Visualizing Protein Movement on DNA at the Single-molecule Level using DNA Curtains

Abstract

A fundamental feature of many nucleic-acid binding proteins is their ability to move along DNA either by diffusion-based mechanisms or by ATP-hydrolysis driven translocation. For example, most site-specific DNA-binding proteins must diffuse to some extent along DNA to either find their target sites, or to otherwise fulfill their biological roles. Similarly, nucleic-acid translocases such as helicases and polymerases must move along DNA to fulfill their functions. In both instances, the proteins must also be capable of moving in crowded environments while navigating through DNA-bound obstacles. These types of behaviors can be challenging to analyze by bulk biochemical methods because of the transient nature of the interactions, and/or heterogeneity of the reaction intermediates. The advent of single-molecule methodologies has overcome some of these problems, and has led to many new insights into the mechanisms that contribute to protein motion along DNA. We have developed DNA curtains as a tool to facilitate single molecule observations of protein-nucleic acid interactions, and we have applied these new research tools to systems involving both diffusive-based motion as well as ATP directed translocation. Here we highlight these studies by first discussing how diffusion contributes to target searches by proteins involved in post-replicative mismatch repair. We then discuss DNA curtain assays of two different DNA translocases, RecBCD and FtsK, which participate in homologous DNA recombination and site-specific DNA recombination, respectively.

Introduction

Accurate repair of DNA damage is of critical importance to cells if they are to faithfully pass their genetic information onto future generations [1–7]. DNA damage comes in a variety of different flavors and cells have evolved different preferred methods of repairing each of them, but all DNA repair mechanisms share certain commonalities. Namely, all repair pathways have proteins that must be able locate the DNA damage within a large excess of undamaged DNA, process the damage to generate a repairable substrate, and then carry out subsequent reactions necessary for the completion of repair. In addition, many DNA repair pathways require the use of DNA motor proteins, such as helicases, to process the damaged DNA or intermediates before repair can be completed [8–10]. While traditional biochemical methods have been the mainstay of studying these processes, more recently several groups have been applying single-molecule approaches to probe the details of these systems (reviewed in reference [11]). Bulk biochemical experiments interrogate large numbers of molecules simultaneously and provide averaged information reflecting the entire population of molecules in a reaction. Single-molecule techniques have the advantage of allowing researchers to distinguish features that are often occluded in these population-averaged measurements. However, single-molecule experimentation can be technically challenging and often suffers from the necessity to visualize large numbers of molecules in order to generate statistically significant information.

To overcome this drawback, our lab has developed a high-throughput single-molecule approach known as DNA curtains, which allows for the simultaneous visualization of hundreds of individual DNA molecules suspended above a lipid bilayer (Figure 1) [12–18]. In brief, metallic patterns are first deposited onto the surface of a fused silica microscope slide by electron-beam lithography, and the resulting slide is assembled into a microfluidic sample chamber. The sample chamber surface is then passivated by deposition of a lipid bilayer in which a fraction of the lipid head groups have been functionalized with a biotin moiety. Singe-molecule methodologies are typically dependent on attaching one or more reaction components, either protein or DNA, to a solid supporting surface. Obtaining biologically relevant data is not possible if the molecules under investigation nonspecifically adsorb to the surface. Therefore a crucial aspect of the lipid bilayer is that it provides an inert environment resembling a normal cellular membrane and is compatible with many different types of biological molecules, including DNA and DNA-binding proteins. The bilayer also allows biotinylated DNA substrates to be attached to the bilayer surface through a streptavidin linkage. The anchored DNA molecules are pushed towards the leading edges of the nanofabricated metallic patterns on the fused silica slide, which act as physical barriers to lipid diffusion. The DNA molecules align along these barriers, enabling visualization of hundreds or even thousands of individual molecules in real time using total internal reflection fluorescence microscopy (TIRFM). Varying the design of the patterned surface allows for the DNA to be anchored in a variety of different configurations as necessary for specific experimental requirements, as will be described below (Figure 1, Video 1, Video 2) [15].

Figure 1. Different types of DNA curtains.

(A) Linear barriers used to align DNA [18]. (B) Zig-zag barriers, used to prevent overlap of adjacent DNA molecules [17]. (C) Double-tethered curtains, used when necessary to terminate buffer flow while viewing the molecules [13]. (D) PARDI curtains (Parallel Array of Double-tethered Isolated DNA molecules), which allow the DNA molecules to be placed at defined distances from one another [37]. (E) Crisscrossed curtains, for looking at movement of proteins from one DNA to another [32]. Reviewed in [15].

The flexibility of DNA curtains as a tool for single-molecule imaging has allowed us to probe the features of a number of DNA repair pathways [19–27]. In this review we will focus on experiments that analyze the movement of proteins along DNA by either diffusion-based mechanisms or by ATP-hydrolysis driven translocation. In the first section we describe our recent work in probing the mechanism by which the yeast mismatch repair (MMR) proteins MutSα and MutLα scan DNA by diffusion-based mechanisms during various stages of post-replicative mismatch repair. In the second section, we will explore how RecBCD and FtsK interact with and translocate along dsDNA, and highlight how RecBCD responds upon colliding with DNA-bound protein roadblocks. Finally, we highlight future challenges facing the field, with emphasis on the need to move single-molecule experimentation towards more complicated biological scenarios that more closely reflect physiological environments.

MMR Proteins as Models for Studying Diffusive Motion on DNA

Site- and structure-specific DNA-binding proteins must be capable of locating specific binding targets among a vast excess of nonspecific DNA, yet basic principles governing these fundamental search mechanisms remain poorly understood [28, 29]. MMR offers an outstanding model system for studying how diffusion-based target searches contribute to protein-nucleic acid interactions because of the number of searches involved in the repair reaction, and the fact that the target the proteins must look for changes during the course of the reaction [2, 6, 7]. Post-replicative mismatch repair (MMR) is a cellular process that corrects errors in DNA synthesis that would otherwise lead to genomic instability and results in a 1,000 fold increase in the fidelity of DNA replication. Defects in proteins involved in MMR result in an increase in spontaneous mutation, and in humans, are associated with hereditary nonpolyposis colorectal cancer and other tumors [30, 31]. MutSα (a heterodimer of Msh2 and Msh6 heterodimer) and MutLα (a heterodimer of Mlh1 and Pms1) are two conserved protein complexes necessary for initiating MMR in eukaryotes [2, 6, 7]. MutSα is responsible for initial recognition of mismatches and small insertion/deletion loops, whereas MutLα harbors an endonuclease activity necessary for cleavage of the lesion-bearing DNA strand.

