of Enhanceosome Assembly
in Embryonic Stem Cells
Jiji Chen,1Zhengjian Zhang,3Li Li,3Bi-Chang Chen,3Andrey Revyakin,1Bassam Hajj,1,5Wesley Legant,3
Maxime Dahan,1,5Timothe ´e Lionnet,1Eric Betzig,3Robert Tjian,1,3,4and Zhe Liu1,2,*
1Transcription Imaging Consortium, Janelia Farm Research Campus, Howard Hughes Medical Institute, 19700 Helix Drive, Ashburn,
VA 20147, USA
2Junior Fellow Program, Janelia Farm Research Campus, Howard Hughes Medical Institute, 19700 Helix Drive, Ashburn, VA 20147, USA
3Janelia Farm Research Campus, Howard Hughes Medical Institute, 19700 Helix Drive, Ashburn, VA 20147, USA
4LKS Bio-Medical and Health Sciences Center, CIRM Center of Excellence, University of California, Berkeley, Berkeley, CA 94720, USA
5Laboratoire Physico-Chimie Curie, Institut Curie, Centre National de la Recherche Scientifique Unite ´ Mixte de Recherche 168, Paris 75794,
Enhancer-binding pluripotency regulators (Sox2 and
Oct4) play a seminal role in embryonic stem (ES) cell-
specific gene regulation. Here, we combine in vivo
and in vitro single-molecule imaging, transcription
factor (TF) mutagenesis, and ChIP-exo mapping
to determine how TFs dynamically search for and
assemble on their cognate DNA target sites. We
find that enhanceosome assembly is hierarchically
ordered with kinetically favored Sox2 engaging the
target DNA first, followed by assisted binding of
Oct4. Sox2/Oct4 follow a trial-and-error sampling
mechanism involving 84–97 events of 3D diffusion
(3.3–3.7 s) interspersed with brief nonspecific colli-
sions (0.75–0.9 s) before acquiring and dwelling at
specific target DNA (12.0–14.6 s). Sox2 employs a
3D diffusion-dominated search mode facilitated by
1D sliding along open DNA to efficiently locate tar-
gets. Our findings also reveal fundamental aspects
of gene and developmental regulation by fine-tuning
TF dynamics and influence of the epigenome on
target search parameters.
Precise spatiotemporal regulation of gene expression underpins
the finely balanced lineage-specification and morphogenetic
de-Leon and Davidson, 2007; Levine and Tjian, 2003; Tam and
Loebel, 2007). Although a powerful combination of classical
biochemistry, genetics, and genomic approaches (ENCODE
Project Consortium et al., 2012; Levine and Tjian, 2003) has re-
vealed many aspects of mammalian gene regulation, the kinetic
principles that govern transcription factor (TF) dynamics as it
searches for specific target sites in the nucleus of living cells
remained elusive. With recent advances in molecular imaging, it
has become possible to track individual protein molecules in sin-
gle live cells (Abrahamsson et al., 2013; Elf et al., 2007; Gebhardt
et al., 2013; Mazza et al., 2012). These rapidly emerging superre-
solution platforms provide a means for elucidating the search
pattern and efficiency of TFs in finding and binding target sites.
It has been appreciated that in a fractal and compact nuclear
environment, a TF must manage to search for its specific binding
siteswhile colliding and navigating past manynonspecific decoy
sites (Fudenberg and Mirny, 2012; Mirny et al., 2009; Mueller
et al., 2013a). However, to date, the dynamic balance between
specific and nonspecific binding events of individual TF mole-
cules in mammalian cells remains largely unknown limiting our
understanding of TF search modalities. Likewise, the order of
events of TF transactions and mechanisms that direct multiple
TFs homing in on a cis-regulatory DNA element to form an
enhanceosome complex have been challenging to dissect.
Here, we report a single-cell single-molecule imaging strategy
that allows us to quantitatively measure in individual live cells
both the specific and nonspecific TF residence times on chro-
matin DNA and thus compute the ‘‘in vivo’’ target search time.
We developed our approach by tracking the DNA-binding dy-
namics of two key pluripotency regulators, Sox2 and Oct4, in
mouse embryonic stem (ES) cells as well as in a reconstituted
in vitro single-molecule-binding assay. These single-molecule
tracking (SMT) results were complemented with an analysis of
TF mutants and genome-wide TF-binding studies to establish
that the search for a specific DNA target follows a trial-and-
error sampling mechanism: each TF undergoes multiple rounds
of short-lived nonspecific chromatin-binding events (tns= 0.75–
0.9 s) punctuated by 3D diffusion episodes (t3D = 3.3–3.7 s)
before eventually encountering a specific DNA target to which
it binds more stably. Interestingly, we also observe that Sox2
slides along short stretches of DNA nonspecifically searching
for target sites in vitro and that Sox2 and Oct4 assemble on their
in vivo enhancer sites in a hierarchically ordered fashion with
1274 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
Sox2 serving as the lead factor that first engages with its specific
DNA target followed by assisted binding of Oct4 that subse-
quently stabilizes the ternary complex. We also show by
inducing specific chromatin modifications that the global epige-
netic state of ES cells influences the search parameters and
kinetics of Sox2 and Oct4 DNA binding. Our results thus reveal
important in vivo kinetic properties controlling TF dynamics
and unmask new enhanceosome assembly mechanisms under-
lying key pluripotency transcription programs.
Imaging TF-DNA Dissociation Kinetics
(Gao et al., 2012; Planchon et al., 2011) for 2D SMT. In order
to selectively track DNA-bound Sox2 molecules, we imaged
live ES cells expressing a fluorescently tagged Sox2 (halo-
TMR labeled) using a combination of low excitation power
(50 W/cm2) and long integration times (500 ms) (Figure 1A). As a
result, the blurred images of fast diffusing molecules blend into
the background, whereas less-mobile Sox2 molecules that are
interacting with chromatin appeared as bright individual diffrac-
online). The dwell time of each ‘‘immobile’’ Sox2 single molecule
was then directly measured as the lifetime of the fluorescence
spot as demonstrated by a single step of photobleaching. We
found that dwell times of engaged Sox2 molecules failed to be
described by a single-component decay model (Figure S1D).
However, a two-component exponential decay model was in
good agreement with our data (Figure S1E), with lifetimes of 0.8
and 12.03 s for the short- and long-lived population, respectively
(after photobleaching correction; Figures S1A and S1B; Equa-
tions S2, S3, and S4). In order to test whether the two classes
of relatively ‘‘immobile’’ particles corresponded to nonspecific
and specific DNA-bound Sox2 molecules, we deleted the Sox2
DNA-binding domain (Sox2-TAD, Sox2 121–319 aa, Figure S2A)
ing domain resulted in the disappearance of the long-lived
Figure 1. Discriminating Specific and Nonspecific TF-DNA Dissociation Kinetics in Live Cells by Single-Molecule Imaging
(A) Selective visualization of immobile Sox2 molecules by 2D imaging using long exposure times (low excitation power, 50 W/cm2; long integration time, 500 ms).
