Skip to main content
Advertisement
  • Loading metrics

Cooperative DNA Recognition Modulated by an Interplay between Protein-Protein Interactions and DNA-Mediated Allostery

  • Felipe Merino,

    Affiliations Computational Structural Biology Group, Department of Cell and Developmental Biology, Max Planck Institute for Molecular Biomedicine, Münster, Germany, Center for Multiscale Theory and Computation, Westfälische Wilhelms University, Münster, Germany

  • Benjamin Bouvier,

    Current address: Laboratoire de Glycochimie, des Antimicrobiens et des Agroressources, CNRS FRE3517, Amiens, France

    Affiliation Bioinformatics: Structures and Interactions, Bases Moléculaires et Structurales des Systèmes Infectieux, Univ. Lyon I/CNRS UMR5086, IBCP, Lyon, France

  • Vlad Cojocaru

    vlad.cojocaru@mpi-muenster.mpg.de

    Affiliations Computational Structural Biology Group, Department of Cell and Developmental Biology, Max Planck Institute for Molecular Biomedicine, Münster, Germany, Center for Multiscale Theory and Computation, Westfälische Wilhelms University, Münster, Germany

Abstract

Highly specific transcriptional regulation depends on the cooperative association of transcription factors into enhanceosomes. Usually, their DNA-binding cooperativity originates from either direct interactions or DNA-mediated allostery. Here, we performed unbiased molecular simulations followed by simulations of protein-DNA unbinding and free energy profiling to study the cooperative DNA recognition by OCT4 and SOX2, key components of enhanceosomes in pluripotent cells. We found that SOX2 influences the orientation and dynamics of the DNA-bound configuration of OCT4. In addition SOX2 modifies the unbinding free energy profiles of both DNA-binding domains of OCT4, the POU specific and POU homeodomain, despite interacting directly only with the first. Thus, we demonstrate that the OCT4-SOX2 cooperativity is modulated by an interplay between protein-protein interactions and DNA-mediated allostery. Further, we estimated the change in OCT4-DNA binding free energy due to the cooperativity with SOX2, observed a good agreement with experimental measurements, and found that SOX2 affects the relative DNA-binding strength of the two OCT4 domains. Based on these findings, we propose that available interaction partners in different biological contexts modulate the DNA exploration routes of multi-domain transcription factors such as OCT4. We consider the OCT4-SOX2 cooperativity as a paradigm of how specificity of transcriptional regulation is achieved through concerted modulation of protein-DNA recognition by different types of interactions.

Author Summary

Pluripotent stem cells can give rise to all somatic lineages. When taken out of the context of the embryo they can be maintained and for this a core transcriptional regulatory circuitry is crucial. OCT4 and SOX2, two factors of this network, are also critical for the induction of pluripotency in somatic cells. In pluripotent cells, OCT4 and SOX2 associate on DNA regulatory regions, enhancing or modifying each other's sequence specificity. In contrast, in the early stages during induction of pluripotency, it was proposed that OCT4 explores the genome independent of SOX2. Here we report the mechanism by which SOX2 influences the orientation, dynamics, and unbinding free energy profile of OCT4. This involves an interplay of protein-protein interactions and DNA-mediated allostery. We consider that this mechanism enables OCT4 to use its DNA binding domains and the interaction partners available in a certain biological context to access alternative genome exploration routes. This study enhances the understanding of the context specific function of OCT4 and provides a general perspective on how DNA-binding cooperativity is modulated by different types of interactions.

Introduction

Transcription factors recognize short DNA sequences found in the regulatory regions of genes. In eukaryotic cells, a large number of biologically irrelevant binding sites are present due to the large size of their genomes. In addition, different transcription factors can share DNA specificities, due to homology or convergence. Therefore, the correct choice of gene targets has to rely on a more sophisticated mechanism than pure DNA-binding specificity. To increase gene regulation specificity, regulatory elements contain a high number of tandem transcription factor binding sites, known as enhanceosomes. Notably, only specific combinations of proteins can bind them to control gene expression [1,2]. Due to the close proximity of their binding sites within the enhanceosomes, transcription factors bind cooperatively, modifying their affinities for these particular sites and creating a transcriptional response that is both highly specific and sensitive [1]. Moreover, their protein-protein interactions can evoke latent DNA specificities, causing them to occupy binding sites rarely bound by the isolated protein [3]. Interestingly, transcription factors can even bind cooperatively in the absence of physical interaction between them due to DNA-mediated allostery [4,5,6]. Despite recent advances in understanding the atomic structure of some enhanceosomes [7], the structural details behind enhanceosome assembly are still poorly understood.

Multiple DNA-binding domains can also regulate the specificity and affinity of modular transcription factors through the increase in the length of their sequence specific binding sites [1]. Furthermore, the intrinsic difference in the DNA-binding affinities of the individual domains allows these proteins to employ alternative DNA-recognition mechanisms [8]. For instance, the members of the POU family are characterized by a DNA-binding region composed of two independent DNA-binding domains, a POU specific (POUS) and a POU homeodomain (POUHD), which are connected by a flexible linker. Both domains contain a helix-turn-helix fold from which one helix docks into the major groove of the DNA, establishing the majority of the sequence specific protein-DNA contacts [9]. In addition, the N-terminal end of the POUHD contains a disordered region that docks into the minor groove and contributes significantly to DNA binding [10]. Combined, the two domains recognize the consensus sequence ATGC(A/T)AAT, where the POUS recognizes the first half and the POUHD the second. The POU factors are involved in the control of a wide variety of biological processes [11]. In particular, the POU factor OCT4 lies in the core of the transcriptional network controlling the maintenance and induction of stem cell pluripotency [10]. Together with other transcription factors in this network, it mediates enhanceosome assembly in pluripotent stem cells [12].

Whole genome chromatin immuno-precipitation (ChIP) analysis in embryonic stem cells have shown that, predominantly, OCT4 binds to DNA in combination with SOX2 (consensus binding site: C(T/A)TTGTT), to a composite motif made by the juxtaposition of their individual binding sites (canonical motif) [13]. They bind cooperatively to this motif forming a protein-protein interaction interface in which only the POUS of OCT4 interacts with SOX2 [14,15]. Moreover, although the OCT4-SOX2 interaction occurs only upon binding to DNA [16], SOX2 assists the in vivo DNA recognition process of OCT4 [12]. Interestingly, they also bind cooperatively to composite motifs with different spacing between their individual sites [14]. Experiments with chimeric proteins have shown that depending of the composite motif, either the POUS or the POUHD are the most relevant for binding, suggesting that SOX2 can influence their relative contribution to the binding affinity [17,18]. In addition, ChIP experiments suggest that OCT4 binds to DNA alone during the initial stages of reprogramming to pluripotency [19]. Therefore, OCT4 may employ alternative DNA-recognition mechanisms depending on the cellular and genomic context. Whereas the atomic structure of some POU-SOX complexes and many POU homodimers bound to semi-palindromic sites are known [15,20,21,22], the structural basis for the inter-molecular communication of these proteins is still not understood. Interestingly, nuclear magnetic resonance (NMR) studies have demonstrated that co-binding with SOX2 to the HOXB1 enhancer, which contains a consensus canonical motif, modifies the DNA-binding mechanism of the OCT4 homolog OCT1, by altering the way in which the individual domains scan the DNA [8].

UTF1 is a key coactivator in pluripotent cells. The enhancer of this gene contains a canonical motif under the control of the OCT4-SOX2 combination [18]. In human, the sequence at the 3' end of the OCT4 binding site (5'-CATTGTTATGCTAGC-3') lacks part of the sequence-specific POUHD binding site, making the binding of the POUHD to this site partly unspecific. Interestingly, the ability to recognize this sequence strongly correlates with the ability to maintain the pluripotent state in stem cells [23], suggesting that the recognition of degenerate sequences is a key component of OCT4's biological function.

