Initially we attempted to directly observe fluorescent RecA nucleoprotein filaments interacting with bacteriophage λ dsDNA in real-time by using total internal reflected fluorescence microscopy (TIRFM)⁹. Fully homologous fluorescent ssDNA that was complementary to three different loci of λ DNA (Fig. 1a) was generated by incorporation of 5-(3-aminoallyl) dUTP into ssDNA comprising a region of λ DNA using PCR, followed with covalent attachment of ATTO565 (see Methods section). RecA nucleoprotein filaments were assembled on these fluorescent ssDNA substrates in ensemble reactions containing ssDNA-binding protein (SSB) and the non-hydrolysable ATP analog, ATPγS (5’-O-3’- thiotriphosphate)⁴. ATPγS was used to maintain the filament in its active form, eliminate filament disassembly, and prevent dissociation of DNA pairing products^7,10-12. Using biochemical assays, we confirmed that the fluorescent ssDNA generated by this procedure was functional for RecA-mediated DNA pairing (Supplementary Fig. 1). The λ dsDNA, biotinylated at each end, was attached under flow to the interior surface of a single-channel microfluidic device (flowcell) (Fig. 1b). Due to sequential attachment of each end to the streptavidin-coated surface, most DNA molecules were extended to nearly (~80%) B-form length, and extension could be maintained in the absence of flow (Figure 1b, i and ii).

To confirm DNA pairing at the homologous λ DNA target site, reactions were conducted under ensemble conditions, and products extended on the surface of a flowcell for analysis by single-molecule, two-color TIRFM; dsDNA was imaged by YOYO-1 binding (green) and ssDNA via ATTO565 (red). DNA pairing products were observed; the sites of interaction coincided with the region of homology within the λ DNA molecule (Fig. 1b, i). For the 430 nt ssDNA, all bound fluorescent ssDNA-RecA filaments were at the homologous locus (observed fractional distance 0.51±0.02; n= 21; Supplementary Fig. 2).

Next, we attempted to detect homologous pairing in real-time using single-molecule TIRFM. Preformed RecA nucleoprotein filaments were introduced into a flowcell to which λ DNA molecules were tethered; buffer flow was stopped; and the reaction monitored in real-time (Fig. 1b, ii). Although the dsDNA was readily visible, we failed to observe any interaction between the fluorescent nucleoprotein filaments and extended λ DNA, even for reaction periods longer than 1 hour. However, we noticed that in addition to the desired doubly-tethered extended λ DNA molecules, some DNA molecules were attached only via one end (Fig. 1b, iii). When flow was stopped to score pairing with the doubly-tethered λ DNA molecules, these singly-tethered molecules relaxed to a randomly coiled state. Unexpectedly, when these unconstrained DNA molecules were subsequently re-extended by buffer flow, 80% (n=20) revealed a stable pairing product (Fig. 1b, iii). This finding suggested that either a free DNA end or random coiled DNA was needed for pairing. In the same field of view, there were also λ DNA molecules that had both ends attached, but at a relatively close end-to-end distance (Fig, 1b, iv). When the flow was stopped, we observed that these molecules also participated in homologous pairing during the time that flow was off, demonstrating that a free DNA end was not required. These unanticipated results revealed that DNA pairing did not occur on DNA that was extended to near its entropic elastic limit, and suggested that the DNA homology search required the 3-dimensional states that are accessible in randomly coiled DNA. Collectively, they suggested that a coiled conformation of the target dsDNA is crucial.

To address this possibility, we developed an alternative single-molecule imaging strategy that permitted reproducible measurement of the effects of dsDNA conformational structure, unperturbed by flow, on the DNA homology search process. This method utilizes a specialized flowcell (Fig. 2a), two optical laser traps operated in position-clamp mode, epifluorescent detection, fluorescent ssDNA-RecA filaments and a λ DNA-dumbbell (a single λ DNA molecule with a 1 μm polystyrene bead attached at each end (see Methods section¹³)). The DNA pairing assay was performed in situ using the dsDNA-dumbbell target, and the dual optical trap configuration was utilized to reliably vary the end-to-end distance of the dsDNA. The flowcell has 4 channels and a flow-free reservoir. Movement of DNA-dumbbells between channels of the flowcell was accomplished via stage translation; manipulation of optical traps relative to one another was accomplished using a steering mirror controlling one of the traps. Each experiment (Fig. 2b; Supplementary Movie 1) consisted of the following steps: 1) In channel 1, a streptavidin-coated bead was trapped in each of the two optical traps. 2) The beads were moved to channel 2 to capture a λ dsDNA molecule (biotinylated on both ends, and stained with YOYO-1) on one bead. 3) The beads were moved into channel 3, and by independent steering of a trap, the distal end of the DNA was attached to the second bead. 4) The DNA-dumbbell was moved to the dye-free channel for de-staining, and the end-to-end distance was fixed. 5) The DNA-dumbbell was moved to the flow-free reservoir containing the fluorescent ssDNA-RecA filaments. 6) After a defined incubation time, the DNA-dumbbell was moved back to channel 4, which is free of nucleoprotein filaments, extended to its contour length (~16 μm), and examined for DNA pairing products.