Fast particles blend into the background, whereas immobile ones appear as spots. The dissociation rate (koff) is extracted from the measured residence time.
(B) Immobile Sox2 molecules imaged with the 2D imaging setup in ES cells. Top-left view is of immobile single Sox2 molecules that are detected as near
diffraction-limited spots in the nucleus. Top-right view shows that at 500 ms, halo-tag NLS (nuclear localization signal) displays negligible stable binding but
diffused fluorescence background. Bottom-left view is of a construct lacking the DNA-binding domain (Sox2-TAD) displaying a great reduction in the number of
immobile molecules. Bottom-right view shows that at faster frame rates (10 ms), freely diffusing halo-tag NLS molecules can now be detected as spots. Yellow
dotted circle represents the nucleus outline.
(C and D) 1-cumulative distribution function (1-CDF) of Sox2 (C) and Sox2-TAD (D) residence time was respectively fitted with a two-component (long-lived and
short-lived component) and asingle-component exponentialdecay model. For WTSox2, thefittedlifetimes are t1= 12.03 ± 1.8 s(long-lived component)and t2=
0.8 ± 0.07 s (short-lived component). In the case of Sox2 DNA-binding domain deletion (Sox2-TAD), t = 0.75 ± 0.03 s.
(E) Residence lifetime of the long-lived component for Sox2 (12.03 ± 1.8 s), a Sox2 construct with mutations on Sox2 DNA-binding surface (Sox2M) (9.11 ±
1.93 s), Sox2D (8.62 ± 0.98 s), and mean residence lifetime for Sox2-TAD (0.75 ± 0.03 s).
(F) Long-lived bound fraction of all bound molecules for Sox2, Sox2M, and Sox2D determined by our 2D dwell time analysis. (See Figure S3 and Equation S1
*p < 0.05. Error bars represent SD. See also Figures S1, S2, S3, and S4, Movies S1 and S2, and Table S1.
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1275
population from the dwell time histograms of the truncated
protein, suggesting that long-lived immobile particles most likely
(Figure 1D). Mutation of amino acids on the Sox2 DNA-binding
surface (Sox2M, M47G:F50:M51G, Figure S2A) also reduced
the fraction and lifetime of the long-lived population (Figures 1E
and 1F; Table S1). Both epi-illumination and Bessel plane illumi-
nation techniques gave convergent results, ruling out any bias
induced by the imaging modality (Figure S3). To test whether
some artifact might have been introduced to our nonspecific
residence time measurements by the 500 ms acquisition time,
we applied a time-lapse imaging method described by Gebhardt
et al. (2013) to independently characterize the nonspecific chro-
matin-binding events of both halo-Sox2 and a control protein,
halo-NLS. We found that halo-Sox2 and halo-NLS nonspecifi-
cally interact with chromatin with residence times of 0.75
and 0.19 s, respectively (Figures S1F and S1G). Importantly,
the halo-Sox2 nonspecific residence time (0.75 s) derived from
this method agreed well with our 500-ms-long acquisition mea-
correlation spectroscopy (FCS) measurements revealed that
compared to ?23.5% bound halo-Sox2 molecules, only ?3%
halo-NLS molecules are bound to chromatin in the live ES cells
(Figure S4E). Both the shorter nonspecific residence time (0.19
s) and the much smaller bound fraction (?3%) of the control
halo-NLS protein suggest that the shorter-lived (?0.8 s) compo-
nent we observed in our 2D SMT experiments (Figures 1A–1C) is
likely due to intrinsic nonspecific interactions of Sox2 with chro-
matin and cannot be accounted for by the presence of the halo
tag. Together, these observations suggest that the long-lived
component (?12 s) likely reflects the residence time of Sox2 at
specific DNA-binding sites, whereas the short-lived component
is most likely due to nonspecific Sox2 protein-DNA or protein-
To further verify this interpretation, we reconstituted a mini-
mal in-vitro-purified TF system to study Sox2-DNA interaction
kinetics with surface-attached specific and nonspecific DNA
at single-molecule resolution by total internal reflection fluores-
containing a canonical Sox2-binding site is 16.9 s, whereas the
average residence time on a nonspecific probe is 0.9 s (Figures
2A and 2B; Movies S3A and S3B). These numbers are remark-
ably consistent with our ‘‘in vivo’’ residence time measurements,
confirming that our 2D-imaging analysis successfully resolved
TF Target Search Mechanism
To dissect the association kinetics of TF binding to DNA, we
turned to multifocus microscopy with fast 3D SMT (Abrahams-
son et al., 2013) that is able to acquire simultaneous whole-
nucleus imaging in live ES cells (nine focal plates, 4 mm in the Z
direction) at a 33 Hz frame rate (Figures 3A and S5A; Movie
rapid diffusion, DNA/chromatin binding, or a combination of
these activities (Figure 3B). In order to attribute the contributions
of the free and DNA-bound states to the observed trajectories,
we developed a kinetic model based on a previously established
sion and DNA/chromatin binding that was then fitted to our SMT
measurements (for details, see Extended Experimental Proce-
dures; Equations S5, S6, S7, S8, S9, S10, S11, S12, and S13).
Our model assumes that TFs alternate between a freely diffusing
and a chromatin-bound state. We set the dissociation rate from
the bound state (koff) to the average value obtained from our
dwell time histograms (Figure 1C). The observed association
rate (k*on), bound fraction in the whole population (Ceq), and
the diffusion coefficients of the free (Df) and bound species (Db)
were derived through fitting (Figure 3C; Table S1). The diffusion
coefficient for the bound population was in the range of 0.1–
0.2 mm2/s, closely matching the diffusion coefficient of immobile
H2B molecules in the nucleus, consistent with a stable DNA-
bound state (Table S1; Figure S5B). Applying this kinetic model
to Sox2 yielded an average duration of the freely diffusing state
of 1/ k*on?3.7 s (Figure 4C; Table S1), which is the average
time (t3D) Sox2 spends diffusing in the nucleus between two
Using the measured ratio of specific-to-nonspecific binding
events (Figure 1C), we computed the average number of random
trials before a TF molecule reaches a cognate target site (Fig-
ure 4D; Table S1; Equations S14, S15, S16, and S17) and found
that for Sox2, it is ?84 collisions. This means that after leaving a
specific target, a Sox2 molecule on average samples ?83
nonspecific sites in ES cells before reaching another cognate
target site where it stably binds. Combining the information
that includes the number of trials, the nonspecific residence
time, and the t3D(Equation S18), we calculated a search time
of ?377.2 s (Figures 4C and 4D) for Sox2 to find and bind to
a specific recognition site in the chromatin of ES cells. It has
previously been proposed that TF target search follows a one-
dimensional (1D)-3D facilitated diffusion model in which TFs un-
dergo multiple rounds of 1D search (local sliding and hopping)
interspersed with 3D jumps before reaching a specific site
(Berg et al., 1981; Hager et al., 2009; Slutsky and Mirny, 2004).