Recently we investigated the OCT4-SOX2 interface on an idealized canonical motif by classical molecular dynamics simulations [24]. Whereas in that study we focused on the protein-protein interface, in this work we explore how OCT4 recognizes degenerate binding sites such as that found in the UTF1 enhancer and the role of SOX2 in this process. For this, we used unbiased molecular dynamics simulations combined with simulations of protein-DNA unbinding and free energy profiling. More generally, we aimed at understanding how DNA-binding cooperativity involving transcription factors with multiple DNA binding domains is modulated. As a result, we provide a mechanism by which the function of such transcription factors may acquire cellular and genomic context specificity.

Results

SOX2 modifies the orientation and dynamics of the DNA-bound configuration of OCT4

To characterize the OCT4-DNA interfaces we performed 2 ensembles of 1.8 μs of unbiased simulations each, one for the OCT4-UTF1 complex and one for the OCT4-SOX2-UTF1 complex (Fig 1A). Each ensemble is composed of 4 independent, 450-ns-long simulations. Unless otherwise specified, all results are derived from the ensemble analysis.

thumbnail
Fig 1. Interactions and structural changes at the protein-DNA interfaces.

(A) The OCT4-SOX2-UTF1 complex. (B) Per-residue number of recurrent protein-DNA contacts. The top plot shows the number of contacts in a model of the OCT4-SOX2-HOXB1 (consensus sequence) complex as reference. The middle and bottom plots show the data from the unbiased simulations of the OCT4-DNA and OCT4-SOX2-DNA complexes respectively. The gray boxes highlight the 8 helices of OCT4. α1 – α4 correspond to the POUS, α6 – α8 to the POUHD. See also S1 and S2 Figs.

https://doi.org/10.1371/journal.pcbi.1004287.g001

To study the protein-DNA interactions, we created contact maps containing all the atom-atom contacts within a distance threshold of 4.5 Å (S1A Fig). We then defined as recurrent contacts those formed in more than 50% of the total simulation time (S1 Table). In the absence of SOX2, most of the recurrent POUS-DNA contacts (S1 Table) cluster around the docking helix, consistent with the orientation of this domain when bound to consensus sequences (Figs 1B and S2A). For the POUHD, most of the recurrent contacts involve the N-terminal tail of the domain (S1 Table), mainly due to the insertion of R95 and R97 into the minor groove (Figs 1B and S2A). Notably, although the globular region of the POUHD remains bound to the major groove (S1B Fig), it forms only few recurrent contacts with the DNA (Figs 1B and S2A, S1 Table). In contrast, when OCT4 binds to consensus sequences, the residues V139, N143, and Q146 from the docking helix of the POUHD form sequence-specific contacts with the DNA bases at the 3' end of binding site (top plot in Fig 1B). As these bases are different in the UTF1 sequence, N143 and Q146 contact the DNA through non-stable interactions with the DNA backbone.

In the presence of SOX2, the OCT4-SOX2 interface is formed, mainly through the hydrophobic contacts between I21 from helix α1 of the POUS and the SOX2 residues A61 and M64, as well as some transient electrostatic interactions with the linker region of OCT4 (S2B Fig). These protein-protein contacts have only minor effects on the POUS-DNA contact map (Figs 1B and S1A), including a small decrease in the number of contacts of T45 with the DNA bases (Figs 1B, S2B and S2C) and of residues in helices α1 and α4 with the DNA backbone (Figs 1B, S2B and S2D). Importantly, the POUS-DNA interface is similar to the one observed when bound to consensus sequences, irrespective of the presence of SOX2. On the other hand, SOX2 modifies the POUHD-DNA contact map (S1A Fig) even in the absence of a direct interaction. SOX2 induces the formation of several recurrent interactions between the POUHD and the DNA backbone both in the tail and globular part of POUHD (Figs 1B and S2, S1 Table). This suggests that an allosteric communication between the POUS-SOX2 interface and the POUHD contributes to the OCT4-SOX2 cooperativity.

To explore the dynamics of the POUS and POUHD domains relative to their binding sites, we calculated the orientation of the docking helices (Fig 1A) around the helical axis (Rock) and inside the binding groove (Tumble) (Fig 2A and 2B). Consistent with the small number of recurrent POUHD-DNA contacts observed in the absence of SOX2 (Fig 1B, S1 Table), the binding orientation of the POUHD fluctuates more than that of the POUS (Fig 2C and 2D). When SOX2 is present, there is a 14% decrease in the fluctuation of the POUS orientation (S2 Table) calculated from the distributions of the Rock and Tumble angles (Fig 2C and 2E). However, the effect of SOX2 on the POUS orientation and dynamics is subtle and further sampling may be necessary for its correct quantification. In addition, SOX2 induces a reorientation and a decrease in the dynamics of the POUHD (Fig 2D and 2F). The diagonal pattern in the Rock versus Tumble histogram (Fig 2F) suggests that SOX2 couples the motion of the POUHD to the major groove of the DNA, reflecting the increased number of protein-DNA interactions of the globular region of the POUHD in the presence of SOX2 (Fig 1B, S1 Table). Importantly, these results were found to be consistent in all individual simulations from the ensembles (S3 Fig).

thumbnail
Fig 2. Orientational dynamics in DNA-bound configurations.

(A,B) Schematic view of the coordinate system (A) and definition (B) of the Rock and Tumble angles describing the orientation of the docking helices (shown as opaque cartoons) of the two domains of OCT4. Rock-Tumble histograms in the absence (C,D) or presence (E,F) of SOX2 for the POUS (C,E) and POUHD (D,F). See also S3 Fig.

https://doi.org/10.1371/journal.pcbi.1004287.g002

SOX2 and OCT4 communicate through DNA-mediated allostery

Our results suggested that the POUHD and the OCT4-SOX2 interface communicate through an allosteric signal. To explore this, we calculated a positional cross-correlation matrix from the simulations. For consistency, we superimposed OCT4 and its binding site in the binary and ternary complexes. In the absence of SOX2, the helices α2 and α3 of the POUS and the N-terminal tail of the POUHD are correlated with the OCT4-binding site (Fig 3A). In addition, the helices α6 and α7 of the POUHD are anticorrelated with the POUS binding site, whereas the helix α8 was only slightly correlated with its binding site. When SOX2 is present, the correlations of DNA regions with the POUS are extended, whereas those involving the POUHD tail and globular part are diminished (Fig 3B). Notably, this correlation pattern in which the tail of the POUHD correlates with the same DNA region as the POUS rather than the globular region of its own domain suggests that OCT4 may be divided into three units with independent roles in DNA binding.

thumbnail
Fig 3. Correlated motions and allosteric communication pathways.

(A,B) Positional cross-correlation between OCT4 and the DNA in the absence (A) or presence (B) of SOX2. Each docking helix is marked with a star. The two DNA strands are labeled with “+” and “-”. (C,D) Shortest communication paths between K57 from SOX2 and K117 from OCT4 (C) and between M64 from SOX2 and K117 from OCT4 (D). The size of the edges between nodes is proportional to the number of paths crossing them. The shortest path is shown in blue while the suboptimal paths in light brown. See also S4 Fig.

https://doi.org/10.1371/journal.pcbi.1004287.g003

To elucidate the pathway used to propagate the allosteric signal from the POUS-SOX2 interface to the POUHD, we performed a network analysis on the trajectories of the ternary complex. For this, a reduced representation of the protein and DNA is generated by defining nodes to represent groups of atoms. Two of these nodes are connected if within the atoms represented by them, an atom-atom pair stays within 4.5 Å for more than 75% of the simulation time. The distance between two connected nodes in this network reflects their positional cross-correlation calculated from the simulation. Finally, optimal signal propagation pathways between two distant nodes are estimated by minimizing their separation distance which is computed by adding the distances between connected nodes along the pathway (see Methods for details). For this, we generated a new cross-correlation matrix, superimposing OCT4, SOX2 and the composite motif. Independent of the end-points, the shortest communication pathways between SOX2 and the POUHD do not cross the SOX2-POUS interface nor the partially structured linker peptide of OCT4, but propagate through either the DNA or the POUS-DNA interface (Fig 3C and 3D). The optimal path travels from SOX2 to DNA and then cross the POUS-DNA interface and through the DNA reach the tail of the POUHD. An alternative path that only threads through the DNA to reach the POUHD tail connects M64SOX2 to K117OCT4 (Fig 3D). Combined, these results suggest that the allosteric interaction between SOX2 and OCT4 is mediated mainly by changes in the DNA structure induced by their DNA-binding domains. Interestingly, the globular region and the tail of the POUHD belong to different communities (S4 Fig) which are regions of the network wherein the correlation between the nodes is higher than to the rest of the network. This is in agreement with the correlation pattern calculated by superimposing only OCT4 and its binding site (Fig 3A and 3B), adding further evidence that the two regions of the POUHD may function independently in DNA recognition.

