cells
were imaged for 5 min. The solid black lines represent linear ts of points
between 0 and 1 min and between 1:45 and 15 min after antigen stimula-
tion, weighted by the inverse of the mean 5SE in D
S
and N for each point.
Points spanning these two regimes are indicated with arrows and labeled
with the time after antigen addition. (B) Average intensity of the cyto-
plasmic Ca
2
indicator Fluo-4 over a population of cells imaged
as described in Materials and Methods in the Supporting Material. The
increase in Fluo-4 intensity after antigen stimulation indicates the onset
of Ca
2
mobilization. The time period coinciding with the timing of the
transition from the rst regime to the second in A is highlighted by the
shaded region. (C) The cumulative distribution of cells exhibiting an initial
Ca
2
response indicates that the majority (>90%) of cells have initial Ca
2
responses between 1 and 4 min after antigen stimulation. The time point
when 50% of the responding cells have exhibited a Ca
2
response is indi-
cated by the open circle. To see this gure in color, go online.
proteins that are not labeled in these experiments. It is
possible that receptor slowing is due to slower movement
between corrals dened by cortical actin, either due to
mechanical occlusion of the growing signaling platforms
with actin-anchored proteins (36,37), or due to increased
coupling of platforms to actin-stabilized, lipid-mediated
heterogeneity (38). Alternatively, the reduced mobility of
receptors at early signaling stages could be a consequence
of direct or indirect tethering to an immobilized component
such as actin. The actin cytoskeleton has been shown to be
partially responsible for the immobilization of cross-linked
IgE-FcRI (12), and it plays a role in the desensitization of
receptors to antigen (39), and in internalization (40). The
increasing clustering of receptors that dominates later
signaling stages could be a mechanism to downregulate
signaling by sequestering receptors, as occurs in B cell
and T cell receptor activation (4144). It is also possible
that the observed receptor clusters represent an early stage
of receptor internalization that has not progressed substan-
tially under these conditions of stimulation time and tem-
perature (11).
Ca
2
mobilization is used here as an approximate indica-
tion of the timing of the onset of cellular signaling, but
consideration should be given to its direct comparison
with superresolution measurements. Clustering and immo-
bilization of IgE receptors on the ventral surface of cells
could begin more slowly than Ca
2
mobilization because
the Ca
2
response originates from receptors on the dorsal
surface of cells, which are expected to be more accessible
to antigen. This may lead to a time delay between the initi-
ation of the Ca
2
response and our observation via super-
resolution measurements of immobilization and clustering.
If this is the case, receptors on the dorsal surface may be
somewhat more clustered at the onset of the Ca
2
response,
but this would not change our conclusion that the rst phase
of clustering and immobilization occurs before Ca
2
mobi-
lization, and the second phase occurs predominantly
afterward. Ca
2
mobilization itself may be considered a
downstream measure of the onset of signaling. Previous
work has shown that receptor phosphorylation by Lyn
kinase proceeds much more rapidly than Ca
2
mobilization
(35), and thus, Ca
2
mobilization represents membrane in-
teractions that initiate a broader cellular response and occur
subsequent to the initiation of signaling at the level of single
receptors.
Single-molecule analysis of receptor diffusion
and clustering
Early and late signaling stages are also distinguished when
receptors are examined as single molecules, and several
representative trajectories are shown in Fig. 4 A. Single re-
ceptor trajectories in unstimulated cells traverse large areas.
Soon after antigen is added, trajectories rapidly condense,
and some receptors appear to sample multiple conned
areas in single trajectories, lasting ~1 s each. After a few mi-
nutes of antigen stimulation, trajectories are compact and
appear highly conned. The ensemble of single-molecule
FIGURE 4 Slower and more conned diffusion of single receptors corre-
lates with regions of high receptor density. (A) Examples of single-molecule
trajectories are shown from the same cell in Fig. 1, A and B, Fig. 2, A and B,
and Fig. 4 C recorded before and after antigen stimulation. The tracks
shown persist for at least 0.5 s for the 1 min (unstimulated) time point
and for 1 s for other time points. (B) Short-time diffusion coefcients
(D
S
) are evaluated from MSD curves tabulated from single-molecule trajec-
tories lasting at least 0.5 s within a 16 s time period and are assembled into
histograms. Histograms are normalized by the total number of tracks
collected to generate each histogram. (C) Single-molecule trajectories per-
sisting for at least 0.5 s are superimposed on a superresolution image recon-
structed from unstimulated data (upper), from data acquired within 1 min of
antigen addition (middle), and from data acquired after several minutes of
stimulation. Track coloring indicates D
S
for each track on a log scale from
10
5
mm
2
/s (blue) to 1 mm
2
/s (red). Images on the right are enlargements of
the boxed regions in the images on the left. (D) Three-dimensional histo-
grams of D
S
versus average receptor density along trajectories lasting at
least 0.5 s. The average receptor density for each trajectory is determined
by averaging the pixelated grayscale values from the time-averaged recon-
structed image over all positions of the trajectory and then normalizing
assuming r
ave
200/mm
2
, as described in Materials and Methods in the
Supporting Material.
2348 Shelby et al.
trajectories is quantied by assembling histograms of D
S
.
Fig. 4 B shows histograms assembled using 16 s of data
acquired in a single cell, which are representative of histo-
grams obtained from other cells examined. Histograms are
well described as single log-normal distributions for all
time points, indicating that a single population of diffusers
is resolved in these measurements. Distributions of D
S
rapidly shift to lower values and broaden soon after antigen
is added, stabilizing after 3 min of stimulation time. These
distributions are broad, in part because diffusion coefcients
are not well dened when obtained from short trajectories
(45). To separate this effect from real heterogeneity, we
compare measured distributions of D
S
to those obtained
by simulating Brownian trajectories with 16 frame (0.5 s)
track length (Fig. S4). In unstimulated cells, the width of
measured D
S
histograms is comparable to those of simu-
lated trajectories. In contrast, measured histograms for
receptors after antigen addition are signicantly broader
than the simulated distributions, indicating that the mem-
brane environment sampled by IgE-FcRI is heterogeneous.
We also investigated how receptor diffusion correlates
with the local surface density of receptors in reconstructed
images. To accomplish this, we reconstructed superresolu-
tion uorescence images as described in Materials and
Methods (see Supporting Material). Representative gray-
scale images for a single cell at various stimulation stages
are shown in Fig. 4 C. In these images, pixel intensity is pro-
portional to the observed receptor density, and trajectories
that persist for >0.5 s are superimposed onto this image.
In the unstimulated cell (Fig. 4 C, upper), individual recep-
tors diffuse over large areas and their mobility is not visually
correlated with roughly random receptor density. Soon after
antigen addition (<1 min), individual receptors appear more
conned, even though the spatial distribution of receptors
remains largely random (Fig. 4 C, middle). At longer
times, (>5 min after antigen addition), diffusing receptors
are conned to regions where receptors are densely packed
(Fig. 4 C, lower).
These visual observations are quantied by calculating
the average pixel intensity over the length of single-mole-
cule trajectories persisting for >0.5 s. Three-dimensional
histograms are displayed as contour plots in Fig. 4 D
for several stimulation times. In unstimulated cells, recep-
tor diffusion appears relatively unconstrained, and the
ensemble of single molecules experiences roughly the
same local environment at the frame rates used in these
experiments (~30/s). This is not surprising given that recep-
tors typically move hundreds of nanometers between
observations, and over several square microns in a typical
trajectory. Lipid-mediated and/or actin-generated obstacles
to diffusion are expected to occur on smaller length-scales
in unstimulated cells, and thus, single receptors are expected
to sample a large number of local environments in a single
trajectory. At long times, we observe a homogeneous but
broad distribution in the D
S
versus receptor density histo-
gram in Fig. 4 D. This is a result of both a distribution of re-
ceptor densities in puncta and individual receptors conned
to sample the local environment of single puncta over their
recorded trajectories.
Soon after stimulation with antigen, histograms are elon-
gated, extending between faster-moving receptors in a low-
density local environment and slower-moving receptors in a
higher-density local environment. In some cases, two peaks
are observed, as indicated by the orange triangles in Fig. 4
D. This elongated distribution could arise from receptors
sampling a heterogeneous membrane environment. We
think it is more likely that this elongated distribution arises
from receptors slowly exchanging between a more mobile,
less aggregated state and a less mobile, cluster-associated
state over the length of the trajectory, as appears to be the
case from visual inspection of single trajectories (Fig. 4 A).
Elongated distributions in Fig. 4 D are observed before
the onset of Ca
2
mobilization, and transient associations
of individual receptors could represent interactions that
result in signaling. This also suggests that receptor aggre-
gation is dynamic, at least in the early signaling stages.
Consistent with this interpretation, it has been shown previ-
ously that readily dissociable cross-linked receptors are
primarily responsible for generating downstream signaling
responses (39). Previous work has also shown that initial
binding of DNP-BSA to IgE is primarily monovalent, and
that cross-linking occurs slowly as DNP haptens on the
receptor-bound antigen subsequently become available for
binding (34). This is in good agreement with our current
observations that suggest transient association of receptors
with receptor clusters soon after antigen addition.
Receptor clustering, immobilization, and
connement is reversible in live cells
A monovalent DNP hapten, DNP-aminocaproyl-L-tyrosine
(DCT), competes with multivalent DNP-BSA for binding
to anti-DNP IgE (39). The addition of an excess of DCT af-
ter antigen stimulation reverses antigen-induced cross-link-
ing and results in the cessation of signaling (39,46,47). The
representative live-cell superresolution images in Fig. 5 A
show uniform distribution of AF647-IgE bound to FcRI
before antigen addition, clustered IgE-FcRI distribution
after 7 min of antigen stimulation, and uniform distribution
of receptors after DCT incubation for 10 min. The average
time dependence of A, N, and D
S
from six live-cell experi-
ments quanties the reversal of clustering and immobiliza-
tion upon DCT exposure (Fig. 5 B). IgE-FcRI clusters are
dispersed on a timescale of several minutes, as shown by
a decrease in A. Likewise, N and D
S
recover close to pres-
timulation levels within 10 min of DCTaddition. These data
indicate that receptor immobilization is reversible and
dependent on receptor cross-linking, as has been shown pre-
viously by FPR measurements that used DCT to reverse
receptor immobilization (5). This reversibility demonstrates
that the densely packed and immobile receptor clusters
formed within 7 min of antigen addition are not stabilized
solely through interactions with other cellular components
or that these interactions are insufcient to stabilize clusters
in the absence of antigen cross-linking.
Receptor organization and mobility in response
to cholesterol perturbation in unstimulated cells
To examine the role of lipid-mediated membrane heteroge-
neity on early and later stages of antigen stimulation, we
investigated the impact of membrane cholesterol levels
on receptor organization and mobility. Receptor diffusion
(D
S
) and clustering (N) upon cross-linking with antigen
are measured for multiple live cells exposed to either
methyl-b-cyclodextrin (MbCD) to reduce plasma mem-
brane cholesterol or MbCD precomplexed with cholesterol
(MbCD chol) to enrich plasma membrane cholesterol
(Figs. 6 and S5). Under our conditions, we expect an
~20% decrease in the total cellular cholesterol content
after 5 min of MbCD addition and an ~50% decrease after
15 min (48).
In the absence of cross-linking by antigen, local receptor
density changes due to variations in membrane cholesterol
concentration. Fig. S5 shows representative superresolu-
tion images of AF647-IgE/FcRI with and without antigen
treatment for individual cells with MbCD or MbCD
chol. IgE-FcRI continues to show largely random distribu-
tion in unstimulated cells with reduced cholesterol levels,
but becomes tightly clustered in unstimulated cells with
elevated cholesterol levels. This visual observation is quan-
tied in Fig. S6, where D
S
, A, and N are plotted as a function
of time and treatment with MbCD or MbCD chol.
It is likely that other cellular processes contribute to the
organization and mobility of receptors in cholesterol-loaded
cells. For example, we observe robust antigen-independent
activation of transient Ca
2
oscillations after cells are incu-
bated with MbCD chol for 2 min, and this persists until
~5 min after MbCD chol is added (Figs. S7 and 6, B
and C). This indicates that the cellular environment changes
dramatically in response to MbCD chol in ways that may
not be directly related to cholesterols effects on lipid-medi-
ated membrane organization. Modulating cellular choles-
terol levels also leads to changes in receptor diffusion
in unstimulated cells. MbCD addition to unstimulated cells
results in a time-dependent decline in D
S
over 15 min, and
incubation with MbCD chol leads to slight increases in
D
S
(Fig. S6), despite the large increases in receptor clus-
tering described above. These observations could be the
result of changes in membrane surface area, changes in
the surface density of immobile obstacles, or induction of
solid-phase domains (49).
Receptor organization and mobility in response
to cholesterol perturbation in stimulated cells
Perturbations of membrane cholesterol also affect the
organization and mobility of IgE-FcRI complexes when
receptors are subsequently cross-linked with multivalent
antigen (Fig. 6 A). For both cholesterol reduction and
enrichment, receptor clustering increases and receptor
diffusion decreases in response to antigen, qualitatively
similar to trends in the absence of perturbation (Figs. 3 A
and 6 A). We observe two distinct regimes in plots of D
S
versus N for points after antigen addition in cells pretreated
with MbCD, as is also observed in cells in the absence of
cholesterol modulation. The crossover between these two
regimes occurs at larger values of N in cells with reduced
cholesterol levels (N < 40 for untreated versus N > 60 for
MbCD-treated cells (Fig. 6 A, upper)), which also corre-
sponds to longer stimulation times at the crossover point
(~1 min for untreated vs. ~3 min for MbCD-treated cells).
FIGURE 5 Antigen-induced changes in receptor clustering and mobility
are reversible. (A) Reconstructed images of an AF647-IgE-labeled living
cell before and after stimulation and subsequent addition of DCT. Each im-
age is reconstructed from 80 s of acquired data. (Insets) Magnications of
the boxed regions in the main image. DNP-BSA (0.1 mg/ml) was added at
0 min, and DCT (200 mM) was added at 7 min. (B) The parameters A, N,
and D
s
are calculated as in Figs. 1 C and 2 C. The average values of A,
N, and D
s
, indicated by black lines, for six live-cell experiments distin-
guished by different colors, over the time course of stimulation and DCT
addition. Antigen addition is indicated by the orange dashed line at
0 min, and DCT addition is indicated by the gray dashed line at 7 min.
To see this gure in color, go online.
2350 Shelby et al.
For cells pretreated with MbCD chol (Fig. 6 A, lower),
only one regime is apparent in plots of D
S
versus N, and
the N is larger before antigen addition.
Antigen-induced functional responses are also affected
in cells pretreated with MbCD or MbCD chol. Antigen-
induced signaling is less effective in cells with reduced
cholesterol levels when stimulated degranulation is assessed
(5053) (Fig. S8). When Ca
2
mobilization is again used as a
rough measure of the onset of cellular signaling, as in Fig. 3,
pretreatment of cells with MbCD results in a Ca
2
response
that is both reduced in magnitude (Fig. 6 B) and delayed
(Fig. 6 C) compared to untreated cells. A fraction (~40%)
of cells pretreated with MbCD fail to show Ca
2
responses
within 10 min after antigen addition. We quantify the timing
of antigen-induced Ca
2
mobilization by measuring the time
taken for 50% of responding cells to show an initial Ca
2
response, t
1/2
, as indicated by the symbols on the cumulative
distributions shown in Fig. 6 C. Antigen-induced signaling is
attenuated in cells pretreated with MbCD chol when they
are assayed by degranulation (Fig. S8) or by measurement of
Ca
2
mobilization (Fig. 6).
For the case of cholesterol enrichment, we observe
an initial Ca
2
response after MbCD chol is added, as
described above, followed by a second, weaker Ca
2
signal
in response to antigen (Fig. 6 B). A large fraction (~60%)
of MbCD chol-treated cells also fail to exhibit an anti-
gen-induced Ca
2
response (Fig. 6 C), although cells that
do respond do so with a minimal time lag after antigen addi-
tion. As a result, t
1/2
is shorter compared to either untreated
or MbCD-treated cells (Fig. 6 D). These differences in
the nature of the antigen-dependent Ca
2
response may be
inuenced by the MbCD chol-induced Ca
2
transient
observed before antigen addition. The shape, frequency,
and duration of Ca
2
oscillations are also severely affected
by changes in cellular cholesterol (Fig. S7).
t
1/2
is correlated with the timing of the crossover
observed in plots of D
S
versus N for the three cholesterol
treatments (Fig. 6 D). Despite the uncertainty in relating
the timing of Ca
2
mobilization to superresolution mea-
surements discussed above, we observe differences in the
relative timing of the crossover and t
1/2
that are both depen-
dent on cholesterol perturbation. This observation supports
our conclusion that the initial, rapid decrease in the diffu-
sion coefcient of IgE-FcRI receptors is a consequence
of interactions that precede Ca
2
mobilization, whereas
the accumulation of receptors into densely packed clusters
represents receptors after the onset of signaling. For the
case of cholesterol reduction, antigen-induced slowing of
receptor diffusion occurs at a slower rate than in untreated
or cholesterol-enriched cells (Fig. 6 A), suggesting that
initial signaling steps occur over a longer time period.
This is consistent with our observations of a slower Ca
2
response in MbCD-treated cells compared to untreated
cells. The crossover between regimes occurs at larger
values of N in MbCD treated versus untreated cells,
FIGURE 6 Perturbations of membrane cholesterol have corresponding
effects on receptor diffusion, receptor clustering, and cellular Ca
2
responses. (A) Average D
S
and N from ve live cells for each of six
treatmentsMbCD (upper), no perturbation (middle), or MbCD chol
(lower)in the presence (solid symbols) and absence (open symbols) of
antigen stimulation (0.1 mg/ml). Changing color from blue to red indicates
advancement in time by 30 s for each time point from 10 to 10 min after
the addition of antigen for stimulated cells or a blank addition of buffer for
unstimulated cells. Cholesterol perturbations are added, where applicable,
at 5 min. For stimulated time points (solid symbols outlined in black),
the t lines shown in black represent linear ts of the time points belonging
to the two regimes of D
S
dependence on N, weighted by the inverse of
the the mean 5SE in D
S
and N. Stimulated time points near the crossover
are indicated by arrows and labeled with time after antigen addition. (B)
Total uorescence intensity is shown as a function of time for populations
of cells (at least 500) loaded with Fluo-4-AM and treated with MbCD,
MbCD chol, or no perturbation. The dotted gray line at 5 min indicates
the addition of MbCD or MbCD chol, and the dotted orange line
indicates the addition of antigen. (C) The cumulative distribution in time
of cells with an initial Ca
2
response as monitored by Fluo-4 uorescence.
The dark blue dotted line indicates the cumulative distribution of cells
treated with MbCD chol that exhibit an additional Ca
2
response after
the addition of antigen. The half-maximum times (t
1/2
) of cumulative curves
are denoted by open symbols. (D) The estimated midpoints of the crossover
times are plotted versus t
1/2
for each cholesterol treatment. Error bars
represent uncertainty in determining the midpoint of the crossover time
by 530 s.
suggesting that more receptors are needed to initiate
downstream signaling events when cholesterol levels are
reduced. Plots of Ds versus N for cholesterol-enriched cells
indicate only a single regime, and antigen-induced Ca
2
responses occur with a minimal time lag after antigen
addition.
Previous work has demonstrated the importance of mem-
brane-lipid-mediated protein targeting for transmembrane
signaling in the FcRI cascade (54,55). Cholesterol reduc-
tion inhibits stimulated receptor phosphorylation by Lyn,
and productive signaling only occurs upon the redistribution
of receptors, kinases, and phosphatases via changes in the
local lipid environment surrounding cross-linked receptors
(53,56,57). In our previous SEM work we found that
cholesterol reduction before antigen addition led to smaller
IgE-FcRI-rich clusters and reduced Lyn partitioning into
receptor-rich clusters when cells were chemically xed
1 min after antigen addition at 37
C. Although it is not
possible to compare absolute numbers between these exper-
iments due to the different labeling strategies employed, our
current ndings are consistent with these previous results.
Specically, we nd that it takes longer for receptors to
assemble into tight clusters when cells are pretreated with
MbCD, so at a given time point after stimulation, we would
expect receptor-rich clusters to be smaller in MbCD-pre-
treated cells compared to untreated cells. Our observations
of a delay in Ca
2
responses in MbCD-pretreated cells rela-
tive to untreated cells are consistent with our past observa-
tions of defects in Lyn recruitment under these conditions.
These ndings are consistent with an inhibitory role for
cholesterol reduction in signaling that has been supported
by previous observations (50,52,53). An alternative expla-
nation for the changes in diffusion versus clustering
behavior and Ca
2
responses observed in MbCD- and
MbCD chol-treated cells are related to more global
effects of cholesterol modulation, such as its perturbation
of the actin cytoskeleton. Cholesterol reduction can disrupt
cytoskeleton-membrane attachment through perturbation of
plasma membrane PIP
2
(58), or because actin is frequently
coupled to the plasma membrane via more ordered regions
(59). Therefore, the changes in receptor clustering, receptor
mobility, and Ca
2
responses caused by MbCD, and by
extension MbCD chol, may be indirect results of mem-
brane cholesterol modulation via its effects on the actin
cytoskeleton.
CONCLUSION
In conclusion, we demonstrate that superresolution uores-
cence localization imaging is a powerful method for quan-
tifying the organization and mobility of immune receptors
in cells undergoing stimulated responses. Simultaneous
measurements of clustering and diffusion enable the
resolution of two distinct temporal phases of receptor clus-
tering and immobilization. At early times after stimulation,
receptor-rich clusters increase marginally in size and recep-
tors slow dramatically when averaged over the population
of receptors. When examined as individual molecules,
either as monomers or as members of clusters, single
receptors appear to reversibly associate with small and
slowly moving receptor clusters soon after the addition of
antigen. These behaviors are observed at stimulation times
preceding Ca
2
mobilization, leading us to conclude that
they arise from interactions associated with initial signaling
steps. At later times, receptors in clusters become increas-
ingly dense and are largely immobile. Since these behav-
iors occur at times after the initial Ca
2
response, we
hypothesize that receptor immobilization into densely
packed clusters leads to subsequent cellular interactions
related to downregulation of signaling. Before these termi-
nating steps, dense receptor clusters are dispersed when the
cross-linking antigen is displaced by a monovalent ligand.
Receptor clustering and dynamics are also altered in cells
with modulated cholesterol levels. Most notably, receptors
cluster in cells with increased cholesterol levels even in
the absence of antigen, and we observe changes in the
duration of the initial phase of receptor clustering and
immobilization in stimulated cells with modulated choles-
terol levels. The onset of Ca
2
mobilization requires asso-
ciation of activated receptors with multiple proteins, and
this signaling complex formation appears to occur before
or during the crossover between the two regimes dened
by our analysis. We observe differences in the timing of
Ca
2
mobilization for cells with modulated cholesterol
levels that correspond to changes in the timing of the
initial phase of clustering. These ndings are motivators
for future work investigating the physical interactions
that give rise to the observed changes in receptor organiza-
tion and mobility, and how these translate into cellular
functions.
SUPPORTING MATERIAL
Eight gures, one movie, References (6062), Supporting Results and Sup-
porting Materials and Methods are available at http://www.biophysj.org/
biophysj/supplemental/S0006-3495(13)01126-0.
The authors thank Amit Singhai for assistance with experiments and for
providing reagents, and Christopher V. Kelly, Matthew B. Stone, and
Benjamin B. Machta for helpful discussions.
Research was supported through grants from the National Institutes of
Health to S.L.V. (R00GM087810) and B.B. and D.H. (RO1 AI018306).
S.A.S. acknowledges partial support from a National Institutes of Health
Molecular Biophysics Training Grant (T32GM008267).
REFERENCES
1. Siraganian, R. P. 2003. Mast cell signal transduction from the high-
afnity IgE receptor. Curr. Opin. Immunol. 15:639646.
2. Blank, U., and J. Rivera. 2004. The ins and outs of IgE-dependent mast-
cell exocytosis. Trends Immunol. 25:266273.
2352 Shelby et al.
3. Gilllan, A. M., and J. Rivera. 2009. The tyrosine kinase network
regulating mast cell activation. Immunol. Rev. 228:149169.
4. Schlessinger, J., W. W. Webb, ., H. Metzger. 1976. Lateral motion
and valence of Fc receptors on rat peritoneal mast cells. Nature. 264:
550552.
5. Menon, A. K., D. A. Holowka, ., B. A. Baird. 1986. Cross-linking of
receptor-bound IgE to aggregates larger than dimers leads to rapid
immobilization. J. Cell Biol. 102:541550.
6. Stump, R. F., J. R. Pfeiffer, ., J. M. Oliver. 1988. Mapping gold-
labeled IgE receptors on mast cells by scanning electron microscopy:
receptor distributions revealed by silver enhancement, backscattered
electron imaging, and digital image analysis. J. Histochem. Cytochem.
36:493502.
7. Wilson, B. S., J. R. Pfeiffer, and J. M. Oliver. 2000. Observing FcRI
signaling from the inside of the mast cell membrane. J. Cell Biol.
149:11311142.
8. Veatch, S. L., E. N. Chiang, ., B. A. Baird. 2012. Quantitative nano-
scale analysis of IgE-FcRI clustering and coupling to early signaling
proteins. J. Phys. Chem. B. 116:69236935.
9. Menon, A. K., D. A. Holowka, ., B. A. Baird. 1986. Clustering,
mobility, and triggering activity of small oligomers of immunoglobulin
E on rat basophilic leukemia cells. J. Cell Biol. 102:534540.
10. Feder, T. J., I. Brust-Mascher, ., W. W. Webb. 1996. Constrained
diffusion or immobile fraction on cell surfaces: a new interpretation.
Biophys. J. 70:27672773.
11. Larson, D. R., J. A. Gosse, ., W. W. Webb. 2005. Temporally resolved
interactions between antigen-stimulated IgE receptors and Lyn kinase
on living cells. J. Cell Biol. 171:527536.
12. Andrews, N. L., K. A. Lidke, ., D. S. Lidke. 2008. Actin restricts
FcepsilonRI diffusion and facilitates antigen-induced receptor immobi-
lization. Nat. Cell Biol. 10:955963.
13. Andrews, N. L., J. R. Pfeiffer, ., D. S. Lidke. 2009. Small,
mobile FcRI receptor aggregates are signaling competent. Immunity.
31:469479.
14. Spendier, K., K. A. Lidke, ., J. L. Thomas. 2012. Single-particle
tracking of immunoglobulin E receptors (FcRI) in micron-sized
clusters and receptor patches. FEBS Lett. 586:416421.
15. Rust, M. J., M. Bates, and X. Zhuang. 2006. Stochastic optical recon-
struction microscopy (STORM) provides sub-diffraction-limit image
resolution. Nat. Methods. 3:793795.
16. Heilemann, M., S. van de Linde, ., M. Sauer. 2008. Subdiffraction-
resolution uorescence imaging with conventional uorescent probes.
Angew. Chem. Int. Ed. Engl. 47:61726176.
17. Betzig, E., G. H. Patterson, ., H. F. Hess. 2006. Imaging intracellular
uorescent proteins at nanometer resolution. Science. 313:16421645.
18. Hess, S. T., T. P. K. Girirajan, and M. D. Mason. 2006. Ultra-high
resolution imaging by uorescence photoactivation localization
microscopy. Biophys. J. 91:42584272.
19. Greeneld, D., A. L. McEvoy, ., J. Liphardt. 2009. Self-organization
of the Escherichia coli chemotaxis network imaged with super-resolu-
tion light microscopy. PLoS Biol. 7:e1000137.
20. Lillemeier, B. F., M. A. Mortelmaier, ., M. M. Davis. 2010. TCR and
Lat are expressed on separate protein islands on T cell membranes and
concatenate during activation. Nat. Immunol. 11:9096.
21. Owen, D. M., C. Rentero, ., K. Gaus. 2010. PALM imaging and clus-
ter analysis of protein heterogeneity at the cell surface. J. Biophotonics.
3:446454.
22. Williamson, D. J., D. M. Owen, ., K. Gaus. 2011. Pre-existing clus-
ters of the adaptor Lat do not participate in early T cell signaling events.
Nat. Immunol. 12:655662.
23. Hsu, C.-J., and T. Baumgart. 2011. Spatial association of signaling
proteins and F-actin effects on cluster assembly analyzed via photoac-
tivation localization microscopy in T cells. PLoS ONE. 6:e23586.
24. Sherman, E., V. Barr, ., L. E. Samelson. 2011. Functional nanoscale
organization of signaling molecules downstream of the T cell antigen
receptor. Immunity. 35:705720.
25. Gunewardene, M. S., F. V. Subach, ., S. T. Hess. 2011. Superresolu-
tion imaging of multiple uorescent proteins with highly overlapping
emission spectra in living cells. Biophys. J. 101:15221528.
26. Jones, S. A., S.-H. Shim, ., X. Zhuang. 2011. Fast, three-dimensional
super-resolution imaging of live cells. Nat. Methods. 8:499508.
27. Manley, S., J. M. Gillette, ., J. Lippincott-Schwartz. 2008. High-den-
sity mapping of single-molecule trajectories with photoactivated local-
ization microscopy. Nat. Methods. 5:155157.
28. Manley, S., J. M. Gillette, and J. Lippincott-Schwartz. 2010. Single-
particle tracking photoactivated localization microscopy for mapping
single-molecule dynamics. Methods Enzymol. 475:109120.
29. Giannone, G., E. Hosy, ., L. Cognet. 2010. Dynamic superresolution
imaging of endogenous proteins on living cells at ultra-high density.
Biophys. J. 99:13031310.
30. Gadi, D., A. Wagenknecht-Wiesner, ., B. Baird. 2011. Sequestration
of phosphoinositides by mutated MARCKS effector domain inhibits
stimulated Ca
2
mobilization and degranulation in mast cells. Mol.
Biol. Cell. 22:49084917.
31. Veatch, S. L., B. B. Machta, ., B. A. Baird. 2012. Correlation func-
tions quantify super-resolution images and estimate apparent clustering
due to over-counting. PLoS ONE. 7:e31457.
32. Erickson, J., B. Goldstein, ., B. Baird. 1987. The effect of receptor
density on the forward rate constant for binding of ligands to cell sur-
face receptors. Biophys. J. 52:657662.
33. Barisas, B. G., S. M. Smith, ., D. A. Roess. 2007. Compartmentaliza-
tion of the Type I Fc receptor and MAFA on mast cell membranes.
Biophys. Chem. 126:209217.
34. Xu, K., B. Goldstein, ., B. Baird. 1998. Kinetics of multivalent anti-
gen DNP-BSA binding to IgE-Fc RI in relationship to the stimulated
tyrosine phosphorylation of Fc RI. J. Immunol. 160:32253235.
35. Paolini, R., M. H. Jouvin, and J. P. Kinet. 1991. Phosphorylation and
dephosphorylation of the high-afnity receptor for immunoglobulin
E immediately after receptor engagement and disengagement. Nature.
353:855858.
36. Ritchie, K., R. Iino, ., A. Kusumi. 2003. The fence and picket
structure of the plasma membrane of live cells as revealed by single
molecule techniques (Review). Mol. Membr. Biol. 20:1318 (Review).
37. Murase, K., T. Fujiwara, ., A. Kusumi. 2004. Ultrane membrane
compartments for molecular diffusion as revealed by single molecule
techniques. Biophys. J. 86:40754093.
38. Machta, B. B., S. Papanikolaou, ., S. L. Veatch. 2011. Minimal model
of plasma membrane heterogeneity requires coupling cortical actin to
criticality. Biophys. J. 100:16681677.
39. Pierini, L., N. T. Harris, ., B. Baird. 1997. Evidence supporting
a role for microlaments in regulating the coupling between poorly
dissociable IgE-Fc RI aggregates downstream signaling pathways.
Biochemistry. 36:74477456.
40. Ra, C., K. Furuichi, ., K. N. White. 1989. Internalization of IgE
receptors on rat basophilic leukemic cells by phorbol ester. Comparison
with endocytosis induced by receptor aggregation. Eur. J. Immunol.
19:17711777.
41. Liu, C., H. Miller, ., W. Song. 2011. A balance of Brutons tyrosine
kinase and SHIP activation regulates B cell receptor cluster formation
by controlling actin remodeling. J. Immunol. 187:230239.
42. Lee, K.-H., A. R. Dinner, ., A. S. Shaw. 2003. The immunological
synapse balances T cell receptor signaling and degradation. Science.
302:12181222.
43. Varma, R., G. Campi, ., M. L. Dustin. 2006. T cell receptor-proximal
signals are sustained in peripheral microclusters and terminated in the
central supramolecular activation cluster. Immunity. 25:117127.
44. Weetall, M., D. Holowka, and B. Baird. 1993. Heterologous desensiti-
zation of the high afnity receptor for IgE (Fc R1) on RBL cells.
J. Immunol. 150:40724083.
45. Saxton, M. J. 1997. Single-particle tracking: the distribution of diffu-
sion coefcients. Biophys. J. 72:17441753.
46. Fewtrell, C. 1985. Calcium in Biological Systems. Plenum Press,
New York.
47. Seagrave, J. C., G. G. Deanin, ., J. M. Oliver. 1987. DNP-phycobili-
proteins, uorescent antigens to study dynamic properties of antigen-
IgE-receptor complexes on RBL-2H3 rat mast cells. Cytometry.
8:287295.
48. Surviladze, Z., L. Draberova, ., P. Draber. 2001. Differential sensi-
tivity to acute cholesterol lowering of activation mediated via the
high-afnity IgE receptor and Thy-1 glycoprotein. Eur. J. Immunol.
31:110.
49. Nishimura, S. Y., M. Vrljic, ., W. E. Moerner. 2006. Cholesterol
depletion induces solid-like regions in the plasma membrane.
Biophys. J. 90:927938.
50. Kato, N., M. Nakanishi, and N. Hirashima. 2003. Cholesterol depletion
inhibits store-operated calcium currents and exocytotic membrane
fusion in RBL-2H3 cells. Biochemistry. 42:1180811814.
51. Yamashita, T., T. Yamaguchi, ., S. Nagasawa. 2001. Detergent-resis-
tant membrane domains are required for mast cell activation but
dispensable for tyrosine phosphorylation upon aggregation of the
high afnity receptor for IgE. J. Biochem. 129:861868.
52. Silveira e Souza, A. M. M., V. M. Mazucato, ., C. Oliver. 2008. The
a-galactosyl derivatives of ganglioside GD(1b) are essential for the
organization of lipid rafts in RBL-2H3 mast cells. Exp. Cell Res.
314:25152528.
53. Sheets, E. D., D. Holowka, and B. Baird. 1999. Critical role for choles-
terol in Lyn-mediated tyrosine phosphorylation of FcRI and their
association with detergent-resistant membranes. J. Cell Biol. 145:
877887.
54. Silveira E Souza, A. M. M., V. M. Mazucato, ., C. Oliver. 2011. Lipid
rafts in mast cell biology. J. Lipids. 2011:752906.
55. Holowka, D., J. A. Gosse, A. T. Hammond, X. Han, P. Sengupta.,
2005. Lipid segregation and IgE receptor signaling: a decade of
progress. Biochim. Biophys. Acta. 1746:252259.
56. Field, K. A., D. Holowka, and B. Baird. 1997. Compartmentalized
activation of the high afnity immunoglobulin E receptor within
membrane domains. J. Biol. Chem. 272:42764280.
57. Young, R. M., D. Holowka, and B. Baird. 2003. A lipid raft environ-
ment enhances Lyn kinase activity by protecting the active site tyrosine
from dephosphorylation. J. Biol. Chem. 278:2074620752.
58. Kwik, J., S. Boyle, ., M. Edidin. 2003. Membrane cholesterol, lateral
mobility, and the phosphatidylinositol 4,5-bisphosphate-dependent
organization of cell actin. Proc. Natl. Acad. Sci. USA. 100:13964
13969.
59. Holowka, D., E. D. Sheets, and B. Baird. 2000. Interactions between
FcRI and lipid raft components are regulated by the actin cytoskel-
eton. J. Cell Sci. 113:10091019.
60. Naal, R. M. Z. G., J. Tabb, ., B. Baird. 2004. In situ measurement of
degranulation as a biosensor based on RBL-2H3 mast cells. Biosens.
Bioelectron. 20:791796.
61. Hardy, R. R. 1986. Handbook of Experimental Immunology, 4th ed.
Blackwell Scientic, Oxford, United Kingdom.
62. Jaqaman, K., D. Loerke, ., G. Danuser. 2008. Robust single-particle
tracking in live-cell time-lapse sequences. Nat. Methods. 5:695702.
2354 Shelby et al.
Building KCNQ1/KCNE1 Channel Models and Probing their Interactions by
Molecular-Dynamics Simulations
Yu Xu, Yuhong Wang, Xuan-Yu Meng, Mei Zhang, Min Jiang, Meng Cui, and Gea-Ny Tseng*
Department of Physiology & Biophysics, Virginia Commonwealth University, Richmond, Virginia
ABSTRACT The slow delayed rectier (I
Ks
) channel is composed of KCNQ1 (pore-forming) and KCNE1 (auxiliary) subunits,
and functions as a repolarization reserve in the human heart. Design of I
Ks
-targeting anti-arrhythmic drugs requires detailed
three-dimensional structures of the KCNQ1/KCNE1 complex, a task made possible by Kv channel crystal structures
(templates for KCNQ1 homology-modeling) and KCNE1 NMR structures. Our goal was to build KCNQ1/KCNE1 models
and extract mechanistic information about their interactions by molecular-dynamics simulations in an explicit lipid/solvent
environment. We validated our models by conrming two sets of model-generated predictions that were independent
from the spatial restraints used in model-building. Detailed analysis of the molecular-dynamics trajectories revealed
previously unrecognized KCNQ1/KCNE1 interactions, whose relevance in I
Ks
channel function was conrmed by voltage-
clamp experiments. Our models and analyses suggest three mechanisms by which KCNE1 slows KCNQ1 activation: by
promoting S6 bending at the Pro hinge that closes the activation gate; by promoting a downward movement of gating
charge on S4; and by establishing a network of electrostatic interactions with KCNQ1 on the extracellular surface that
stabilizes the channel in a pre-open activated state. Our data also suggest how KCNE1 may affect the KCNQ1 pore
conductance.
INTRODUCTION
KCNQ1 is a typical voltage-gated potassium (Kv) channel
and KCNE1 is a type-I transmembrane peptide (Fig. 1 A).
They associate to form the slow delayed rectier (I
Ks
)
channel (Fig. 1 B) expressed in human atrial and ventricular
myocytes (14). The importance of I
Ks
in maintaining
the electrical stability of human heart is highlighted by
the linkage between loss-of-function mutations in KCNQ1
or KCNE1 to long QT syndromes (LQT1 or LQT5) (5),
and the linkage between their gain-of-function mutations
to congenital atrial brillation, and for KCNQ1, short QT
syndrome (68).
There has been long-standing interest in understand-
ing how the I
Ks
channel works: how does KCNE1 slow
KCNQ1 activation, increase the current amplitude through
the KCNQ1 pore, and suppress KCNQ1 inactivation
(effects ac in Fig. 1 B)? Structural information is required
for rational design of I
Ks
modulators, whose clinical
applications include treating congenital and acquired
long-QT syndromes (9). Our goal was to build and
validate three-dimensional models of the KCNQ1/
KCNE1 channel complex, subject the models to mole-
cular-dynamics (MD) simulations, and extract novel in-
sights into the structure-dynamics-function relationship
of the I
Ks
channel from detailed analysis of the MD
trajectories.
Homology modeling of KCNQ1
Forty three-dimensional KCNQ1 models were generated using the
program MODELLER (University of California at San Francisco,
San Francisco, CA), and the one with the highest G-score from
PROCHECK (https://www.ebi.ac.uk/thornton-srv/software/PROCHECK/)
and the highest percent of residues in favored regions was selected.
We rebuilt the S1-S2 and S2-S3 linkers using the SYBYL loop structure
template database (JPR Technologies; http://www.jprtechnologies.com.
au/tripos/discovery-informatics/sybyl/), and constrained the distance
between positions 142 and 228 (C
z
-C
z
< 5 A
) using a CHARMM
simulation (Martin Karplus, Harvard University, Boston, MA; http://
www.charmm.org/). A Ramachandran plot showed that 93.2% of the
residues were in the most favored regions, and 6.0% in the allowed
regions.
Rening KCNE1 NMR structure
We adjusted the following peptide backbone dihedral angles in the
KCNE1 NMR structure: j at E43 (from 26.1
to 139.8
), 4 at H73
(86.2
to 122.9
), S74 (112.4
to 60.0
), and D76 (111.8
to
79.9
). The adjusted KCNE1 structure was rened by MD simulations

