The Escherichia coli RecBCD is a helicase and nuclease complex that plays a critical role in repair of DSBs in bacteria by homologous recombination, whose function in promoting recombination is regulated by the Chi (χ, crossover hotspot instigator) sequence in DNA (Dillingham and Kowalczykowski, 2008; Smith, 2012). Biophysical properties of RecBCD have been extensively characterized over the past two decades using a plethora of different single-molecule techniques. Though known to be a highly processive helicase, the actions of individual RecBCD complexes had not been directly observed (Roman et al., 1992). Using YOYO-stained λ-DNA conjugated to an optically trapped bead and extended by flow (Figure 3A, left), velocity and processivity of single RecBCD enzymes were measured by quantifying the loss of YOYO signal as the dsDNA was converted to ssDNA through actions of the enzyme (Figure 3A, right) (Bianco et al., 2001). Using the same imaging technique, the mechanism of regulation for the recombination hotspot χ sequence was elegantly elucidated. Single RecBCD complexes were observed to pause precisely upon encountering the χ site and slow down afterwards, which was revealed to be due to a change in the lead helicase from RecD to RecB, rather than the loss of RecD as previously believed (Spies et al., 2003; Handa et al., 2005; Spies et al., 2007). Moving from naked DNA towards a more physiological environment, further studies on RecBCD using the high throughput DNA curtains showed that the powerful complex is capable of ejecting stably bound proteins from DNA (Finkelstein et al., 2010; Terakawa et al., 2017).

An early responder in mammalian DNA damage response, 53BP1 is an enigmatic factor that is known to prevent the formation of long 3′ overhangs by limiting 5′ end resection at DSB sites in G1 cells (Mirman and de Lange, 2020). The 53BP1-mediated block to end resection mediated is accomplished through effector proteins RIF1 and PTIP, whose recruitment depends on 53BP1 N-terminal phosphorylation by ATM (Callen et al., 2013; Zimmermann et al., 2013). For HR to proceed, restoration of end resection in S/G2 cells relies on the antagonistic functions of BRCA1 towards 53BP1 (Chapman et al., 2012; Densham et al., 2016; Hustedt and Durocher, 2016). Super-resolution light microscopy techniques such as the single-molecule localization microscopy (SMLM) has proven invaluable in elucidating the behavior of 53BP1 in response to DNA damage by ionizing radiation (Depes et al., 2018). SMLM measurements have shown cell type specific recruitment patterns of 53BP1, as well as dynamic changes in chromatin architecture, after high and low linear energy transfer irradiations (Bobkova et al., 2018; Hausmann et al., 2020).

End resection in human cells begins with the short-range resection initiated by the MRN complex with its phosphorylated cofactor CtIP, which produces a nick ∼20 nt away from the end (Anand et al., 2016; Cannavo and Cejka, 2014). Single-molecule imaging of fluorescently labeled MRN showed that the protein utilizes facilitated diffusion to reach the DNA ends (Figure 3B, left) (Myler et al., 2017). Because the NHEJ initiating factor Ku also binds tightly to DNA ends, it raises the question of how MRN behaves when encountering DNA-bound Ku, as it also relates to the problem of pathway choice in DSB repair (Scully et al., 2019). Myler et al. showed in the same study that MRN is able to release DNA-bound Ku via an Mre11-dependent nucleolytic reaction (Figure 3B, right), thus providing a mechanism for initiation of HR even when Ku is the first to arrive at DSB sites (Myler et al., 2017). Recent follow-up on the topic from the same groups corroborated and extended the initial finding by including CtIP and DNA-PKcs in the DNA curtain assay, showing nucleolytic release of DNA-PK by MRN/CtIP (Deshpande et al., 2020).

