n-DNA Interactions by Unzipping DNA
Michelle D. Wang*
ate Physics, Cornell University, Ithaca, New York 14853, USA
ly 2003; publisher error corrected 17 July 2003)
olecule characterization of the prominent activation
ding complex. We achieved this new capability by
pping force analysis of protein association and used
e binding to specific DNA sites. Analysis revealed
ein-DNA interactions. This new method is able to
site-specific, single-molecule basis.
PACS numbers: 87.14.Ee, 05.70.Ln, 82.37.Rs, 87.14.Gg
In this Lett
quantitative
(DFS) in ord
acteristic int
vation barrie
We also show
us to distin
plexes. To de
DFS is straightforward. A bound molecule will unbind
under thermal agitation if given sufficient time. However,
ein
tion (not to
P H Y S I C A L R E V I E W L E T T E R S week ending11 JULY 2003VOLUME 91, NUMBER 2to study the binding of restriction enzymes to their rec-
ognition sites.
DFS is a powerful method for mechanical investigation
of the interactions of biomolecules. The basic idea behind
scale). Applied tension unzips the two strands of the DNA
molecule. The location of the unzipping fork is indicated by an
unzipping index j. The presence of a DNA-binding protein at
the unzipping fork is observed as an increase in the force
required to separate the strands.028103-1er we add essential features to UFAPA by
application of dynamic force spectroscopy
er to determine the lifetimes (toff) and char-
eraction distances (d) of the prominent acti-
rs of site-specific protein-DNA interactions.
that analysis of the disruption forces allows
guish between different protein-DNA com-
monstrate these ideas, we used this approach
Bound Prot
Force Force
dsDNA
FIG. 1. Schematic of the unzipping configuraDynamic Force Spectroscopy of Protei
Steven J. Koch and
Department of Physics, Laboratory of Atomic and Solid St
(Received 18 December 2002; published 10 Ju
We demonstrate the first site-specific single-m
barrier for the disruption of a protein-DNA bin
combining dynamic force spectroscopy with unzi
the combination to investigate restriction enzym
lifetimes and interaction distances for three prot
distinguish protein-DNA binding complexes on a
DOI: 10.1103/PhysRevLett.91.028103
Protein-DNA interactions are central to many major
cellular processes, including transcription, replica-
tion, and packaging of DNA into chromatin. Indeed,
recent sequencing of the human genome shows that

15% of the
30 000 genes encode proteins which bind
to nucleic acids [1]. Basic parameters of protein-DNA
interactions include binding location (sequence specific-
ity), binding affinity (equilibrium association constant),
binding rate constants (on and off rates), and in some
cases catalytic rate constants. While traditional bio-
chemical (bulk) methods have been successful in eluci-
dating some of these parameters, the new single-molecule
method that we describe in this Letter provides special-
ized advantages and in some cases will enable measure-
ments that thus far have been inaccessible.
This method, called unzipping force analysis of pro-
tein association (UFAPA), is a novel and versatile method
for probing protein-DNA interactions [2](see Fig. 1). In
this method, a single DNA double helix is unzipped [3] in
the presence of DNA-binding proteins using a feedback-
enhanced optical trap. One strand of the DNA is anchored
to a microscope coverslip while the other strand is at-
tached to a microsphere held in an optical trap. The DNA
is unzipped as the microscope coverslip is moved away
from the trapped microsphere. When the unzipping fork
in a DNA reaches a bound protein, a dramatic increase in
the tension in the DNA, followed by a sudden tension
reduction, is detected. Analyses of the unzipping forces
and lengths of the DNA tether reveal the locations of the
bound proteins and the equilibrium association constants.0031-9007=03=91(2)=028103(4)$20.00 it is often not experimentally feasible to wait for unbind-
ing in cases where the lifetime (toff) of the bound state is
exceedingly long. To circumvent this, DFS reduces the
lifetime by tilting the energy landscape with an external
force (F) to encourage unbinding. Analysis of constant
force lifetimes allows determination of the lifetime under
no applied force. While this is the simplest approach, the
lifetime can also be determined by using a method in
which the applied force is increased at a constant rate
(r  dF=dt) to encourage unbinding [5]. In this method,
as activation barriers are lowered, the unbinding force
distribution from many measurements gives a measure of
the natural lifetime. The unbinding force probability den-
sity function (PDF) is well defined if there is one pre-
dominant activation barrier for unbinding:
pF; r  1
toffr
exp


kBT
toffrd

exp

Fd
kBT
 kBT
toffrd
exp


Fd
kBT

;
(1)
where kBT is the thermal energy and d is the distance
along the direction of the applied force between the
bound state and the activation barrier peak [4]. A series
of these distributions at different r values provides a
relation between the most-probable unbinding force (F)
and r:
ssDNA1 2 3 j2003 The American Physical Society 028103-1