(details below).
Docking KCNE1 to KCNQ1 using Brownian
dynamics simulations
The program package MACRODOX (Ver. 3.2.2 used, latest version is
4.6.1; available at http://iweb.tntech.edu/macrodox/macrodox.html) was
used to assign charges, solve the linearized Poisson-Boltzmann equation,
and run the Brownian dynamics protein-docking simulations. The nal
docking conformations were rened by CHARMM simulation for 20 ps,
with KCNQ1 C
a
atoms restrained harmonically and KCNE1 residues
46 and 71 constrained at C
a
-C
a
distance 38.4 A
.
Submitted August 2, 2013, and accepted for publication September 11,
2013.
*Correspondence: gtseng@vcu.edu
Yu Xu, Yuhong Wang, and Yuan-Yu Meng contributed equally to the article.
Editor: Randall Rasmusson.
Biophysical Journal Volume 105 December 2013 24612473 2461
MD simulations
We conducted MD simulations using GROMACS, Ver. 4.5.3 with the
GROMOS96 53a6 force eld (www.gromacs.org). Using the VMD
membrane package, the protein structure was immersed in an explicit
POPC (palmitoyloleoyl-phosphatidylcholine) bilayer, and solvated with
single-point-charge water molecules. Two sets of MD simulations were
performed on Q1, Q1Ea, and Q1Eb systems. In the rst set of MD simu-
lations (MDS#1), we applied a constant electric eld of 0.128 V,nm
1
(corresponding to transmembrane voltage of 435 mV) with 600 mM
KCl. The total numbers of atoms in the Q1, Q1Ea, and Q1Eb systems
were 74,681, 96,394, and 119,970. In the second set of MD simulations
(MDS#2), there was no electrical eld and only four K
ions were placed