OCT4 and SOX2 modify DNA structural properties

To explore the DNA structural changes induced upon binding of OCT4 and SOX2 and how these may contribute to cooperativity and DNA-mediated allostery, we performed two additional ensembles of 1.5 μs unbiased simulations each, one for the SOX2-UTF1 complex and one for the free UTF1 DNA. Each ensemble was composed of 2 independent, 750-ns-long simulations. Then, we analyzed changes in the major (Fig 4A) and minor (Fig 4B) groove widths and axis bending (Fig 4C) in all the simulations relative to the free DNA.

thumbnail
Fig 4. Effect of OCT4 and SOX2 binding on the DNA structural properties.

(A) Major groove width (B) Minor groove width (C) Axis bending. On the left schematic representations of analyzed properties are drawn. The plots in the middle show the average values with standard errors (see Methods), whereas the plots on the right the inter base pair correlations. The data was collected from the four ensembles of unbiased simulations (free DNA, and SOX2-DNA, OCT4-DNA, OCT4-SOX2-DNA complexes).

https://doi.org/10.1371/journal.pcbi.1004287.g004

SOX2 modifies the structure of the DNA by binding to the minor groove and inserting a methionine side chain between two base pairs, which significantly bends the DNA [22] (Fig 4C). In the absence of OCT4, SOX2 widens the major groove, and changes the inter base pair correlation of the major groove width in the region of the POUS binding site (Fig 4A). These changes may enhance the DNA-binding affinity of the POUS in the presence of SOX2. In addition, SOX2 very slightly narrows the major groove at the end of the POUS and beginning of the POUHD binding sites, indicating that the binding of SOX2 propagates a signal through the DNA structure up to eight base pairs away from the most distorted base pair of its own binding site. Remarkably, the combined effect of OCT4 and SOX2 involves mainly changes in the inter base pair correlations of all DNA-structural properties analyzed (Fig 4). Importantly, only the ternary complex shows correlations between the SOX2 and POUHD binding sites. While this effect could be POUS-mediated, it is most likely due to the DNA-mediated communication observed in the network analysis (Fig 3).

OCT4 also modifies the structure of the DNA, causing alternate narrowing and widening of the major groove (Fig 4A), and an overall narrowing of the minor groove (Fig 4B). This is consistent with the binding of its globular domains to the major groove and the preference of the highly positive N-terminal tail of homeodomains for narrow minor grooves due to the strong negative electrostatic potential found in these regions of DNA [25]. In addition, the POUHD slightly bends the last bases of the POU binding site (Fig 4C). Notably, although the presence of SOX2 modifies the DNA-bound configuration of the POUHD, it does not affect the POUHD-induced changes in the structure of the composite motif. Similarly, it has been reported that other POU factors bend the DNA [26]. However, the isolated POUHD did not bend DNA, and therefore the bending was attributed to the POUS or the interaction between the two domains.

SOX2 influences the unbinding profiles of both OCT4 domains

To characterize the unbinding process of OCT4, we performed umbrella sampling simulations to dissociate each domain of OCT4 from the DNA in the absence and presence of SOX2. For this, we used the minimal interatomic distance between the pulled domain and the DNA (dmin) as collective variable to describe the dissociation process. We only simulated the unbinding process when the other domain remained bound to the DNA because simulations with both domains detached are unlikely to converge on a reasonable timescale. To monitor the unbinding process we calculated how the recurrent and the non-stable (formed in less than 50% of the simulation time) interactions break during the simulations.

Irrespective of the domain analyzed or the presence of SOX2, most of the contacts were lost between 3.1 and 3.5 Å minimal distance separation (Fig 5). This suggests that the OCT4-UTF1 interface is dominated by hydrogen bond interactions. The POUS domain detaches from the DNA in a cooperative fashion, where all the recurrent interactions break simultaneously in the region with dmin between 3.0 and 3.2 Å (Fig 5A). The presence of SOX2 moves the upper limit of this region further to 3.4 Å (Fig 5A). On the other hand, SOX2 has no effect on the non-stable contacts between the POUS and the DNA (Fig 5B).

thumbnail
Fig 5. Unbinding profiles of the OCT4 domains.

The change in the number of protein-DNA contacts is shown during the unbinding simulations while pulling the POUS (A,B) or the POUHD (C-F). (C,D) Globular region of the POUHD. (E,F) N-terminal tail of the POUHD. The contacts were divided into recurrent (A,C,E) and non-stable (B,D,F). The fraction of recurrent contacts was calculated using the number of recurrent contacts from the unbiased simulations as reference.

https://doi.org/10.1371/journal.pcbi.1004287.g005

Similar to the unbinding of the POUS, the unbinding of the globular region of the POUHD is very cooperative, with all interactions breaking simultaneously (Fig 5C and 5D). Most of the recurrent contacts between the POUHD tail and the DNA break in the region between 3.0 and 3.45 Å when SOX2 is absent (Fig 5E). Interestingly, the presence of SOX2 slightly decreases the lower limit of this region to 2.9 Å. In addition, there is an increase in the number of non-stable contacts in the region between 3.5 and 4.0 Å (Fig 5F). This shows that the protein-DNA interactions formed by the tail break at a larger separation than those formed by the globular region. The difference in the unbinding process of the POUHD tail and the globular region is in agreement with our observation that these regions are independent in the correlation and communities analysis of the unbiased simulations (Fig 3A, 3B, and S4).

Next, we analyzed the effect of the unbinding of one domain of OCT4 on the domain that remained bound to DNA (Fig 6). The unbinding of the POUHD has no impact on the number of recurrent or non-stable contacts of the POUS (Fig 6A and 6B). On the other hand, the unbinding of the POUS has a strong effect on the DNA interaction of the globular region of the POUHD when SOX2 is present (Fig 6C and 6D). A significant decrease in the number of recurrent interactions is accompanied by an increase in non-stable interactions. This suggests that the POUHD is reorienting, but not detaching from the DNA. Indeed, the analysis of the Rock and Tumble angles during dissociation further confirms this (S5 Fig). Importantly, the absence of this phenomenon in the simulation of the OCT4-UTF1 complex suggests that this reorientation is induced by the allosteric communication between the POUHD and SOX2. Conversely, the unbinding of the POUS does not affect the number of recurrent and non-stable contacts between the POUHD tail and the DNA (Fig 6E and 6F).

thumbnail
Fig 6. Interactions with the DNA of the remained-bound domain.

(A,B) Change in recurrent (A) and non-stable (B) POUS-DNA contacts when pulling the POUHD. (C-F) Change in recurrent (C,E) and non-stable (D,F) contacts between the globular region (C,D) and the N-terminal tail (E,F) of the POUHD when pulling the POUS. The fraction of recurrent contacts defined as in Fig 4. See also S5 Fig.

https://doi.org/10.1371/journal.pcbi.1004287.g006

Unbinding of either domain of OCT4 induces distal changes in DNA structure

To study the DNA structural changes upon unbinding of the OCT4 domains, we monitored the relaxation of the minor groove width and the bending angle for each base pair along the unbinding simulations.

In the absence of SOX2, the unbinding of the POUS produces a small widening of the minor groove in the region between the SOX and POUS binding regions (Fig 7A) without affecting the bending angle (Fig 7B). Conversely, the unbinding of the POUHD has a strong effect on the DNA structure. Around a minimal distance of 3.25 Å, the minor groove width at the 3' region of the POUHD binding site increases towards the average value from ideal B-DNA (~ 5.4 Å) (Fig 7C), while the POUHD-induced bending of the DNA disappears as the POUHD unbinds (Fig 7D). In addition, in the region between 3.5 and 4.5 Å, the detachment of the POUHD tail from the DNA widens the minor groove at the beginning of the POUHD binding site (Fig 7C).

thumbnail
Fig 7. Changes in DNA structure during the biased simulations.