F  kBT
d
ln


toffrd
kBT

: (2)
Therefore, toff and d can be determined either from
pF; r at a given r [Eq. (1)] or from the F vs r relation
[Eq. (2)]. If there is truly a single activation barrier, these
two methods should yield the same result and the plot of
F vs lnr should give a straight line for all values of r. A
nonlinear plot of F vs lnr then indicates the presence of
more than one activation barrier, and a crossover in F vs
lnr from one linear regime to another indicates a transi-
tion from one dominant activation barrier to another [4].
In order to incorporate DFS into UFAPA, we designed a
novel digital loading rate clamp, r  dF=dt  const, in
the optical trapping setup. In this implementation, unzip-
ping proceeded under the control of an algorithm that
effectively became a loading rate clamp when the unzip-
ping fork encountered a bound protein [6]. An example of
data taken with the loading rate clamp is shown in Fig. 2.
We have previously described [2] the DNA construct [7],
the optical trapping instrument and calibration methods,
the unzipping buffer conditions, and the proteins [8] used
in these experiments. Briefly, a single DNA double helix
was unzipped in buffers containing the restriction en-
zymes (BsoBI or XhoI) but without Mg2, so that the
enzymes would bind to the DNA without cutting it. For
the data shown in Fig. 2, BsoBI molecules were bound
to a number of sites on a pCP681-derived DNA construct.
In Fig. 2(a), the presence of bound complexes is revealed
by the prominent peaks seen in this force versus time
graph. Stochastic unbinding events occurred when the
force reached 25–50 pN and are indicated by the sudden
drops in force. As Fig. 2(a) clearly shows, preceding each
unbinding event the force increased linearly with time as
prescribed by the loading rate clamp. In Fig. 2(b), a force
ramp preceding an unbinding in Fig. 2(a) corresponds to a
horizontal plateau due to the inhibition of unzipping by
the bound protein complex.
As shown in Fig. 2, unbinding events are easily dis-
tinguished from the baseline unzipping forces. A novel
automated event detection scheme located each event, and
determined an event starting force, the disruption force,
and the average loading rate during the event [9]. An
example of an automatically detected event is highlighted
in Fig. 2. At a given protein-DNA complex, under the
action of the loading rate clamp, the actual loading rate
was estimated from a linear fit of the data [Fig. 2(a)]. The
observed force at any time, including Brownian noise,
remained within 1:5 pN of the force predicted from the
fit. The actual loading rates were distributed around the
specified value with a standard deviation of
10% of
P H Y S I C A L R E V I E W L E T T E R S week ending11 JULY 2003VOLUME 91, NUMBER 2FIG. 2 (color online). Example of unzipping data taken using
loading rate clamp (59 pN=s). (a) Force versus time. The graph
demonstrates the uniform force loading rate for forces greater
than 15 pN. One of the eight events from the automated event
detection is highlighted. The dotted line shows the loading rate
fit for the event. (b) Calculated unzipping index, j, vs time.
Each horizontal step represents data where a restriction enzyme
pins the unzipping index at a certain value until the complex
disrupts. The same event as in (a) is highlighted. Horizontal
dashed lines mark unzipping indices which are predicted
BsoBI binding sites, of types  (ttcCTCGGGaat) and 
(aaaCTCGAGact).
028103-2FIG. 3. Dynamic force spectroscopy for BsoBI unbinding
from  sites (N  449 events total). Each histogram shows
unbinding force distribution with bins of 2.5 pN width at a
given force loading rate. Dashed curves represent predicted
force probability density functions resulting from the determi-
nation of local values of d and toff (see text). Solid curves
represent predicted force probability density functions resulting
from the determination of global values of d and toff (also see
text). Vertical dashed lines designate the experimentally acces-
sible force range of the current implementation of UFAPA.
028103-2