MMR and the target search problem

The difficulty of the target search problem faced during MMR can be illustrated by considering that S. cerevisiae, which has a diploid genome of ~2.4×10⁷ base pairs, will normally only incur just 2 mismatches per round of replication. Furthermore, unlike many other types of DNA lesions, single base mismatches are minimally distorting and are chemically identical to undamaged DNA. This means that MutSα must somehow survey millions of undamaged bases, and identify these exceedingly rare (and subtle) lesions with high specificity. MutLα must then search the genome for lesion-bound MutSα, and must avoid nonproductive interactions with MutSα molecules not bound to mismatches. Finally, the lesion-bound MutSα/MutLα complex must search for signals on the flanking DNA that distinguish the parental and daughter in order to ensure that the correct strand is repaired.

There are at least four diffusion-based models that describe how DNA-binding proteins might search for targets: random 3-dimensional (3D) collision, 1D hopping, 1D sliding, or intersegment transfer [28, 29]. Search mechanisms that employ hopping, sliding, or intersegmental transfer are collectively referred to as facilitated diffusion, because the lower dimensionality brought about through use of these mechanisms allows for target association rates that exceed the limits imposed by 3D collisions. It should be noted that these diffusion-based mechanisms are not mutually exclusive, and that proteins can potentially use different combinations of mechanism to locate their target sites within the genome. However, no previous study had visualized a protein locating a specific target within a DNA substrate, and likewise, no study had directly shown facilitated diffusion leading to target binding. Therefore, it remained a long-standing question as to whether and how facilitated diffusion actually contributes to targeting mechanisms.

To address these questions, we established a single-tethered DNA curtain assay that allowed us to visualize DNA binding by quantum dot-tagged MutSα, and then developed mismatch-containing substrates and used these to map the location of the proteins to the known position of an engineered mismatch (Figure 2A) [20, 32]. Single-tethered curtains permit the use of a simple control to verify that the proteins are bound to the DNA and not adsorbed to the slide surface: when buffer flow is terminated, the DNA and proteins simultaneously diffuse away from the sample chamber surface and out of the detection volume defined by the penetration depth of the evanescent field (Figure 2B, Video 1, Video 3). Use of the single-tethered curtains in these initial binding measurements also allowed us to very quickly assess the binding distributions of the proteins along the DNA, which revealed a preference for binding mismatched DNA (Figure 2C, Video 3). We then used a double-tethered DNA curtain assay (Figure 1D, Video 2) to watch single molecules of MutSα in real time as they search single mismatches engineered into a ~50-kb lambda DNA substrate (Figure 3A, Video 4) [32]. The double-tethered curtains are used in this instance such that the proteins can be quickly injected into the sample chamber, at which point buffer flow is terminated. This strategy allowed us to then visualize the proteins in real time as they searched the DNA for the mismatch, and eliminated the need for continuous buffer flow, which would have otherwise perturb their diffusive properties. These experiments demonstrated that MutSα could locate mismatches either by first diffusing in 1D along the undamaged flanking DNA, or by binding directly to the mismatches from solution consistent with a targeting mechanism based on 3D diffusion (see below) (Figure 3B–C). For the 1D pathway, MutSα could diffuse along undamaged DNA for long distances (up to ~14.6 kbp in our assays on naked, mismatch-bearing DNA).

Figure 2. MutSα binding to DNA curtains.

(A) Schematic of single-tethered DNA curtains illustrating the mismatched base (MM) located at a specific site in the DNA. (B) Images of a 3-tiered DNA curtain with flow on (left panel), during a transient pause in flow (middle panel), and after flow has been resumed (right panel). Flow is from top to bottom, DNA is green, proteins are magenta. The location of the 3 tandem G/T mismatches (MM) is indicated. (C) Distribution of MutSα bound to mismatch-containing DNA. Adapted with permission from reference [32].

Figure 3. Diffusion of MutSα on DNA before and after mismatch recognition.

(A) Schematic of the double-tethered DNA curtains. DNA substrates are anchored by one end to the lipid bilayer, aligned along the nanofabricated barriers, and then anchored at their downstream ends through a digoxigenin-antibody linkage. (B) Example of MutSα undergoing 1D diffusion until encountering the lesion. MutSα is magenta, the DNA is not labeled, and gaps in the trajectories reflect QD blinking. (C) Example of MutSα capturing the mismatch through a direct 3D-diffusion. (D) Kymograph showing the response of mismatch-bound MutSα upon injection of 1 mM ATP. Experiments were conducted with double-tethered curtains, MutSα was prebound to the mismatch, ATP was injected at 0.1 ml min⁻¹, and flow is terminated after ATP enters the sample chamber. (E) MutSα structural changes predicted upon ATP-triggered release from a lesion. The structures on the left represent mismatch-bound human MutSα in which the protein complex (gray) is wrapped around the DNA (green) with domain I of Msh6 (magenta) engaged with the mismatched base [41–43]. The hypothetical structure on the right was obtained by rigid body rotation of Msh6 domain I out of the major groove to illustrate how retraction of Msh6 domain I might allow for the release of MutSα from the mismatch, and still allow the protein to remain tightly wrapped around the DNA while enabling 1D diffusion in the absence of obligatory rotational component. Adapted with permission from reference [32].

