Long-distance lateral diffusion of human Rad51 on double-stranded DNA

Abstract

Rad51 is the primary eukaryotic recombinase responsible for initiating DNA strand exchange during homologous recombination. Although the subject of intense study for over a decade, many molecular details of the reactions promoted by Rad51 and related recombinases remain unknown. Using total internal reflection fluorescence microscopy, we directly visualized the behavior of individual Rad51 complexes on double-stranded DNA (dsDNA) molecules suspended in an extended configuration above a lipid bilayer. Here we show that complexes of Rad51 can bind to and slide freely along the helical axis of dsDNA. Sliding is bidirectional, does not require ATP hydrolysis, and displays properties consistent with a 1D random walk driven solely by thermal diffusion. The proteins move freely on the DNA for long periods of time; however, sliding terminates and the proteins become immobile upon encountering the free end of a linear dsDNA molecule. This study provides previously uncharacterized insights into the behaviors of human Rad51, which may apply to other members of the RecA-like family of recombinases.

The repair of double-stranded DNA breaks by homologous recombination is essential for maintaining genome integrity in most organisms (1–3). The importance of homologous recombination is highlighted by the finding that Rad51 null mutations are lethal in mice (4). Furthermore, defects in this repair pathway are associated with a variety of human cancers (5, 6). In eukaryotes, the broken ends of chromosomes are processed by 5′ to 3′ exonucleases to yield long single-stranded DNA overhangs (2, 3). Rad51, a DNA-dependent ATPase, assembles onto these overhangs, forming a nucleoprotein filament that is a key intermediate in homologous recombination (1, 2, 7, 8). The primary functions of this filament are to locate homologous sequence that can be used as a template to repair the damaged DNA strand and to initiate strand exchange (1, 7).

The structure and function of the complexes formed by Rad51 and the other RecA-like recombinases are conserved throughout evolution (8, 9). In their active states, Rad51 and related recombinases form a helical filament on DNA that induces a 50% extension of the bound DNA molecule (8). The extended nucleoprotein filament is correlated with DNA recombination activity; however, these proteins also form inactive filaments with shortened pitches (≈65–85 Å versus ≈90–130 Å) (10). Rad51 and related recombinases also form octameric rings with a central pore large enough to accommodate a double-stranded (dsDNA) molecule (11–16). These ring-like recombinase structures do not appear to be the form of the protein that is active during the strand exchange phase of homologous recombination. Although the biological role of these rings remains unknown, it has been suggested that they may function as DNA “pumps,” allowing the proteins to move along DNA (12, 13).

Here we have developed a unique total internal reflection fluorescence microscopy (TIRFM)-based assay to investigate the behavior of single fluorescent Rad51 complexes bound to dsDNA. We show that human Rad51 can bind stably to dsDNA and diffuse long distances in one dimension along the helical axis. The sliding proteins are likely in the ring-like conformation or bound to the DNA as short inactive filaments with reduced pitch. This study provides insights into the range of behaviors attributed to Rad51 and also presents a general approach that can easily be adapted to investigate the lateral movements of other protein molecules bound to DNA.

Results

Construction and Characterization of Fluorescent Rad51. Human Rad51 has five cysteine residues: C31, 137, 144, 312, and 319. Of these residues, C31, 312, and 319 are exposed on the surface, whereas 137 and 144 are buried within the interior of the protein. Modification of the wild-type protein with thiol reactive fluorescent dyes decreased the biochemical activity of the protein in an ensemble assay for DNA recombination (data not shown). Therefore, the three surface-accessible cysteine residues were mutated to serine, and a nonnative cysteine was added near the N terminus (A11C). This protein (referred to as A11C Rad51) was labeled with Alexa Fluor 555. Fig. 1A shows results from typical labeling protocols with the wild-type protein, the cysteine minus mutant, and the A11C version of Rad51. As illustrated in Fig. 1 A, removal of the surface exposed cysteines eliminated fluorescent labeling with maleimide-fluorophore conjugates, and the addition of the A11C mutation to the cysteine minus mutant allowed site-specific labeling of the protein with the fluorescent dye.