To test whether 1D sliding might also contribute to nonspecific
Sox2 interactions with DNA, we performed in vitro Sox2 single-
molecule-binding assays with varying lengths of nonspecific
DNA templates (30, 213, and 443 bp). Consistent with a 1D
sliding mode, we observed that Sox2 dwell times increased as
a function of DNA template length (Figures 2B, 2C, and S6B;
Movies S3C and S3D). These in-vitro-binding results suggest
that the in vivo nonspecific chromatin-binding events we
observed in ES cells may also contain a 1D sliding component
that could contribute to a more efficient target search by sam-
pling short stretches of native open chromatin regions (i.e.,
DNAase-hypersensitive sites) during nonspecific collisions
(Figures 2B and 2C).
Complementing SMT with High-Resolution ChIP-Exo
and Sox2 Mutants
To validate our kinetic model, we next performed 2D/3D SMT
experiments with Sox2 mutant proteins. As expected, mutations
interfering with the protein-DNA interaction domain of Sox2
(Sox2M) significantly increased the t3D(from 3.6 to 5.6 s), the
number of encounters (from 83 to 95 trials), and the search
time (from 377 to 603.2 s) (Figure 4C; Table S1). Consistent
1276 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
with these observations, the Sox2M DNA-bound fraction (Ceq=
23.5%) and long-lived bound fraction (Fl= 11.3%) were also
significantly decreased compared with the wild-type (WT) Sox2
protein(Ceq=36.9%,Fl=15.1%) (TableS1). These data suggest
that the number of Sox2M-specific binding sites is substantially
decreased in the cells. By contrast, deletion of the trans-activa-
tion domain (Sox2D, Sox2 1–120 aa) accelerated the t3Dfrom
?3.7 to ?2.3 s and decreased the number of trials from 83 to
?55, which together result in a much shortened search time of
170.2 s. As a result, the Sox2D DNA-bound fraction (Ceq=
43.6%) was also significantly increased (Figure 1F; Table S1).
We next confirmed that the shortened t3D and search times
observed for Sox2D were not merely due to a change in the
DNA association rate (kon) as determined independently by sur-
Figure 2. Distinct Sox2 Residence Time
Distributions at Specific versus Nonspecific
DNA Sequences and Sox2 Sliding on DNA
(A) In vitro single-molecule imaging to determine
Sox2-specific and -nonspecific residence time on
DNA. Single-molecule fluorescence traces of
halo-TMR Sox2 molecule interacting with surface-
tethered WT DNA (top) and mutant DNA (bottom)
(N.F., normalized fluorescence) are presented
along with the corresponding raw images (the red
arrow on the graph indicates the time interval
displayed on top of each plot).
(B) Schematic of Sox2 interaction with WT DNA
probe (30 bp) and different lengths of mutant DNA
probes (30, 213, and 443 bp) measured by in vitro
(C) Left: colocalization of WT DNA probe (top) or
mutant DNA probe (bottom) (cy5 channel) with
TMR-halo-Sox2 molecules (TMR channel). Right:
a histogram of TMR-halo-Sox2 residence time on
30 bp (green), 213 bp (blue), and 443 bp mutant
DNA (gray) and 30 bp WT (red). Mean residence
time of Sox2 on the 30 bp WT DNA is 16.9 s; that
on the 30, 213, and 443 bp mutant DNA are 0.9,
1.6, and 4.5 s, respectively.
See also Figure S6, Movie S3, and Table S1.
face plasmon resonance measurements
(Figure S7). These results taken together
domain deletion mutant binds to many
more ‘‘pseudotarget’’ sites than WT
Sox2 in the nucleus of ES cells.
As an independent andcomplementary
test of the findings from our imaging ex-
periments, we carried out high-resolution
genome-wide mapping of TF-binding
sites for Sox2, Sox2M, and Sox2D with
the recently developed chromatin immu-
noprecipitation exonuclease (ChIP-exo)
technique (Rhee and Pugh, 2011). We de-
tected 6,558 sites bound by halo-Sox2,
Sox2 sites (7,153) detected by a Sox2 antibody. Our ChIP-exo-
derived halo-Sox2-binding enrichment profile also correlated
well with endogenous Sox2-binding sites (Figures 4A and 4B),
suggesting that halo-Sox2 binds to DNA in a manner similar to
WT endogenous Sox2. By contrast, we found only 1,711 sites
bound by the DNA-binding mutant Sox2M, ?3-fold less than the
vation, a DNA sequence motif analysis revealed that, compared
with Sox2, Sox2D binds with greater frequency to degenerate
genome-wide binding studies underscore a remarkable consis-
tency between our single-molecule imaging results and more
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1277
conventional genome-wide bulk binding assays thus confirming
the robustness of the 3D SMT kinetic model. The SMT-imaging
analysispresentedherethereforeenabled usto fullycharacterize
kinetics in live cells.
Hierarchically Ordered Enhanceosome Assembly
DNA-binding site that can serve as a nucleating element for the
assembly of an enhanceosome complex and thus function coor-
(Avilion etal.,2003; Chenetal.,2008;Nicholsetal.,1998;Reme ´-
nyietal.,2003).Oursingle-molecule imaging techniqueprovides
a direct way to dissect a potential order of events during enhan-
ceosome assembly on chromatin/DNA in the nucleus of living
cells. To discern whether Sox2 and Oct4 interact with their tar-
gets in a random or ordered fashion (Hager et al., 2009), we first
designed a 3T3 cell line that stably expressed halo-Sox2 protein
Because the 3T3 cell line does not normally express either Sox2
or Oct4, it was possible to express Sox2 and/or Oct4 without
any background of competing endogenous TFs. In our inducible
tion unit as Oct4 by use of an internal ribosome entry site (IRES)
that provided a convenient way to correlate green fluorescence
intensities in each cell with levels of Oct4 expression. We then
tracked Sox2-binding behavior in cells containing or lacking
Oct4. With Oct4 overexpression, we only observed a modest
increase in Sox2-specific residence time (from 11.6 to 14.1 s);
the Sox2 long-lived bound fraction largely remained the same,
whereas its t3Dand search times showed slight increases (Fig-
ure 5A; Movie S5). These results suggested that Oct4 mainly
helps stabilize the binding of Sox2 to DNA but has little ability to
Figure 3. TF-Chromatin Association Kinetics Determined by Fast 3D Single-Molecule Imaging
(A) Fast 3D tracking of TF movement by simultaneous multifocus microscopy (axial coverage, 4 mm; 33 Hz). The average association rate (k*on) is determined
through fitting the displacement histogram with a 3D kinetic model (for details, see Extended Experimental Procedures; Equations S5, S6, S7, S8, S9, S10, S11,
S12, and S13).