The effect of the unbinding on the conformation of the composite DNA motif is shown in the absence (A-D) or the presence (E-H) of SOX2. The pulled domain was the POUS (A,B,E,F) or the POUHD (C,D,G.H). The properties analyzed were the minor groove width (A,C,E,G) and the bending of the DNA axis (B,D,F,H). For each plot, the average obtained from the unbiased simulations is shown as a color bar reference.

https://doi.org/10.1371/journal.pcbi.1004287.g007

Strikingly, when SOX2 is present, the changes in DNA structure induced by unbinding of both domains of OCT4 are affected. For instance, the POUS-induced narrowing of the minor groove seen in the absence of SOX2 (Fig 7E) is no longer present. Instead, when the POUS unbinds from the DNA, the minor groove deformations induced by the POUHD tail and SOX2 propagate into the POUS binding region (Fig 7E). In addition, there is an increase in the POUHD bending angle (Fig 7F), suggesting that the POUS modulates the changes in DNA structure induced by SOX2 and the POUHD. The presence of SOX2 also significantly amplifies the widening of the minor groove induced by the unbinding of the POUHD tail (Fig 7G), in agreement with the increase of non-stable contacts observed between this region and the DNA (Fig 5F). Remarkably, the unbinding of either domain from OCT4 causes a decrease in the bending angle induced by SOX2 (Fig 7F and 7H). Although this effect is stronger during the unbinding of the POUS (Fig 7F), it is also present when the POUHD dissociates (Fig 7H). Therefore, the changes in DNA structure induced upon unbinding of the POUS and POUHD propagate away from their binding regions, further demonstrating that the communication between the POUS-SOX2 interface and the POUHD is DNA-mediated.

SOX2 affects the relative DNA-binding strength of the OCT4 domains

To quantify the effect of SOX2 on the DNA-binding affinity of OCT4, we calculated the unbinding free energy profile for each of its domains from the biased simulations. Then, we calculated a macroscopic binding free energy from the total probability of finding the domains in the bound or unbound configurations (see Methods and S1 Text).

Because an experimentally measured affinity of OCT4 for this enhancer is not available, a direct comparison with the absolute affinities calculated here is not possible. However, an OCT4-SOX2 cooperativity of -1.56 kcal/mol has been measured for an idealized canonical motif [14,24], similar to the OCT1-SOX2 cooperativity of -1.7 kcal/mol measured for the enhancer of HOXB1 [27]. Assuming that the presence of SOX2 neither modifies the cooperativity between the POUS and the POUHD, nor the effect of the linker between them, we can calculate the OCT4-SOX2 cooperativity from our simulations as the sum of the SOX2-induced changes in the affinity of both domains (See S1 Text). For this, we tested several definitions of the bound-unbound threshold (Table 1).

thumbnail
Table 1. Estimates of the DNA-binding affinity of the POUS and the POUHD.

https://doi.org/10.1371/journal.pcbi.1004287.t001

As all our free energy profiles show a sharp transition close to the hydrogen bond distance threshold (3.2–3.4 Å) (Fig 8), which reflects the breaking of most recurrent contacts around the same values (Fig 5), we consider that the bound-unbound transition is well defined at thresholds between 3.3 and 3.5 Å. Using these values, we calculated cooperativities of -2.0 and -2.59 kcal/mol respectively (Table 1), in good agreement with experiments [14,24,27,28]. Lower values for the threshold are not appropriate as they correspond to the steep part of the free energy profiles, whereas higher values still provide the correct sign for the cooperativity although they overestimate it, possibly due to the poor sampling of the unbound state (Table 1, S6 Fig). Therefore, we focus on the results obtained with threshold values of 3.3 and 3.5 Å. We stress that the comparison with experiments is only approximate, as neither the DNA-binding element nor the ionic strength were the same.

thumbnail
Fig 8. Unbinding free energy profiles.

(A) POUS. (B) POUHD. The black and red curves show the profiles in absence and presence of SOX2 respectively. The dotted vertical lines mark the different bound-unbound thresholds tested (Table 1.). The shaded lines represent the error calculated as described in Methods. See also S6 Fig.

https://doi.org/10.1371/journal.pcbi.1004287.g008

In the absence of SOX2, the POUHD binds to DNA stronger than the POUS by 1.8–2.2 kcal/mol despite the reduced number of sequence-specific POUHD-DNA interactions (Table 1, Fig 8). This is consistent with the affinities measured for the isolated domains of OCT1, where the POUHD binds 2.9 kcal/mol stronger than the POUS to a DNA with consensus sequence [29]. Although the difference between our estimation and the experimental value in the case of OCT1 may be attributed to differences between the two proteins, it may also reflect the absence of sequence-specific POUHD-DNA interactions in the OCT4-UTF1 complex. The presence of SOX2 results in an increase in the unbinding free energy profile of the POUS (Fig 8A), corresponding to an increase in DNA-binding affinity of 2.4–2.7 kcal/mol (Table 1). This agrees with the shift towards a higher minimal distance for the breaking of recurrent POUS-DNA contacts in the presence of SOX2 (Fig 5A). Importantly, SOX2 also modifies the unbinding free energy profile of the POUHD (Fig 8B). This involves a small decrease in the region between 3.2 and 3.75 Å of the free energy profile combined with an increase in the region between 3.75 and 4.0 Å (Fig 8B). The first effect coincides with the shift in the breaking of the recurrent POUHD tail-DNA contacts (Fig 5E), whereas the second correlates with the increase of non-stable POUHD tail-DNA contacts (Fig 5F) and the widening of the DNA minor groove (Fig 7G). Overall, SOX2 decreases the DNA-binding strength of the POUHD by 0.4 kcal/mol at an unbound/bound threshold of 3.3 Å or 0.15 kcal/mol at 3.5 Å (Table 1). Consequently, the DNA-binding strength of the POUS becomes larger by 0.7–1.0 kcal/mol than that of the POUHD in the presence of SOX2 (Table 1). Interestingly, the strong effect of SOX2 on the unbinding free energy profile of POUS corresponds to a modest effect on its orientation and dynamics, whereas a large effect on the POUHD dynamics corresponds to a smaller effect on its unbinding free energy profile (Figs 2 and 8). Remarkably, the increase in the DNA-binding affinity of the POUS is larger than the measured OCT4-SOX2 cooperativity. The estimated cooperative binding free energy approaches the experimental value only with the additional decrease in the POUHD affinity (Table 1), which further supports the finding of an allosteric component that modulates the OCT4-SOX2 cooperativity and suggests that the allostery has a slight detrimental effect on the cooperativity.

Our estimates of the DNA-binding affinity are higher than expected for typical transcription factors, and thus unlikely to represent the real values. In principle, if we account for the inter-domain cooperativity and the effect of the linker peptide, we can compare our estimations with the previously measured affinities of the POUS and POUHD domains of OCT1 which are -7.9 and -10.8 kcal/mol respectively [29]. These values only account for the inter-domain cooperativity. Although we cannot compare directly the effects of the linker peptides of OCT4 and OCT1 due to their different length and structure [20] (See S1 Text), the much lower affinities measured for the isolated domains of OCT1 indicate that we overestimate the absolute binding free energies

The analysis of the consistency in the density of states shows that we achieved convergence at minimal distances lower than 4 Å, but the quality of the sampling decreases at high minimal distances (S6 Fig). This is expected, since the volume of the conformational space increases significantly with the protein-DNA separation distance and is therefore a common problem in this type of calculations [30]. A confinement scheme in which conformational, positional and rotational restraints are used and removed after the induced dissociation has successfully been used to alleviate such issues in protein-ligand systems [30,31,32]. However, the very high degeneracy of RMSD-based conformational restraints makes this approach unlikely to converge in our case, for which both partners are very large and flexible. The imperfect convergence at higher separation as well as the presence of the second domain bound to the DNA may explain the inaccuracy in the estimation of binding free energies. However, our estimates for the OCT4-SOX2 cooperativity as well for the difference in the affinities of the POUS and POUHD in the absence of SOX2 are in good agreement with experiment, suggesting that the error in the calculation is of similar magnitude among the different free energy profiles.