The short overhang generated by MRN allows long-range resection factors BLM/DNA2 or EXO1 to assemble and carry out extensive resection. Significant insights into the biophysical characteristics of these enzymes as well as their regulation have been gained from single-molecule studies. Fluorescence imaging on DNA curtains demonstrated that human and yeast Exo1 are both processive nucleases that are susceptible to displacement by multivalent ssDNA binding proteins such as RPA, though extensive resection by human EXO1 was supported by the SOSS1, another ssDNA binding complex essential for HR in human cells (Myler et al., 2016). The coordination and regulation of long-range end resection among its participants was well illustrated in a recent DNA curtain study focused on Sgs1, the yeast ortholog of BLM (Figure 3C) (Xue et al., 2019b). The authors showed that Sgs1 unwound dsDNA from internal positions in the presence of RPA (Figure 3Ci) and can be targeted to dsDNA ends in either Top3-Rmi1-dependent or independent manner (Figure 3Cii). However, Sgs1 remained inactive at DSBs until the addition of Dna2, which activated long-range translocation by Sgs1 from DNA ends (Figure 3Ciii). Furthermore, this complex lacked nucleolytic activity, which was only triggered through addition of RPA (Figure 3Civ), thus underscoring the importance of RPA in end resection as previously reported (Cejka et al., 2010a; Niu et al., 2010). Simultaneously, Sgs1 functions were also being studied using magnetic tweezers, where dsDNA unwinding initiated from a ssDNA gap with a 5′ flap produced comparable velocities to those from DNA curtain measurements, though rewinding of dsDNA, suggested to involve strand switching by Sgs1, was also observed (Kasaciunaite et al., 2019). In addition to the role in end recruitment of Sgs1 observed on DNA curtains, Top3-Rmi1 was shown to increase Sgs1 velocity when initiating translocation from flapped gap substrate, consistent with previously observed stimulatory effects (Cejka et al., 2010b; Kasaciunaite et al., 2019). The role of RPA in human RECQ helicase BLM-mediated resection was examined by two recent DNA curtain studies. BLM exhibited high speed and robustness in DNA unwinding regardless of the presence of RPA, while its end resection activity was dependent on the phosphorylation status of RPA (Xue et al., 2019a; Soniat et al., 2019). In the latter paper, resection by BLM/EXO2 or BLM/DNA2 on single-tethered DNA curtain was quantified in the presence of RPA or its phosphomimetic or phosphoblocking mutants. Phosphorylation on residues in RPA32 was found to reduce velocity and processivity of end resection by both BLM/EXO2 and BLM/DNA2, as well as inhibit their resection past individual nucleosomes, therefore acting as a negative regulator of resection (Soniat et al., 2019).

As the 3′-ssDNA tail is being generated by long-range resection, it is rapidly bound and protected from nucleases by single strand binding proteins, which must then be replaced by recombinases. The bacterial SSB and recombinase RecA were among the first to be studied using single-molecule methods. Efforts to characterize fundamental behavior of SSB typically favored smFRET experiments with surface-immobilized single-stranded DNA that is labeled at the ends with a donor/acceptor fluorophore pair. Wrapping of ssDNA around the SSB tetramer during binding would bring the donor and acceptor closer and allow FRET efficiency to be used as a main observable. It was elegantly shown that tetrameric SSB could spontaneously diffuse on ssDNA, capable of removing secondary structure such as a small stem-loop hairpin and promoting formation of RecA filaments (Roy et al., 2009). Coupling an optically trapped bead to the smFRET substrate to apply pN-level of force on the complex, Zhou et al. was able to discern the molecular mechanism for SSB sliding as reptation, where the motion is facilitated by the formation of a DNA bulge and its propagation around the protein opposite its direction of sliding (Zhou et al., 2011).

Following binding of SSB to the 3′-ssDNA overhang, the E. coli recombinase RecA must be loaded to form a nucleoprotein filament capable of homology search and strand invasion. Observation of this process at the single-molecule level was first reported using smFRET and DNA substrates with short ssDNA overhang, where the donor/acceptor pair was placed at the junction and end of ssDNA (Figure 4A) (Joo et al., 2006). Five monomers of RecA were determined necessary for nucleation and dynamic binding and dissociation of single monomers from both ends of the filament contributed to filament growth. Notably, it was shown that RecA could displace SSB from ssDNA when a preformed nucleation cluster was present (Joo et al., 2006). Having first shown RecA binding of flow-extended double-stranded λ-DNA tethered to an optically trapped bead, Kowalczykowski et al. then reported formation of fluorescently labeled RecA filament on SSB-coated surface-tethered ssDNA (Galletto et al., 2006; Bell et al., 2012). In this work, Bell et al. observed that a RecA dimer was required for filament nucleation through titration of RecA concentration and its relationship with nucleation frequency. Using two-color labelling of RecA, it was demonstrated that RecA filament growth was bidirectional but faster in the 5′-3′ direction, consistent with previous findings (Galletto et al., 2006; Joo et al., 2006). Furthermore, E. coli recombination mediator proteins RecOR were shown to stimulate both RecA nucleation and filament growth (Bell et al., 2012).