One conclusion drawn from this data was that the movement of MutSα was consistent with a model wherein MutSα rotates along the DNA as it tracks the backbone of the helix while scanning for mismatches [20, 32], and this type of motion appears to be fairly coming among different proteins that can scan DNA by 1D diffusion [33, 34]. This conclusion was based upon the close agreement between the experimentally observed diffusion coefficient for MutSα and the diffusion coefficient obtained from a theoretical calculation that was originally developed by Michael Schurr to explain how backbone tracking might influence the 1D diffusion of lac repressor along DNA [35, 36]. Similar assays developed to determine how MutLα searched for lesion-bound MutSα revealed a striking difference in the diffusive properties of these two proteins [21, 32]. While MutSα moved by 1D sliding, the movement of MutLα was best characterized by a mechanism involving 1D hopping, possibly while the protein was wrapped around the DNA in a large ring-like configuration. This key difference in diffusion mechanism was reflected as an order of magnitude increase in the 1D diffusion coefficient of MutLα relative to that of MutSα, enabling MutLα to move much more rapidly along DNA relative to MutSα [21].

Influence of protein concentration on path-to-target

One often unrecognized challenge with single-molecule techniques is that the protein concentrations used for the experiments can be limited to relatively narrow windows if the protein under study is fluorescently tagged: if there is too little protein, then no reactions will be observed; in contrast, too much protein results in higher background. Because of these practical considerations, experiments with quantum dot-tagged MMR proteins were restricted to concentrations in the low nanomolar range [20, 21, 32]. Under these conditions, the MMR proteins could locate their targets by either 1D or 3D pathways (Figure 3B–C) [32]. For MutSα, ~42% of the mismatch binding events occurred through the 1D pathway and the remaining ~58% occurred through the 3D pathway. Similarly, 55% of MutLα found mismatch-bound MutSα through pathway involving a 1D search, whereas the remaining 45% used a 3D association pathway. This ability to engage lesions through either 1D or 3D pathways is not surprising or even unexpected; moreover it is crucial to recognize that the ratio of 1D to 3D binding events is expected to change as a function of protein concentration [37, 38]. The fraction of proteins utilizing a 1D search will always increase at low protein concentrations, whereas the fraction of those that bind through a 3D pathway will always increase at higher protein concentrations [37, 38]. This concentration-dependent effect on target searches needs to be considered when trying to interpret how results obtained from a single-molecule measurement of 1D diffusion might pertain to in vivo scenarios where the protein concentrations (among other things; see below) might differ substantially from in vitro conditions. For example, although MutSα is clearly capable of sliding along naked DNA for very long distances, it seems exceedingly unlikely that the proteins moves over similar distance scales in vivo (see below), nor would they need to slide long distances given the relatively high concentration of MutSα in the cell (>1,000 molecules of MutSα per yeast cell, corresponding to an intranuclear concentration of at least 0.6 μM) [39, 40].

Lesion recognition alters the diffusive characteristics of MutSα

Once bound, MutSα remained stably associated with the mismatch, but an injection of ATP triggered mismatch release and caused the proteins to diffuse in 1D away from the lesion (Figure 3D, Video 5) [32]. Remarkably, the diffusion of MutSα after ATP-triggered release from the mismatches differed from that observed prior to mismatch binding, as reflected by a 6-fold increase in the diffusion coefficient, exceeding the upper limit imposed by rotation coupled diffusion. Intriguingly, the proteins also failed to re-recognize mismatches after release, even though they must re-cross the mismatches up to hundreds of times due to the highly redundant nature of diffusion [32]. Structural data provides a possible explanation for these differences (Figure 3E) [41–43]. MutSα completely encircles DNA and domain I of Msh6 lies within the major groove, allowing a conserved phenylalanine and glutamic acid to engage the mismatch; all remaining contacts with the DNA lie along the phosphate backbone (PDB: 1EWR; 2O8B; 1E3M) [41–43]. This configuration of Msh6 domain I would impose steric constraints requiring MutSα to track the helical pitch of the DNA during any 1D-diffusion. Retraction of domain I from the major groove would allow MutSα to diffuse on DNA without obligatory rotation. Therefore, we hypothesize that domain I of Msh6 is inserted into the major groove prior to lesion recognition, remains within the major groove upon binding the lesion, but is retracted from the major groove after ATP-triggered release from the mismatch. Retraction of Msh6 domain I from the major groove would also provide an explanation for how MutSα is released from the mismatch upon binding ATP, and would explain how it avoids re-binding the mismatch while searching for strand discrimination signals. Similar effects on diffusion behavior were observed for the MutSα/MutLα complexes, with one striking difference: the MutSα/MutLα complex became highly resistant to salt-induced dissociation from the DNA after ATP-triggered release from the mispaired bases [32]. Together, these findings appear to reflect at least two as yet uncharacterized structural transitions whose ultimate outcomes are to alter the ability of MutSα to recognize its target and to prevent the proteins from coming off of the damaged DNA strand once they are released from the lesion.

Intersite transfer during MMR

Intersite transfer can allow proteins to move quickly between DNA segments otherwise separated by long regions of linear sequence [28, 29]. Intersite transfer might assist MMR proteins in their respective target searches, but would impede the process after lesion recognition, because the proteins would lose track of the damaged DNA. We developed crisscrossed DNA curtains as a tool for visualizing intersite transfer at the single molecule level, which is revealed as 90° turns in the diffusion trajectories at intersections between the crisscrossed DNA molecules (Figure 4A) [32]. These results demonstrated that MutLα can undergo intersite transfer (Figure 4B–D), whereas MutSα and the MutSα/MutLα complex cannot (Figure 4E–G), which suggests that MutLα would be able to search for mismatch-bound MutSα through a combination of 1D-hopping and intersite transfer [32]. Importantly, the intersite transfer activity of MutLα was suppressed upon association with MutSα. Moreover, as indicated above the MutSα/MutLα complex was uniquely resistant to salt-induced dissociation from DNA after ATP-triggered release from the mismatches. Both of these observations indicate that protein-protein interactions within the context of the complex modify the behavior of the MutSα/MutLα complex. Interestingly, our prior work also showed that MutLα displayed DNA end-dependent dissociation, suggesting that MutLα travels while wrapped around DNA in a large ring-like configuration [21], in agreement with predictions based on crystallography data of isolated MutL fragments (PDB: 1X97) [44]. But, if MutLα travels while wrapped around DNA in a ring-like configuration, then this ring must transiently open to allow intersite transfer, and might also contribute to the ability of MutLα to diffuse past nucleosomes on DNA (see below).