Discussion

Our simulations demonstrate that the mechanism by which OCT4 recognizes the DNA is modified by SOX2. Interestingly, a similar phenomenon has been described for OCT1. NMR studies have shown that the POUS of OCT1 is involved in hopping between DNA segments, while the POUHD scans the DNA through 1D sliding. Interestingly, co-binding with SOX2 to the HOXB1 enhancer increases the affinity of the POUS locking it on its cognate site [27,33]. This has also been inferred from steered molecular dynamics of the complex between SOX2 and the POUS from OCT1 [34]. As a result, the POUHD is now the most likely to transfer to another region of the DNA through a “slide and transfer” mechanism [33]. From our simulations, we observed that SOX2 also locks the POUS of OCT4 on its binding site, as its DNA binding affinity increases by 2.4–2.7 kcal/mol (Table 1). In addition, we show that the presence of SOX2 generates an allosteric signal that propagates through DNA, couples the POUHD motions with the major groove of the DNA (Fig 2) and likely decreases the POUHD affinity beyond that of the POUS (Table 1). This suggests that the POUHD of OCT4 also adopts the exploratory role when SOX2 is present. Therefore, our findings suggest that, by modulating the affinity and specificity of the individual domains, SOX2 can affect the mechanism by which many POU factors recognize the DNA. However, a mechanism involving DNA-mediated allostery has not been yet described for any other POU-SOX2 complex. Further research is necessary to understand if such a mechanism is common among POU factors. Nevertheless, the POU-SOX2 cooperativity is likely to change the identity of their target genes, and the transcriptional programs they promote.

For OCT4, this regulatory mechanism may have important functional implications. A ChIP-based analysis has suggested that OCT4 does not bind in combination with SOX2 during the early stages of reprogramming to pluripotency [19]. Remarkably, the analysis of the binding sites shows that most of the DNA specificity comes from the POUHD. On the other hand, the same analysis in pluripotent cells shows that the POUS has a much stronger specificity than the POUHD [13]. Therefore, the selection among different DNA recognition mechanisms induced by SOX2 may give cellular and genomic context specificity to the biological function of OCT4.

Our simulations show that the unbinding of the POUS involves the breaking of all the protein-DNA contacts simultaneously (Fig 5A and 5B). On the other hand, the POUHD shows a modular behavior where the domain can be further sub-divided into the N-terminal tail and the globular region. When binding to a consensus sequences, the globular region is the one that contributes most to the DNA specificity, because its docking helix forms sequence-specific contacts with DNA bases. However, this is likely to be a small contribution to the total DNA-binding affinity given that estimates of the affinity of some homeodomains for unspecific sequences have shown that they can strongly bind the DNA in the absence of their cognate site (Kd ~ 300 nM) [35]. Notably, the interaction between the N-terminal tail and the minor groove seems to determine part of the DNA-binding specificity of homeodomains [25], since different tail sequences prefer DNA sequences with different minor groove widths. In turn, this correlates with the sequence preferred by the globular region of the homeodomain. Furthermore, protein-protein interactions involving the N-terminal tail of homeodomains are known to drastically change their sequence preferences, thus evoking latent specificities [3]. Theoretical and experimental studies have shown that the tail of homeodomains has a major contribution to the overall binding affinity and is key for the DNA-recognition mechanism, as it can speed up the binding process [36,37]. We showed that SOX2 modifies the dynamics of the POUHD by enhancing its interaction with the DNA backbone through an allosteric signal that propagates through its N-terminal tail (Fig 3C). It is possible that this changes the sequence selectivity of OCT4, thus promoting the binding to degenerated binding sites such as the UTF1 sequence. Similarly, the presence of SOX2 allows the OCT4 homolog OCT1 to bind a composite motif where the POUHD half-site has been removed [38]. In addition, our findings suggest that this region contributes the most to the affinity of the POUHD. Interestingly, the POUHD tail serves as the nuclear localization signal of OCT4 [39], and the region around it is subject to several post-translational modifications known to alter its biological activity [10]. Therefore, the allosteric communication is likely to be a key aspect of the regulation of OCT4's function in vivo.

A key difference between OCT4 and most POU factors is the presence of a significantly more structured linker peptide that contains a defined helix (α5). Previously, we have demonstrated that the structure of this region is important for protein-protein interactions [20,24]. Thus, our initial hypothesis was that this region serves as a communication route between the two domains of OCT4 and between OCT4 and SOX2 in ternary complexes. However, the network analysis shows that the communication occurs mainly through the DNA and not the protein (Figs 3C and 3D and S4). This suggests that the allosteric communication pathway is not OCT4-specific and may represent a general mode of communication between POU and SOX proteins.

Furthermore, the unbinding of the domains reveals subtle modifications of the DNA structure that propagate to the adjacent biding sites, consistent with a DNA-mediated allosteric signal that is likely to modify protein-DNA binding affinities. Interestingly, a similar effect has been described for other assemblies of transcription factors. For instance, the expression of interferon-β depends on the cooperative assembly of 8 proteins to a 55 bp binding site [5]. Structural studies have suggested that their cooperativity arise from DNA-mediated effect, as their protein-protein interfaces are very small and flexible [7,40]. In addition, allosteric effects through the DNA have been systematically explored using synthetic binding sites, and they have been shown to be key to the binding and activity of other transcription factors [4,6]. In one case, it has been observed that an interplay between direct and allosteric interactions is key for the assembly of protein-protein-DNA complexes [41]. However, allostery accounted for a positive contribution to DNA-binding cooperativity and has not been related to the selection of alternative DNA-recognition mechanisms.

In the case of the OCT4-SOX2 interaction upon binding to the UTF1 enhancer, the allosteric component may have a small negative contribution to the cooperativity. Therefore, DNA-mediated allostery can determine the cooperativity of proteins that bind to DNA not close enough to form physical interactions or modulate cooperativity in an interplay with protein-protein interactions. The latter provides a mechanism to modify DNA exploration pathways, which in turn may give an additional layer of specificity to enhanceosome assembly and transcriptional regulation.

Methods

Molecular dynamics simulations

We built models of the OCT4-SOX2-UTF1 ternary complex with the human UTF1 sequence (5'-CAGGCATTGTTATGCTAGCGGAACTCC-3') using our previous models of OCT4-SOX2 complexes bound to a consensus canonical motif (HOXB1 enhancer) [20]. These were built based on the structure of the OCT1-SOX2-HOXB1 complex (pdbid 1O4X) [15] and the OCT4-OCT4-DNA homodimer complex [20] (see S1 Text for details). Two alternative models of the OCT4-SOX2-UTF1 complex, identical in the DNA-binding interface but slightly differing in the unstructured part of the linker region, were chosen to provide slightly different strating coordinates. These and the corresponding models of the OCT4-UTF1 complex were equilibrated in explicit TIP3P water and 150 mM NaCl under periodic boundary conditions (see S1 Text for details). Typically, the systems contained ~ 100.000 atoms. Four independent, 450 ns-long simulations of the ternary complex and the corresponding simulations of the binary complex were performed in the canonical (NPT) ensemble. Two simulations were performed for each model, each of them with different initial velocities, using a standard protocol in NAMD [42] (details in S1 Text).

To generate the starting configuration for the simulations of the SOX2-UTF1 complex and the free UTF1 DNA, OCT4 and 13 Cl- ions compensating for the OCT4 net charge of +13 were stripped from the starting configurations of the solvated OCT4-SOX2-UTF1 and OCT4-UTF1 systems respectively. The newly generated SOX2-UTF1 and UTF1 systems were equilibrated (see S1 Text for details) and two independent 750-ns-long simulations, with different initial velocities, were performed using the same protocol as described above.

For proteins and DNA we used the amber force field [43] modified for DNA (ff99) [44] and proteins (ff99SB) [45] with further corrections for protein side-chains (ILDN) [46], protein backbone (NMR-based) [47] and DNA backbone [48]. For the ions we used the Smith-Dang parameters [49]. The integration step was 1.5 fs and coordinates were saved every 3 ps.