Fig. 1.

Fluorescent human Rad51 protein. (A Left) A Coomassie-stained gel showing wt Rad51, the cysteine minus Rad51 mutant, and A11C Rad51, which has a single surface cysteine. All three proteins were subjected to the same labeling and purification procedure (Supporting Materials and Methods, which is published as supporting information on the PNAS web site). (A Right) A fluorescence image of the same gel is shown. (B) Unlabeled wt or fluorescently tagged Rad51 was assembled onto circular single-stranded DNA and reacted with homologous linear duplex DNA. The resulting reaction products were deproteinized, resolved on an agarose gel, and stained with ethidium bromide. (C) ATPase assays with wt Rad51, unlabeled A11C Rad51, and the fluorescently tagged version of A11C Rad51.

The ensemble-level biochemical characterization of the fluorescent Rad51 was performed by using an in vitro homologous recombination assay with plasmid-sized DNA substrates as described in ref. 17. As shown in Fig. 2B, the fluorescently labeled version of human Rad51 exhibits recombination activity comparable to that of its unlabeled, wild-type counterpart. Similar results were obtained by using oligonucleotide substrates (data not shown; ref. 18). Rad51 also has a DNA-dependent ATPase activity (8). As shown in Fig. 1C, wt Rad51, unlabeled mutant Rad51, and the fluorescently tagged version of Rad51 all displayed similar levels of ATPase activity. These results indicated that neither the mutagenesis nor the fluorescent labeling had a drastic effect on the ability of the protein to hydrolyze ATP. Taken together, our data indicated that the fluorescent version of human Rad51 behaved similarly to the wt protein in all tested ensemble-level assays.

Fig. 2.

Human Rad51 slides on dsDNA. (A) Biotinylated λ-DNA molecules were tethered to a sample chamber surface protected with a lipid bilayer. Hydrodynamic force was used to extend the DNA molecules parallel to the surface. A 532-nm laser provided illumination, and data were collected by using an electron-multiplying charge-coupled device (CCD). (B) An image sequence showing Rad51 movement on an individual DNA molecule. Five nanomolar fluorescent Rad51 and 2 mM ATP were injected into the sample chamber, and images collected at 40-sec intervals for the duration of the experiment. The DNA is oriented vertically in the center of each frame. Off-axis fluorescent signals are due to protein molecules nonspecifically adhered to the surface and serve as stationary reference points. The numbers at the bottom show elapsed time, arrows indicate the direction of flow and highlight the movement of Rad51, and the tethered (T) and free (F) ends of the DNA molecule are indicated (refer to Movie 1 for a movie of Rad51 sliding).

Single-Molecule Assay for Rad51 Binding to DNA. An unappreciated aspect of TIRFM is the need for an inert environment that eliminates nonspecific interactions between the biomolecules under investigation and the surface of the sample chamber. For TIRFM, the microfluidic sample chamber surface was prepared by deposition of a supported lipid bilayer onto a fused silica slide sparsely coated with neutravidin (19). The bilayer formed on the surface and surrounded the immobilized molecules of neutravidin, which serve as fixed anchor points for the biotinylated λ-DNA, and provided an inert microenvironment mimicking the interior of the cell (19, 20).

The experimental design used to visualize fluorescent Rad51 on single molecules of dsDNA is illustrated in Fig. 2 A. Here the λ-DNA was tethered to the surface by one end, and the application of a hydrodynamic force was used to maintain the molecules in an extended configuration, parallel to the sample chamber surface and within the evanescent field (19, 21). This experimental design allowed continual visual inspection of fluorescent proteins bound to the DNA molecules at any point along their entire contour lengths. To initiate binding, Rad51 and ATP were flushed into the sample chamber. At high concentrations of Rad51 (≥50 nM), the DNA becomes coated by the fluorescent protein, but, surprisingly, the length of the DNA did not increase (data not shown), suggesting that the proteins were in a conformation inactive for recombination. Interestingly, reactions performed at lower concentrations of Rad51 (5 nM) revealed small, individual complexes bound to the DNA. Contrary to initial expectations, the fluorescent Rad51 complexes bound to the DNA were not stationary, rather they appeared to slide freely along the entire length of the tethered λ-DNA (Fig. 2B; see also Movie 1, which is published as supporting information on the PNAS web site). The movement always occurred in the direction of flow, the velocity was proportional to the flow rate, and the direction of sliding was not influenced by which end (left or right) of the λ-DNA was immobilized to the surface (data not shown). Similar sliding behavior was observed by using Rad51 labeled at different positions on its surface, with GFP-tagged Rad51 and with the Alexa Fluor 555-labeled Rad51 mixed with a 4-fold molar excess of unlabeled wild-type Rad51 (data not shown). This same behavior has been observed for thousands of Rad51 complexes on hundreds of different dsDNA molecules (see below).