(B) Volume rendering of 3D Sox2 single-molecule image (purple) superimposed with single-molecule trajectories. Three molecules with distinct behaviors were
selectively displayed on the right (from top to bottom: freely diffusing particle, particle undergoing a free/bound transition, and immobile molecule). Color bar
shows the corresponding frame number.
(C) 3D displacement histogram fitted by our 3D kinetic model (Equation S11) for the indicated factor. Histogram bin was set as 30 nm. The t3Dequal to 1=k?
different TFs was calculated through Equations S9, S10, S11, and S12.
See also Figure S5, Movie S4, and Table S1.
1278 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
binding. By contrast, when we performed the converse experi-
ment by inducing Sox2 expression, we found substantial de-
creases in Oct4 t3D(from 3.3 to 2.6 s) and search times (from
366.6 to 158.9 s), as well as significant enhancement of its
long-lived bound fraction (from 10.7% to 18.4%) (Figure 5B;
Movie S6). Taken together, these results support a model in
which Sox2 engages with the chromatin first and primes the
target site for subsequent Oct4 binding (Figure 6E). The Sox2-
assisted binding of Oct4, in turn, appears to help stabilize the
Sox2-Oct4 complex at composite recognition sites. This step-
gests that Sox2 and Oct4 do not form a complex prior to binding
DNA, consistent with a previous report of DNA-dependent inter-
actions between Sox2 and Oct4 in live cells (Lam et al., 2012).
To test if this ordered enhanceosome assembly also occurs in
EScells, wemutatedamino acidson theinteraction surfacesbe-
tween Sox2 and Oct4 (Figures 6A and 6C) that do not directly
affect the ability of Sox2 or Oct4 to bind DNA (Reme ´nyi et al.,
2003). Consistent with our induction experiments, mutations
on Sox2 that disrupted its interaction with Oct4 only moderately
decreased the Sox2 residence time but did not significantly
affect its t3Dand search timesin ES cells. Bycontrast, reciprocal
mutations on Oct4 that impeded Sox2 interactions not only
reduced Oct4 residence times (from 14.6 to 12.66 s) but also
(from 407 to 567.9 s) (Figure 6D).
Next, we performed lentivirus-mediated small hairpin RNA
(shRNA) knockdown (KD) depletion of Sox2 in ES cells that sta-
bly express halo-Oct4 and vice versa. Results from subsequent
SMT experiments after the loss of either Sox2 or Oct4 in ES cells
ceosome assembly process (Figures 6B, 6C, and 6D). Specif-
ically, the loss of Sox2 by KD in ES cells greatly increased
Oct4 t3D (from 3.3 to 4.7 s) and search times (from 407 to
659 s). However, Sox2 search times and t3Dwere only modestly
altered after depletion of Oct4 by KD. It is unlikely that the hier-
archical assembly mechanism we observed here is due to
some artifact of the halo tagging of TFs. First, untagged Sox2
decreased Oct4 search time, whereas untagged Oct4 did not
have the same effect on Sox2, suggesting that this asymmetric
regulation is independent of the halo tag (Figures 5A and 5B).
Also, halo-Oct4 showed a normal pattern of activities during its
target search in ES cells, but as expected, the mutant halo-
Oct4S displayed a compromised search mode (Figure 6D).
Figure 4. A Trial and Error TF Target Search Mechanism
(A) Representative ChIP-exo tracks for Sox2 (antibody [A]), Sox2 (Halo [H]), Sox2D (Halo), and Sox2M (Halo) at two different enhancer regions. Sox2 ChIP-seq
data were previously published by Chen et al. (2008). Autoscale was applied to each track. Chr2, chromosome 2; Chr6, chromosome 6.
(B) Correlation analysis of ChIP-exo data. Upper view is an enrichment heatmap for all factors ranked by descending order of Sox2 (antibody) enrichment
(top 2,000 loci). Bottom view is a heatmap for Sox2 (halo), Sox2D (halo), and Sox2M (halo) showing their own enrichment in the same data set. The read counts
within ±100 bp regions of exo-peaks were taken into account in this analysis.
(C) Calculated duration of the 3D free diffusion state (t3D), the number of trials, the target search time, the ratio that a TF spends in 3D diffusion (3D%), and the
number of binding sites determined from genomic analysis (Exo peaks) for Sox2, Sox2D, Sox2M, and Oct4. The enrichment cutoff for ChIP-exo peak pairs is >12
reads for both the left and the right peak (also see Figure S2C).
(D) The model of an in vivo target search. A TF goes through on average ‘‘n’’ episodes of 3D diffusion (t3D= 2.0–5.6 s, Table S1) interspersed by nonspecific
binding at random accessible chromatin sites (tns= 0.75–0.9 s, Table S1) before reaching a specific site. tsand tsRdenote the search time and the specific
residence time, respectively.
See also Figures S2, S7, and Table S1.
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1279
These results confirm that the ordered binding and cooperativity
among Sox2, Oct4, and DNA contribute to the ability of Oct4 to
execute a productive target search with little or no interference
by the halo tag. Based on our induction experiments in 3T3 cells,
we could estimate the probabilities for each enhanceosome as-
sembly pathway. We determined that minimally, the preferred
pathway (Sox2 first) occurs ?75.3% of the time, whereas the
reverse order (Oct4 first) occurs at most 24.7% of the time (Fig-
ure 6E; Equations S19, S20, S21, S22, S23, S24, S25, and S26).
Epigenetic Regulation of TF Target Search
matin and TF search dynamics, we imaged ES cells that were
treated with either a histone deacetylase (HDAC) inhibitor (tri-
chostatin A [TSA]) or a DNA methylation inhibitor (5-azacytidine
[5-AZA]). Both treatments are expected to open chromatin glob-
ally and thus potentially influence TF search modes: (1) the num-
ber and quality of accessible nonspecific sites are expected to
increase, which could potentially require a greater number of
random trials needed to find a cognate target and thus increase
the search time; (2) the number and quality of accessible specific
binding sites are likely to increase as well, which should limit
the number of trials before a successful engagement and thus
decrease the search time. With both drug treatments, we actu-
ally observed much shortened t3D times, consistent with the
notion that the total number of available DNA-binding sites
increased (Figures 7A and 7B). Importantly, we also observed
significant decreases in the search time (Figures 7A and 7B),
suggesting that an increased accessibility of specific binding
sites and a shortened t3Dtime account for the observed chro-
matin-modification effects on TF search behavior.