UFAPA also provides a new ability to distinguish
between different protein-DNA complexes on a single-
molecule site-specific basis. In addition to BsoBI unbind-
ing from the  site, we also examined BsoBI and XhoI
unbinding from  sites (Fig. 4). Interestingly, these two
binding species also show linear but distinguishable F vs
lnr plots. We used the same methods described earlier to
find the best-fit d and toff , and all results are summarized
in Table I, along with values of the equilibrium associa-
tion constant KA for these three species, measured with
P H Y S I C A L R E V I E W L E T T E R S week ending11 JULY 2003VOLUME 91, NUMBER 2the specified value. In the context of DFS, this spread is
insignificant, due to the logarithmic relationship between
the force-loading rate and the expected unbinding force
distribution [see Eqs. (1) and (2)] [4].
In Fig. 3, we show a summary of the unbinding force
distributions for BsoBI from  sites at various force-
loading rates r. For a given r, the local fit values of d
and toff were obtained using the maximum-likelihood
method [10], based on the assumed form of the PDF
[11] [Eq. (1)]. The parameter search was expedited by
the use of the Nelder-Mead simplex method. Once the
best-fit PDF (a dashed curve in Fig. 3) for a given r was
FIG. 4. Dynamic force spectroscopy for three binding spe-
cies. Data points represent the most-probable unbinding force
for a given force-loading rate, obtained from the maximum
likelihood method described in the text. Error bars were
determined using a Monte Carlo method [12]. Open circles
represent BsoBI unbinding from  sites, open squares represent
BsoBI unbinding from  sites, and filled squares represent
XhoI unbinding from  sites. Solid lines are linear fits of Eq. (2)
to the data. Parameters d and toff obtained from the fits are
listed in Table I.obtained, F was calculated analytically from Eq. (2).
After repeating for all r, a plot of F vs lnr was
generated from these results (see Fig. 4, open circles). A
linear least-squares fit to this plot using Eq. (2) yielded
the so-called global fit values of the parameters d and toff .
These values were then used to generate the global fit
PDFs in Fig. 3 (solid curves). Figure 3 shows a good
agreement among the measured PDF, its local fit PDF,
and its global fit PDF, and Fig. 4 shows the expected linear
relation for F vs lnr. All these are evidence that (for the
loading rates investigated), a single activation barrier
dictates the behavior of BsoBI unbinding from  sites.
TABLE I. Summary of results for three different binding species
equilibrium association constant measured from the occupancy of
included in the DFS analysis, while NEQ represents the total numb
Protein DNA binding site d (nm) l
BsoBI , ttcCTCGGGaat 0:98 0:04 4
BsoBI , aaaCTCGAGact 0:80 0:09
XhoI , aaaCTCGAGact 1:18 0:12
028103-3UFAPA as described previously [2].
Figure 4 and Table I show that the three binding species
have many potential distinguishable dynamic signatures,
including the characteristics of the dominant activation
barriers (d and toff), the F vs lnr behavior, and even the
force distributions themselves. This is true for all the spe-
cies examined: between the same protein (BsoBI) binding
to two different sites ( vs ), and between two different
proteins (BsoBI vs XhoI) binding to the same DNA site
(). The former comparison is further illustrated in Fig. 5,
where unbinding force distributions are shown for BsoBI
unbinding from  (filled bars) and  (lined bars) sites at a
force-loading rate of
60 pN=s. Figure 5 shows clearly
distinguishable distributions with only
19% overlap.
Therefore, under these conditions, a single measurement
of the unbinding force is nearly sufficient to distinguish
between the two species. For example, one could set a
threshold at 33 pN such that a measurement < 33 pN is
considered to correspond to an  site and a measurement
> 33 pN is considered to correspond to a  site; then an
assessment based on this single measurement will yield a
correct conclusion 90% of the time. This capability can
lead to novel assays which screen for multiple proteins
and multiple binding sites simultaneously and in parallel.
We have shown that when probed with UFAPA, dis-
ruption of three protein-DNA binding species conformed
well to the theory of DFS. Analysis revealed a prominent
activation barrier for disruption of each site-specific
protein-DNA complex. Note that it is possible that dis-
ruption of protein-DNA interactions by unzipping may
not proceed along the ‘‘natural’’ zero-force dissociation
pathways for the binding species examined. Indeed, for
XhoI disruption from beta sites, it is likely that the
natural lifetime is much lower than the apparent life-
time of 6000 s obtained here [13]. While it may be
difficult to relate the apparent lifetime to the natural
zero-force lifetime, it is nevertheless important to note
. The parameters d and toff are obtained from Fig. 4. KA is the
sites at
60 pN=s. NDFS represents the total number of events
er of sites counted for the KA measurements.
n toff s
NDFS log10 KA M1
NEQ
:53 0:27 449 9:15 0:07 194
6:1 0:7 82 9:34 0:19 30
8:7 0:9 47 8:94 0:27 18
028103-3