Rad51 Stops Sliding and Binds Tightly to DNA Ends. The sliding proteins appeared to stop at the free end of the DNA molecules, where they remained tightly bound and accumulated over time (Fig. 2B and Movie 1). To verify this observation, we used a new technology that allows assembly of parallel arrays of aligned DNA molecules (19). In brief, the DNA molecules were tethered by one end directly to the fluid lipid bilayer that coats the fused silica surface. Hydrodynamic force was then used to organize the DNA molecules along the leading edge of a microscale barrier to lipid diffusion (19). This procedure yielded parallel arrays of DNA molecules aligned at defined positions on the surface of the sample chamber (Fig. 3 A and B).

Fig. 3.

Human Rad51 stops sliding at the free ends of the DNA. (A) Shows a depiction of the tethered DNA molecules and their response to changes in hydrodynamic force. (B) Tethered DNA molecules were assembled into an aligned array by using a combination of hydrodynamic force and microscale barriers to lipid diffusion. The DNA molecules were stained with YOYO1 and illuminated at 488 nm. In the presence of flow, all of the molecules can be view across their full contour lengths. In the absence of flow, the DNA molecules experience an increase in conformational entropy and diffuse out of the evanescent field. The free (F) and tethered (T) ends of the DNA molecules are indicated. (C) A DNA array bound by fluorescent Rad51 is shown. Each fluorescent spot within the array corresponds to a single-protein complex sliding down a DNA molecule, and the accumulation of Rad51 at the free end of the DNA is evident as a fluorescent “line” of protein extending across the array. In the absence of flow, the DNA molecules and the DNA-bound proteins diffuse out of view, confirming that they did not nonspecifically adhere to the lipid bilayer. Each image represents a single 100-msec frame taken from a real-time video (Movie 2). (Scale bar: 10 μm.)

The DNA molecules within the array are physically aligned with one another. Therefore, application of a fluorescent sequence- or structure-specific DNA binding protein is predicted to yield a fluorescent line extending across the array at a position corresponding to the binding site for that particular protein. When Rad51 (5 nM) was injected into the sample chamber with a DNA array, virtually all of the protein moved down the DNA molecules and accumulated at the free ends, which yielded a line of protein that extended across the array (Fig. 3C and Movie 2, which is published as supporting information on the PNAS web site).

Lateral Movement of Rad51 on DNA in the Absence of Buffer Flow. To allow observation over the entire length of the λ-DNA with TIRFM, it was necessary to maintain a constant flow of buffer through the sample chamber; otherwise, the DNA molecules experience an increase in conformational entropy and diffuse out of the evanescent field (21). As indicated above, the fluorescent proteins always appeared to move in the direction of buffer flow. Therefore, it was reasonable to presume that the movement of Rad51 was being driven by hydrodynamic force. To determine whether movement could occur in the absence of buffer flow, the λ-DNA was biotinylated at both ends and applied to the surface of the sample chamber under a constant, moderate flow force (19). Under these conditions, one end of the DNA binds to neutravidin immobilized on the surface, and the DNA is immediately extended by hydrodynamic force. Once fully extended, the second biotinylated end of the DNA molecule can bind to the neutravidin on the surface. This approach yielded DNA molecules that were suspended above the inert lipid bilayer and yet confined within the detection volume defined by the evanescent field (ref. 19; see also Movie 3, which is published as supporting information on the PNAS web site).