in the pore with Cl
ions added to neutralize net charges of the system

(nominally 0 mM ions). The total numbers of atoms in the Q1, Q1Ea,
and Q1Eb systems were 76,329, 98,418, and 122,450. The E1-alone MD
simulation was run under the second set of conditions, with a total of
60,092 atoms. Bond lengths were constrained with the LINCS algorithm.
Electrostatic interactions were calculated by the particle-mesh Ewald
method with 12 A
cutoff. The van der Waal interactions were modeled

using Lennard-Jones 6-12 potentials with 14 A
cutoff. All simulations

were conducted at a constant temperature (300
K) and constant pressure

(1 bar) using the Berendsen method. The neighborhood list was updated
every 20 fs.
After 100 (E1 alone) or 3000 (Q1 alone, Q1Ea, and Q1Eb) steps of
energy minimization using the steepest-descent algorithm, each system
was subjected to a 0.5-ns two-step dynamics simulation with the restraint
on positions gradually weakened. To permit water and ions to relax
about the protein(s), the restraints on the protein(s) and K
ions were
set to 1000 kJ/mol/nm
2
for 0.2 ns, and 10 kJ/mol/nm
2
for 0.3 ns, respec-
tively. A 100-ns production run was conducted on each system under the
conditions described above and coordinates were saved every 10 ps for
analysis.
Analysis of MD trajectories
Root mean-square deviation (RMSD) values of protein C
a
atoms during
whole MD simulations were generated by GROMACS, Ver. 4.5.3.
The following analyses were conducted on the second halves of MD simu-
lations (50100 ns), when the systems had reached or were approaching
equilibrium based on their RMSD values:
1. Clustering structures and analysis of side-chain/backbone interactions,
including hydrogen bonds, salt bridges, and hydrophobic contacts
(using the SIMULAID online data base, http://www.freechemical.info/
freeSoftware/Simulaid.html);
2. Calculation of backbone root mean-square uctuations (RMSFs,
determined with the software GROMACS, Ver. 4.5.3); and
3. Principal component analysis (GROMACS, Ver. 4.5.3) and visualization
(using VMD, available from the University of Illinois, Urbana-Cham-
paign, IL, https://www-s.ks.uiuc.edu/Research/vmd/; and CHIMERA,
available from the University of California at San Francisco, San
Francisco, CA; http://www.cgl.ucsf.edu/chimera/).
Methods in the Supporting Material
Details of site-directed mutagenesis, oocytes expression and voltage-clamp,
COS-7 culture, and immunoblot experiments are provided in the
Supporting Material.
A
C
E
B
D
FIGURE 1 KCNQ1 and KCNE1 associate to form
the I
Ks
channel. (A) Two-dimensional diagram of
KCNQ1 and KCNE1 subunits. Each KCNQ1 subunit
has six transmembrane segments (S1S6) and a pore-
lining P-loop, which can be functionally divided into
a voltage-sensing domain (VSD) and a pore domain
(PD) linked by the S4-S5 linker. Each KCNE1 subunit
has one transmembrane domain (TMD), with extracel-
lular amino-terminus (NT) and intracellular carboxy-
terminus (CT). (B) KCNE1 slows KCNQ1 activation
(a), increases the current amplitude (b), and suppresses
KCNQ1 inactivation (abolishing the hooked phase
of tail current, c). (C) Cartoon of a Kv channel crystal
structure (PDB:2R9R, viewed from the extracellular
side of the membrane), with two KCNE subunits in
diagonal KCNE-binding pockets (gray shades). As a
reference for Fig. 7, the four Kv channel subunits are
designated as chains AD, and the two KCNE subunits
as chains E and F. (D) KCNE1 NMR structure in
LMPG micelles (PDB:2K21), with NT-, TM-, and
CT-helices and loops marked. (Dotted circle) Putative
LMPG micelle. (E) Sequence alignment between
PDB:2R9R and KCNQ1 (PDB:2R9R position numbers
based on rat Kv1.2, accession number: P63142).
(Dashes) Gaps. Helical regions (S1S6, and pore-
helix) are noted. The following residues are high-
lighted: T184 of PDB:2R9R and T144 of KCNQ1
(asterisks), I331 of PDB:2R9R and Y299 of KCNQ1
(asterisks), L142 and R228 of KCNQ1 (pound signs).
2462 Xu et al.
RESULTS
Constraining the relationship between the
extracellular end of S1 and pore domain
or S4 in the KCNQ1 channel
We used a high-resolution Kv channel crystal structure
as template for building a KCNQ1 homology model
(PDB:2R9R, Fig. 1 C) (10). Amino-acid sequence align-
ment indicated a serious challenge in modeling the
KCNQ1 S1-S2 linker, which is much shorter than the tem-
plate without sequence homology (Fig. 1 E). Therefore,
we rst sought experimental restraints to position the extra-
cellular end of S1 with respect to other domains in the
KCNQ1 channel.
Lee et al. (11) identied a contact between the extra-
cellular end of S1 and the beginning of the pore helix of
adjacent subunits. In PDB:2R9R, these are T184 and I361.
The equivalent positions in KCNQ1 are T144 and Y299
(highlighted by asterisks in Fig. 1 E). We tested whether
these two positions, when occupied by cysteine (Cys), could
come close enough to allow intersubunit disulde forma-
tion, which would produce a KCNQ1 dimer. Cys was
engineered into a Cys-free KCNQ1 background, designated
as Q1*, and our strategy is diagrammed in Fig. 2 Aa. Fig. 2
Ab depicts a representative immunoblot image, and Fig. 2
Ac summarizes densitometry data. When T144C was paired
with S298C, Y299C, or A300C, dimer formation occurred
spontaneously (oxidized by air O
2
). Dimer formation was
markedly enhanced by incubation with H
2
O
2
and abolished
by subsequent dithiothreitol (DTT) treatment. These data
constrained the relationship between the extracellular end
of S1 and the beginning of pore-helix in the KCNQ1
channel.
We tested whether L142 (at the extracellular end of S1)
could come close to R228 (the rst Arg on S4) to allow
disulde formation (highlighted by the pound signs in
Fig. 1 E). We also wanted to know whether this occurred
in the closed or open state (two scenarios in Fig. 2 Ba).
L142 and R228 were replaced by Cys, singly or simulta-
neously, in the same Q1* subunit, and we used oocyte
expression to test the effects of DTT on the channel
gating function. The immunoblot method described above
was not suitable because the disulde bond between
142C and 228C would be intrasubunit, i.e., Q1*
remained a monomer and its mobility in sodium do-
decyl-sulfate polyacrylamide gel electrophoresis would
change little. On the other hand, based on the effects of
DTT on the channel gating kinetics, we could infer
whether the spontaneous disulde bond was preferentially
formed in the open state (DTT treatment destabilized the
open state, leading to a slowing of activation, acceleration
of deactivation, and/or a decrease in the instantaneous
component, depending on the voltage-clamp protocol
used), or in the closed state (opposite effect[s]) (12).
Fig. 2 Bb shows that Q1*-L142C/R228C exhibited a large
constitutive component under the control conditions.
DTT treatment removed the constitutive component,
revealing a slowly activating kinetics. The development
of the constitutive component in Q1*-L142C/R228C
was enhanced by depolarizing pulses that activated the
channel, and was prevented by DTT treatment (Fig. 2
Bc). Negative control (single Cys mutants: L142C or
R228C) did not exhibit this gating behavior or sensitivity
to DTT (see Fig. S1 in the Supporting Material). These
data supported the second scenario in Fig. 2 Ba, and con-
strained the open-state position of S1 with respect to S4
in the KCNQ1 channel.
Creating a homology KCNQ1 model and docking KCNE1
in two stages
Fig. 3 A depicts a side view of the nal KCNQ1 homology
model. The S4 positions equivalent to the rst four gating
charge-bearing positions of Shaker (R228, R231, Q234,
and R237) were above a putative hydrophobic seal signied
by Y167 on S2 (13,14), conrming the activated status of
the model. Consistent with our experimental restraints,
L142 and R228 of the same subunit were close to each other.
T144 and S298 of adjacent subunits were also close to each
other (not shown).
There are issues with the KCNE1 NMR structure,
PDB:2K21 (15), in terms of the conformations of transmem-
brane helix and loops between helices (Fig. 1 D) (16). It
needed renement before the docking exercise. We manu-
ally adjusted selected dihedral angles of the KCNE1 NMR
structure, so that the amino-terminus (NT)- and carboxy-ter-
minus (CT)-domains did not fold back into the membrane.
The adjusted KCNE1 structure was rened by 100-ns MD
simulations in lipid and solvent environment (see Fig. S2).
Analysis of snapshots between 50 and 100 ns of the MD
trajectory revealed ve clusters of structures. Fig. 3 B
depicts representative structures from the ve clusters
superimposed by their transmembrane (TM) helices. While
the E1-NT, -TM, and -CT helices were maintained, their
relative positions were variable due to the highly exible
loops connecting them. Based on this analysis, the following
docking strategy was devised: we would dock the helical
regions of KCNE1 to the KCNQ1 homology model in
separate steps, select the most favorable docking conforma-
tions based on available experimental restraints, build the
loops connecting the helices, and then allow the systems
to adjust themselves by MD simulations. The E1-CT was
not included, because the KCNQ1 homology model does
not include the cytoplasmic regions that interact with
E1-CT (17). Furthermore, to our knowledge, there are not
sufcient experimental data to constrain E1-CT with
respect to the intracellular surface of the KCNQ1 homology
model (18).
Fig. S3 lists the procedures of the rst stage: docking
E1-TMD (amino acids (aa) 4071) to the KCNQ1
homology model. The nal docking model, designated
Structure and Function of the I
Ks
Channel 2463
as (Q1)
4
/(E1-TMD), satised modeling criteria and all
available experimental restraints based on disulde for-
mation between Cys side chains engineered into specic
KCNQ1 and KCNE1 positions (12,1923). These spatial
restraints positioned the KCNE1 transmembrane domain
(TMD) relative to the KCNQ1 voltage-sensing domain
and pore-domain (Fig. 3 C).
Fig. S4 outlines the second stage: docking E1-NT
(aa 134) to the above structure. Our strategy was based
on experimental data suggesting a preferred spatial
relationship between E1-NT positions 14, 22, and 34
and the KCNQ1 external pore entrance (see detailed
description in the Supporting Material) (24). Selections
based on modeling criteria and these spatial restraints
led to 22 distinct conformations that could be roughly
classied as having the KCNE1 NT-helix (aa 1123)
interacting with, or away from, the external surface of
the KCNQ1 channel (designated as Q1Ea and Q1Eb,
respectively, Fig. 3 D). We inspected each of these 22
structures and removed those violating further modeling
criteria (lter 4; see Fig. S4). After this step, four
structures remained: two each of the Q1Ea and Q1Eb
FIGURE 2 Relationship between the extracellular end of S1 and other domains in KCNQ1. (A) Testing S1 and pore-helix contact in KCNQ1.
(a) Experimental design. T144C and pore-helix Cys mutants were engineered into separate subunits of KCNQ1 whose native Cys had been replaced
by Ala (designated Q1*), and coexpressed at 1:1 ratio. Disulde formation between the Cys side chains would produce Q1* dimer. To minimize the inter-
ference of KCNQ1 self-oligomerization, we included wild-type KCNE1 in the transfection. Pilot experiments showed that KCNE1 coexpression reduced
KCNQ1 self-oligomerization, likely due to KCNQ1/KCNE1 interactions in their C-terminal domains (17). (b) Representative immunoblot image of
whole-cell lysates from COS-7 cells expressing Q1*-T144C paired with Q1*-G297C, -S298C, -Y299C, -A300C, or -D301C, along with KCNE1.
COS-7 cells were treated with H
2
O
2
(0.1%), or with H
2
O
2
followed by DTT (50 mM) before NEM (protection of free thiol groups) and protein solubi-
lization. Whole-cell lysates were fractionated by nonreducing sodium dodecyl-sulfate polyacrylamide gel electrophoresis, and probed for Q1* (immuno-
blot: Q1). (Left) Q1* monomer and dimer bands and size-marker bands. (c) Ratios of dimer/monomer band intensities when Q1*-T144C was paired with
pore-helix Cys mutants listed along the abscissa. As negative controls, single Cys mutants (pore-helix Cys mutants without Q1*-T144C, or Q1*-T144C
alone) were also included. All were coexpressed with KCNE1. COS-7 cells were treated with H
2
O
2
(triangles), H
2
O
2
followed by DTT (inverted
triangles), or without any treatment (oxidized by air O
2
, circles). (B) Testing contacts between positions 142 and 228 in KCNQ1. (a) Two scenarios
of L142/R228 contact. (b) Current traces recorded from an oocyte expressing Q1*-L142C/R228C, elicited by the voltage-clamp protocol diagrammed
(inset). The recordings were made under the control conditions and then in the presence of DTT (10 mM). (c) Effects of repetitive depolarizing pulses
on the constitutive component of Q1*-L142C/R228C. Shown are changes in the constitutive current amplitude during pulsing (to 20 mV for 2 s) applied
once every 30 s under the control conditions and in the presence of DTT. (Inset) Current traces elicited by the rst and last pulses. (Dotted rectangles)
Constitutive current component.
2464 Xu et al.
conformations. We selected one Q1Ea and one Q1Eb for
further analysis.
MD simulations and model validation
We subjected the homology model of KCNQ1 (Q1 alone),
Q1Ea and Q1Eb docking conformations to 100-ns MD
simulations (Fig. 4 A) under two in silico conditions:
1. In 600 mM [KCl] with 435 mV transmembrane
voltage; and
2. In nominally 0 mM [KCl] (4 K
ions in the pore) and