Figure 4. Using crisscrossed DNA curtains to visualize intersite transfer.

(A) Schematic of a crisscrossed DNA curtain. (B) Integrated trajectory showing behavior of MutLα (magenta) upon encountering a crisscrossed DNA junction. (C) Corresponding tracking data, and (D) tracking data super-imposed on the DNA axes. In (D), DNA molecules are shown as blue lines; the green circle highlights the DNA junction within a 90% confidence interval; tracking data are color-coded according location relative to the crisscross; the color-coded bar shows the relative location of protein over time. (E) MutSα before encountering a lesion. (F) MutSα after ATP-triggered release from a mismatch. (G) MutSα/MutLα after ATP-triggered mismatch release; MutLα was QD-tagged and MutS was untagged. In (F) and (G) the zero time points correspond to the location of the lesions, and the longer time trajectories for these data sets reflect the longer DNA-binding lifetimes of MutSα and MutSα/MutLα after ATP-triggered release from mismatches. Adapted with permission from reference [32].

Diffusive searches on crowded DNA

As the diffusive motions of MutSα, MutLα and other DNA repair proteins are studied it is important to recognize that in vivo, most genomic DNA of eukaryotes exists primarily as compacted chromatin wrapped by nucleosomes [45–47]. To understand how MutSα and MutLα might navigate such a substrate, we studied the diffusive motions using DNA curtains on bound with varied densities of nucleosomes (Figure 5) [21]. Experiments performed at low nucleosome densities demonstrated that MutSα was unable to efficiently bypass individual nucleosomes and frequently got stuck diffusing back and forth between two adjacent nucleosomes (Figure 5A, upper panel). At higher nucleosome densities MutSα was unable to diffuse along the DNA, indicating it was again trapped between adjacent nucleosomes (Figure 5A, lower panel). These findings were consistent with the both the structure MutSα and the observation that MutSα diffuses on DNA by sliding on DNA while rotating and retaining constant (or near constant) contact the phosphate backbone. Moreover, the rare nucleosome bypass by MutSα observed at low nucleosome densities was consistent with an occasional “hop” by the protein as it diffused along the DNA. This ability of MutSα to occasionally bypass nucleosomes serves as an important reminder that any observed 1D diffusion trajectory (for any protein) is likely comprised of multiple diffusive modes, rather than just a single type of diffusion.

Figure 5. Protein diffusion on crowded DNA substrates.

(A) MutSα (red) is shown diffusing on a DNA molecule with unlabeled nucleosomes (green), and regions of overlap appear yellow. (B) MutLα (magenta) diffusion on DNA bound by recombinant nucleosomes (green), and regions of overlapping signal are white. Examples of bypass and bounded diffusion are highlighted. (C) Illustrates how the structures of MutLα (left) and MutSα (right) may influence nucleosomal encounters. The molecules are drawn to scale. The trajectory of the DNA leaving the nucleosome surface has been modified for illustrative purposes. Mlh1-Pms1 steps over the nucleosome (solid arrow). Nucleosome bypass by Msh2-Msh6 might occur by occasional hopping or 2D sliding (dashed arrows), but neither mechanism allows free mobility on a nucleosomal array. Adapted with permission from reference [52].

The diffusion of MutLα on nucleosome-bound DNA was remarkably different that that of MutSα [21]. Rather than being restricted by the presence of nucleosomes, MutLα was able to diffuse past nucleosomes with no apparent impediment (Figure 5B). This differential response to nucleosomes is likely a reflection of key mechanistic differences regarding how MutSα and MutLα diffuse along DNA (Figure 5C): MutSα diffuses primary through a sliding mechanism, whereas MutLα appears to primarily use a hopping mechanism. Simply put, MutLα can “hop” over nucleosomes, whereas MutSα cannot. These outcomes may reflect general mechanistic attributes of their respective modes of 1D-diffusion, which may also apply to other proteins that diffuse on DNA: proteins that track the phosphate backbone by 1D sliding will experience a barrier upon encountering obstacles such as nucleosomes and must either disengage the DNA to continue searching for targets, or the DNA must be cleared of obstacles before hand to allow unhindered access to the DNA; in contrast, proteins that do not track the backbone can bypass obstacles without experiencing significant barriers. This hypothesis awaits experimental tests of other diffusing proteins on nucleosome-bound DNA substrates.

The inability of MutSα to diffuse past nucleosomes provides a clear indication that the search for mismatches in vivo is unlikely to involve extensive 1D sliding. There are two potential mechanisms that might allow MutSα to conduct effect target searches on DNA bound by nucleosomes: either something must remove the nucleosomes from the DNA, and/or the high in vivo concentration of MutS may obviate the need for 1D sliding during the target search by driving the preference for mismatch binding through a 3D pathway [37, 38]. Recent work by Hombauer and colleagues suggest that both possibilities may MMR in vivo [48]. For example, ~10–15% MMR arises from MutSα that is physically associated with replication factories, perhaps enabling MutSα to conduct a search for replication induced mismatches on naked DNA behind the replication fork [48]. The remaining 80–90% of MMR does not involve any direct association of MutSα with the replication machinery, suggesting that these proteins locate their targets through a simple 3D pathway.

Visualizing Translocation of Molecular Motor Proteins along DNA

In contrast to proteins that passively diffuse along DNA, molecular motor protein use the energy derived from ATP-binding and hydrolysis to drive directed motion along DNA. Motor proteins participate in a wide-range of biological processes involving DNA, including reaction critical for DNA repair. For example, motor proteins play crucial roles in nearly every step of homologous DNA recombination in both prokaryotes and eukaryotes [5, 8, 9, 49–51]. DNA curtains provide a means of directly visualizing the movement of motor proteins along DNA [22, 24, 52], and below we highlight two recent studies from our laboratory looking at the translocation behavior of two model prokaryotic translocases.