Rad51 and ATP were injected into a flow cell containing double-tethered λ-DNA, incubated for a brief period, and the unbound protein was then flushed from the sample chamber (Fig. 4). Once data collection was initiated, there was little or no free Rad51 remaining in the sample chamber of the microfluidic flow cell and removal of the unbound protein ensured that the fluorescent signals on the DNA were due only to protein molecules bound at the outset of the experiment. Flow was then terminated, and the complexes were monitored in the absence of the perturbing hydrodynamic force by capturing images at 20-sec intervals over a period of 33 min. As shown in Fig. 4B, Rad51 appeared to move long distances on the λ-DNA, even in the absence of the externally applied force (Movie 4, which is published as supporting information on the PNAS web site). Movement in the absence of flow was bidirectional, and the movements of different complexes on the same DNA were completely uncorrelated, precluding the possibility that pump drift or convection currents played a role in the observed behavior. This data suggested that the movement of Rad51 on the DNA occurred via a 1D random walk and was driven by thermal diffusion.

Fig. 4.

The movement of Rad51 on dsDNA occurs via a 1D-diffusion mechanism. (A) λ-DNA was tethered by both ends to a fused silica surface coated with a supported lipid bilayer. Rad51 and ATP were then injected into the sample chamber and allowed a brief period for binding. Unbound protein and ATP were then flushed from the sample chamber. (B) An image sequence showing the movement of Rad51 on dsDNA in the absence of flow force and ATP. Images were collected at 20-sec intervals for a total of 33 min, and the individual Rad51 complexes on the DNA are highlighted with arrowheads. Three different complexes are highlighted, although two occasionally come into close proximity with one another and can no longer be spatially resolved. Fluorescent signals not aligned with the axis of the DNA are due to proteins adsorbed to the sample chamber surface and serve as stationary reference points (see also Movie 4).

In Fig. 4B, there were three distinct complexes of Rad51 bound to the DNA, and the Rad51 in the center of the DNA displayed a decreased fluorescence signal relative to the two flanking complexes (Movies 3 and 4). This difference in emission intensity allowed us to distinguish between the different complexes, and we observed no evidence that the proteins could bypass one another as they moved back and forth along the DNA molecule. This outcome is expected if the diffusing complexes were unable to freely pass by one another as they moved along the DNA because they would be limited to a single-file diffusion mechanism (22). Although separate complexes on the same DNA occasionally merged for brief periods of time, we saw no evidence suggesting persistent interactions between the adjacent Rad51 complexes bound to the same DNA molecule (Fig. 4B). Rad51 could remain bound to the double-tethered λ-DNA for several hours even though they exhibited unrestricted lateral mobility along the helical axis.

To determine whether ATP hydrolysis was required, fluorescent Rad51 was mixed with ADP and injected into the sample chamber containing tethered molecules of DNA. When ADP was the only nucleotide cofactor present in the reaction mixture, Rad51 still bound to and diffused along the DNA (data not shown). Similar sliding behavior was observed when ATP was flushed from the sample chamber and replaced with ADP, ATPγS, or buffer lacking nucleotide cofactor (data not shown). The fact that the complexes continued to slide freely on the DNA in the absence of hydrolyzable ATP supported the hypothesis that the movement occurred via a 1D-diffusion mechanism.

The Movement of Rad51 on DNA Occurs via a 1D-Random Walk Mechanism. The lateral motion was also analyzed by single-particle tracking (please refer to Materials and Methods for details of the particle tracking procedures). For this analysis, data collected at 8.3 frames per sec were fit to 2D Gaussian functions to locate the centroid position of the fluorescent proteins (23). A graphical representation of this analysis is presented in Fig. 5. The sliding in the y direction along the DNA (parallel to the helical axis) was characterized by a series of short-distance oscillations, as predicted for a 1D random walk, rather than long continuous movements and could span regions encompassing several microns (Fig. 5A). In contrast, movement of the fluorescent protein in the x direction (perpendicular to the helical axis of the DNA) was highly restricted (Fig. 5B). We attribute this horizontal motion to a combination of noise in the measurements and entropically driven transverse fluctuations of the DNA molecules in the x-y plane relative to the sample chamber surface. Importantly, the amplitude of these fluctuations was at least an order of magnitude less than the movements observed in the y direction, ruling out the possibility that the movement observed for Rad51 along the helical axis was due to motion of the DNA itself.