0 mV transmembrane voltage.
These are designated as MDS#1 and MDS#2. Before
detailed analysis of these MD trajectories (Fig. 4 B), we
checked the quality of our models by testing two sets of
model-generated predictions observed during both MDS#1
and MDS#2. Both sets of predictions were independent
from the spatial restraints used in the above model-building
processes described in Fig. S3 and Fig. S4.
Prediction No. 1: salt-bridge formation between
oppositely charged side chains on KCNE1-NT
and KCNQ1 S5-P linker
Table S1 in the Supporting Material shows that the pre-
dictions of salt-bridge formation between KCNQ1 and
KCNE1 were similar between MDS#1 and MDS#2. We
checked nine of these predictions (diagrammed at the top
of Fig. 5) by engineering Cys into these positions in the
Cys-free Q1* and E1 background and testing whether
they could form a disulde bond. The procedures were
modied from those used in Fig. 2 A (see Fig. 5s legend).
Fig. 5 A shows that a clear 80-kDa band was seen in each of
these nine Cys-substituted Q1*/E1 pairs, but not in the
negative controls (Q1* constructs alone, without the E1 part-
ners). Furthermore, DTT treatment abolished the 80-kDa
bands, conrming that they represented disulde-linked
Q1*-S-S-E1.
Because the peptide backbones of KCNE1-NT and
KCNQ1 S5-P linker are highly exible (Figs. 6 A and 7 A),
FIGURE 3 Intermediate structures created during
the process of building KCNQ1/KCNE1 docking
models. (A) Homology model of KCNQ1 in the acti-
vated state, designated as (Q1)
4
. For clarity, only
two diagonal subunits are shown (green and light-
blue ribbons). (Boxed area) Relationships between
S4 (R228, R231, Q234, R237, H240, and R243),
S1 (L142), and S2 (E160, F167, and E170). (B)
Five representative KCNE1 structures rened from
the adjusted NMR structure (see Fig. S2 in the Sup-
porting Material), superimposed based on their TM
helices. (C) Final model of E1-TMD(aa 4071, blue
ribbon) docked to (Q1)
4
(green, cyan and white
ribbons), designated as (Q1)
4
/(E1-TMD) (see
Fig. S3). (Enlarged view of boxed area) KCNE1
T58 far from KCNQ1 S338, F339, or F340. (D)
(Left) Top view of an ensemble of 22 models with
E1-NT (aa 135, multicolor ribbons) docked to
(Q1)
4
/(E1-TMD) (white ribbons), designated as
(Q1)
4
/(E1-TMD)(E1-NT) (see Fig. S4). (Right)
Side views of two nal structures of KCNE1
(171) docked to the KCNQ1 homology model in
a 2:4 stoichiometry, designated as Q1Ea and Q1Eb
(KCNQ1 as gray ribbons, and KCNE1 as red or
blue ribbons).
Ks
Channel 2465
we further tested whether the above disulde bonds resulted
from random encounters between KCNQ1 and KCNE1 in
this region. If this were the case, we expected to see similar
degrees of disulde formation involving Cys side chains
engineered into neighboring positions. Fig. 5 B shows that
E1-R32C formed stronger disulde-linked bands with
Q1*-E284C, -D286C, -E290C, and -E295C, than with Cys
at anking KCNQ1 positions. The same was true for E1-
E19C/Q1*-R293C: its disulde-linked band was stronger
than that of E1-E19C/Q1*-G292C or V294C.
Experiments shown in Fig. 5, A and B, were repeated
multiple times with similar results and densitometry quanti-
cation is summarized in Fig. S5 and Fig. S6. These obser-
vations indicated that the multiple salt-bridges predicted
to occur between the KCNE1-NT and KCNQ1 S5-P linker
stabilized the loop conformations sufciently to support
the preferred contact pattern, even when two of the charged
side chains were replaced by neutral Cys. Because the pre-
dicted salt-bridge interactions were largely nonredundant
between Q1Ea and Q1Eb (Fig. 5 A, top, see also Fig. 6
B), these data not only validated our models but also
conrmed that both docking conformations are possible in
KCNQ1/KCNE1 complexes.
Prediction No. 2: KCNE1 position 58 is too
far from KCNQ1 positions 338340 to allow
disulde formation
Based on indirect evidence, it has been proposed that
KCNE1 T58 (in the middle of the TMD) interacts with
KCNQ1 F340 (in the middle of S6) (25,26). This has since
been used as a major spatial restraint in building KCNQ1/
KCNE1-TMD docking models (15,27), although in the
latter case MD simulations suggested a preferred interac-
tion between KCNE1 T38 and KCNQ1 S338, instead of
F340. We docked KCNE1-TMD to a KCNQ1 homology
model based on disulde trapping data, i.e., evidence of
direct contacts (see Fig. S3) (12,1923), without any
assumption about KCNE1 T58 relative to KCNQ1. In the
nal docking model, KCNE1 T58 was far from KCNQ1
S338, F339, or F340 (Fig. 3 C). Inspecting the MD trajec-
tories revealed that T58 did not make contact with KCNQ1
S6 at all.
To check this model prediction, we engineered Cys
into KCNE1 position 58 and anking positions (57 and
59) and paired them with each of Q1*-S338C, -F339C,
and -F340C. Because these were transmembrane positions
and the hydrophobic environment would not favor disulde
formation, it was critical to guard against false-negative
results. The experimental conditions were as previously
described in Wang et al. (28) (see also Fig. S7s legend).
We took three precautions:
1. We used two positive controls to ensure our ability to
detect disulde bonds if they occurred: Q1*-331C/
KCNE2-M59C (transmembrane disulde-forming pair,
producing a 160-kDa band; see Fig. S7 B), and Q1*-
Q147C/KCNE1-G40C (extracellular disulde-forming
pair, producing an 80-kDa Q1*-S-S-E1 band, similar to
those shown in Fig. 5, and sometimes higher molecular
weight bands, as seen in Fig. S7 B) (12,28).
2. We included one of the positive controls in all our
immunoblots to ensure the nonreducing conditions
(see Fig. S7 A).
3. We conrmed the expression of Cys-substituted E1
variants (see Fig. S7 A, lower panel), so that failure
to detect disulde formation was not due to failure
of expressing disulde-forming partners.
FIGURE 4 MD simulations. (A) Proteins (green ribbons) embedded in
lipids (gray wires) were immersed in water molecules (red) with K
ions
(purple spheres). For clarity, the KCNQ1 subunit closest to the viewer is
removed. (B) Trajectories of C
a
RMSD of Q1 alone, Q1Ea, and Q1Eb
(gray, red, and blue traces, respectively) under two sets of conditions:
MDS#1 (in 600 mM KCl, with transmembrane voltage of 435 mV) and
MDS#2 (with four K
ions in the pore, no transmembrane voltage). (C)