J. T. Lis, C. L. Peterson, and M. D. Wang, Proc. Natl.
Acad. Sci. U.S.A. 99, 1960 (2002).
[6] To generate a loading rate clamp, the microscope cover-
P H Y S I C A L R E V I E W L E T T E R S week ending11 JULY 2003VOLUME 91, NUMBER 2that the observed activation barrier represents an im-
portant physical aspect of protein-DNA interaction
landscape. This is the first direct experimental access to
site-specific protein-DNA interaction landscape with pre-
vious experiments relying on gel mobility-shift, filter-
binding, or other assays which are not site specific,
or inextricably include nonspecific protein-DNA
interactions.
Possible future applications for UFAPA range from
qualitative assays for mapping protein binding sites and
distinguishing between binding species to quantitative
assays which can probe the energetics of the protein-
DNA interactions to access information that was previ-
ously unobtainable. Thus, UFAPA combined with DFS
presents a powerful new tool for probing specific protein-
DNA interactions.
We thank R. C. Yeh, A. Shundrovsky, A. La Porta,
K. Adelman, B. D. Brower-Toland, and D. S. Johnson for
helpful technical advice and scientific discussions. This
work was supported by grants from the NIH, the
Beckman Foundation, the Sloan Foundation, and the
Keck Foundation.
highly distinct unbinding signatures, as shown by both the data
and the predicted PDF.*
[1]
[2]
[3]
[4]
[5]
02810FIG. 5. Distinct unbinding force distributions. Solid gray
bars represent BsoBI unbinding from  sites (N  141), while
lined bars represent BsoBI unbinding from  sites (N  35).
Solid lines are predicted distributions based on the global fit
parameters from Table I. All data are for a force loading rate of

60 pN=s. At this particular stretch rate, the two sites produceCorresponding author.
Email address: mwang@physics.cornell.edu
E. S. Lander et al., Nature (London) 409, 860 (2001); J.C.
Venter et al., Science 291, 1304 (2001).
S. J. Koch, A. Shundrovsky, B. C. Jantzen, and M. D.
Wang, Biophys. J. 83, 1098 (2002).
U. Bockelmann, B. Essevaz-Roulet, and F. Heslot, Phys.
Rev. Lett. 79, 4489 (1997).
E. Evans, Annu. Rev. Biophys. Biomol. Struct. 30, 105
(2001).
R. Merkel, P. Nassoy, A. Leung, K. Ritchie, and
E. Evans, Nature (London) 397, 50 (1999); X. Zhang,
E. Wojcikiewicz, and V.T. Moy, Biophys. J. 83, 2270
(2002); B. D. Brower-Toland, C. L. Smith, R. C. Yeh,
3-4slip velocity (controlled by a piezostage) was modulated
to produce a desired linear force-loading rate. During
unzipping, the force, F, and extension were measured,
and a freely jointed chain (FJC) model was used to
calculate in real time the number of bases unzipped, j.
The instantaneous stiffness of the ssDNA (k; pN=nm)was
calculated using the FJC model and the known F and j.
Throughout the experiment, the piezostretch rate
(; nm=s) was modulated according to k in order to
produce the desired linear force-loading rate (k). The
calculations were performed while still maintaining a
feedback loop rate on the order of 10 kHz.
[7] The unzipping construct has 17 nearly identical repeats
of
200 bp. Slight variations in the sequence produced
two different binding sites for the enzymes used in this
report. The first, designated  sites, have the sequence
CTCGGG and are bound only by BsoBI. The second,
designated  sites, have the sequence CTCGAG and may
be bound by both BsoBI and XhoI.
[8] The restriction enzymes used were BsoBI (350 pM) and
XhoI (2300 pM). Experiments were performed at 23 C
in a buffer containing 50 mM sodium phosphate pH 7.0,
50 mM NaCl, 0.02% Tween-20, 10 mM EDTA.
[9] The time series data for j were converted to jp , followed
by median filtering, a derivative with respect to time, and
finally threshold detection of events. Disruptions show up
as large positive derivatives of

j
p
. Events with a dis-
ruption force less than 19 pN are discarded to avoid
events that may be naked DNA disruption.
[10] P. R. Bevington and D. K. Robinson, in Data Reduction
and Error Analysis for the Physical Sciences, edited by
S. J. Tubb and J. M. Morris (McGraw-Hill, Inc., New
York, 1992), 2nd ed., Vol. 1, Chap. 10, p. 180.
[11] This distribution assumes that all forces from 0 to 1 are
experimentally accessible, whereas in our UFAPA im-
plementation, we have both a lower and an upper force
cutoff (shown as vertical dashed lines in Fig. 3). To
account for this, we modified Eq. (1) to set pF; r to
zero outside of the experimental range and to normalize
the remaining probability so the integral remains unity.
The lower force threshold depends on the starting force
for the particular event, while the upper cutoff force is set
to 51 pN, to prevent overstretching of the dsDNA handle.
[12] To obtain the error bars, the data points in Fig. 4 first
were fit (without error bars) to estimate the global d and
toff . These parameters and Eq. (1) were used to perform a
Monte Carlo simulation of the same number of events as
were in the original data set. The simulated data set was
then analyzed in the same way as the original data, and
F was computed from Eq. (2). The simulation was
repeated 1000 times, and the standard deviation of F
was used for the error bar.
[13] Our observed KA of about 109 M1 would require an
extremely slow on rate of about 105 M1 s1 —or, at the
1 nM concentration of protein used, an on time of about
10 000 s. Experimentally we observed an on time on the
order of 100 s or less.
028103-4