RecBCD as a model DNA translocase

Prokaryotes rely predominantly upon homologous recombination (HR) to repair potentially lethal DNA double strand breaks (DSBs). In E. coli, HR is initiated by the heterotrimeric RecBCD protein complex, which binds the free ends of the DSB and processes them to make ssDNA suitable for downstream reactions in the repair pathway [8]. The biochemical, biophysical and structural properties of RecBCD have been extensively characterized at both the ensemble and single-molecule levels (PDB: 1W36) [53–63]. Together, this work has revealed that RecBCD is a highly processive helicase/nuclease that can translocate rapidly along DNA, and this activity is conferred by its two ATP-dependent helicases, RecB and RecD [55, 60]. These two SF1-family helicases track along DNA strands of opposite polarity: RecB translocates with 3′→5′ polarity, and RecD translocates with 5′→3′ polarity. RecC lacks helicase activity, despite having a canonical SF1 helicase fold, but plays an important structural role in harboring the “pin” domain that separates the incoming dsDNA strands before feeding the two separate strands to RecB and RecD [64]. Additionally, RecC is responsible for recognizing the 8-nucleotide Chi sequence (5′-GCTGGTGG-3′), which causes RecBCD to switch lead motors from RecD to RecB. This change is marked by a reduction in apparent translocation velocity [59], and Chi recognition also acts as a signal for RecBCD to begin loading RecA onto the ssDNA for the next step of HR [63].

RecBCD and protein-protein collisions on DNA

All DNA translocases must encounter other proteins as they translocate along chromosomes in living cells [65–69]. For example, in prokaryotes, it has been estimated that about 30% of the genome is occupied by DNA-binding proteins, and that protein-free regions of DNA are unlikely to exceed more than 100-base pairs in length [70–72]. In addition, during chromosome replication, collisions between the replication machinery and protein obstacles are mitigated by non-replicative accessory helicases such as Rep, UvrD, PcrA, Dda, which are capable of removing DNA-bound proteins and allowing replication to continue unhindered [66, 73]. Moreover, the consequences of protein-protein collisions on DNA are so important and RNA polymerase is such a potent roadblock, that evolution has driven the organization of bacterial genomes to minimize head-on collisions between the replication and transcription machineries [67, 68, 74, 75]. However, while the broad implications of protein-protein collisions on DNA are becoming well established, particularly with respect to their influence on genome stability [65, 76, 77], the underlying mechanistic attributes of motor protein collisions with other DNA-bound proteins remain poorly understood.

We used RecBCD as a model for understanding the mechanistic basis for how motor protein might deal with DNA-bound obstacles [19]. Although it is clear that RecBCD must encounter protein-bound obstacles with high frequency during end resection, the effects of these obstacles on RecBCD progression were unknown. To address the effects of protein-protein collisions during RecBCD translocation, we used a DNA curtain assay to monitor RecBCD activity and the outcome of its collisions with different protein roadblocks (Figure 6A) [19]. RecBCD translocation is revealed as an ATP-dependent reduction in DNA length, as the protein degrades the fluorescently-stained DNA (Figure 6B, Video 6). Assays with naked DNA confirmed that RecBCD behaved as expected in the DNA curtain measurements, and responded appropriately to the cis-acting Chi DNA sequence (Figure 6B).

Figure 6. DNA curtain assay for RecBCD and examples of collisions.

(A) Schematic illustrations of DNA curtain assays for RecBCD. RecBCD bound to the ends of aligned DNA molecules in the absence of ATP (left panel); translocation of in the presence of ATP RecBCD causes the DNA to shorten (middle panel); and translocation of RecBCD on DNA bound by other proteins (right panel). (B) Kymographs showing examples of RecBCD on naked DNA substrates with (lower panel) and without (upper panel) chi sequences. The DNA is labeled with YOYO1 and is shown in green, and the location of chi is indicated. (C) Kymographs showing examples of RecBCD translocation on DNA bound by various protein roadblocks (as indicted). Adapted with permission from reference [52].

For protein-protein collision experiments, we challenged RecBCD with four different tight-binding fluorescently-labeled roadblock proteins: E. coli RNA polymerase (RNAP), catalytically inactive EcoRI^E111Q, lac repressor, and nucleosomes. Each of these roadblocks was selected for a specific reason. RNAP is highly abundant [78], and is itself a powerful translocase that transcribes through roadblocks and can move under an applied load of ~14–25 pN [79]. RNAP survives collisions with replication forks and stalls fork progression in head-on collisions [80–84]. Bacterial genomes have evolved to avoid head-on collisions between the transcription and replication machineries [67, 68], and RNAP has been suggested to be one of the most formidable roadblocks encountered in vivo. EcoRI^E111Q is a catalytically inactive version of the restriction enzyme EcoRI and is one of the tightest DNA binding proteins known to exist (Kd ≈ 1 pM) [85]. EcoRI^E111Q can halt the progression of E. coli RNAP [86, 87], T7 and SP6 RNA polymerases [88], SV40 large T-antigen, UvrD, DnaB, and Dda helicases, SV40 replication forks [89], and E. coli replication forks [73]. Lac repressor is representative of a large family of bacterial transcription factors that binds tightly to DNA and can block both RNAP and replication forks [86, 90]. Nucleosomes are the most frequently encountered roadblocks in eukaryotes, and replisomes, transcription machinery, and ATP-dependent chromatin remodelers all act on nucleosomes through mechanisms requiring force generation [91–93]. Although bacterial RecBCD will never encounter eukaryotic nucleosomes in normal scenarios, heterologous systems have profoundly influenced our understanding of principles underlying these processes [94–99], arguing that RecBCD could also serve as a good model for studying the fate of nucleosomes during encounters with a DNA translocase.

Remarkably, RecBCD could displace and evict all of these different protein roadblocks from DNA with little or no apparent effects on translocation velocity or processivity (Figure 6C, Video 7). RecBCD was also able to push the proteins for long distances along DNA prior to their eviction. Notably, the results were similar for RNAP core complexes, holoenzymes, stalled elongation complexes, and active elongation complexes in both co-directional and head-to-head collisions, which indicate that RecBCD can displace this potent physiological roadblock regardless of its directionality or functional state. These results show that RecBCD is largely unhindered by the presence of tightly bound proteins, and can push and evict them from DNA as it processes DSBs during the early stages of HR.