TF Dynamics and Temporal Patterns of Target Site
Our data reveal that TFs Sox2 and Oct4 execute a trial and error
Figure 5. Sox2 Assists Oct4 Target Search
(IRES) was under the control of an inducible Cumate switch. Thus, GFP is an expression indicator for Oct4. Left images are of Sox2 2D SMT experiments
performed with GFP negative/positive cells without or with Oct4 induction. Right images are measured values for the long-lived bound fraction of all bound
molecules (from 2D SMT analysis), the specific residence time, the t3D, the number of trials, and the target search time for Sox2. The results from the converse
experiment with Sox2 induction are presented in (B). *p < 0.05; **p < 0.01. Error bars represent SD. See also Movies S5, S6, and Table S1.
1280 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
nonspecific sites in the nucleus of mammalian cells before
acquiring a cognate binding site. These results suggest that in
an ES cell nucleus containing >7,000 potential Sox2-binding
sites (Figure S2C), a Sox2 molecule would require ?6 min
(377 s) to find and bind to a nonallele-specific target. Thus, if
there is only one cognate binding site available in the ES cell
genome, it would take a Sox2 molecule ?31 days to find that
site. From the perspective of DNA sites, if one target needs to
besampledbySox2 every minute,ourdata indicatethat EScells
must contain at least ?40,000 Sox2 molecules to carry out its
function in the appropriate timescale. Simulations of TF dy-
namics based on our SMT data (Figures 7C and 7D) strongly
ofacognatetarget bySox2andOct4molecules wouldbeexqui-
sitely sensitive to their nuclear concentrations. To better under-
stand Sox2 and Oct4 assembly dynamics at target sites, we
combined FCS measurements and western blot analysis to
quantify the endogenous Sox2 concentrations in live ES cells
(Figure S4). We found that Sox2 concentrations are in the micro-
molar range (?0.73 mM). If we assume that the ES cell nucleus
approximates a 10 3 10 3 5 mm ellipsoid, one ES cell nucleus
would generally have ?115,000 Sox2 (protein) molecules (Equa-
tion S30). Thus, one cognate Sox2-Oct4 site is sampled by Sox2
approximately every 24s(Equation S31).The Sox2-specific resi-
dence times are about 12–16 s, and nonspecific residence time
is about 0.7–0.9 s. These results suggest a 50%–70% Sox2 tem-
Equations S32 and S33). In short, these measurements and cal-
culations informusof theexquisite and differential concentration
dependence of key TFs in ES cells.
Mechanisms of TF Target Search
The TF target search process for binding to endogenous single-
copy genes has been the subject of considerable interest and
theoretical discussion over the last ?40 years, but direct mea-
surements particularly in animal cells have proven elusive
(reviewed in Halford, 2009; Zakrzewska and Lavery, 2012). By
harnessing apowerful combinationofrecentadvancesinmolec-
ular imaging, the multifaceted and cross-validated measure-
ments of TF behavior reported here offer insights into several
long-standing TF target search questions: we first describe a
Figure 6. Hierarchically Ordered Enhanceosome Assembly
(A) Structural illustration of mutations (Sox2O and Oct4S) that selectively disrupt Sox2-Oct4 interaction surface. Crystal structures of Sox2 HMG domain (blue)
andOct1 POUdomain (red)bindingtotwo different enhancer DNAs (yellow)(FGF4, 1GT0; UTF1,1O4X;Reme ´nyi etal.,2003)arepresentedinacartoon model by
respectively. Notably, these mutations interfere with Sox2-Oct4 interaction in both conformations.
(B) Sox2 and Oct4 western blot analysis to examine Sox2 KD and Oct4 KD efficiencies in ES cells. Tubulin (Tub) served as a loading control.
(C and D) Changes in the long-lived fraction of all bound events, the long-lived residence time, the t3D, the number of trials, and the search time of Sox2 and Oct4
after the indicated perturbation experiment. Sox2O and Oct4Sare as described in(A); Oct4 KD and Sox2 KDare as described in (B). Errorbars representSD. *p<
0.05; **p < 0.01.
(E) Sox2 and Oct4 assemble on DNA in an asymmetrically regulated fashion in live cells. We calculated the probability for each reaction route based on Oct4 and
Sox2 DNA dissociation and association kinetic information obtained from the 3T3 cell induction experiments (see Equations S19, S20, S21, S22, S23, S24, S25,
and S26 for calculation details).
See also Table S1.
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1281
interactions; these studies also describe several key kinetic fea-
tures (3D diffusion periods, number of trials, search times, resi-
dence times, etc.) associated with in vivo TF dynamics. Broadly
consistent with previously proposed facilitated diffusion models
(Berg et al., 1981; Hager et al., 2009; Slutsky and Mirny, 2004),
our imaging data indicate the involvement of at least two kineti-
a dominant phase involving multiple intervals of 3D diffusion
events (tns= 0.75–0.9 s) prior to finding a cognate recognition
site. Our in vitro single-molecule results suggest that Sox2 likely
spends at least some time nonspecifically sliding along naked
DNA searching for its target. If we assume that some periods of
nonspecific TF chromatin binding (tns) reflect the 1D search
time (t1D), the time partition for the 3D search is then calculated
to be 72%–88%, significantly greater than 50%—the 3D search
time partition that gives rise theoretically to the fastest target
search (Slutsky and Mirny, 2004). It is worth noting that live-
cell two-photon FCS measurements independently confirmed
nated by 3D diffusion. Thus, in contrast to single-molecule
studies in E. coli in which the LacI TF was found to spend more
than 90% of its time nonspecifically bound to and diffusing along
DNA in a 1D search (Elf et al., 2007; Hammar et al., 2012), our re-
sults record a significantly greater contribution of 3D diffusion to
search strategies/mechanisms given their distinct nuclear envi-
ronment and genomic landscape relative to bacteria. Indeed,
reducing 3D search time partition or/and number of trials—thus
linking changes in chromatin to TF function in a manner that
Figure 7. Epigenetic Modulation of TF Dynamics and Target Site Occupancy
(A and B) Changes in specific residence time, t3D, number of trials, and search time of Sox2 and Oct4 after TSA and 5-AZA treatment in ES cells. ES cells were
treated with 50 nM TSA for 6 hr or 5mM 5-AZA for 12 hr before the imaging experiment. **p < 0.01. Error bars represent SD.