Fig. 5.

Characteristics of diffusing complexes. (A) A graphical representation of the y displacement (i.e., parallel to the helical axis of the tethered DNA) is shown of three typical Rad51 complexes bound to dsDNA molecules monitored over a period of 124 sec. Measurements were made with double-tethered DNA in the absence of buffer flow, images were collected at 8.3 frames per sec, the positions of the protein complexes were measured by fitting the images to a 2D-Gaussian function, and movement was measured as the frame-to-frame change in position for each individual protein complex. (B) Shows the x displacement (i.e., perpendicular to the helical axis of the tethered DNA) for the three diffusing protein complexes. Time-dependent variations in the amplitude reflect local flexibility of the extended DNA molecules as well as changes in the signal-to-noise ratio as the fluorophores bleach over time. (C) The mean squared displacement for these three complexes was plotted as a function of time interval for a period up to 12 sec. ○, calculated data points; solid lines, linear fits to the data points. (D) A graph of the total distance versus time for the same three Rad51 complexes is shown. In D, the points on the graph represent the absolute value of the change in position; direction of movement is not implied. Each trace represents the total distance traversed by a single Rad51 complex during the indicated time interval. Like colors correspond to the same Rad51 particles in each of the four graphs.

We further analyzed the lateral movement of Rad51 along the DNA by measuring the squared displacements (in the y direction) between pairs of positions whose time interval ranged from 0.124 sec (1 frame) to 12.4 sec (Fig. 5C). The arithmetic average of the square displacements of the pairs separated by the same time interval was calculated and plotted versus the time interval (Fig. 5B). In most cases (47 of 50), the mean square displacement (MSD) plots yielded a linear curve, as expected for unbounded 1D diffusion (23–25). In the three examples shown in Fig. 5, the complexes displayed an average 1D-diffusion coefficient of 0.042 ± 0.054 μm²/sec. The cumulative apparent distance covered by the proteins was calculated as the running sum of the individual step sizes and ranged from ≈50–150 μm in a period of just 124 sec (Fig. 5D). This calculation revealed apparent velocities that ranged from 0.46 to 1.1 μm/sec (or 1.1–3.5 kb/sec). These movements allowed the proteins to scan back and forth across regions of the DNA spanning several micrometers over the 2-min duration of the observation (Fig. 5 A and D). Analysis of 50 different protein complexes revealed an average observed step size (i.e., the observed change in position of the protein complexes between adjacent frames within a data set) of 0.095 μm (≈300–400 bp) and diffusion coefficients that ranged from 0.001 to 0.21 μm²/sec (Fig. 6, which is published as supporting information on the PNAS web site; refs. 26 and 27). It should be noted that these measurements indicate upper bounds for the diffusion values and steps sizes, and smaller scale motions are likely to occur below the spatial and temporal detection limitations of our optical microscope.

Discussion

Direct Visual Detection of 1D Diffusion of Proteins on DNA. One-dimensional diffusion of proteins on DNA is a commonly invoked mechanism of facilitated target location (28, 29). However, movement that involves passive diffusion is difficult to measure in bulk assays and often requires theoretical assumptions to interpret the behavior of the proteins under investigation (28, 29). Single-molecule measurements of passive diffusion offer an attractive alternative to bulk measurements because they allow direct observation of the diffusing entities. Despite this potential, single-molecule approaches for measuring 1D diffusion of proteins on DNA have been very limited (27, 30). The use of TIRFM and single-particle tracking, along with long DNA substrates maintained in an extended conformation above an inert lipid bilayer, provides an experimental approach for probing the diffusion of proteins on DNA, which can be applied to virtually any protein that binds to DNA.