Mechanism of roadblock removal by RecBCD

Based on our observation that RecBCD could displace proteins from DNA, we proposed four generic mechanisms by which any translocase might displace a protein obstacle from DNA, each of which makes distinct experimental predictions (Figure 7A) [19]. These models included: (i) Passive Release, where the proteins are dislodged from a high-affinity specific site, and pushed onto lower-affinity nonspecific DNA. This model predicts the rates of RecBCD-induced dissociation would be similar to spontaneous dissociation from nonspecific DNA in the absence of RecBCD; (ii) Preferred site release, where the proteins encounter sequences of exceptionally low-affinity such that they preferentially dissociate from these sites; (iii) Structural disruption in which collisions alter the proteins, such that they persist as structurally perturbed complexes with a characteristic lifetime dictated by their weakened affinity. In this case, the distance over which proteins are pushed will be proportional to RecBCD velocity, and the rate of RecBCD-induced dissociation would be greater than the rate of spontaneous dissociation from nonspecific DNA; and (iv) Transition state ejection, which predicts the existence of a more weakly bound state as proteins are pushed from one position to the next, with dissociation taking place specifically during this step of the reaction. The time required for a protein to be pushed from one nonspecific site to the next during each round of the chemomechanical cycle is equivalent to the time required for the translocase to take a single step, which is a fixed intrinsic value independent of ATP concentration and overall velocity. Therefore, the time it takes the roadblock to pass through the weakly bound state is independent of overall velocity, but the cumulative time spent in the weakly bound state increases linearly with step number, and therefore the probability of dissociation increases with step number irrespective of velocity.

Figure 7. Mechanism of protein removal by RecBCD.

Features and key predictions of each model are highlighted, and full details can be found in reference [52]. (1) Passive Release: Proteins are dislodged from a high-affinity specific site (S), and pushed from one nonspecific site (N) to the next. Dissociation occurs because the proteins are bound to lower-affinity nonspecific DNA. This model predicts the rates of RecBCD-induced dissociation (k_off,obs) would be similar to spontaneous dissociation from nonspecific DNA in the absence of RecBCD (k_off,obs≈k_off,N). (2) Preferred Site Release: Proteins encounter sequences of exceptionally low-affinity (N′) such that they preferentially dissociate from these sites (k_off,N′ ≫k_off,N). (3) Structural Disruption: Collisions alter the proteins, such that they persist as structurally perturbed complexes (X) with a characteristic lifetime (τ_x) dictated by their weakened affinity. The distance (d) over which proteins are pushed will be proportional to velocity (V), and the rate of RecBCD-induced dissociation (k_off,obs) would be greater than the rate of spontaneous dissociation from nonspecific DNA (k_off,obs≈k_off,X ≫k_off,N). (4) Transition State Ejection: Proteins pass through a weakly bound state (T) as they are pushed from one position to the next. This model predicts dissociation occurs during the transition state (k_off,T ≫k_off,N). The time required to pass through the transition state during one round of the chemomechanical cycle is equivalent to the time required for the translocase to take a single step (k_step), which is a fixed intrinsic value. The time it takes the roadblock to pass through the transition state is therefore independent of [ATP], but the cumulative time spent in the transition state increases linearly with step number (n), and the probability of dissociation increases with step number irrespective of velocity. Adapted with permission from reference [52].

With this framework in hand, we then sought to determine which of these was most consistent with the behavior of RecBCD. Roadblock lifetime measurements, either with or without RecBCD collisions, revealed that RecBCD-induced dissociation was ≥200-fold faster than spontaneous dissociation from nonspecific sites, which is also inconsistent with the passive release model. There was no evidence for dissociation at preferential sites as the roadblocks were pushed and dislodged by RecBCD. To distinguish between structural disruption and transition state eviction we compared protein lifetimes and pushing distances at different translocation velocities. These experiments were made possible because the velocity of RecBCD can be tuned over a broad range (from ~80 up to ~1,500 kbp per second) simply by adjusting the concentration of ATP (Figure 7B); this ability to tune the velocity of RecBCD was the critial component of this analysis. Remarkably, a decrease in RecBCD velocity led to a corresponding increase in the post-collision half-life of EcoRI^E111Q, RNAP, and nucleosomes, while the distribution of distances over which the proteins were pushed remained largely unaltered (Figure 7C–D). This result indicated that roadblock dissociation was dictated by the number of steps the proteins were forced to take while being pushed by RecBCD rather than the cumulative time it took to be pushed a given distance, which is most consistent with transition state ejection. While our experiments did not reveal any evidence for a structural disruption mechanism of eviction, this does not rule out the possibility that EcoRI^E111Q, RNAP, and nucleosomes are structurally altered when acted upon by RecBCD. However, if they are structurally perturbed, this alone does not result in their eventual dissociation from DNA.

These experiments have revealed the mechanism by which the RecBCD translocase can disrupt proteins that lie in its path during the initial steps of HR. One question that remains is whether the mechanism of protein dissociation employed by RecBCD is general, or whether different translocases might use different mechanisms to displace DNA-bound protein obstacles. Our DNA curtain platform is amenable to future experiments with other translocases and other roadblocks to determine the generality of the mechanisms that we have proposed for resolving protein-protein collisions.

FtsK is a hexamer molecular motor that translocates on DNA

Homologous recombination can lead to the formation of chromosome dimers, which if left unrepaired can prevent chromosome segregation and lead to cell death. In E. coli, the tyrosine recombinase XerCD promotes chromosome dimer resolution at a specific site (dif) within the replication termination region [100–102]. FtsK is hexameric DNA translocase that localizes to the division septum in bacteria and is essential for activating XerCD (Figure 8A) [100–102]. FtsK has an N-terminal integral membrane domain and a C-terminal motor domain separated by a long (~600 amino acids) linker region. The motor domain encircles DNA and is comprised of α, β, and γ subdomains: the αβ forms a RecA-like ATPase, which couples the energy derived from ATP hydrolysis to translocation along DNA [100–102]; and γ forms a winged helix that binds KOPS (FtsK Orienting Polar Sequences), an 8 base-pair sequence that is overrepresented in the E. coli genome and is preferentially oriented towards dif [103, 104]. FtsK is oriented by KOPS, allowing it to translocate towards XerCD bound to dif.