(C) Simulation heatmap to illustrate the relationship among TF residence time, TF concentration (binding site-sampling frequency), and TF target site temporal
occupancy. For each pair of mean TF residence time (x axis) and sampling frequency (y axis), 1,000 continuous binding events were simulated and then the
temporal occupancy of the target site was calculated based on the total binding on-off durations. The expected temporal occupancies of Sox2 at specific and
nonspecific sites derived based on data from Figure S4 were marked with the indicated black lines. The percentage of temporal occupancy is presented in ‘‘Jet’’
color map. See ‘‘TF Dynamics Simulation’’ in the Extended Experimental Procedures for details of the parameter setup.
(D) Individual simulation tracks of distinct TF dynamics regimes (indicated in A) are presented. Specifically, track 1 represents high-frequency nonspecific
chromatin binding. Track 5 represents the TF temporal occupancy of the target site by a TF with a relatively long TF residence time but at low concentrations
in the cells.
See also Figure S4 and Table S1.
1282 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
has no likely parallel in bacteria. Together, these results give us a
glimpse of TF target search dynamics in mammalian cells and
how its evolutionarily adapted mechanisms may operate in a
developmental context (Figure 4).
Despite the obvious differences between bacterial and
mammalian TF search mechanisms, our findings suggest
that the less-dominant 1D sliding mode may, nevertheless,
contribute to an efficient target search process. The haploid
mouse genome is ?2.5 Gbp (Mouse Genome Sequencing Con-
sortium et al., 2002) and bears ?7,000 potential specific Sox2-
binding sites. If we assumed that only1%–2% of the mammalian
genome is accessible to TF binding due to nucleosome imped-
ance according to hypersensitive site mapping experiments
(Natarajan et al., 2012; Sabo et al., 2006), it would still require
Sox2 to effectively scan and sample a DNA length of 43–86 bp
(2:5Gbp=7000ðsitesÞ=84ðtrialsÞ31% ? 2%) during each non-
specific chromatin collision to reach a nonallele-specific target
within ?84 trials. Our in vitro single-molecule experiments re-
the naked DNA nonspecifically searching for its target. If we as-
sumethatSox2 hassimilar sliding speedsasLacI(43105bp2/s)
(Hammar et al., 2012) or P53 (2.6 3 106bp2/s) (Tafvizi et al.,
2008), during each nonspecific binding transaction (0.7–0.9 s),
Sox2 would cover the expected sliding length (43–86 bp) multi-
ple times to ensure a specific target site recognition and acqui-
sition. We note that multiple rounds of sliding and sampling
DNA have been reported as requirements for efficient TF target
site recognition, consistent with our calculations (Hammar
et al., 2012).
Biological Implications of TF Function in Stem Cells
Critical lineage-commitment events during animal development
must be exquisitely timed and orchestrated by dynamic molec-
ular transactions of multiple TFs in the nucleus. However, to
date, there have been few adequate tools to characterize the
kinetic features that can discriminate specific from nonspecific
TF chromatin-binding events at endogenous diploid loci in living
cells. Equally important has been a paucity of reliable means to
determine the order of multiple TF-binding events in live
mammalian cells. The desirability of single-cell single-molecule
imaging and tracking is particularly acute because the complex
multistep process of transcription is highly stochastic with dy-
namic behaviors relevant to function that are difficult to discern
by conventional ensemble methods.
With the suite of new imaging modalities adopted here, we
sembly at endogenous loci in live mammalian cells with minimal
perturbation. Unexpectedly, we observed that Sox2 is the lead
TF that assists Oct4 to assemble on its in vivo targets in a hierar-
chically ordered mechanism. Oct4, in turn, appears to stabilize
the Sox2/Oct4 complex on what are presumably composite
cognate DNA recognition sites. We note that although Sox2
evidently performs a rather important function as a lead TF, we
are not sure whether it qualifies, strictly speaking, as a ‘‘pioneer’’
factor because it is difficult to reconcile a 12 s residence time
with presumably highly stable chromatin interactions with
condensed mitotic chromosomes that are a defining property
of pioneer factors.
The dynamic behavior of individual TFs in living cells revealed
by our studies establishes that the sampling frequency of a
cognate target by Sox2 and Oct4 molecules would be exqui-
sitely sensitive to their nuclear concentrations in order to
achieve coordinated cobinding at composite target sites to
nucleate enhanceosome assembly and trigger the appropriate
transcriptional programs underlying pluripotency (Figure 7).
Specifically, upon ES cell differentiation, the concentrations of
both Sox2 and Oct4 become gradually downregulated. Conse-
quently, the target site-sampling frequency will be reduced
accordingly. A critical time and TF concentration would be
reached when a bound Sox2 molecule would no longer meet
an Oct4 molecule within the duration of its average residence
time on chromatin (?12 s). The direct consequence is that a
Sox2-Oct4-dependent enhanceosome could no longer be
formed efficiently, and the maintenance of the ES cell self-re-
newing pluripotent transcriptional circuitry would be disrupted.
Thus, this critical tip in the balance of just a few TFs would
then lead to cellular differentiation. We can also now explain
how it is that merely reducing the Oct4 concentration but largely
preserving Sox2 levels in ES cells is sufficient to trigger a major
change in the pluripotency program given the distinct division of
labor and yet codependence between Sox2/Oct4. These data
also revealed that TFs with different target search and chro-
matin-binding kinetics would display distinct temporal patterns
of target site occupancy (Figure 7). These results thus highlight
the importance of dissecting the behavior and kinetic parame-
ters intrinsic to different types of TFs as they carry out critical
time-dependent activities in living cells. These essential mea-
surements of TF search and binding modes provide new in-
sights into the dynamic and stochastic behavior of TF-DNA
transactions that form the basis for the production of mRNA
outputs and cell fate determination.
A key aspect of this study is that we complemented SMT ex-
periments with a range of genetic loss-of-function studies to
probe biologically relevant regulatory aspects of TF behavior.