Top views of superimposed snapshots of Q1Ea and Q1Eb taken at every
10th ns during the 100-ns MDS under conditions #1.
2466 Xu et al.
The representative immunoblots shown in Fig. S7 A
and data summarized in Fig. S7 C led us to conclude
that our model prediction was conrmed: none of the
Cys-substituted pairs (E1-F57C, -T58C, or -L59C paired
with Q1*-S338C, -F339C, or -F340C) could form a
disulde bond.
Analyzing KCNQ1/KCNE1 interactions during MDtrajectories
To seek mechanistic insights into how KCNE1 and
KCNQ1 interact with each other, we analyzed how they
inuenced each others backbone exibility (C
a
RMSF)
and the degree of their contacts during MD trajectories
(denition and calculation in Fig. 6s legend). We used
principal component analysis to deduce KCNQ1 backbone
displacements induced by KCNE1 docking (2932).
The analysis was performed on the equilibrium phase
(50100 ns) of MD simulations under conditions 1 and
2. Figs. 6 and 7 present analysis based on MDS#1.
Analysis based on MDS#2 is presented in Fig. S8 and
Fig. S10.
KCNE1 interactions with KCNQ1
It has been proposed that the I
Ks
channel complex is formed
by binding of KCNE1 TMD to KCNQ1 (33). When KCNE1
was alone in lipid bilayer, its TMD showed a distinct
helical pattern of local peaks in backbone exibility.
Docking to KCNQ1 stabilized the KCNE1 TMD backbone,
so that all the local peaks disappeared (Fig. 6 A). In both
Q1Ea and Q1Eb docking conformations, the KCNE1
TMD displayed a helical pattern of contact with KCNQ1
(Fig. 6 B), that almost perfectly coincided with the helical
pattern of local peaks in backbone exibility observed in
lone KCNE1 (marked by the dotted lines through Fig. 6,
A and B). These observations suggest a scenario for
KCNQ1/KCNE1 docking, during which the KCNE1 TMD
adopts multiple conformations until it can snuggly t into
the space between KCNQ1 subunits and make multiple
contacts with KCNQ1. Fig. 6 C shows that in both Q1Ea
and Q1Eb, the rst half of KCNE1 TMD (positions 44
55) made extensive contacts with KCNQ1 in the voltage-
sensing domain (extracellular ends of S1, S1-S2, and
8
4
8
6
9
0
9
5
9
3
TMD
E
1
9
R
3
3
R
3
2
KCNE1
Q1Ea interaction
Q1Eb interaction
A - DTT + DTT
-
3
2
C
3
3
C
-
3
2
C
3
3
C
-
3
2
C
3
3
C
-
3
2
C
3
3
C
-
1
9
C
3
2
C
3
3
C
3
2
C
3
3
C
3
2
C
3
3
C
3
2
C
3
3
C
E1
1
9
C
S5
E
2
8
D
2
8
E
2
9
E
2
9
R
2
9
P-helix KCNQ1
E
2
8
4
C
E
2
8
4
C
R
3
E
2
8
4
C
R
3
D
2
8
6
C
D
2
8
6
C
R
3
D
2
8
6
C
R
3
E
2
9
0
C
E
2
9
0
C
R
3
E
2
9
0
C
R
3
E
2
9
5
C
E
2
9
5
C
R
3
E
2
9
5
C
R
3
R
2
9
3
C
R
2
9
3
C
E
1
E
2
8
4
C
R
3
E
2
8
4
C
R
3
D
2
8
6
C
R
3
D
2
8
6
C
R
3
E
2
9
0
C
R
3
E
2
9
0
C
R
3
E
2
9
5
C
R
3
E
2
9
5
C
R
3
Q1*
R
2
9
3
C
E
1
100 -
Q1* S S E1
75 -
50 -
C + E1-R32C + E1-E19C
Q1 -S-S-E1
Q1*
100 -
B
Q1*
A
2
8
3
C
D
2
8
6
C
E
2
8
4
C
K
2
8
5
C
A
2
8
7
C
S
2
9
1
C
V
2
9
4
C
N
2
8
9
C
E
2
9
0
C
E
2
9
5
C
F
2
9
6
C
G
2
9
2
C
R
2
9
3
C
V
2
9
4
C
+ E1-R32C + E1-E19C
Dep
E1
Q1*
E19C
R293C
+ -
R32C
E290C
+ -
R32C
E295C
+ -
R33C
E295C
+ -
75 -
50 -
Q1*-S-S-E1
Q1*
50
FIGURE 5 Using disulde trapping to conrm model predictions of KCNQ1/KCNE1 interactions. (Top) Predicted salt-bridge interactions between
oppositely charged side chains on KCNE1 NT and KCNQ1 S5-P linker. The interacting pairs are linked (red or blue lines, occurring in Q1Ea or Q1Eb
conformation, respectively). (A) Detecting disulde formation between specied Cys-substituted Q1*/E1 pairs under nonreducing conditions (DTT).
DTT treatment (DTT) and single Cys-substituted Q1* without E1 partners served as positive and negative controls. (B) Preference of disulde formation
between KCNQ1 and KCNE1 positions bearing oppositely charged side chains in the native state. (C) Effects of depolarization (Dep, by 100 mM [K]
o
)
on the degree of disulde formation between specied Q1*/E1 pairs. In experiments shown in panels A and B, COS-7 cells were incubated with 0.01%
H
2
O
2
in regular medium ([K]
o
5 mM) for 10 min, followed by raising [K]
o
to 100 mM in the continuous presence of H
2
O
2
for 10 min before NEM
(N-ethylmaleimide) (protection of free thiol groups) and protein solubilization. In panel C, cells with or without 100 mM [K]
o
treatment are labeled
as or Dep. To illustrate the presence of disulde-linked Q1*-S-S-E1 bands (80 kDa) and loading control (Q1*, 60 kDa), immunoblot images in
the 75100 kDa range (long exposure) and in the 5070 kDa range (short exposure) are shown separately. The corresponding complete immunoblot
(long exposure) images are shown in Fig. S5 and Fig. S6.
Ks
Channel 2467
S3-S4 linkers) and the pore-domain (S5-P and P-S6 linkers,
and the extracellular ends of S5 and S6). The second half
of KCNE1 TMD (positions 5671) made more limited con-
tacts with KCNQ1: S5 and S4-S5 linker. In Q1Ea, there was
also a small degree of contact with KCNQ1 S1 and S6
CT
.
Docking to KCNQ1 also reduced the backbone exibility
of KCNE1 extracellular domain, more so in Q1Ea than
Q1Eb. This is consistent with snapshots during MD trajec-
tories shown in Fig. 4 C: the E1-NT was much more
dynamic in Q1Eb than Q1Ea. In the Q1Ea conformation,
the major contact was between the E1-NT helix (F12,
L16, E19, and Q23) and the KCNQ1 S5-P linker (open
red arrow linking Fig. 6 B to the top panel of Fig. 6 C
and D). In the Q1Eb conformation, the major contact was
between E1-NT loop and KCNQ1 S5-P linker (blue open
arrow linking Fig. 6 B to the bottom panel of Fig. 6 C),
where electrostatic interactions between KCNE1 R32 and
R33 and negatively charged side chains on KCNQ1 S5-P
linker were important (Fig. 6 D).
Fig. S8 A shows that the main features of KCNE1 interac-
tions with KCNQ1 during MDS#2 were similar to those
described above for MDS#1.
KCNQ1 interactions with KCNE1
KCNQ1 alone in lipid bilayer exhibited distinct features of
peptide backbone exibility (Fig. 7 A). The transmembrane
helices, S4-S5 linker, P-loop, and P-S6 linker were stable
(34). On the other hand, the S1-S2, S2-S3, S3-S4, and
S5-P linkers were highly dynamic. KCNE1 association
in both Q1Ea and Q1Eb conformations only modestly
perturbed the KCNQ1 backbone exibility. The most
sensitive region was the S5-P linker, whose backbone
was stabilized by KCNE1 in Q1Ea conformation during
MDS#1 (Fig. 7 A), and in both Q1Ea and Q1Eb during
MDS#2 (see Fig. S8 Ba).
The pattern of KCNQ1 contacts with KCNE1 during
MDS#1 was similar between Q1Ea and Q1Eb (Fig. 7 B).
In the pore domain, the S5-P and P-S6 linkers, the beginning
of the S5 helix (intracellular), and the beginning of the S6
helix (extracellular) made frequent contacts with KCNE1.
In the voltage-sensing domain, the following KCNQ1
regions made various degrees of contacts with KCNE1: S1
helix, S1-S2 linker, S3-S4 linker, and the intracellular end
of S4 helix. The pattern of KCNQ1 contacts with KCNE1
observed during MDS#2 was similar, except that the
KCNQ1 S2-S3 linker also made a small degree of contact
with KCNE1 (Fig. S8 Bb).
To probe how KCNE1 association perturbed the KCNQ1
backbone conformation, we applied principal component
analysis to combined MD trajectories: the equilibrium
phase of MD trajectory of Q1 alone was combined with
that of Q1Ea or Q1Eb (designated as Q1-to-Q1Ea and
Q1-to-Q1Eb trajectories) (2932). Correlated molecular
motions in the combined trajectories were analyzed by
covariance matrices. Diagonalizing the covariance matrices
decomposed the molecular motions into different principal
components, based on their eigenvectors (describing the
directions of motions) and corresponding eigenvalues
(describing the magnitudes of motions). The principal com-
ponents were ranked by their eigenvalues in descend-
ing order, so that the rst principal component had the
largest magnitude of motions. In our analysis, the rst
principal components contributed to 80 and 70% of the
A
B
C
D
FIGURE 6 Analysis of KCNE1 interaction with
KCNQ1 based on MDS#1. (A) RMSFs of KCNE1
C
a
atoms when alone or associated with KCNQ1 in
Q1Ea or Q1Eb conformation. (B) Degree of contact
with KCNQ1 in Q1Ea or Q1Eb conformation.
Contact was dened as C
a
-C
a
distance % 9 A
.
For each of the KCNE1 positions (171), the frac-
tions of 5000 poses during MD trajectories in
which contacts occurred between the KCNE1 C
a
atom and any of the KCNQ1 C
a
atoms were
summed and dened as degree of contact with
Q1 (A.U., arbitrary unit). KCNE1 domains are
marked (top), with helical regions (NT-helix: aa
1123, TM-helix: aa 4471) highlighted (gray
shading). (Dotted vertical lines) Transmembrane
positions that showed local peaks of C
a
RMSF
(A), or degree of contact with Q1 (B). (C)
KCNQ1 regions with which each of the KCNE1
positions was making contact in Q1Ea (top) or
Q1Eb (bottom) conformation. The KCNE1
sequence is listed above. (Shown on left) KCNQ1
was divided into 13 color-coded regions. (D)
Closeup views of key KCNE1/KCNQ1 interactions
in Q1Ea or Q1Eb conformation.
2468 Xu et al.
molecular motions observed in Q1-to-Q1Ea and Q1-to-
Q1Eb trajectories, respectively (see Fig. S9 A). Projection
of MD trajectories along eigenvectors 13 showed that
although the path of motions in Q1 alone and that of Q1Ea
or Q1Eb were well separated along eigenvector 1 (see
Fig. S9 B), they were not separated along eigenvectors 2 or
3 (see Fig. S9 C). This analysis suggested that the rst prin-
cipal components of the combined MD trajectories likely
represented the displacements of KCNQ1 backbone induced
by KCNE1 association in Q1Ea and Q1Eb conformations.
Fig. S10 shows that KCNE1-induced KCNQ1 C
a
displacements were asymmetrical among the four KCNQ1
subunits. The patterns varied between Q1Ea and Q1Eb as
well as between MDS#1 and MDS#2. We averaged the
KCNQ1 C
a
displacements over the four KCNQ1 subunits
in both Q1Ea and Q1Eb conformations based on MDS#1.
The average KCNQ1 backbone displacements displayed
a distinct pattern (Fig. 7 C), which tracks the pattern of
KCNQ1 C
a
RMSF (Fig. 7 A) but not the pattern of
KCNQ1 contacts with KCNE1 (Fig. 7 B) during MD trajec-
tories. These comparisons indicate that KCNE1 can modu-
late KCNQ1 channel function by allosteric mechanisms:
conformational changes resulting from KCNQ1/KCNE1
contacts were transmitted to other KCNQ1 areas that
directly controlled its channel function. For example,
despite a lack of direct contacts, KCNE1 docking caused
conformational changes in the KCNQ1 pore loop around
T312 (Thr of the selectivity lter, TIGYG) and around the
S6 hinge (L342/P343) (35). These secondary conforma-
tional changes are likely to contribute to KCNE1 effects
on KCNQ1 pore conductance, and the probability of
opening of the activation gate (S6
CT
). Movie S1 (in the
Supporting Material) depicts the KCNQ1 backbone dis-
placements induced by KCNE1 in the Q1Ea and Q1Eb
FIGURE 7 Analysis of KCNQ1 interaction with
KCNE1 based on MDS#1. (A) RMSF of KCNQ1
C
a
atoms (averaged over chains AD) when
KCNQ1 was alone, or associated with KCNE1 in
Q1Ea or Q1Eb conformation. (B) Degree of contact
with KCNE1 indexed by KCNQ1 position numbers
(107359) in Q1Ea or Q1Eb conformation. The
calculation of degree of contact was similar to that
described for Fig. 6 B. (C) KCNQ1 C
a
displace-
ments related to KCNE1 association deduced from
principal component analysis (see Fig. S9). The
KCNQ1 C
a
displacement values of principal
component 1 were averaged over chains AD of
both Q1Ea and Q1Eb conformations and plotted as
mean 5 SE against KCNQ1 position numbers.
Selected residues and position numbers are marked
for local peaks. (D) Views of putative KCNE1-
induced motions in KCNQ1 peptide backbone in
the Q1Ea conformation. (Gray traces) KCNQ1
backbones; (tubes) regions of interest, extramem-
brane linkers, and S6
CT
. (Cyan tubes) C
a
atom posi-
tions of Q1 alone; (red tubes) Q1/Ea linked by
principal component 1. The magnitudes and direc-
tions of KCNQ1 motions in the regions of interest
are signied by lines (cyan, green, red) connecting
corresponding C
a
atoms of the two tubes. The four
Q1 subunits are marked as AD (peripheral
voltage-sensing domains) or ad (central pore
domain). (Yellow and magenta triangles) Putative
direction of forces exerted by KCNE1 on KCNQ1
on the extracellular (solid triangles) or intracellular
(open triangles) side of the membrane.
Ks
Channel 2469
conformations during MDS#1. The functional implication
related to Fig. 7 D (enlarged views in Fig. S11) is discussed
below.
Functional relevance of novel KCNQ1/KCNE1 interactions
revealed by MD trajectories
Interactions between KCNQ1 S5-P linker and KCNE1-NT
stabilize a pre-open activated state of the KCNQ1/KCNE1
channel. Both KCNQ1 S5-P linker and KCNE1-NT are
extracellular loops, far from the voltage-sensor (S4) and
the activation gate (S6
CT
). Therefore, their interactions are
not expected to directly impact on the activation gating
process. However, our models predicted that KCNE1/
KCNQ1 interactions in the extracellular region could
perturb a network of salt-bridges and hydrogen bonds
among KCNQ1 residues within the voltage-sensing domain
(VSD) and between VSD and the pore domain (PD) (see
Table S1 and Fig. S12). These interactions involve gating
charges on the S4 (R228 and R231), and thus should impact
on the stability of the channel in activated versus deactivated
states. If this was the case, then disulde formation between
the KCNQ1 S5-P linker and KCNE1-NT might be gating-
state-dependent. To test this possibility, we compared the
degree of disulde formation between two conditions:
during H
2
O
2
incubation COS-7 cells were treated with
100 mM [K]
o
for 10 min (depolarization to 0 mV, favoring
channel activation) or not ([K]
o
5 mM). Fig. 5 C shows
that depolarization reduced disulde formation in the
following pairs: Q1*-E290C/E1-R32C, Q1*-R293C/E1-
E19C, Q1*-E295C/E1-R32C, and Q1*-E295C/E1-R33C
(two independent experiments with similar results). Because
these disulde formations were predicted by our KCNQ1/
KCNE1 docking conformations built on the activated-state
KCNQ1 homology model (Fig. 3 A), we suggest that the
docking conformations represent a pre-open activated state
of the channel, which could be destabilized by prolonged
depolarization. Depolarization did not alter the degree of
disulde formation in the following pairs: Q1*-E284C/
E1-R32C and R33C, Q1*-D286C/E1-R32C and R33C.
We used oocyte voltage-clamp to probe further. Disulde
formation should stabilize the gating-state in which it was
formed. If disulde occurred in a pre-open activated state,
then H
2
O
2
by promoting disulde formation should lead
to a positive shift in the voltage-dependence of activation.
Fig. 8 shows that this was indeed the case with Q1*-
E290C/E1-R32C or -R33C, Q1*-E295C/E1-R32C or
-R33C. As a comparison, H
2
O
2
treatment also right-shifted
V
0.5
of activation of Q1*-E284C/E1-R33C and Q1*-D286C/
E1-R32C. However, the degree of shift was much smaller
than the above four pairs. We could not test the effects of
H
2
O
2
on Q1*-R293C/E1-E19C, because H
2
O
2
treatment
shifted V
0.5
of activation in Q1*-WT/E1-E19C, suggesting
that the Cys side chain at KCNE1 position 19 could form
disulde bond(s) with unidentied partner(s) in the oocyte
cell membrane.
Side-chain properties at KCNE1 position 46 impact on
ion conduction through the KCNQ1/KCNE1 pore. Previ-
ously we showed that Cys substitution of KCNE1 Y46
signicantly increased the Rb:K conductance ratio (G
Rb
/
G
K
) from 0.74 5 0.03 (KCNQ1*/KCNE1-WT) to 2.02 5
0.09 (KCNQ1*/KCNE1-Y46C) (28). This was unique
among the 25 positions in the KCNE1 TMD and NT loop
tested, although the mechanism was not clear. Our docking
models suggested that KCNE1 Y46 was packed against
KCNQ1 S1-S2, S5-P, and P-S6 linkers (Fig. 9 A). Analysis
of MD trajectories revealed dynamic interactions between
KCNE1 Y46 and KCNQ1 side chains or peptide backbone
in these linker regions (see Table S2). These observations
suggested that the bulky Y46 side chain might exert steric
pressure on the KCNQ1 pore loop to restrict the conduc-
tance to Rb
versus K
ions (ionic radii 1.48 and 1.33 A
).
Substituting Y46 with much smaller Cys side chain relieved
the pressure, thus increasing G
Rb
/G
K
.
To check this possibility, we mutated Y46 to the smallest
residue (Y46G), a similar aromatic residue (Y46F), or a
bulky aromatic residue (Y46W), and tested their effects on
FIGURE 8 Effects of H
2
O
2
on voltage-dependence of activation of Cys-
substituted disulde-forming KCNQ1/KCNE1 pairs. Oocytes expressing
specied constructs were voltage-clamped under control conditions and
then after treatment with H
2
O
2
(0.1%, 1015 min) using pulse protocols
similar to that described for Fig. 2 Bb. Isochronal (2-s) activation curves
were constructed by tting the relationship between peak tail current
amplitude (I
tail
) and test pulse voltage (V
t
) with a simple Boltzmann
function: I
tail
I
max
/(1 exp[(V
0.5
V
t
)/k]) to estimate the maximal I
tail
value (I
max
), half-maximum activation voltage (V
0.5
), and slope factor (k).
Fraction activated I
tail
/I
max
; DV
0.5
H
2
O
2
-induced shift in V
0.5
value;
(n) number of oocytes studied.
2470 Xu et al.
Cs:K conductance ratio (G
Cs
/G
K
, Cs
ionic radius 1.69 A
).
Fig. 9 B shows that KCNQ1* expressed alone has a G
Cs
/G
K
value of 2.34 50.15, and KCNE1 association reduces G
Cs
/
G
K
to 0.11 5 0.02. Whereas Y46F and Y46W resembled
KCNE1-WT in reducing the G
Cs
/G
K
value, Y46G and
Y46C signicantly increased the G
Cs
/G
K
value. Surpris-
ingly, making the Y46C side chain bulkier but posi-
tively charged (2-(TRIMETHYLAMMONIUM)ETHYL]
METHANETHIOSULFONATE, MTSET modication)
further increased the G
Cs
/G
K
value. We suggest that the
KCNQ1/KCNE1 pore conductance to the bulky Cs
relative
to K
ions could be increased by relieving the steric pres-

sure on the channel pore by removing the aromatic ring at
KCNE1 position 46. It could be further increased by
creating an aqueous cleft around side-chain 46 by making
it positively charged.
DISCUSSION
Insights into the structure-dynamics-function
relationship of the I
Ks
channel
How does KCNE1 slow KCNQ1 activation?
There has been a debate as to how KCNE1 slows KCNQ1
activation: whether by weakening the coupling between
the activation gate (S6
CT
) and voltage-sensor movement
(via the S4-S5 linker), by slowing S4 outward movement,
or by a combination of both (20,3638)? Three observations
in our study offer insights into this issue:
1. Principal component analysis of combined Q1-to-Q1Ea
trajectories revealed that KCNE1 induced correlated
motions in the S4-S5 linker and S6
CT
(Fig. 7 D, enlarged
views in Fig. S11): KCNE1 TMD pushed the KCNQ1
S4-S5 linker of subunit B toward the pore center, which
caused the S6
CT
of subunit A to bend at L342/P343. As a
result, S6
CT
of subunit A moved to reduce the opening of
intracellular pore entrance, i.e., causing the closure of the
activation gate. Closure of the activation gate was also
observed in Q1Eb during MD trajectory but not in Q1
alone (see Fig. S13).
2. As shown in Table S1, in Q1Eb during MDS#2 (0 mV)
the gating charge on S4, R237, ipped from interacting
with E160 on S2 (above the F167 hydrophobic seal,
Fig. 3 A), to interacting with E170 below F167. This
did not occur during MDS#1 (435 mV). We suggest
that at 0 mV transmembrane voltage the S4 was dynamic
enough to allow a downward transfer of R237 across the
hydrophobic seal in the presence of KCNE1. This is
consistent with the KCNE1-mediated remodeling of
KCNQ1 voltage sensor observed experimentally (39,40).
3. Oocyte voltage-clamp experiments, in conjunction with
COS-7 disulde-trapping experiments, suggested that a
FIGURE 9 Side-chain properties at KCNE1 po-
sition 46 inuence ion-conduction through the
KCNQ1/KCNE1 pore. (A) (Top left) Side view of
KCNQ1/KCNE1 docking conformation, high-
lighting KCNE1 (cyan ribbon) and Y46 side chain
(sphere-lled model), KCNQ1 S1/S1-S2 linker
(blue ribbon/loop), S5-P linker (magenta loop),
and P-S6 linker (brown loop). The rest of KCNQ1
is shown as gray ribbons. (Top right) Top view of
the boxed region, with the four selectivity lters
(SF) lining the pore marked. (Lower) Closeup
views of proximity between Y46 and selected
side chains on S1-S2, S5-P, and P-S6 linkers. (B)
Cs:K conductance ratio (G
Cs
/G
K
) of KCNQ1*
expressed in oocytes alone, or coexpressed with
specied KCNE1 variants: WT or Y46 substituted
by Phe, Trp, Gly, or Cys. For Y46C, the value
after 2-(TRIMETHYLAMMONIUM)ETHYL]
METHANETHIOSULFONATE (MTSET) modi-
cation is included. The method of quantifying G
Cs
/
G
K
has been described in Wang et al. (28).
Ks
Channel 2471
network of electrostatic interactions between KCNQ1
S5-P linker and KCNE1-NT stabilized the channel in a
closed state. Because this network of interactions was
predicted by our Q1Ea and Q1Eb models whose S4
voltage sensors were in the UP or activated state
(Fig. 3 A), we suggest that these docking conformations
represent a pre-open activated state of the KCNQ1/
KCNE1 channel.
How does KCNE1 modify KCNQ1 pore conductance?
We did not detect any direct contact between the KCNE1
and the KCNQ1 pore loop (Fig. 7 B). However, principal
component analysis revealed that KCNE1 induced
KCNQ1 backbone displacement around T312 (part of the
selectivity lter), whose backbone carbonyl oxygen and
side-chain hydroxyl oxygen are expected to contribute to
K
ion coordination in the pore (41). Furthermore, reducing

the volume or adding a positive charge to the side chain
at KCNE1 position 46 markedly increased the pore con-
ductance to bulky Cs
ions relative to K
ions. These
observations suggest that KCNE1 allosterically inuences
the backbone conformation of the KCNQ1 selectivity
lter. It also exerts steric pressure on the KCNQ1 pore
domain. These effects combined lead to an optimization
of the pore conductance to K
ions, while limiting the