Unbinding free energy profiles

We performed umbrella sampling simulations to calculate the free energy profiles for the unbinding of the OCT4 domains from the DNA. Most of the errors in unbinding free energy profiles come from the incomplete sampling of the unbound state [30]. In the case of multi-domain proteins such as OCT4, obtaining converged sampling of the state in which all domains are dissociated from the DNA is particularly challenging. Therefore, we only estimated the free energy profiles for the dissociation of one domain, while the other remained bound. We defined the POUHD as residues 95 to 152 and the POUS domain as residues 1 to 88, which includes the helical region of the linker (helix α5; residues 76 to 88) (Fig 1A). The collective variable chosen to describe the dissociation process was the minimal interatomic distance between the pulled domain and the DNA (dmin) [50]. In each window, dmin was restrained by a biasing potential of the form where is dmin at the center of the window. When at least one protein-DNA heavy atom pair has a distance shorter than then all atom-atom pairs ij with distances dij below this threshold are subjected to the biasing potential. Otherwise, only the pair ij with the minimal distance is biased. This definition of the potential limits the bias imposed on the dissociation mechanism. While no atom pair can be closer than the current value of dmin, atom pairs can potentially be further apart, which does allow for progressive, hierarchical unbinding. This is a reasonable assumption considering that short-range Van der Waals contacts typically break before stronger long-range electrostatic interactions. Indeed, such hierarchical behavior has been seen in previous studies on other protein-DNA [50] and protein-protein [51] systems using the same enhanced sampling methodology.

Each domain was pulled away from the DNA in the presence and absence of SOX2. We started the first window from a configuration taken after 57.8 ns of unbiased simulation. For each free energy profile, we used 60 windows centered every 0.05 Å, covering a range for dmin between 2.55 and 5.5 Å.

To minimize the equilibration time, each biased simulation was started after 4.5 ns of simulations of the previous window. For each window we performed 22.5 ns of biased simulation, summing up to 1.35 μs simulation time per domain, per system. We used a force constant of 300 kcal /mol ∙ Å2 for most windows. For the unbinding of the POUHD in the absence of SOX2, we used a force constant of 250 kcal /mol ∙ Å2 at separations above 4.75 Å, given that the free energy profile is essentially flat, and large biases are no longer necessary. For all further analysis, the initial 6 ns of simulation in each window were discarded as equilibration.

The free energy profiles were reconstructed from the umbrella sampling simulations using the weighted histogram method (WHAM) [52]. The error associated with each profile was calculated as described in [53]. The convergence of the simulations was assessed by comparing the density of states calculated for each pair of consecutive windows (See S1 Text and [53] for details).

A macroscopic binding free energy was calculated from the free energy profile, using the total probability of the domain being either bound or unbound. From this, the affinity can be calculated as, where ρ(dmin) can be estimated from the relation ρ(dmin) ∝ exp(G(dmin)/kB T).

Given the absence of a clear transition state in the free energy profiles, we defined the boundary between the bound and unbound states at different values of dmin. As an upper integration limit we always used 5.5 Å. Importantly, due to the exponential nature of the integration, the lower region of the unbound state contributes the most to the final result, and the choice of the upper integration limit does not change the results significantly. However, the definition of the integration limits is arbitrary and can modify the final result [54]. The errors on the computed binding free energies were obtained from the errors on the PMF by applying linear error propagation theory to the equation used to calculate ΔG (see above), as implemented in the ‘Uncertainites’ Python package (http://pythonhosted.org/uncertainties).

Analysis of structural properties

The structural properties of the DNA were analyzed using Curves+ [55]. The standard errors were generated using the block averaging procedure [56].

All other properties were analyzed using VMD [57]. To measure the orientation of the binding domains on the DNA, we created a DNA-based coordinate system as described previously [24]. We defined vx as the vector between the centers of mass of the first and last base pairs of the OCT4 binding site, vt as the vector between the backbone of the bases from the first base pair of the OCT4 binding site, vz as the cross product of vx and vt, and vy as the cross product of vx and vz. The “Rock” angle is the angle between the axis of the docking helix of the domain analyzed and vy projected on the vy/vz plane. The “Tumble” angle is the angle between the axis of the docking helix of the domain analyzed and vx projected on the vx/vz plane (Fig 2A and 2B). The conformational volume sampled inside the two-dimensional rock/tumble subspace was estimated from a principal component analysis of the rock/tumble data sets for the POUS for which the corresponding population density is quasiharmonic. The ratio of conformational volumes sampled in the absence and presence of SOX2 was computed as the ratio of the areas of the corresponding confidence ellipses, which only depends on the product of the square roots of the covariance matrix eigenvalues if the same (arbitrary) confidence threshold is chosen in both cases.

The average and standard errors for the structural properties of the biased simulations were calculated without re-weighting. Although this influences the absolute values of the properties, it will not affect the comparative analysis.

Network analysis

We performed a contact network analysis with the VMD network plugin [58] which uses Carma [59] to calculate positional cross-correlations. For this, we defined nodes on the Cα and Cβ of the proteins. For prolines, glycines, and alanine only one node on the Cα was defined. For DNA we defined two nodes per nucleotide, one representing the backbone centered on the C3' atom and one representing the base centered on N3 in adenines and guanines, or C4 in cytosines and thymines. This selection of nodes is representative of the description of the dynamics of the backbone, protein side-chains and DNA bases and represents both the major and minor groove faces of DNA. To define the edges of the network, an atom-atom contact map was calculated where only those contacts present in more than 75% of the simulation time were kept. A contact was defined as a pair of atoms separated by 4.5 Å or less. Then, an edge was added between two nodes when at least 1 atom-atom contact exists between the atoms represented by the beads. Edges were not allowed within the same protein residue and between neighboring protein and DNA backbone beads. To explore possible communication pathways between the POUHD and SOX2, we calculated the shortest collection of paths between the POUS-SOX2 and POUHD-DNA interfaces. Here, the inter-nodal distance dmn is defined as dmn = -log|Cmn|, where Cmn is the positional cross-correlation coefficient between two nodes of the network. Then, we included only paths with inter-nodal distances within 5 of the optimal path.

Supporting Information

S1 Text. Supporting document with detailed methods.

References are cited with their numbers either from the main article or from the additional list at the end of this document which contains those not cited in the main article.

https://doi.org/10.1371/journal.pcbi.1004287.s001

(PDF)

S1 Table. Number of protein-DNA stable contacts seen in the unbiased simulations of the OCT4-UTF1 and OCT4-SOX2-UTF1 systems

https://doi.org/10.1371/journal.pcbi.1004287.s002

(DOC)

S2 Table. Principal component analysis of the Rock-Tumble data sets for the POUS domain

https://doi.org/10.1371/journal.pcbi.1004287.s003

(DOC)

S1 Fig. OCT4-DNA contacts during the unbiased simulations.

(A) Average number of OCT4-DNA contacts per-residue. The graph on top shows the number of protein-DNA contacts present in a model of the OCT4-SOX2-HOXB1 complex. The gray boxes highlight the 8 helices of OCT4. α1 – α4 correspond to the POUS, while α6 – α8 to the POUHD. (B,C) Evolution of the number of DNA contacts made by the globular region of the POUHD in the absence (B) and presence (C) of SOX2 during the six independent 450 ns-long unbiased simulations. See also Fig 1.

https://doi.org/10.1371/journal.pcbi.1004287.s004

(TIF)

S2 Fig. Recurrent OCT4-DNA and the OCT4-SOX2 interactions during the unbiased simulations.

(A,B) Recurrent protein-DNA interactions of the POUS and POUHD in the absence (A) and presence (B middle, right) of SOX2. (B) Protein-protein recurrent contacts at the OCT4-SOX2 interface (left). (C,D) SOX2-induced changes in recurrent protein-DNA interactions with the DNA bases (C) or backbone (D) mapped on the structure of the OCT4-SOX2-UTF1 complex. The color scale shows the difference in recurrent contacts, measured as Q+SOX2—Q−SOX2. See also Fig 1.

https://doi.org/10.1371/journal.pcbi.1004287.s005

(TIF)

S3 Fig. Contribution of each simulation to the overall orientational dynamics in the DNA-bound configurations.