For example, we gained new insights into how TF target search
dynamics could be affected by the presence of trans-activation
domains, by the specific interplay of two functionally linked
TFs (i.e., Sox2/Oct4), and by changes in the global epigenetic
state of the cell. A combination of these studies directly un-
masked the molecular underpinnings that give rise to a specific
target search mode. For instance, deletion of the Sox2 transcrip-
tion activation domain substantially decreased the target search
time by reducing the number of nonspecific binding trials, sug-
gesting that removal of a transcription activation domain might
lead to reduced nonspecific decoy protein-protein tethering
and thus significantly truncate the target search process. This
surprising finding also suggests that from the standpoint of
maximal efficiency in the search process, TFs with multiple
high-affinity activation domains might actually be at a disadvan-
tage for target search. Thus, one might posit that combinatorial
regulation may in part derive from the need for different TFs to
assume distinct and complementary targeting duties: with
search but perhaps limited in diversity of activation domains,
whereas other partner factors like Oct4 may be more suited for
executing multifaceted interactions with coactivators and the
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1283
core promoter complex, etc. that require a greater panoply of
activation surfaces but at the expense of a less efficient search
capacity. Our studies also provide evidence that the acetylation
alter the search mode of TFs, suggesting that epigenetic regula-
tion may be critical in affecting DNA binding and also influence
the TF target search pattern.
Potential Impact on Interpreting Genome-wide
Our data revealed several general and fundamental aspects
of TF binding to endogenous chromatin in live cells that may
impact the way we interpret results from widely adopted ChIP
experiments (ChIP-qPCR, ChIP sequencing [ChIP-seq], and
ChIP-exo). Specifically, our SMT data establish that nonspecific
chromatin binding via either protein:protein or protein:DNA
transactions is a major component of the TF target search pro-
cess. Although the nonspecific binding residence time of Sox2
(0.75–0.9 s) is 15-fold shorter than its residence time at specific
sites (?12 s), when TF concentrations are relatively high as
we have measured in ES cells (in the micromolar range), even
nonspecific sites will be sampled frequently enough to be effi-
ciently crosslinked by commonly used agents such as formal-
dehyde (Figures 7 and S4). Thus, high-frequency sampling
explain the large number of nonconsensus decoy or false
TF-binding sites detected by ensemble assays such as those
reported by ENCODE Project Consortium et al. (2012) and Ger-
stein et al. (2012). Many putative target sites identified by such
genome-wide binding studies when tested more rigorously
were indeed found to be nonfunctional (Fisher et al., 2012).
The preponderance of transient nonspecific interactions de-
tected by SMT is also consistent with the proposed weak pro-
tein-protein-tethering mechanisms reported by the ENCODE
Project Consortium et al. (2012) and Gerstein et al. (2012).
Indeed, our Sox2 ChIP-exo data generated using ultrahigh
sequencing depth (40 million mapped reads) also detected
many weak or degenerate Sox2-binding sites in ES cells (Fig-
ure S2C). These SMT experiments, thus, underscore the need
to be cautious in interpreting high-throughput ensemble-binding
assays with respect to identifying meaningful or functionally
relevant binding interactions.
Culture Condition for Live-Cell Imaging
ES cell-imaging experiments were performed in the imaging medium:
Dulbecco’s modified Eagle’s medium (DMEM) without phenol-red (Invitrogen),
15% FBS, 1 mM GlutaMAX, 0.1 mM nonessential amino acids, 0.1 mM 2-mer-
captoethanol, and 1,000 U of LIF. We used a Tokai-hit PI live-cell chamber and
GM-8000 digital gas mixer to maintain cell-culturing condition (37?C, 5% CO2,
and humidity) during the imaging experiment.
Stable Cell Line Generation
Stable cell lines were generated by cotransfection of ES cells or 3T3 cells with
the PiggyBac vector and a helper plasmid that overexpresses PiggyBac trans-
posase (Supper PiggyBac Transposase; System Biosciences). At 48 hr post-
transfection, cells were subjected to G418 (Invitrogen) selection (500 mg/ml).
After a week of selection, cells were maintained in their culturing medium
with a 250 mg/ml final concentration of G418.
Live-Cell Single-Molecule Imaging
2D single-molecule experiments were conducted on a Nikon Eclipse Ti micro-
scope equipped with a 1003 oil-immersion objective lens (N.A., 1.4). 3D SMT
experiments were performed on the same instrument as 2D. The multifocus
optical elements are appended after the primary image plane. The details for
the multifocus microscopy instrumentation were described in our previous
dures for the procedures of single-molecule localization and tracking.
In Vitro TIRF Single-Molecule Imaging
TIRF microscope-imaging system setup and fluorescent molecule spots’
colocalization analysis were essentially as described previously (Revyakin
et al., 2012). See the Extended Experimental Procedures for experimental
details and imaging analysis.
The ChIP-exo library was prepared by following the published protocol with
TruSeq Small RNA System. Primer information is in Table S1. See the Extended
Experimental Procedures for the procedures of genomic data analysis.
The Gene Expression Omnibus (GEO) accession number for the genomic data
reported in this paper is GSE54103.
figures, six movies, and one table and can be found with this article online at
Z.Z. and L.L. contributed to this work equally. Z.Z. and L.L. performed in vitro
single-molecule and ChIP-exo experiments, respectively.
We thank Xavier Darzacq, Carl Wu, Robert Singer, Yick W. Fong, Wulan Deng,
Michael Levine, and Sean B. Carroll for general discussion and proof reading
the manuscript, Yick W. Fong for the Sox2 shRNA vector, Florian Mueller for
John Macklin for the assistance of FCS, and C. Morkunas and S. Moorehead
Received: September 20, 2013
Revised: December 16, 2013
Accepted: January 27, 2014
Published: March 13, 2014
Abrahamsson, S., Chen, J., Hajj, B., Stallinga, S., Katsov, A.Y., Wisniewski, J.,
Mizuguchi, G., Soule, P., Mueller, F., Dugast Darzacq, C., et al. (2013). Fast
multicolor 3D imaging using aberration-corrected multifocus microscopy.
Nat. Methods 10, 60–63.
Avilion, A.A., Nicolis, S.K., Pevny, L.H., Perez, L., Vivian, N., and Lovell-Badge,
R. (2003). Multipotent cell lineages in early mouse development depend on
SOX2 function. Genes Dev. 17, 126–140.
Ben-Tabou de-Leon, S., and Davidson, E.H. (2007). Gene regulation: gene
control network in development. Annu. Rev. Biophys. Biomol. Struct. 36, 191.
Berg, O.G., Winter, R.B., and von Hippel, P.H. (1981). Diffusion-driven mech-
anisms of protein translocation on nucleic acids. 1. Models and theory.
Biochemistry 20, 6929–6948.
1284 Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc.
Chen,X.,Xu,H.,Yuan,P.,Fang,F.,Huss,M.,Vega,V.B.,Wong,E.,Orlov,Y.L., Download full-text
Zhang, W., Jiang, J., et al. (2008). Integration of external signaling pathways
with the core transcriptional network in embryonic stem cells. Cell 133,
Elf, J., Li, G.W., and Xie, X.S. (2007). Probing transcription factor dynamics at
the single-molecule level in a living cell. Science 316, 1191–1194.