Figure 8. Visualizing FtsK translocation on DNA.

(A) Crystal structure of the FtsKαβ motor domain hexamer from P. aeruginosa [106]. (B) Kymograph illustrating ATP- dependent translocation of QD-tagged FtsK (magenta) along a single DNA molecule (unlabeled). Transient gaps in the magenta signal correspond to QD blinking. The location of the KOPS sites is illustrated schematically on the left, and arrowheads indicate the arbitrarily assigned (+) and (−) designations for translocation direction. (C) Example of FtsK bound to DNA in the presence of ATPγS. (D) Velocity distribution histogram comprised of the combined (+) and (−) velocity data sets, revealing a mean velocity of V=4.66±1.3 kb/sec. (E) Scatter plot showing the relationship between (+) and (−) direction velocities for individual molecules of tranlocating FtsK. The red line illustrates a fit to the data and yields a slope of 1. (F) Velocity distributions for data collected at 37°C and 5 mM ATP, revealing a mean velocity of V=17.5±3.5 kb/sec. (G) Histograms showing positions at which FtsK changed direction during translocation. Adapted with permission from reference [22].

The FtsKαβγ motor domain and the motor domain from the related protein SpoIIIE have been extensively characterized by a combination of bulk biochemical and single-molecule studies (PDB: 2IUU) [105–116]. FtsKαβγ is a DNA-dependent ATPase, which can translocate along DNA at ~5 kb per second (at 20°C) and resist stalling at forces up to 65 piconewtons (pN) [104, 111, 112, 117]. Several studies have investigated how FtsK is guided towards dif through interactions with KOPS [103, 104, 113, 114, 117]. To help shed light on the activities of FtsK, we visualized quantum dot-tagged FtsK using DNA curtains, which enable us to directly visualize DNA binding independent of translocation [22]. Visualization of FtsK translocation was not possible with previous approaches, which were all force-based measurements. Our studies also utilized a linked trimeric version of the FtsK motor, which dimerizes to form a hexamer and displayed better solubility compared to previous FtsK constructs [22, 118]. The data obtained from these translocation experiments was largely consistent with prior work, confirming that the quantum dots did not grossly perturb the protein activity (Figure 8B–E).

FtsK also exhibited an unusual ability to abruptly change direction while translocating along DNA (Figure 8B), which has also been reported in all single-molecule studies of FtsK translocation to date. How is it possible for a hexameric motor that encircles DNA to suddenly reverse direction? While this remains an open question, we can envision at least two possibilities. Either the FtsK motors under observation are actually two hexamers organized in opposing orientations (as opposed to a single hexamer), or a single hexamer somehow changes direction, perhaps by coming off the DNA and then re-orienting in the opposing direction. Future work will be necessary to distinguish between these and other possibilities. Perhaps more importantly, the ability of FtsK to reverse direction must be suppressed in living cells otherwise it would not be able to promote unidirectional chromosome segregation. Therefore, it will be essential to understand how the FtsK translocation is regulated in physiological settings were the organization and geometry of the division septum may play a dominating role in regulating its activity (see below).

One remarkable result from these studies was that FtsK could travel along DNA at speeds approaching 18 kbp per second when the reaction temperature was raised to 37°C, making it the fastest known DNA translocase. By comparison, the Lockheed SR-71 Blackbird is nearly 32 meters long and can reach speeds of up to 980 meters per second (~2,200 miles per hour), corresponding to ~30 body-lengths per second. By comparison, FtsK is just ~6 nanometers in length, but can travel at velocities approaching ~1000 body-lengths per second, making it 30-times faster than the worlds fastest man-made aircraft, based on relative body-lengths per second, but not absolute velocity.

KOPS recognition by FtsK

The polar organization of KOPS on the bacterial chromosome provided an early hint that this DNA sequence must somehow be involved in directing FtsK towards the terminus region. Initial reports suggested that FtsK could recognize KOPS during translocation, enabling it to re-orient while translocating [103, 104, 113, 114], but other studies suggested that KOPS acts only as a loading site for FtsK [108, 117]. Our results seem to indicate that FtsK does not recognize KOPS during translocation, nor does FtsK recognize KOPS when ATP is present as the sole nucleotide co-factor under our reaction conditions [22]. The only conditions where we were able to see FtsK binding to KOPS were those under which ATP hydrolysis could not occur. For instance, efficient KOPS binding was observed with ADP, nonhydrolyzable AMP-PNP or ATPγS, or with ATPase-deficient mutant proteins, or in the presence of ATP when MgCl₂ was omitted from the reactions (Figure 9A). However, when initial binding was conducted in ADP, then FtsK did initiate translocation in the direction dictated by the orientation of KOPS when chased with ATP (Figure 9B). These findings support a model where KOPS binding in the presence of ADP dictates the initial direction of FtsK movement, but that subsequent encounters with KOPS have no influence on translocation.

Figure 9. Interactions between FtsK and KOPS.

(A) FtsK binding distribution histograms for QD-tagged FtsK in the presence of ADP. (B) Examples of tracking data showing that FtsK leaves KOPS in the direction dictated by the orientation of KOPS. The relative orientation of each is depicted with the arrows, and the location of each KOPS site is indicated with a dashed orange line. 15 seconds of each trajectory are shown. (C) Kymograph showing the initial association of FtsK with the λ-phage substrate in a reaction containing 1 mM ADP. Examples of nonspecific and KOPS-specific binding are highlighted, and shown along with the corresponding particle tracking data. Adapted with permission from reference [22].