Shown in Fig. 2c are representative products of reactions where the DNA-dumbbells were initially held at a center-to-center bead distance of 2 μm, and incubated for 2 minutes in the reservoir that contained RecA nucleoprotein filaments. For the two homologous ssDNA nucleoprotein filaments shown (430 nt and 1,762 nt), the pairing is clearly at the homologous locus. For a 2 minute incubation with dsDNA at a bead-to-bead distance of 2 μm and the 430 nt substrate, 90% of the dsDNA molecules (n=29) contained a nucleoprotein filament stably bound to the expected region of homology (Fig. 3a). To determine the effect of end-to-end distance (i.e., three-dimensional conformation) on the RecA-mediated DNA pairing reaction, the reactions were performed at increasing bead separations (Fig. 3a). As the bead distance was increased from 2 to 8 μm, the efficiency of DNA pairing period decreased to near zero, extrapolating to zero at ~9 μm; for comparison, in the TIRFM experiments where no DNA pairing was detected in situ, the DNA end-to-end distance was ~13 μm.

We compared the time-course of homologous pairing for fixed center-to-center bead distances of 2 μm and 6 μm (Fig. 3b), to determine the effect of decreasing DNA conformational states on the rate of the reaction. For the 2 μm separation, the rate of DNA pairing increased with half-time of ~30 seconds, and approached a yield of 100%. When the separation was increased to 6 μm, the rate slowed 4-fold to a half-time of ~125 seconds, but nonetheless approached 100% (Fig. 3b). To establish the kinetic reaction order, we conducted single-molecule DNA pairing assays as a function of RecA nucleoprotein filament concentration (Supplementary Fig. S3). The reaction rate was independent of nucleoprotein filament concentration, showing that DNA pairing under these conditions is not diffusion-limited, but rather, limited by a rate-determining unimolecular step, as in the ensemble studies¹⁴, but that was dependent on dsDNA conformation and not the pairing step itself.

To understand the nature of the complex that limits the rate of DNA pairing, we varied the length of RecA nucleoprotein filaments. Shown in Fig. 3c is a comparison of the time-courses for 162, 430, and 1,762 nt nucleoprotein filaments. Increasing the ssDNA length ~4-fold, from 430 to 1762 nt, increased the observed rate of pairing ~3.8-fold. However, when the length of the ssDNA was decreased to 162 nt, we did not observe any stably bound homologously paired products after incubations for 10 minutes at the closest bead-to-bead distance possible (2 μm), even though this substrate was active in ensemble DNA pairing reactions (Supplementary Fig. 2). We conclude that that the length of the RecA nucleoprotein filament is crucial factor in rate-limiting step of homologous pairing.

In addition to the anticipated stable, homologously paired end products, short-lived non-homologous interactions were observed (Fig. 4a). These events, which occurred outside of the homologous regions, were relatively unstable and dissociated during the movement of the molecule from the reservoir to the observation channel, during the separation of beads, or after the λ DNA molecule was extended (Supplementary Movie 2). At most, these heterologous events lasted for a few tens of seconds and never persisted on the minute timescale. When the molecules from the 2 μm dataset were analyzed, 22% of the reactions with the 430 nt ssDNA and 40% of reactions with the 1,762 nt ssDNA had these unstable heterologously paired intermediates (Fig. 4b); for the 162 nt ssDNA, only 1 heterologously bound filament was seen out of 28 molecules.

Some intermediates of the pairing process had a second filament bound non-specifically to spatially-separated regions of the λ DNA molecule. For such a heterologously-bound nucleoprotein filament, when the relaxed DNA molecule was moved into the observation channel and the beads were separated for observation, the existence of a loop could be inferred from sudden recoil of the homologously paired spot. As the beads were separated, the weaker of the two heterologous interactions was released, and there was a simultaneous movement (“jump”) of the fluorescence at the homologous pairing locus (Fig. 4a; Supplementary Movie 3) resulting from the release of DNA that was constrained in the loop. Approximately ~12% (n = 50) of the DNA-dumbbells displayed loop release events for the 430 nt nucleoprotein filament and, consistent with expectations, when the length of the nucleoprotein filament was increased to 1,762 nt, the number of molecules with transient loop structures increased to 47% (n = 30) (Fig. 4c).