ENCODE Project Consortium, Bernstein, B.E., Birney, E., Dunham, I., Green,
E.D., Gunter, C., and Snyder, M. (2012). An integrated encyclopedia of DNA
elements in the human genome. Nature 489, 57–74.
Fisher, W.W., Li, J.J., Hammonds, A.S., Brown, J.B., Pfeiffer, B.D., Weisz-
mann, R., MacArthur, S., Thomas, S., Stamatoyannopoulos, J.A., Eisen,
M.B., et al. (2012). DNA regions bound at low occupancy by transcription
factors do not drive patterned reporter gene expression in Drosophila. Proc.
Natl. Acad. Sci. USA 109, 21330–21335.
Fudenberg, G., and Mirny, L.A. (2012). Higher-order chromatin structure:
bridging physics and biology. Curr. Opin. Genet. Dev. 22, 115–124.
Gao, L., Shao, L., Higgins, C.D., Poulton, J.S., Peifer, M., Davidson, M.W., Wu,
X., Goldstein, B., and Betzig, E. (2012). Noninvasive imaging beyond the
diffraction limit of 3D dynamics in thickly fluorescent specimens. Cell 151,
Gebhardt, J.C., Suter, D.M., Roy, R., Zhao, Z.W., Chapman, A.R., Basu, S.,
Maniatis, T., and Xie, X.S. (2013). Single-molecule imaging of transcription
factor binding to DNA in live mammalian cells. Nat. Methods 10, 421–426.
Gerstein, M.B., Kundaje, A., Hariharan, M., Landt, S.G., Yan, K.K., Cheng, C.,
Mu, X.J., Khurana, E., Rozowsky, J., Alexander, R., et al. (2012). Architecture
of the human regulatory network derived from ENCODE data. Nature 489,
Hager, G.L., McNally,J.G.,and Misteli, T. (2009). Transcriptiondynamics.Mol.
Cell 35, 741–753.
Halford, S.E.(2009).Anendto40yearsofmistakes inDNA-proteinassociation
kinetics? Biochem. Soc. Trans. 37, 343–348.
Hammar, P., Leroy, P., Mahmutovic, A., Marklund, E.G., Berg, O.G., and Elf, J.
(2012). The lac repressor displays facilitated diffusion in living cells. Science
Lam, C.S., Mistri, T.K., Foo, Y.H., Sudhaharan, T., Gan, H.T., Rodda, D., Lim,
L.H., Chou, C., Robson, P., Wohland, T., and Ahmed, S. (2012). DNA-
dependent Oct4-Sox2 interaction and diffusion properties characteristic of
the pluripotent cell state revealed by fluorescence spectroscopy. Biochem.
J. 448, 21–33.
Levine, M., and Tjian, R. (2003). Transcription regulation and animal diversity.
Nature 424, 147–151.
Mazza, D., Abernathy, A., Golob, N., Morisaki, T., and McNally, J.G. (2012). A
benchmark for chromatin binding measurements in live cells. Nucleic Acids
Res. 40, e119.
Mirny, L., Slutsky, M., Wunderlich, Z., Tafvizi, A., Leith, J., and Kosmrlj, A.
diffusion. J. Phys. A. Math. Theor. 42, 434013.
Mouse Genome Sequencing Consortium, Waterston, R.H., Lindblad-Toh, K.,
Birney, E., Rogers, J., Abril, J.F., Agarwal, P., Agarwala, R., Ainscough, R.,
Alexandersson, M., et al. (2002). Initial sequencing and comparative analysis
of the mouse genome. Nature 420, 520–562.
Mueller, F., Stasevich, T.J., Mazza, D., and McNally, J.G. (2013a). Quantifying
transcription factor kinetics: at work or at play? Crit. Rev. Biochem. Mol. Biol.
Natarajan, A., Yardimci, G.G., Sheffield, N.C., Crawford, G.E., and Ohler, U.
(2012). Predicting cell-type-specific gene expression from regions of open
chromatin. Genome Res. 22, 1711–1722.
Nichols, J., Zevnik, B., Anastassiadis, K., Niwa, H., Klewe-Nebenius, D.,
Chambers, I., Scho ¨ler, H., and Smith, A. (1998). Formation of pluripotent
stem cells in the mammalian embryo depends on the POU transcription factor
Oct4. Cell 95, 379–391.
Planchon, T.A., Gao, L., Milkie, D.E., Davidson, M.W., Galbraith, J.A., Gal-
braith, C.G., and Betzig, E. (2011). Rapid three-dimensional isotropic imaging
of living cells using Bessel beam plane illumination. Nat. Methods 8, 417–423.
Reme ´nyi, A., Lins, K., Nissen, L.J., Reinbold, R., Scho ¨ler, H.R., and Wilmanns,
M. (2003). Crystal structure of a POU/HMG/DNA ternary complex suggests
differential assembly of Oct4 and Sox2 on two enhancers. Genes Dev. 17,
Revyakin, A., Zhang, Z., Coleman, R.A., Li, Y., Inouye, C., Lucas, J.K., Park,
S.R., Chu, S., and Tjian, R. (2012). Transcription initiation by human RNA
polymerase II visualized at single-molecule resolution. Genes Dev. 26, 1691–
Rhee, H.S., and Pugh, B.F.(2011). Comprehensive genome-wide protein-DNA
interactions detected at single-nucleotide resolution. Cell 147, 1408–1419.
Sabo, P.J., Kuehn, M.S., Thurman, R., Johnson, B.E., Johnson, E.M., Cao, H.,
Yu, M., Rosenzweig, E., Goldy, J., Haydock, A., et al. (2006). Genome-scale
mapping of DNase I sensitivity in vivo using tiling DNA microarrays. Nat.
Methods 3, 511–518.
Slutsky, M., and Mirny, L.A. (2004). Kinetics of protein-DNA interaction: facili-
tated target location in sequence-dependent potential. Biophys. J. 87, 4021–
Tafvizi, A., Huang, F., Leith, J.S., Fersht, A.R., Mirny, L.A., and van Oijen, A.M.
(2008). Tumor suppressor p53 slides on DNA with low friction and high stabil-
ity. Biophys. J. 95, L01–L03.
Tam, P.P., and Loebel, D.A. (2007). Gene function in mouse embryogenesis:
get set for gastrulation. Nat. Rev. Genet. 8, 368–381.
Zakrzewska, K., and Lavery, R. (2012). Towards a molecular view of transcrip-
tional control. Curr. Opin. Struct. Biol. 22, 160–167.
Cell 156, 1274–1285, March 13, 2014 ª2014 Elsevier Inc. 1285