Recombinase Rings and Filaments. Several lines of evidence suggested the possibility that the Rad51 observed moving on the DNA was in the ring-like conformation and encircled the DNA molecules. First, in the TIRFM experiments, we do not see 50% extension of the DNA substrates, which is expected for the helical form of the nucleoprotein filament. We do, however, observe the expected 50% extension in experiments with fluorescently labeled DNA and unlabeled wt Rad51 (data not shown). Second, the diffusing complexes bound to the DNA were highly stable, often remaining bound for several hours, and we observed no evidence that proteins bound to the same DNA molecule were able to bypass one another as they moved back and forth along the DNA. Third, it is unlikely that the active helical form of the protein could maintain the DNA in an extended and untwisted conformation yet still be capable of sliding freely along the helical axis. Taken together, these data indicate that the fluorescent protein is capable of forming helical filaments under standard experimental conditions, but that in our TIRFM experiments, it is likely present as a ring structure similar to that reported for Dmc1 (13, 31). Although we suspect that the protein is bound to the DNA in a ring-like conformation, it is also possible that it was bound as an inactive helical filament with a reduced pitch.

Members of the RecA-like recombinase family can form filaments of varying pitch as well as ring-like structures. However, it is not entirely clear what molecular determinants influence filament pitch, the function of the rings has remained elusive, and it is not known what controls the ring-to-filament transition (11, 13, 15, 32). Reaction conditions that favor filament formation also support recombination, whereas reaction conditions that favor rings or reduced-pitch filaments do not support DNA recombination (15, 32). These observations suggest that the ring-like recombinase structures may represent an inactive form of the protein, which must somehow be activated before the formation of an active helical filament. Alternatively, it has also been previously proposed that the ring form of the proteins may enable the recombinase to translocate or pump DNA (13). Although we have observed no evidence for active translocation, our data clearly demonstrate that human Rad51 exhibits unrestricted lateral movement along the helical axis of dsDNA, yet the proteins do not slide off the free ends of the DNA molecules. This end-binding specificity suggests that the freely diffusing proteins undergo a conformational change upon encountering the DNA ends, which causes them to become tightly bound and unable to diffuse. It is possible that the ends of the dsDNA become transiently unpaired, and the diffusing proteins recognize and bind tightly to single-stranded DNA at the ends of the molecules.

Conclusion

Rad51, Dmc1, RecA, and RadA all form ring-like structures, and the evolutionary conservation of these recombinase rings from bacteria to humans strongly implies biological function (11–16, 31, 33). We have demonstrated that human Rad51 is capable of free lateral diffusion along the helical axis of dsDNA, consistent with a ring-like structure topologically linked to the DNA, and we have also shown that sliding stops at sites resembling a double-stranded break. To our knowledge, sliding of Rad51 (or any related protein) on DNA has never been experimentally detected, nor has 1D diffusion of this magnitude and duration ever been directly visualized for any other DNA-bound protein. A hypothesis suggested by these observations is that lateral diffusion of the protein may enable the recombinase to scan DNA for regions in need of repair. Alternatively, sliding may facilitate delivery of the recombinase to sites of damage with the aid of other DNA repair proteins. This previously undescribed behavior of human Rad51 highlights the significant advantages of single-molecule fluorescence-based approaches for examining the dynamics of macromolecular complexes, and the approaches presented here are applicable to the study of a wide range of proteins that move on DNA.

Materials and Methods

Protein Labeling and Ensemble Characterization. Purification and fluorescent labeling of human Rad51 was performed as described in the Supporting Materials and Methods. Ensemble level characterization of ATPase assays and in vitro recombination reactions were performed as described in refs. 17 and 34, and full details are also available in the Supporting Materials and Methods.

Total Internal Reflection Fluorescence Microscopy. The TIRFM, flow cell, sample delivery system, and surface modifications used here were described in refs. 19, 21, and 35. Complete experimental details are provided as Supporting Materials and Methods.

Data Analysis and Single-Particle Tracking. The procedures for tracking single particles have been described as have the methods for calculating 1D-diffusion parameters in refs. 23, 25, 26, 34, and 36. Details of these procedures are provided in Supporting Materials and Methods.