Rock-Tumble measurements in the absence (C,D) or presence (E,F) of SOX2 for the POUS (C,E) and POUHD (D,F). The black lines represent the histograms shown in Fig 2. and are located at 4, 40, 400, 4000 counts. See also Fig 2.

https://doi.org/10.1371/journal.pcbi.1004287.s006

(TIF)

S4 Fig. Network analysis of correlated motions.

The subnetwork communitites are shown in different colors. View (B) corresponds to view (A) rotated by 180° around the axis shown. See also Fig 3.

https://doi.org/10.1371/journal.pcbi.1004287.s007

(TIF)

S5 Fig. Effect of the unbinding of the domains of OCT4 on the orientation of the domain that remains bound to the DNA.

Effect of the unbinding of the POUHD on the orientation of the POUS (A,B). Effect of the unbinding of the POUS on the orientation of the POUHD (C,D). The simulations were performed in the absence (A,C) or presence (B,D) of SOX2. The points show the average and the standard deviation of the Rock and Tumble values from each umbrella window. The color scale represents the protein-DNA separation of the domain being pulled. See also Fig 6.

https://doi.org/10.1371/journal.pcbi.1004287.s008

(TIF)

S6 Fig. Sampling inconsistency, θ1,2, between successive umbrella windows as a function of the inter-partner distance.

(A,B) POUS in the absence (A) or presence (B) of SOX2. (C,D) POUHD in the absence (C) or presence (D) of SOX2. See also Fig 8.

https://doi.org/10.1371/journal.pcbi.1004287.s009

(TIF)

Acknowledgments

We thank Hans Schöler and Richard Lavery for support and discussions. Felipe Merino and Vlad Cojocaru are part of the “Cells in Motion” cluster of excellence at the University of Münster.

Author Contributions

Conceived and designed the experiments: FM BB VC. Performed the experiments: FM BB VC. Analyzed the data: FM BB VC. Contributed reagents/materials/analysis tools: FM BB VC. Wrote the paper: FM BB VC.