Our results clearly establish that both the 3-dimensional conformation of dsDNA and the length of the nucleoprotein filament are important determinants of the rate for DNA homologous pairing. These findings lead us to propose a model termed “intersegmental contact sampling” to describe the search for homology by a RecA nucleoprotein filament (Fig. 4d). One of the key features of the model is that the RecA nucleoprotein filament has a polyvalent interaction surface capable of binding simultaneously and non-specifically, but weakly, with non-contiguous segments of dsDNA. The second related feature of this model is that 3-dimensional conformational entropy of the dsDNA greatly enhances the probability that DNA sequence homology will be found through iterated homology sampling, via multiple weak contacts, by this polyvalent filament. This model is compatible both with our key experimental findings, which we expect would apply to the search in the presence of ATP as well, and with the involvement of heterologously-bound intermediates that have been inferred from biochemical studies^15,16. Our data show that dsDNA extended to near contour length fails to produce homologously paired products. This observation provides an explanation for the observation that the formation of stable DNA pairing products in single-molecule studies utilizing magnetic tweezers required negative plectonemic supercoils in the DNA target^7,8. In contrast, when a ssDNA-RecA filament was extended to near its contour length, homologous pairing with fully homologous coiled dsDNA occurred ⁷, which is compatible with our finding that the coiled structure of dsDNA is essential to the homology search. Here we established that as the end-to-end distance of the dsDNA was decreased, allowing it to assume a more random coil-like 3-dimensional conformation, the rate of DNA pairing increased because the local DNA concentration increases, and the likelihood that DNA segments will be in close proximity also greatly increases. The increased local DNA concentration results in a greater statistical probability that a single nucleoprotein filament can simultaneously interact with and sample multiple regions of the same DNA molecule. This, in turn, is manifest as a kinetically more efficient homology sampling process. In further support of the intersegmental contact sampling model, when the length of the ssDNA in the nucleoprotein filament is increased the observed rate of pairing is increased, as well as the number of nucleoprotein filaments with multiple, transient, heterologous intersegmental interactions. This shows that longer nucleoprotein filaments can simultaneously and independently sample more segments of the target dsDNA than shorter nucleoprotein filaments. Kinetically, our findings are consistent with the following two-step scheme: where K_het is the equilibrium constant for the binding of a RecA nucleoprotein filament (NPF) to heterologous dsDNA (the kinetic steps comprising K_het are rapid relative to k_s) and k_s is the rate-limiting unimolecular rate constant for intersegmental homology searching step within the dsDNA molecule or domain. In general, this kinetic formalism predicts a hyperbolic dependence of homologous pairing on the component concentrations unless the equilibrium constant for formation of the heterologous complex is large; when this is case, the observed rate is defined by the first-order rate constant, k_s. Given that the rate of target location is independent of nucleoprotein filament concentration, this implies that the heterologously-bound complex is saturated at a filament concentration of 100 pM (Supplementary Fig. 3), placing a limit on the equilibrium dissociation constant of <10 pM (i.e., K_het > 10¹¹ M^-1). In the context of this kinetic model, values for k_s are defined by the experiments in Fig. 3b & c, which show that the rate of the intersegmental homology search decreases 4-fold when the DNA end-to-end distance increases from 1 μm to 5 μm, and increases ~4-fold when the ssDNA length increases ~4-fold. The latter is suggests that the intradomainal search is enhanced proportionately by the increase in either heterologous contacts or the reach of the longer ssDNA. In many regards, the homology search by RecA has parallels to target location by sequence-specific DNA binding proteins, with the notable exception that the specificity of the RecA filament is determined by the sequence of the associated ssDNA. Seminal work on the DNA target selection by transcriptional regulatory proteins identified sliding, hopping, and intersegmental transfer as potentially facilitating mechanisms^17,18. Here we have established intersegmental transfer as the operative pathway used by RecA to find DNA sequence homology; this behavior is distinct from the sliding and hopping used to enhance the rate of target location by most regulatory proteins, which are typically univalent or bivalent with regard to site binding¹⁸. Our approach now provides a framework for future studies on the previously mysterious homology search by recombination proteins. It is applicable to studies of more complex systems such as eukaryotic Rad51, where it can provide insight into the function of the many accessory proteins that enhance DNA pairing⁹. Finally, more broadly, the imaging strategy and flow-free cell design can easily be adapted to visualize target location and mechanism of processes as diverse as DNA replication and repair, RNA interference, transcription, and protein translation, where the 3-dimenensional conformations of nucleic acids are undoubtedly important.

For DNA pairing using TIRFM, biotinylated λ DNA (1 pM, molecules) in SM2 (SM and 50 mM NaCl) was bound to the flowcell and then washed to remove free DNA, and to attach the second DNA end. Reactions were started by addition of 0.2 nM nucleoprotein filaments. For ensemble experiments visualized by TIRFM, nucleoprotein filaments and λ DNA were incubated either1 hour (162 nt substrate) or 30 minutes (430 nt substrate) at 37°C.

1

2

3

4