conductance to bulkier Rb
and Cs
ions. They may also

restrict the backbone dynamics required for KCNQ1
inactivation (42).
Comparison with previous studies
KCNQ1 homology models (43), docking of KCNE1 TMD
to KCNQ1 homology models (15), and MD simulations
of a KCNQ1/KCNE1-TMD docking model (27) have been
reported before. Our work differs from the previous studies
in four major aspects:
1. We did not assume that KCNE1 position 58 is close to
KCNQ1 positions 338340. This distinction matters
because it inuences how the KCNE TMD is oriented
with respect to the KCNQ1 VSD and the PD.
2. Our models included not only the KCNE1 TMD, but also
the extracellular NT, whose interactions with KCNQ1
S5-P linker may stabilize the I
Ks
channel in a pre-open,
activated state.
3. Our models were validated by disulde experiments,
conrming model-predicted novel KCNQ1/KCNE1 in-
teractions.
4. We performed extensive MD simulations followed by
detailed analysis of the MD trajectories, to gain new in-
sights into the structure-dynamics-function relationship
of the I
Ks
channel.
A more recent study used elastic network models to
predict C
a
backbone conformational dynamics in KCNQ1
and KCNQ1/KCNE1 (44). These methods assume that
major conformational changes in proteins can be predicted
by C
a
atoms as nodes connected by elastic springs with
identical spring constants. Thus, side-chain interactions
were not considered. Instead, our detailed MD simulations
and analyses revealed that side-chain interactions play a
major role in determining how KCNQ1 and KCNE1 interact
in their extracellular domains, which has impact on the
voltage-dependence of I
Ks
channel activation.
Study limitations
The intracellular domains of KCNE1 and KCNQ1 were not
included in the models. The lack of disulde formation
between KCNE1-T58C and KCNQ1 S338C/F339C/F340C
could be due to our experimental conditions; thus we cannot
denitely rule out their interactions. Our models were
incorporated into POPC-membrane without PIP
2
(phospha-
tidylinositol 4,5-bisphosphate), which could have affected
the I
Ks
channel conformations (45).
CONCLUSION
We conclude that KCNE1 can be docked to KCNQ1 in
two possible conformations and a network of electrostatic
interactions between their extracellular domains stabilizes
the channel in a pre-open, activated state. Analysis of the
MD trajectories also provides insights into how KCNE1
transforms KCNQ1 into the unique, slowly activating I
Ks
channel.
SUPPORTING MATERIAL
Three tables, thirteen (13) gures, one movie and additional supple-
mental information are available at http://www.biophysj.org/biophysj/
supplemental/S0006-3495(13)01188-0.
We thank Drs. W. R. Kobertz and T. Morin for sharing unpublished data.
This study was supported by grant No. HL107294 from the National Heart,
Lung and Blood Institute/National Institutes of Health (to G.-N.T.), and
grant No. S10RR027411 from National Center for Research Resources/
National Institutes of Health (to M.C.).
REFERENCES
1. Barhanin, J., F. Lesage, ., G. Romey. 1996. KvLQT1 and I
sK
(minK)
proteins associate to form the I
Ks
cardiac potassium current. Nature.
384:7880.
2. Sanguinetti, M. C., M. E. Curran, ., M. T. Keating. 1996. Coassembly
of KvLQT1 and minK (I
sK
) proteins to form cardiac I
Ks
potassium
channel. Nature. 384:8083.
3. Wang, Z., B. Fermini, and S. Nattel. 1994. Rapid and slow components
of delayed rectier current in human atrial myocytes. Cardiovasc. Res.
28:15401546.
4. Li, G.-R., J. Feng, ., S. Nattel. 1996. Evidence for two components of
delayed rectier K
current in human ventricular myocytes. Circ. Res.

78:689696.
2472 Xu et al.
5. Splawski, I., J. Shen, ., M. T. Keating. 2000. Spectrumof mutations in
long-QT syndrome genes. KVLQT1, HERG, SCN5A, KCNE1, and
KCNE2. Circulation. 102:11781185.
6. Chen, Y.-H., S.-J. Xu, ., W. Huang. 2003. KCNQ1 gain-of-function
mutation in familial atrial brillation. Science. 299:251254.
7. Hong, K., D. R. Piper, ., R. Brugada. 2005. De novo KCNQ1 muta-
tion responsible for atrial brillation and short QT syndrome in utero.
Cardiovasc. Res. 68:433440.
8. Olesen, M. S., B. H. Bentzen, ., N. Schmitt. 2012. Mutations in the
potassium channel subunit KCNE1 are associated with early-onset
familial atrial brillation. BMC Med. Genet. 13:2432.
9. Xiong, Q., Z. Gao, ., M. Li. 2008. Activation of Kv7 (KCNQ)
voltage-gated potassium channels by synthetic compounds. Trends
Pharmacol. Sci. 29:99107.
10. Long, S. B., X. Tao, ., R. MacKinnon. 2007. Atomic structure of a
voltage-dependent K
channel in a lipid membrane-like environment.

Nature. 450:376382.
11. Lee, S.-Y., A. Banerjee, and R. MacKinnon. 2009. Two separate inter-
faces between the voltage sensor and pore are required for the function
of voltage-dependent K
channels. PLoS Biol. 7:e47.

12. Wang, Y.-H., M. Jiang, ., G.-N. Tseng. 2011. Gating-related molecu-
lar motions in the extracellular domain of the I
Ks
channel: implications
for I
Ks
channelopathy. J. Membr. Biol. 239:137156.
13. Bezanilla, F. 2008. How membrane proteins sense voltage. Nat. Rev.
14. Tao, X., A. Lee, ., R. MacKinnon. 2010. A gating charge transfer
center in voltage sensors. Science. 328:6773.
15. Kang, C., C. Tian, ., C. R. Sanders. 2008. Structure of KCNE1 and
implications for how it modulates the KCNQ1 potassium channel.
Biochemistry. 47:79998006.
16. Coey, A. T., I. D. Sahu, ., G. A. Lorigan. 2011. Reconstitution of
KCNE1 into lipid bilayers: comparing the structural, dynamic, and
activity differences in micelle and vesicle environments. Biochemistry.
50:1085110859.
17. Haitin, Y., and B. Attali. 2008. The C-terminus of Kv7 channels: a
multifunctional module. J. Physiol. 586:18031810.
18. Lvov, A., S. D. Gage, ., W. R. Kobertz. 2010. Identication of a
protein-protein interaction between KCNE1 and the activation gate
machinery of KCNQ1. J. Gen. Physiol. 135:607618.
19. Tapper, A. R., and A. L. George, Jr. 2001. Location and orientation
of minK within the I
Ks
potassium channel complex. J. Biol. Chem.
276:3824938254.
20. Nakajo, K., and Y. Kubo. 2007. KCNE1 and KCNE3 stabilize and/or
slow voltage sensing S4 segment of KCNQ1 channel. J. Gen. Physiol.
130:269281.
21. Xu, X.-L., M. Jiang, ., G.-N. Tseng. 2008. KCNQ1 and KCNE1 in
the I
Ks
channel complex make state-dependent contacts in their extra-
cellular domains. J. Gen. Physiol. 131:589603.
22. Chung, D. Y., P. J. Chan, ., R. S. Kass. 2009. Location of KCNE1
relative to KCNQ1 in the I
Ks
potassium channel by disulde cross-link-
ing of substituted cysteines. Proc. Natl. Acad. Sci. USA. 106:743748.
23. Chan, P. J., J. D. Osteen, ., R. S. Kass. 2012. Characterization of
KCNQ1 atrial brillation mutations reveals distinct dependence on
KCNE1. J. Gen. Physiol. 139:135144.
24. Morin, T. J., and W. R. Kobertz. 2008. Counting membrane-embedded
KCNE b-subunits in functioning K
channel complexes. Proc. Natl.

Acad. Sci. USA. 105:14781482.
25. Melman, Y. F., A. Krumerman, and T. V. McDonald. 2002. A single
transmembrane site in the KCNE-encoded proteins controls the speci-
city of KvLQT1 channel gating. J. Biol. Chem. 277:2518725194.
26. Panaghie, G., K. K. Tai, and G. W. Abbott. 2006. Interaction of KCNE
subunits with the KCNQ1 K
channel pore. J. Physiol. 570:455467.

27. Strutz-Seebohm, N., M. Pusch, ., G. Seebohm. 2011. Structural basis
of slow activation gating in the cardiac I
Ks
channel complex. Cell.
Physiol. Biochem. 27:443452.
28. Wang, Y.-H., M. Zhang, ., G.-N. Tseng. 2012. Probing the structural
basis for differential KCNQ1 modulation by KCNE1 and KCNE2.
J. Gen. Physiol. 140:653669.
29. Marni, F., S. Wu, ., L. Zhou. 2012. Normal-mode-analysis-guided
investigation of crucial intersubunit contacts in the cAMP-dependent
gating in HCN channels. Biophys. J. 103:1928.
30. Amadei, A., A. B. M. Linssen, and H. J. C. Berendsen. 1993. Essential
dynamics of proteins. Proteins. 17:412425.
31. Ng, H. W., C. A. Laughton, and S. W. Doughty. 2013. Molecular dy-
namics simulations of the adenosine A2a receptor: structural stability,
sampling, and convergence. J. Chem. Inf. Model. 53:11681178.
32. Zhou, L., and S. A. Siegelbaum. 2007. Gating of HCN channels by
cyclic nucleotides: residue contacts that underlie ligand binding, selec-
tivity, and efcacy. Structure. 15:655670.
33. Tapper, A. R., and A. L. George, Jr. 2000. MinK subdomains that
mediate modulation of and association with KvLQT1. J. Gen. Physiol.
116:379390.
34. Shrivastava, I. H., and I. Bahar. 2006. Common mechanism of pore
opening shared by ve different potassium channels. Biophys. J.
90:39293940.
35. Seebohm, G., N. Strutz-Seebohm, ., F. Lang. 2006. Differential roles
of S6 domain hinges in the gating of KCNQ potassium channels.
Biophys. J. 90:22352244.
36. Rocheleau, J. M., and W. R. Kobertz. 2008. KCNE peptides differ-
ently affect voltage sensor equilibrium and equilibration rates in
KCNQ1 K
channels. J. Gen. Physiol. 131:5968.

37. Osteen, J. D., C. Gonzalez, ., R. S. Kass. 2010. KCNE1 alters the
voltage sensor movements necessary to open the KCNQ1 channel
gate. Proc. Natl. Acad. Sci. USA. 107:2271022715.
38. Ruscic, K. J., F. Miceli, ., S. A. N. Goldstein. 2013. I
Ks
channels open
slowly because KCNE1 accessory subunits slow the movement of S4
voltage sensors in KCNQ1 pore-forming subunits. Proc. Natl. Acad.
Sci. USA. 110:E559E566.
39. Wu, D., K. Delaloye, ., J. Cui. 2010. State-dependent electrostatic
interactions of S4 arginines with E1 in S2 during Kv7.1 activation.
J. Gen. Physiol. 135:595606.
40. Wu, D., H. Pan, ., J. Cui. 2010. KCNE1 remodels the voltage sensor
of Kv7.1 to modulate channel function. Biophys. J. 99:35993608.
41. Zhou, Y., J. H. Morais-Cabral, ., R. MacKinnon. 2001. Chemistry of
ion coordination and hydration revealed by a K
channel-Fab complex
at 2.0 A
resolution. Nature. 414:4348.

42. Seebohm, G., M. C. Sanguinetti, and M. Pusch. 2003. Tight coupling of
rubidium conductance and inactivation in human KCNQ1 potassium
channels. J. Physiol. 552:369378.
43. Smith, J. A., C. G. Vanoye, ., C. R. Sanders. 2007. Structural models
for the KCNQ1 voltage-gated potassium channel. Biochemistry.
46:1414114152.
44. Gofman, Y., S. Shats, ., N. Ben-Tal. 2012. How does KCNE1 regulate
the Kv7.1 potassium channel? Model-structure, mutations, and dy-
namics of the Kv7.1-KCNE1 complex. Structure. 20:13431352.
45. Zaydman, M. A., J. R. Silva, ., J. Cui. 2013. Kv7.1 ion channels
require a lipid to couple voltage sensing to pore opening. Proc. Natl.
Acad. Sci. USA. 110:1318013185.
Ks
Channel 2473
Two-Photon Excitation STED Microscopy in Two Colors in Acute Brain
Slices
Philipp Bethge,
Ronan Che reau,
Elena Avignone,
Giovanni Marsicano,
and U. Valentin Na gerl
Interdisciplinary Institute for Neuroscience, Universite de Bordeaux, Bordeaux, France;

UMR 5297, Centre National de la Recherche
Scientique, Bordeaux, France; and

Universite de Bordeaux, INSERM U862 NeuroCentre Magendie, Bordeaux, France
ABSTRACT Many cellular structures and organelles are too small to be properly resolved by conventional light microscopy.
This is particularly true for dendritic spines and glial processes, which are very small, dynamic, and embedded in dense tissue,
making it difcult to image them under realistic experimental conditions. Two-photon microscopy is currently the method of
choice for imaging in thick living tissue preparations, both in acute brain slices and in vivo. However, the spatial resolution
of a two-photon microscope, which is limited to ~350 nm by the diffraction of light, is not sufcient for resolving many important
details of neural morphology, such as the width of spine necks or thin glial processes. Recently developed superresolution
approaches, such as stimulated emission depletion microscopy, have set new standards of optical resolution in imaging
living tissue. However, the important goal of superresolution imaging with signicant subdiffraction resolution has not yet
been accomplished in acute brain slices. To overcome this limitation, we have developed a new microscope based on
two-photon excitation and pulsed stimulated emission depletion microscopy, which provides unprecedented spatial resolution
and excellent experimental access in acute brain slices using a long-working distance objective. The new microscope
improves on the spatial resolution of a regular two-photon microscope by a factor of four to six, and it is compatible with
time-lapse and simultaneous two-color superresolution imaging in living cells. We demonstrate the potential of this nanoscopy
approach for brain slice physiology by imaging the morphology of dendritic spines and microglial cells well below the surface
of acute brain slices.
INTRODUCTION
Most neurons in the mammalian brain are studded with
hundreds to thousands of dendritic spines (1). Spines are
tiny specializations of the postsynaptic membrane that are
packed with receptors, ion channels, and other signaling
complexes and mediate fast excitatory synaptic transmis-
sion in the brain. Their structural dynamics are thought to
be critical for brain development and experience-dependent
synaptic plasticity throughout life (2).
Thus, quantitative measurements of the structure and
function of spines have been a primary endeavor in neuro-
science research for a long time. Getting reasonable optical
access is a huge challenge, because they are embedded in
dense brain tissue and are oftentimes smaller than what
can be resolved by diffraction-limited light microscopy,
including two-photon (2P) microscopy.
Moreover, recent studies indicate that spines also interact
directly with surrounding glial cells, like astrocytes and
microglia (3,4), which extend extremely thin processes.
Hence, to study how spines operate and interact with their
partners in a complex and dynamic environment, it is essen-
tial to image several cell types at the same time with suf-
cient spatial resolution below the diffraction limit. In
recent years, several big steps have been made in developing
methods to make this possible.
Starting a new era in optical microscopy, stimulated emis-
sion depletion (STED) microscopy (57) has allowed for
nanoscale imaging of dynamic events in living tissue,
imaging synaptic vesicles (8) and spine morphology (9).
Furthermore, several strategies were recently developed
for multicolor STED imaging (1013).
Most STED studies to date have been done in culture
systems using oil objectives (e.g., Westphal et al. (8), Nagerl
et al. (9), and Willig et al. (14)), but the experimental prep-
aration of choice for synaptic physiologists is acute brain
slices. To record healthy cells in acute brain slices, it is
necessary to image a few tens of microns deep in the tissue,
which cannot be done with oil objectives. Although glycerol
objectives can substantially extend optical penetration with-
out compromising spatial resolution (15), they are not com-
patible with acute brain slices, which call for an upright
microscope design and water-immersion objectives with
long working distances to accommodate electrophysiolog-
ical recording electrodes.
The combination of 2P excitation and STED (2P-STED)
microscopy (16,17) opened new perspectives for superre-
solution imaging in deep tissue. However, the use of contin-
uous wave (CW) STED lasers limited the spatial resolution
to 250 nm in living tissue (16).
Submitted August 27, 2012, and accepted for publication December 11,
2012.
*Correspondence: valentin.nagerl@u-bordeaux2.fr
This is an Open Access article distributed under the terms of the Creative
Commons-Attribution Noncommercial License (http://creativecommons.
org/licenses/by-nc/2.0/), which permits unrestricted noncommercial use,
distribution, and reproduction in any medium, provided the original work
is properly cited.
Editor: Paul Wiseman.
778 Biophysical Journal Volume 104 February 2013 778785
In this study, our goal was to overcome these limitations
and to achieve true subdiffraction imaging in acute brain
slices a few cell layers below the tissue surface, which corre-
sponds to a depth usually targeted in patch-clamp electro-
physiology experiments. To this end, we developed a new
2P-STED microscope incorporating a pulsed STED laser,
a long-working distance water objective and spectral detec-
tion for two-color imaging.
We characterize the 2P-STED microscope and illustrate
its potential by imaging dendritic spines in acute brain
slices. The spatial resolution in the focal plane of the new
microscope is four to six times better than that of a regular,
diffraction-limited 2P microscope. Furthermore, we show
that the microscope is compatible with time-lapse and
two-color superresolution imaging, using transgenic mice
in which neurons and microglia are labeled with yellow
(YFP) and green uorescent protein (GFP), respectively.
Lasers
A femtosecond mode-locked Ti:Sapphire laser (Mai Tai, Spectra-Physics,
Darmstadt, Germany) operating at ~80 MHz and a wavelength of 797 nm
was used in combination with an optical parametric oscillator (OPO BASIC
Ring fs, APE, Berlin, Germany) to produce pulses at a wavelength of
592 nm (STED laser). The pulses of originally ~200-fs duration were
stretched to >68 ps by passing them through a 25-cm dispersive glass
rod (high-refractive-index int glass) and a 20-m-long polarization-
preserving ber (Schafter & Kirchhoff, Hamburg, Germany). A reection
from the STED laser was used to synchronize a second mode-locked ultra-
fast Ti:Sapphire laser (Tsunami, Spectra-Physics) operating at 910 nm for
2P excitation (2P laser).
Synchronization and ne pulse delay (<2 ns) was performed via phase-
locked loop electronics (3930, Lok-to-Clock, Spectra-Physics), while the
coarse delay was set by varying the length of the BNC cable, using a fast
photodiode (3932-LX, Spectra-Physics) placed below the microscope
objective for readout.
Laser intensities were controlled via dedicated electro-optical modula-
tors (Conoptics, Danbury, CT) for the 2P and STED laser beams. The
time-averaged power at the back focal plane (BFP) of the objective was
between 15 and 25 mW for the 2P light and between 20 and 40 mW for
the STED light, depending on imaging depth and sample brightness.
Microscope setup
The microscope was built around a standard commercial research micro-
scope (BX51WI, Olympus, Hamburg, Germany) using scan and tube lenses
from the microscope manufacturer. The telecentric scanner (Yanus IV,
TILL Photonics, Grafelng, Germany) was placed so that both scan axes
are projected into the BFP of the objective, ensuring that the 2P and
STED laser beams stay stationary at the BFP during scanning.
A water-immersion objective with a long working distance (1.5 mm)
and equipped with a correction collar was used for all experiments
(60X LUMFI, 1.1 NA, Olympus). The correction collar was adjusted
to optimize the STED doughnut using gold nanospheres (diameter
150 nm; BBInternational, Cardiff, United Kingdom) and slightly re-
adjusted for particular imaging depths, using the 2P uorescence signal
as readout. The z-position of the objective was controlled via a piezo actu-
ator (P-721 PIFOC, PI Physik Instrumente, Karlsruhe, Germany). Signal
detection and peripheral hardware were controlled by the Imspector scan-
ning software (18) via data acquisition cards (6259 M, 2090A, National
Instruments, Austin, TX).
A polymeric phase plate (RPC Photonics, Rochester, NY) was used to
create the STED doughnut and a bandpass lter (593/40, AHF Analysen-
technik, Tubingen, Germany) was used to spectrally clean up the STED
laser beam. The 2P and STED laser beams were combined using a dichroic
mirror (F73-700UV, AHF) before the scanner.
The uorescence signal was detected in descanned mode and separated
from the excitation and STED light by a longpass dichroic (580 DCXRUV,
AHF). The detectors were protected from the 2P and STED light by suitable
blocking and emission lters (680SP-25, 594S-25, 520-70, Semrock,
Rochester, NY). The signal was spectrally divided into two channels by
a dichroic mirror (514RS, Semrock) before being focused onto multimode
bers (100-mm core diameter, which corresponds to 120% of the back-pro-
jected Airy disk), terminating on avalanche photodiodes (SPCM-AQR-13-
FC, PerkinElmer, Waltham, MA).
Microscope alignment
For spatial alignment of the 2P and STED lasers and quality control of
the doughnut (Fig. 1 D), a pellicle beam splitter (Thorlabs, Maisons-
Laftte, France) was ipped into the beam path so that the reection
from gold nanospheres could be detected by a photomultiplier tube
(MD963, PerkinElmer). A piezo-controlled motorized mirror (AG-
M100N, Newport, Beaune la Rolande, France) and a telescope were used
to align the doughnut on the excitation spot in all three dimensions.
Doughnut quality (shape and null) was optimized via achromatic l/2 and
l/4 wave plates (Qioptiq, Paris, France). Optical resolution (Fig. 2 D)
was assessed using uorescent nanospheres (Fluo Spheres, yellow-green,
diameter 0.04 mm, Invitrogen).
Animals and labeling
Twotransgenic mouse lines (C57BL/6background) were used: Thy1
eYFP/eYFP
mice that express YFP in a subpopulation of principal neurons in the
hippocampus as well as in layer 4/5 of cortex, and CX3CR1
eGFP/eGFP
mice
that express GFP in microglial cells (Jackson Labs, Bar Harbor, ME). In
experiments where only neurons were imaged, heterozygous Thy1
/eYFP
mice were used. In experiments where neurons and microglia were
imaged, mice obtained from the crossbreeding of the two mouse lines
were used (CX3CR1
/eGFP
; Thy1
/eYFP
). All experiments were carried
out in accordance with the National Code of Ethics on Animal Experi-
mentation (Carte Nationale dethique sur lexperimentation animale;
Ministe`re de lenseignement et de la recherche, Ministe`re de lagriculture
et de la peche) and approved by the Committee of Ethics of Bordeaux
(No. 3306001).
Acute brain slices
Animals 2140 days old were killed by cervical dislocation, and their
brains were quickly removed and placed in ice-cold sucrose-based articial
cerebrospinal uid (ACSF) containing (in mM) 210 sucrose, 10 glucose,
2 KCl, 26 NaHCO
3
, 1.25 NaH
2
PO
4
, 0.1 CaCl
2
, and 6 MgCl (pH 7.4,
osmolarity ~320 mOsm/L), which was bubbled with carbogen (95%
O
2
/5% CO
2
). Sagittal 350-mm-thick slices were cut using a vibratome
(VT1200, Leica, Mannheim, Germany) and transferred to a heated
(32
C) holding chamber with NaCl-based ACSF bubbled with carbogen,