Our data also revealed that FtsK locates KOPS through a random 3D target search with no evidence of long-distance 1D diffusion within our optical resolution limits (±30-nm, equivalent to ~230-bp) (Figure 9C). Importantly, target search processes involving site-specific DNA-binding proteins are typically thought of within the context of proteins that can undergo free diffusion. However, FtsK is normally anchored to the division septum through its N-terminal integral membrane domain, and chromosomes undergoing segregation are threaded through this septum as they pass into new daughter cells. This geometry would seemingly preclude FtsK from undergoing a free diffusive search throughout the bacterial genome, and therefore we envision that FtsK searches for KOPS through a mechanism involving random sampling of the DNA as the genome is being pulled through the division septum. Thus the organization of the division septum, along with the locally high protein and DNA concentrations expected within the septum, and the high density of KOPS sites within the dif activation zone could all help contribute to enhance KOPS-specific targeting in vivo. Moreover, we anticipate the ability of FtsK to reverse translocation direction may also be suppressed within the division septum otherwise unidirectional chromosome separation cannot occur. Although challenging, future work will be essential for understanding how full-length FtsK behaves on DNA, and how its behavior is influenced by the spatial organization of the division septum.

Future Challenges

Single-molecule methodologies have provided unprecedented access to reaction mechanisms, and the continued evolution of the field is likely to provide new and important insights. However, one of the greatest challenges in the field will be to move towards increasingly more complex reactions that more closely reflect physiological scenarios. Many studies to date have been restricted to relatively simple, single component systems, or in some cases reactions involving two or three protein components. In addition, most studies have only been able to address one specific aspect of any particular system, without necessarily being able to follow the entire course of a DNA repair reaction from start to finish. These limitations reflect both the need for more readily accessible single-molecule methodologies, as well as further development of bulk biochemical assays that can recapitulate reaction pathways in their entirety, and their concurrent adaptation to conditions amenable to single molecule detection methods. As an example, we have been able to analyze the early stages of MMR in reactions involving MutSα and/or MutLα. Despite the challenges faced in developing these early experiments, from a bigger picture perspective they still only reflect relatively simple aspects of the overall reaction pathway, and do not begin to approach the full complexity of the reaction as it occurs in living cells. The full repair reaction can in fact be recapitulated in vitro using a complement of nine recombinant proteins [119–123], and it will be of important interest to translate these assays into single-molecule experiments. In addition, it is known that MMR must be coordinated with DNA replication, and at least some of the proteins must associate with the replication forks, and that MMR is influenced by both nucleosomes and histone modifications. Although far in the future, an ideal experiment would be able to recapitulate all of these more complex aspects of the post-replicative mismatch repair reaction. Similarly, our work with RecBCD and FtsK only recapitulate a small component of each motor protein’s function, and it will remain important to move towards understanding more fully how each contributes their respective biological pathways.

Supplementary Material

02. Video 1. Single-tethered DNA curtain.

This video shows an example of a single-tethered DNA curtain. The DNA (shown in green) is stained with the fluorescent dye YOYO1 and the molecules are anchored to a lipid bilayer and aligned along the leading edges of nanofabricated chromium barriers. Buffer flow is temporarily paused during the video causing the DNA molecules to diffuse up out of the evanescent field and disappear from view. The molecules reappear when flow is resumed. See Figure 1A and Figure 1B for a schematic image of a single-tethered DNA curtain, and references [15, 17, 18] for additional details.

03. Video 2. Double-tethered DNA curtain.

This video shows an example of a double-tethered DNA curtain. The DNA (shown in green) is stained with the fluorescent dye YOYO1 and the molecules are aligned along the leading edges of nanofabricated chromium barriers. The downstream ends of the DNA are anchored to chromium pentagons that are coated with an antibody directed against a hapten attached to the end of the DNA (digoxigenin). There was no buffer flowing through the sample chamber during data acquisition. See Figure 1C for a schematic image of a double-tethered DNA curtain, and references [13, 15] for additional details.

04. Video 3. Mismatch binding by MutSα.

Example of quantum dot-tagged MutSα bound to a single-tethered DNA curtain made using a lambda DNA substrate bearing a mismatch at a defined location. The DNA is stained with YOYO1 and is shown in green, and the proteins are shown in magenta. The proteins and DNA disappear simultaneously when buffer flow is terminated, verifying that the proteins are bound to the DNA and not absorbed to the sample chamber surface. See Figure 2 and reference [32] for additional details.

05. Video 4. One-dimensional diffusion along DNA by MutLα.

Example of quantum dot- tagged MutLα bound to a double-tethered DNA curtain. The DNA is stained with YOYO1 and is shown in green, the proteins are shown in magenta, and there was no buffer flowing through the sample chamber during data acquisition. For additional details, see reference [21].

06. Video 5. ATP-triggered release of MutSα from a mismatch.

Video and corresponding particle tracking data for MutSα bound to a mismatch on single, double-tethered DNA molecule. MutSα is labeled with a quantum dot (shown in white), and the DNA is unlabeled because the presence of YOYO1 was found to inhibit mismatch release by the protein. Buffer flow is turned on as indicated to permit the injection of ATP, and the time point corresponding to the entry of ATP into the sample chamber is also indicated. See Figure 3D and reference [21] for additional information.

07. Video 6. DNA translocation activity of RecBCD.

Example of a single DNA being digested by RecBCD and corresponding tracking data. The DNA is stained with YOYO1 and the protein is unlabeled. YOYO1 does not perturb the activity of RecBCD and the dye is displaced as the protein translocates and concurrently digests the DNA. RecBCD binds to the end of the DNA, and protein translocation is revealed as a reduction in the length of the DNA when ATP is present. See Figure 6 and reference [19] for additional information.

08. Video 7. RecBCD collisions with RNA polymerase.

Example of a single DNA bound by quantum-dot tagged E. coli RNA polymerase being digested by RecBCD and corresponding tracking data. The DNA is labeled with YOYO1 and is shown in green, and RNA polymerase is shown in magenta. See Figure 6 and reference [19] for additional information.

09. Video 8. DNA binding and translocation by FtsK.

Example of a double-tethered DNA curtain bound by quantum-dot tagged FtsK (shown in magenta) in the presence of ATP. The DNA is unlabeled because the presence of YOYO1 was found to inhibit the translocation activity of FtsK. See Figure 8 and reference [22] for additional information.