References

  1. 1. Georges AB, Benayoun BA, Caburet S, Veitia RA (2010) Generic binding sites, generic DNA-binding domains: where does specific promoter recognition come from? FASEB J 24: 346–356. pmid:19762556
  2. 2. Lelli KM, Slattery M, Mann RS (2012) Disentangling the many layers of eukaryotic transcriptional regulation. Annu Rev Genet 46: 43–68. pmid:22934649
  3. 3. Slattery M, Riley T, Liu P, Abe N, Gomez-Alcala P, et al. (2011) Cofactor Binding Evokes Latent Differences in DNA Binding Specificity between Hox Proteins. Cell 147: 1270–1282. pmid:22153072
  4. 4. Kim S, Brostromer E, Xing D, Jin JS, Chong SS, et al. (2013) Probing Allostery Through DNA. Science 339: 816–819. pmid:23413354
  5. 5. Panne D (2008) The enhanceosome. Curr Opin Struc Biol 18: 236–242. pmid:18206362
  6. 6. Narasimhan K, Pillay S, Huang YH, Jayabal S, Udayasuryan B, et al. (2015). DNA-mediated cooperativity facilitates the co-selection of cryptic enhancer sequences by SOX2 and PAX6 transcription factors. Nucleic Acids Res 43:1513–28 pmid:25578969
  7. 7. Panne D, Maniatis T, Harrison SC (2007) An atomic model of the interferon-beta enhanceosome. Cell 129: 1111–1123. pmid:17574024
  8. 8. Takayama Y, Clore GM (2012) Impact of protein/protein interactions on global intermolecular translocation rates of the transcription factors Sox2 and Oct1 between DNA cognate sites analyzed by z-exchange NMR spectroscopy. J Biol Chem 287: 26962–26970. pmid:22718759
  9. 9. Phillips K, Luisi B (2000) The virtuoso of versatility: POU proteins that flex to fit. J Mol Biol 302: 1023–1039. pmid:11183772
  10. 10. Jerabek S, Merino F, Scholer HR, Cojocaru V (2014) OCT4: Dynamic DNA binding pioneers stem cell pluripotency. Bba-Gene Regul Mech 1839: 138–154.
  11. 11. Tantin D (2013) Oct transcription factors in development and stem cells: insights and mechanisms. Development 140: 2857–2866. pmid:23821033
  12. 12. Chen JJ, Zhang ZJ, Li L, Chen BC, Revyakin A, et al. (2014) Single-Molecule Dynamics of Enhanceosome Assembly in Embryonic Stem Cells. Cell 156: 1274–1285. pmid:24630727
  13. 13. Chen X, Xu H, Yuan P, Fang F, Huss M, et al. (2008) Integration of external signaling pathways with the core transcriptional network in embryonic stem cells. Cell 133: 1106–1117. pmid:18555785
  14. 14. Ng CKL, Li NX, Chee S, Prabhakar S, Kolatkar PR, et al. (2012) Deciphering the Sox-Oct partner code by quantitative cooperativity measurements. Nucleic Acids Res 40: 4933–4941. pmid:22344693
  15. 15. Williams DC Jr., Cai M, Clore GM (2004) Molecular basis for synergistic transcriptional activation by Oct1 and Sox2 revealed from the solution structure of the 42-kDa Oct1.Sox2.Hoxb1-DNA ternary transcription factor complex. J Biol Chem 279: 1449–1457. pmid:14559893
  16. 16. Lam CS, Mistri TK, Foo YH, Sudhaharan T, Gan HT, et al. (2012) DNA-dependent Oct4-Sox2 interaction and diffusion properties characteristic of the pluripotent cell state revealed by fluorescence spectroscopy. Biochem J 448: 21–33. pmid:22909387
  17. 17. Ambrosetti DC, Scholer HR, Dailey L, Basilico C (2000) Modulation of the activity of multiple transcriptional activation domains by the DNA binding domains mediates the synergistic action of Sox2 and Oct-3 on the Fibroblast growth factor-4 enhancer. Journal of Biological Chemistry 275: 23387–23397. pmid:10801796
  18. 18. Nishimoto M, Fukushima A, Okuda A, Muramatsu M (1999) The gene for the embryonic stem cell coactivator UTF1 carries a regulatory element which selectively interacts with a complex composed of Oct-3/4 and Sox-2. Molecular and Cellular Biology 19: 5453–5465. pmid:10409735
  19. 19. Soufi A, Donahue G, Zaret KS (2012) Facilitators and Impediments of the Pluripotency Reprogramming Factors' Initial Engagement with the Genome. Cell 151: 994–1004. pmid:23159369
  20. 20. Esch D, Vahokoski J, Groves MR, Pogenberg V, Cojocaru V, et al. (2013) A unique Oct4 interface is crucial for reprogramming to pluripotency. Nature Cell Biology 15: 295–301. pmid:23376973
  21. 21. Remenyi A, Tomilin A, Pohl E, Lins K, Philippsen A, et al. (2001) Differential dimer activities of the transcription factor Oct-1 by DNA-induced interface swapping. Molecular cell 8: 569–580. pmid:11583619
  22. 22. Remenyi A, Lins K, Nissen LJ, Reinbold R, Scholer HR, et al. (2003) Crystal structure of a POU/HMG/DNA ternary complex suggests differential assembly of Oct4 and Sox2 on two enhancers. Genes Dev 17: 2048–2059. pmid:12923055
  23. 23. Nishimoto M, Miyagi S, Yamagishi T, Sakaguchi T, Niwa H, et al. (2005) Oct-3/4 maintains the proliferative embryonic stem cell state via specific binding to a variant octamer sequence in the regulatory region of the UTF1 locus. Mol Cell Biol 25: 5084–5094. pmid:15923625
  24. 24. Merino F, Ng CK, Veerapandian V, Scholer HR, Jauch R, et al. (2014) Structural basis for the SOX-dependent genomic redistribution of OCT4 in stem cell differentiation. Structure 22: 1274–1286. pmid:25126959
  25. 25. Dror I, Zhou TY, Mandel-Gutfreund Y, Rohs R (2014) Covariation between homeodomain transcription factors and the shape of their DNA binding sites. Nucleic Acids Res 42: 430–441. pmid:24078250
  26. 26. Verrijzer CP, Vanoosterhout JAWM, Vanweperen WW, Vandervliet PC (1991) Pou Proteins Bend DNA Via the Pou-Specific Domain. EMBO J 10: 3007–3014. pmid:1915275
  27. 27. Doucleff M, Clore GM (2008) Global jumping and domain-specific intersegment transfer between DNA cognate sites of the multidomain transcription factor Oct-1. P Natl Acad Sci USA 105: 13871–13876. pmid:18772384
  28. 28. Jauch R, Aksoy I, Hutchins AP, Ng CKL, Tian XF, et al. (2011) Conversion of Sox17 into a Pluripotency Reprogramming Factor by Reengineering Its Association with Oct4 on DNA. Stem Cells 29: 940–951. pmid:21472822
  29. 29. Klemm JD, Pabo CO (1996) Oct-1 POU domain DNA interactions: Cooperative binding of isolated subdomains and effects of covalent linkage. Gene Dev 10: 27–36. pmid:8557192
  30. 30. Gumbart JC, Roux B, Chipot C (2013) Standard Binding Free Energies from Computer Simulations: What Is the Best Strategy? J Chem Theory Comput 9: 794–802. pmid:23794960
  31. 31. Gumbart JC, Roux B, Chipot C (2013) Efficient determination of protein-protein standard binding free energies from first principles. J Chem Theory Comput 9: 3789–3798
  32. 32. Wang J, Deng Y, Roux B (2006) Absolute binding free energy calculations using molecular dynamics simulations with restraining potentials. Biophys J 91: 2798–2814. pmid:16844742
  33. 33. Takayama Y, Clore GM (2012) Interplay between minor and major groove-binding transcription factors Sox2 and Oct1 in translocation on DNA studied by paramagnetic and diamagnetic NMR. J Biol Chem 287: 14349–14363. pmid:22396547
  34. 34. Lian P, Liu LA, Shi YX, Bu YX, Wei DQ (2010) Tethered-Hopping Model for Protein-DNA Binding and Unbinding Based on Sox2-Oct1-Hoxb1 Ternary Complex Simulations. Biophys J 98: 1285–1293. pmid:20371328
  35. 35. Iwahara J, Zweckstetter M, Clore GM (2006) NMR structural and kinetic characterization of a homeodomain diffusing and hopping on nonspecific DNA. P Natl Acad Sci USA 103: 15062–15067. pmid:17008406
  36. 36. Dragan AI, Li ZL, Makeyeva EN, Milgotina EI, Liu YY, et al. (2006) Forces driving the binding of homeodomains to DNA. Biochemistry-Us 45: 141–151.
  37. 37. Toth-Petroczy A, Simon I, Fuxreiter M, Levy Y (2009) Disordered Tails of Homeodomains Facilitate DNA Recognition by Providing a Trade-Off between Folding and Specific Binding. J Am Chem Soc 131: 15084–15085. pmid:19919153
  38. 38. Di Rocco G, Gavalas A, Popperl H, Krumlauf R, Mavilio F, et al. (2001) The recruitment of SOX/OCT complexes and the differential activity of HOXA1 and HOXB1 modulate the Hoxb1 auto-regulatory enhancer function. Journal of Biological Chemistry 276: 20506–20515. pmid:11278854
  39. 39. Pan GJ, Qin BM, Liu N, Scholer HR, Pei DQ (2004) Identification of a nuclear localization signal in OCT4 and generation of a dominant negative mutant by its ablation. Journal of Biological Chemistry 279: 37013–37020. pmid:15218026
  40. 40. Pan YP, Nussinov R (2011) The Role of Response Elements Organization in Transcription Factor Selectivity: The IFN-beta Enhanceosome Example. Plos Comput Biol 7: e1002077. pmid:21698143
  41. 41. Moretti R, Donato LJ, Brezinski ML, Stafford RL, Hoff H, et al. (2008) Targeted chemical wedges reveal the role of allosteric DNA modulation in protein-DNA assembly. ACS Chem Biol 3: 220–229. pmid:18422304
  42. 42. Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, et al. (2005) Scalable molecular dynamics with NAMD. J Comput Chem 26: 1781–1802. pmid:16222654
  43. 43. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz KM Jr, et al. (1995) A second generation force field for the simulation of proteins, nucleic acids, and organic molecules. J Am Chem Soc 117: 5179–5197.
  44. 44. Cheatham TE III, Cieplak P, Kollman PA (1999) A modified version of the Cornell et al. force field with improved sugar pucker phases and helical repeat. J Biomol Struct Dyn 16: 845–862. pmid:10217454
  45. 45. Hornak V, Abel R, Okur A, Strockbine B, Roitberg A, et al. (2006) Comparison of multiple Amber force fields and development of improved protein backbone parameters. Proteins 65: 712–725. pmid:16981200
  46. 46. Lindorff-Larsen K, Piana S, Palmo K, Maragakis P, Klepeis J, et al. (2010) Improved side-chain torsion potentials for the Amber ff99SB protein force field. Proteins 78: 1950–1958. pmid:20408171
  47. 47. Li DW, Bruschweiler R (2010) NMR-Based Protein Potentials. Angewandte Chemie-International Edition 49: 6778–6780. pmid:20715028
  48. 48. Perez A, Marchan I, Svozil D, Sponer J, Cheatham TE 3rd, et al. (2007) Refinement of the AMBER force field for nucleic acids: improving the description of alpha/gamma conformers. Biophys J 92: 3817–3829. pmid:17351000
  49. 49. Dang LX (1995) Mechanism and Thermodynamics of Ion Selectivity in Aqueous-Solutions of 18-Crown-6 Ether—a Molecular-Dynamics Study. J Am Chem Soc 117: 6954–6960.
  50. 50. Bouvier B, Lavery R (2009) A Free Energy Pathway for the Interaction of the SRY Protein with Its Binding Site on DNA from Atomistic Simulations. J Am Chem Soc 131: 9864–9865. pmid:19580270
  51. 51. Bouvier B (2014) Decoding the patterns of ubiquitin recognition by ubiquitin-associated domains from free energy simulations. Physical chemistry chemical physics: PCCP 16: 48–60. pmid:24216748
  52. 52. Kumar S, Bouzida D, Swendsen RH, Kollman PA, Rosenberg JM (1992) The Weighted Histogram Analysis Method for Free-Energy Calculations on Biomolecules. 1. The Method. J Comput Chem 13: 1011–1021.
  53. 53. Zhu FQ, Hummer G (2012) Convergence and error estimation in free energy calculations using the weighted histogram analysis method. J Comput Chem 33: 453–465. pmid:22109354
  54. 54. Wieczor M, Tobiszewski A, Wityk P, Tomiczek B, Czub J (2014) Molecular Recognition in Complexes of TRF Proteins with Telomeric DNA. Plos One 9: e89460. pmid:24586793
  55. 55. Lavery R, Moakher M, Maddocks JH, Petkeviciute D, Zakrzewska K (2009) Conformational analysis of nucleic acids revisited: Curves+. Nucleic Acids Res 37: 5917–5929. pmid:19625494
  56. 56. Grossfield A, Zuckerman DM (2009) Quantifying uncertainty and sampling quality in biomolecular simulations. Annual reports in computational chemistry 5: 23–48. pmid:20454547
  57. 57. Humphrey W, Dalke A, Schulten K (1996) VMD: Visual molecular dynamics. Journal of Molecular Graphics & Modelling 14: 33–38.
  58. 58. Sethi A, Eargle J, Black AA, Luthey-Schulten Z (2009) Dynamical networks in tRNA: protein complexes. P Natl Acad Sci USA 106: 6620–6625. pmid:19351898
  59. 59. Glykos NM (2006) Carma: a molecular dynamics analysis program. J Comput Chem 27: 1765–1768. pmid:16917862