which consisted of (in mM) 124 NaCl, 3 KCl, 26 NaHCO
3
,
1.25 NaH
2
PO
4
, 10 glucose, 2 CaCl
2
, 1 MgCl, and 0.6 Trolox (pH 7.4,
osmolarity ~305 mOsm/L) for 1 h. These slices were subsequently main-
tained at room temperature for a maximum of 4 h. For the imaging
experiments, the slices were transferred to a submerged recording chamber,
where they were continuously perfused (2.1 mL/min) with ACSF at room
temperature.
Superresolution Imaging in Acute Brain Slices 779
Image acquisition and analysis
All images were acquired with a pixel size of 19.5 nm (512 512 pixels,
10 mm 10 mm, except see Fig. 4 B3, which is 1024 1024 pixels,
40 mm 40 mm, with a pixel size of 39 nm) and a pixel dwell time of
30 ms, which corresponds to about 8 s acquisition time for a 10 mm
10 mm eld of view. Imaging depth into the slice was determined by the
piezo z-focus, the uorescence signal on top of the slice dening the zero
z-position. Image analysis was done on raw data using ImageJ. Images
presented in the gures were ltered by a 1-pixel Gaussian lter to reduce
noise. To quantify and compare the line proles from 2P and 2P-STED
images, we used the Lorentzian function:
y y
0
2A
p
G
4x x
c
2
G
2
;
where y
0
and A are constants, x
c
is the center, and G is the width of the
curve.
Two-color images are shown merged with and without linear unmixing
using a plugin for ImageJ as described before (12). Unless stated otherwise,
single frames from a z-stack (Dz typically 400 nm) are shown.
Statistics
Data are expressed as the mean 5 SE. A two-tailed unpaired t-test was
used to compare spine neck widths (G) measured for 2P and 2P-STED in
the CA1, cortex, and for the two groups pooled. Multiple comparisons
were post-hoc Bonferroni-corrected. A nonparametric Kolmogorov-Smir-
nov test was additionally used for groups containing <17 data points,
conrming the results of the parametric t-test. Data from 12 mice were
included in this study. Tests marked with (***) in Fig. 3 C are signicant,
with p < 10
12
.
RESULTS
Construction of the microscope
We built a 2P-STED microscope for two-color sub-
diffraction imaging in brain slices using a long-working
distance, water-immersion objective as outlined in Fig. 1,
AC. For coupling the lasers into the microscope, we
rst checked the reection of the 2P and STED light using
gold beads (Fig. 1 D). We optimized the symmetry and
central minimum of the STED doughnut by adjusting a
l/4 wave plate placed in front of the scanner. The
intensity of the STED beam at the doughnut center was
<1% of the intensity measured at the rim of the doughnut
(Fig. 1 D), which was indistinguishable from background
noise.
Performance of the microscope
Because we used pulsed lasers for 2P excitation and the
STED effect, the lasers had to be synchronized and their
pulses had to arrive at the sample with an optimal delay to
achieve efcient quenching of the 2P uorescence by the
STED laser. Indeed, quenching depended on how much of
the STED pulse overlaps with the time that the molecules
spent in the excited state after 2P excitation. Fig. 2, A
and B, illustrates that 80% quenching efciency in a sea
of dye uorescence (~1 mM calcein) can be achieved.
FIGURE 1 Principle and design of a pulsed
2P-STED microscope. (A) Simplied Jablonski
diagram of the molecular excitation states in 2P-
STED microscopy. The molecule is excited to the
excited state (S1) by two-photon absorption and re-
turns from there to the ground state (S0) by the
emission of uorescence. The incidence of STED
light quenches the uorescence and returns the
molecule to S0 before uorescence can occur
(curved dashed arrows show internal conversion).
(B) Two-photon excitation action cross section
and emission spectra for GFP and YFP (Warren
Zipfel, Cornell University, Ithaca, NY). Simulta-
neous quenching of GFP and YFP by a single laser
beam of 592 nm is possible because of the highly
overlapping tails of the emission spectra. Two-
photon excitation is performed at 910 nm. The
emission signal is spectrally separated and de-
tected in two channels. (C) Schematic of beam
path. The femtosecond pulsed Ti:Sa laser used
for 2P excitation is routed through a beam scanner
into an upright microscope and synchronized with
the STED laser (Ti:Sa/optical parametric oscillator
(OPO)). Femtosecond pulses emitted from the
Ti:Sa/OPO are broadened by a 20-m-long polariza-
tion-preserving single-mode ber. The doughnut is
formed by a helical phase mask. A long-working distance, water-immersion objective is used. It is equipped with a correction ring to correct spherical aber-
rations due to mismatches in refractive index at the lens-sample interface. l/2, half-wave plate; l/4, quarter-wave plate; DC, dichroic mirrors, NA, numerical
aperture; Dt, pulse broadening ber; xy-scan, scanner for x and y dimension; APD, avalanche photodiode; EOMelectro-optical modulator. (D) Reections
of the laser beams from gold particles used for visualization and spatial alignment of the excitation and STED beams. The laser beams are routed through
a pellicle beam splitter so that the reections can be detected by a photomultiplier tube. This allows for the characterization of the excitation and STED beams
and illustrates the doughnut-like intensity distribution of the STED laser. Scale bar, 500 nm.
780 Bethge et al.
Conversely, when the synchronization is turned off, the
quenching effect is strongly diminished (Fig. 2 B).
Fig. 2 C illustrates the uorescence quenching efciency
by the STED laser as a function of the relative delay
between the laser pulses, where zero refers to the optimal
delay. Changing the relative delay between pulses results
in a rapid increase of quenching to a maximum of 80%
before it decreases again gradually. Whereas the rising
phase of the curve should be mostly dened by the duration
of the STED pulse, the decay should be dened by the life-
time of the uorophore (19).
Based on the rising phase (tting a Gaussian error func-
tion), we estimated the duration of the STED pulse to be
at least 68.5 ps (full width at half maximum (FWHM))
using a 20-m-long polarization-preserving optical ber,
which is consistent with pulse durations of 200300 ps re-
ported for 120-m-long optical bers (20).
To check the performance of the 2P-STED micro-
scope, we imaged subdiffraction-sized uorescent beads
(diameter 40 nm) and determined the point-spread func-
tions (PSFs) for the 2P and 2P-STED case (Fig. 2 D). The
width of the PSF is an important indicator of the spatial
resolution of the microscope. Using a Lorentzian function
to t line proles drawn through the centers of individual
beads, we measured a PSF width of 368 nm for the 2P
case, which corresponds well to the expected spatial resolu-
tion of a diffraction-limited 2P microscope given the wave-
length of the 2P light (910 nm) and the NA of the objective
(1.1) (21). Imaging the same bead in STED mode, the PSF
width was 62 nm, which corresponds to about a sixfold
A
B C
D E
FIGURE 2 Fluorescence quenching and spatial resolution in 2P-STED microscopy. (A) Line scans in a uorescent solution excited by the 2P laser.
The signal is quenched by a pulsed STED laser. Desynchronization strongly attenuates this effect. (B) Quantication of the signal intensity along the rect-
angle indicated in A. Desynchronization greatly reduces quenching efciency and increases variability. Fluorescence quenching is 80% when the 2P and
STED pulses are synchronized and aligned in space and time. (C) 2P-STED requires a precise relative temporal delay of the synchronized laser pulses,
which can be used to probe the duration of the STED laser pulse. Quenching efciency is plotted as a function of the relative delay between the excitation
and the STED beam. Fitting the data with a Gaussian error function (red) indicates a STED pulse duration (FWHM) of at least 68.5 ps. (D) 2P and
2P-STED images of uorescent beads (diameter 40 nm). A clear resolution enhancement can be observed in 2P-STED relative to 2P excitation. (E)
Quantication of the line indicated in D and tting with a Lorentzian function returns a width of 62 nm for 2P-STED (red) as compared with 368 nm
for 2P (black).
improvement in spatial resolution over the 2P case. The
peak photon count at the center of the bead was only slightly
(~10%) reduced (Fig. 2 E), conrming that the STED light
intensity in the center of the doughnut is very low.
Subdiffraction imaging of dendritic spines in
acute slices
To check whether we could achieve subdiffraction spatial
resolution in acute brain slices, we imaged dendritic spines
from transgenic mice expressing YFP as a volume label in
a subset of hippocampal and cortical neurons. We compared
the performance of the 2P-STED with the 2P microscope by
imaging spines with or without the STED light. In general,
the spine necks appeared substantially wider without the
STED light, whereas the appearance of the 2P-STED
images was much crisper (Fig. 3 and Fig. S1).
To conrm this impression quantitatively, we measured
line proles across the spine necks and tted them with
a Lorentzian function as above. In the example illustrated
in Fig. 3 A (a spine of a hippocampal CA1 pyramidal neuron
imaged at a depth of 38.5 mm), the neck appeared to be
362 nm wide in the 2P image but only 89 nm in the 2P-
STED image. This fourfold difference demonstrates that
A
B
C D
FIGURE 3 2P-STED microscopy in acute brain slices. (A and B) 2P and 2P-STED images of dendritic spines of YFP-labeled CA1 (A) and cortical (B)
pyramidal neurons in acute slices from Thy1-YFP transgenic mice. Spine necks appear much thinner in 2P-STED compared to 2P images. Dotted lines indi-
cate spine neck widths (Lorentzian t of the line prole of the line indicated). The difference in width between 2P imaging (G 362 nm) and 2P-STED
imaging (G 89 nm) demonstrates the resolution enhancement by 2P-STED. (C) Quantication of spine neck widths of CA1 and cortex imaged in 2P
and 2P-STED modes. No difference between CA1 and cortex was detected. 2P imaging clearly overestimates spine neck widths when compared to the
2P-STED mode. Boxplot indicate Q1 and Q3 (rst and third quartile), median and mean (large and small lines, respectively). (D) Plot of 2P-STED measure-
ments versus the ratio between 2P and 2P-STED of the same object. As expected, the ratio is 1 when the structures are larger than the resolution limit of 2P
microscopy (~350 nm), but it increases steeply when the structures are <350 nm in size.
782 Bethge et al.
the STED approach can be used to achieve a substantial
gain in spatial resolution, well below the surface of living
brain tissue, using a long-working distance water-immer-
sion objective. It is important to note that the peak signal
intensity in 2P-STED mode was only reduced by ~30%
compared with the 2P signal (Fig. 3 A), suggesting that
the minimum of the STED doughnut was fairly intact at
this imaging depth.
We also imaged spines from layer 4/5 of cortical pyra-
midal neurons, imaged in the molecular layer of visual
and parietal cortices at about the same depth (Fig. 3 B).
Similar to spines of hippocampal neurons, the spine necks
of cortical neurons appear much thinner in 2P-STED than
in 2P images (G 97 nm for 2P-STED vs. G 409 nm
for 2P). The differences between 2P and 2P-STED imaging
were clear-cut and highly signicant (p < 0.0001, Fig. 3 C).
We did not detect any signicant differences in width
between spine necks in hippocampus and cortex, irrespec-
tive of whether 2P or 2P-STED was used (G
2P CA1

352 5 12 nm, n 11; G
2P-STED CA1
153 5 17 nm,
n 15; G
2P Cortex
372 5 15, n 6; G
2P-STED Cortex

178 5 8 nm, n 49) (Fig. 3 C).
Taken together, 2P-STED yields substantially better
resolved images of dendritic spines compared to 2P micros-
copy, well below the surface of acute brain slices. To illus-
trate the gain in resolution quantitatively, we plotted the
ratio of 2P to 2P-STED measurements as a function of
the widths of assorted structures measured by 2P-STED
(Fig. 3 D). This ratio equals 1 for structures larger than
the 2P diffraction limit, but increases sharply for structures
below it.
Time-lapse imaging
Next, we addressed the potential issue of phototoxicity
and bleaching during 2P-STED imaging. To this end, we
repeatedly (20 times) imaged the same stretch of dendrite
for 30 min (Dt 10 min) and looked for signs of photo-
damage such as blebbing. Our experiments demonstrate
that it is possible to repeatedly acquire superresolved images
with our 2P-STED microscope without obvious photodam-
age (Fig. 4 A). However, to avoid or reduce deleterious
effects, which are a general concern in uorescence micros-
copy, it was important to optimize the imaging parameters
A
B1
B2
B3
FIGURE 4 Time-lapse and dual-color 2P-STED
imaging. (A) Time-lapse images of a cortical spine
acquired at 38.5 mm below the tissue surface. Indi-
vidual time points are average projections of two
frames based on multiple sections (5 frames/stack,
four stacks, Dz 400 nm). (B) Two-color 2P-
STED imaging of neurons and microglia. Trans-
genic mice (CX3CR1
/eGFP
; Thy1
/eYFP
) express
YFP in neurons and GFP in microglia. Superre-
solved microglial processes (G 149 nm) can be
observed (B1), as can a microglial process contact-
ing dendritic spines (B2) and a maximum-intensity
projection of a z-stack of images (19 frames, 40
40 mm, from49.5 to 56.5 mm, Dz 368 nm) in
the cortex (B3). The high magnication images (B1
and B2) are merges of both color channels (green
(GFP) and yellow (YFP)); the overview image
(B3) is linearly unmixed, effectively separating
both channels (green (GFP) and red (YFP)). Dotted
lines indicate spine neck widths (G, Lorentzian t
of raw data).
(e.g., laser powers, pixel dwell times, image sizes, etc.),
which are linked to several factors, including slice quality,
sample brightness, and imaging depth.
Two-color imaging
Finally, we performed two-color 2P-STED imaging using
an approach we recently developed for STED microscopy,
which is based on a single laser pair for uorescence exci-
tation and STED quenching (12). To this end, we modied
the microscope by adding a dichroic mirror and a second
detection channel for spectral detection. To illustrate the
potential of using these techniques in combination, we
imaged brain slices from transgenic mice in which neurons
and microglia are uorescently labeled with YFP and
GFP, respectively. The images in Fig. 4 B demonstrate the
ability of the new approach to acquire superresolved images
of volume-labeled dendritic spines and microglial processes
well below the surface of acute brain slices. Although
spectral detection already provides for reasonable color
contrast (Fig. 4, B1 and B2), linear unmixing of the uores-
cence channels can improve color separation substantially
(Fig. 4 B3).
DISCUSSION
We present a novel kind of microscope for two-color super-
resolution imaging of neural morphology well below the
surface of acute brain slices using YFP and GFP as volume
labels. It is based on 2P excitation and pulsed STED micros-
copy and hence reconciles many powerful features, such as
subdiffraction spatial resolution, two-color imaging, optical
sectioning, depth penetration, the use of living samples, and
the possibility of combining it with electrophysiological
approaches.
We employed a long-working distance water-immersion
objective with an NA of 1.1, which is relatively low
compared to the glycerol- and oil-immersion objectives
(1.3 and 1.4, respectively) generally used in STED micros-
copy (9,14,15,22). Using spine necks to estimate our
resolution, the spatial resolution at an imaging depth of
~4050 mm is likely to be around 60-70 nm, considering
that spine necks are nite in size (>50 nm, as indicated
by electron microscopy (23)) and thus do not mimic point
sources of light.
STED microscopy, which was initially used in neurobi-
ology to acquire superresolved images of protein distribu-
tions inside cells by immunohistochemistry (14,24), has
recently been extended to live-cell imaging of dendritic
spine morphology (9,16), greatly facilitating investigation
of the structure and function of synapses, which is a major
research topic in neuroscience.
Most previous studies have used high-NA oil or glycerol
objectives with short working distances, which is incom-
patible with acute brain slices, because this preparation
requires water-immersion objectives with long working
distances and an upright microscope design. Acute brain sli-
ces are the preparation of choice for most synaptic physiol-
ogists because of the experimental access they provide for
pharmacological and electrophysiological experiments.
Initially, STED microscopy was based on a confocal
design and single-photon excitation, but recently the use
of 2P excitation has been reported. The rst two studies
demonstrated the principle of 2P-STED using CW lasers
for the STED beam (16,17), which are easier to implement
than pulsed lasers. However, as the CW laser is much less
efcient at quenching uorescence than the pulsed laser,
much more laser power (>10-fold) is needed to achieve
a comparable gain in spatial resolution, which is usually
prohibitive for living tissue. As a consequence, 2P-STED
microscopy in living tissue has been limited to >250 nm
in spatial resolution. Subsequent studies have used pulsed
STED lasers in conjunction with 2P excitation (25,26), but
those studies used inverted microscopes with oil objectives
on xed samples.
Here, we demonstrate 2P-STED time-lapse imaging in
two colors in acute brain slices with a four- to sixfold
improvement in spatial resolution over the 2P case. The
STED laser intensities required to achieve the gain in reso-
lution were substantially less than those reported previously
(16,26), which is important for reducing phototoxicity and
bleaching.
We speculate that the quality of the doughnut, which
featured a very low minimum at its center, and the efciency
of quenching GFP and YFP with short laser pulses afforded
the use of relatively modest STED laser powers.
Illustrating the potential of the new 2P-STED approach,
we imaged dendritic spines in acute brain slices and
measured superresolved spine necks at a few tens of microns
below the tissue surface. The spine-neck widths of pyra-
midal neurons in the CA1 region of the hippocampus and
cortex correspond well to those reported by electron micros-
copy (23,27), in our previous work (9), and by other recent
superresolution approaches (28,29), and they are much
thinner than is indicated by confocal measurements. For
this reason, spine necks are thought to represent a diffusion
barrier for signaling molecules, allowing for biophysical
compartmentalization of synapses (15,3033). Moreover,
the sizes of the microglial processes we have observed are
comparable with those of STED images in perfusion-xed
brain-tissue sections (data not shown).
Interestingly, the ability to measure the level of STED
quenching as a function of the delay between the laser
pulses provides a straightforward way to estimate the dura-
tion of the STED laser pulse, which otherwise requires
special equipment. Likewise, uorescence quenching by
the STED laser can also be used to estimate the uorescence
lifetime of the uorophores, as suggested previously (19),
which is normally done using time-correlated single-photon
counting instrumentation (34).
784 Bethge et al.
In summary, we have developed a new 2P-STED micro-
scope that can be assembled using all commercial compo-
nents and demonstrated its potential for investigating
synapses and glial cells with unprecedented spatial resolu-
tion in acute brain slices using genetically encoded
uorophores.
SUPPORTING MATERIAL
Fig. S1 is available at http://www.biophysj.org/biophysj/supplemental/
S0006-3495(13)00074-X.
This work was supported by a Ph.D. fellowship from the 7th Framework
Program (FP7) Marie Curie ITN SyMBaD to P.B. (cosupervised by G.M.
and U.V.N.), and grants from the Regional Council of Aquitaine (CRA),
Inserm, the French National Research Agency (ANR), and the Human
Frontier Science Program, to U.V.N.
We thank S. Berning for discussions on the design of the microscope and S.
W. Hell for providing optomechanical components (both afliated with the
Max Planck Institute for Biophysical Chemistry, Gottingen, Germany). We
thank members of the lab (M.O. Lenz, A. Panatier, and J. Tonnesen) and L.
S. Wijetunge (University of Edinburgh, Edinburgh, Scotland) for discus-
sions and comments on the manuscript and J. Angibaud for technical
support.
REFERENCES
1. Ramon y Cajal, S. 1995. Histology of the nervous system of man and
vertebrates. Oxford University Press, New York.
2. Holtmaat, A., and K. Svoboda. 2009. Experience-dependent structural
synaptic plasticity in the mammalian brain. Nat. Rev. Neurosci. 10:
647658.
3. Araque, A., G. Carmignoto, and P. G. Haydon. 2001. Dynamic
signaling between astrocytes and neurons. Annu. Rev. Physiol. 63:
795813.
4. Tremblay, M. E., B. Stevens, ., A. Nimmerjahn. 2011. The role of
microglia in the healthy brain. J. Neurosci. 31:1606416069.
5. Hell, S. W. 2007. Far-eld optical nanoscopy. Science. 316:11531158.
6. Hell, S. W., and J. Wichmann. 1994. Breaking the diffraction resolution
limit by stimulated emission: stimulated-emission-depletion uores-
cence microscopy. Opt. Lett. 19:780782.
7. Klar, T. A., S. Jakobs, ., S. W. Hell. 2000. Fluorescence microscopy
with diffraction resolution barrier broken by stimulated emission. Proc.
Natl. Acad. Sci. USA. 97:82068210.
8. Westphal, V., S. O. Rizzoli, ., S. W. Hell. 2008. Video-rate far-eld
optical nanoscopy dissects synaptic vesicle movement. Science. 320:
246249.
9. Nagerl, U. V., K. I. Willig, ., T. Bonhoeffer. 2008. Live-cell imaging
of dendritic spines by STED microscopy. Proc. Natl. Acad. Sci. USA.
105:1898218987.
10. Buckers, J., D. Wildanger, ., S. W. Hell. 2011. Simultaneous multi-
lifetime multi-color STED imaging for colocalization analyses. Opt.
Express. 19:31303143.
11. Pellett, P. A., X. Sun, ., J. Bewersdorf. 2011. Two-color STED
microscopy in living cells. Biomed. Opt. Express. 2:23642371.
12. Tnnesen, J., F. Nadrigny, ., U. V. Nagerl. 2011. Two-color STED
microscopy of living synapses using a single laser-beam pair. Bio-
phys. J. 101:25452552.
13. Willig, K. I., A. C. Stiel, ., S. W. Hell. 2011. Dual-label STED nano-
scopy of living cells using photochromism. Nano Lett. 11:39703973.
14. Willig, K. I., S. O. Rizzoli, ., S. W. Hell. 2006. STED microscopy
reveals that synaptotagmin remains clustered after synaptic vesicle
exocytosis. Nature. 440:935939.
15. Urban, N. T., K. I. Willig, ., U. V. Nagerl. 2011. STED nanoscopy of
actin dynamics in synapses deep inside living brain slices. Biophys. J.
101:12771284.
16. Ding, J. B., K. T. Takasaki, and B. L. Sabatini. 2009. Supraresolution
imaging in brain slices using stimulated-emission depletion two-
photon laser scanning microscopy. Neuron. 63:429437.
17. Moneron, G., and S. W. Hell. 2009. Two-photon excitation STED
microscopy. Opt. Express. 17:1456714573.
18. Imspector. www.max-planck-innovation.de/de/industrie/technologie
angebote/software/.
19. Schrader, M., F. Meinecke, ., S. W. Hell. 1995. Monitoring the
excited state of a uorophore in a microscope by stimulated emission.
Bioimaging. 3:147153.
20. Eggeling, C., C. Ringemann, ., S. W. Hell. 2009. Direct observation
of the nanoscale dynamics of membrane lipids in a living cell. Nature.
457:11591162.
21. Zipfel, W. R., R. M. Williams, and W. W. Webb. 2003. Nonlinear
magic: multiphoton microscopy in the biosciences. Nat. Biotechnol.
21:13691377.
22. Berning, S., K. I. Willig, ., S. W. Hell. 2012. Nanoscopy in a living
mouse brain. Science. 335:551.
23. Harris, K. M., F. E. Jensen, and B. Tsao. 1992. Three-dimensional
structure of dendritic spines and synapses in rat hippocampus (CA1)
at postnatal day 15 and adult ages: implications for the maturation of
synaptic physiology and long-term potentiation. J. Neurosci. 12:
26852705.
24. Sieber, J. J., K. I. Willig, ., T. Lang. 2007. Anatomy and dynamics of
a supramolecular membrane protein cluster. Science. 317:10721076.
25. Li, Q., S. S. Wu, and K. C. Chou. 2009. Subdiffraction-limit two-
photon uorescence microscopy for GFP-tagged cell imaging.
Biophys. J. 97:32243228.
26. Bianchini, P., B. Harke, ., A. Diaspro. 2012. Single-wavelength
two-photon excitation-stimulated emission depletion (SW2PE-STED)
superresolution imaging. Proc. Natl. Acad. Sci. USA. 109:63906393.
27. Arellano, J. I., R. Benavides-Piccione, ., R. Yuste. 2007. Ultrastruc-
ture of dendritic spines: correlation between synaptic and spine
morphologies. Front Neurosci. 1:131143.
28. Testa, I., N. T. Urban, ., S. W. Hell. 2012. Nanoscopy of living brain
slices with low light levels. Neuron. 75:9921000.
29. Izeddin, I., C. G. Specht, ., M. Dahan. 2011. Super-resolution
dynamic imaging of dendritic spines using a low-afnity photoconver-
tible actin probe. PLoS ONE. 6:e15611.
30. Svoboda, K., D. W. Tank, and W. Denk. 1996. Direct measurement of
coupling between dendritic spines and shafts. Science. 272:716719.
31. Grunditz, A., N. Holbro, ., T. G. Oertner. 2008. Spine neck plasticity
controls postsynaptic calcium signals through electrical compartmen-
talization. J. Neurosci. 28:1345713466.
32. Bloodgood, B. L., and B. L. Sabatini. 2005. Neuronal activity regulates
diffusion across the neck of dendritic spines. Science. 310:866869.
33. Araya, R., J. Jiang, ., R. Yuste. 2006. The spine neck lters membrane
potentials. Proc. Natl. Acad. Sci. USA. 103:1796117966.
34. Lakowicz, J. R. 2006. Principles of Fluorescence Spectroscopy.
Springer, New York.
Biophysical Journal will publish a special issue of the Journal with a focus on Quantitative Cell Biology.
The Journal welcomes submissions that report on studies of biological molecules
and structures with a focus on mechanisms at the cellular level using the
concepts and methods of physics, chemistry, mathematics, engineering, and
computational science. Studies should further our understanding of the
interactions between the various systems of a cell as well as how these
interactions are regulated. The journal aims to publish the
highest quality work and articles should have sufcient
importance to be of general interest to biophysicists,
regardless of their research specialty.
Special Issue: Focus on Quantitative Cell Biology
Biophysical Journal
Biophysical Society
For publication December 2, 2014, Volume 107, Number 11
Deadline for submission: July 1, 2014
Please include a cover leller slaling lhal you would like lo be parl ol lhe
special issue on Quanlilalive Cell 8iology
Selecl "Special lssue. locus on Quanlilalive Cell 8iology" when up-
loading your submission.
lnslruclions lor aulhors can be lound al. hllp.//download.cell.com/
images/edimages/8iophys/lnslruclions_lo_Aulhors.pdl
Queslions can be direcled lo lhe BJ Ldilorial Ollce
al 8J@biophysics.org or (240) 290-5545.
PUBLISHED BY
For more information, go to www.biophysj.org
Journal publicalion lees will apply
Call for Papers
Biophysical Journal Award
For more information, go to www.biophysj.org
Biophysical Society
Biophysical Journal (BJ) is pleased to announce the new Biophysical Journal
Paper of the Year Award. The award will be presented to the correspond-
ing author of an outstanding paper published in the Journal during
the previous 12 months, beginning with January 2014 submissions.
The award is given for an original regular article in BJ and will
spotlight the high-quality work of young investigators.
The award recipient will
have received lheir PhD or MD wilhin lhe pasl !2 years,
be selecled by a commillee ol Associale Ldilors and
lhe Ldilor-in-Chiel,
receive $!,000, and
speak al lhe Sociely's Annual Meeling.
FINE SURGICAL INSTRUMENTS
FOR RESEARCH
Quality surgical instruments
as pure as gold.
Visit us at finescience.com to explore our complete product line,
and to locate our offices and dealers around the world.
Fine Science Tools has been shipping world-renowned surgical and
microsurgical instruments globally since 1974. With offices and dealers
throughout the world, FST conveys convenience, expedient and superb
customer service with no boundaries.

B21a7e87 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

B21a7e87 1

Uploaded by

Copyright:

Available Formats

NanoDrop 2000c

technology to evaluate their

* and Leonor Saiz

IKERBASQUE, Basque Foundation for Science, Bilbao, Spain; and

Center for Research in Biological Systems and

The Hong Phong Nguyen,

and Elena P. Ivanova

Departament dEnginyeria Quimica, Universitat Rovira i Virgili, Tarragona, Spain;

C in 5 mL of nutrient broth (Oxoid) or

C in 5 mL of marine broth (Difco) with shaking (120 rpm). Bacterial

C at all times. Each sample

), and the average

and Yale E. Goldman

Department of Physics and Astronomy,

) used for single-molecule myosin V motility

relative to the plane

linear polarized excitations, enable unambiguous determination of

is usually reported by the dipole

. In previous work, this artifact was not noticed because

)) may apply due to the twofold symmetry of

) (Fig. S4). In Fig. 2, four major step-

(where and indicate the state before and after step-

(Fig. 2 E). The wobble parameter

during the steps when b changes

and a high value of 81

. The magnitude of the total angular displacement z between change-point intervals is

. (D) The extent of rapid

) and seven showed it

). Neither the ATP concentra-

) compared with the distribu-

). (B) Distributions of b (red) and a (green) during the

) relative to the plane dened by the input

for the leading and trailing congurations when

, respectively (5SE). The small

C, where many molecular pro-

/bp (or a superhelical density of Dsx 0:03) in the interval

C (4,5)), the question arises of how the properties

Tilt Roll Twist Shift Slide Rise

All values given are at 300 K, s

. The relation between the living temperature of or-

C, the typical melting temperature of uncon-

C. In vivo, the double helix is therefore not

Chigher than the melting

/bp, close to the average twist

/bp for stiff basepair steps to 2.5

to the spontaneous twist per basepair

in the interval under consideration, between

C and the critical temperature.

/bp increase in the

resolution. J. Mol. Biol. 319:10971113.

C and isotopically heavy CH4 production by a hyperthermophilic

chsner, A., C. Henninger, and J. Gegner. 2003. Modellierung mikros-

and Francisco Bezanilla

Committee on Neurobiology and

currents during the falling phase of

C in a solution containing (in mM)

channels increases the stability of voltage sensors in the relaxed state.

in the resting state. This

and Robert Blumenthal

Basic Research Laboratory, Center for Cancer Research,

C followed by cooling at room temperature. Any insoluble mate-

calculated 2229.96]; MAX35 ESI-MS: 2287.4 [(MH)

relative to the pore axis and then a 10-fold rotational