Many of the same characteristics exhibited by SSB and RecA are conserved in their eukaryotic counterparts. Human RPA has been shown by smFRET to diffuse on ssDNA and melt secondary structures (Nguyen et al., 2014). Dynamics of RPA filament were thoroughly investigated using single-stranded DNA curtains (Gibb et al., 2012; Ma et al., 2017b). RPA filament formed on ssDNA was shown to be stable for over 2 h when unbound proteins were flushed out (Gibb et al., 2014a; Deng et al., 2014; Ma et al., 2016). When challenged with free protein in solution, it was observed that ssDNA-bound RPA could be exchanged with those in solution in a manner dependent on concentrations of the free RPA, consistent with facilitated dissociation previously reported for DNA binding proteins with multiple contacts (Graham et al., 2011; Gibb et al., 2014a; Deng et al., 2014; Ma et al., 2016). Direct visualization of fluorescently labeled human RAD51 showed conserved end-biased bidirectional filament growth on dsDNA, initiated by nucleation of ∼2–3 monomers (Hilario et al., 2009). Assembly of RAD51 filament on the physiologically relevant RPA-coated DNA as well as its disassembly characteristics were also recapitulated on ssDNA curtains (Ma et al., 2016).

To overcome RPA-mediated inhibition of RAD51 filament formation, mediator proteins such as BRCA2 and RAD51 paralogs are needed (Bonilla et al., 2020). Effects of yeast Rad52, considered a possible functional ortholog of human BRCA2, on the dynamics of presynaptic filaments were revealed using ssDNA curtains (Figure 4B, left) (Gibb et al., 2014b). Fluorescently labeled Rad52 was shown to nucleate on RPA-coated ssDNA and promote bidirectional growth for additional Rad52 binding (Figure 4B, right). Rad52-RPA clusters were also observed to remain after formation of extended Rad51 filaments and served as nucleation sites for additional binding of RPA and Rad52 (Gibb et al., 2014b). Recent smFRET work suggests that the Rad52 destabilizes the DBD-D DNA binding domain of RPA, thereby increasing access to ssDNA previously occluded by RPA (Pokhrel et al., 2019). Many of these same characteristics of Rad52 were also recapitulated in a DNA curtain study of human RAD52, whose deletion in vertebrates does not produce a strong phenotype, with the exception that human RAD52 and RPA could not rebind to remaining clusters after assembly of human RAD51 filaments (Ma et al., 2017a). In addition, effects of RAD51 paralogs on presynaptic filaments have been the subject of several recent single-molecule studies. In a smFRET study of C. elegans RAD-51 paralogs RFS-1/RIP-1, surfaced-immobilized substrates were labeled with donor and acceptor dyes seven nucleotides apart in the ssDNA region (Taylor et al., 2015). Addition of RAD-51 to naked ssDNA led to transition from high FRET to low FRET, reflecting the stretching of ssDNA upon RAD-51 binding. RFS-1/RIP-1 bound RAD-51 filament exhibited intermediate FRET value along with broadening of the FRET signal distribution, suggesting that these paralogs remodeled RAD-51 filaments to a more flexible conformation (Taylor et al., 2015). Most recently, optical tweezers with confocal fluorescence imaging (Lumicks C-trap) and DNA curtains were separately applied to better understand the actions of RFS-1/RIP-1 and the yeast Rad51 paralogs Rad55-Rad77, respectively (Belan et al., 2021; Roy et al., 2021). Both studies showed that the paralogs promote RAD51 filament assembly through transient interactions, dissociating rapidly by hydrolyzing ATP. In addition, Belan et al. found that RFS-1/RIP-1 synergize with BRC-2 (human BRCA2 homolog) in promoting presynaptic filament assembly, specifically by engaging with the 5′ end of the RAD-51 filament to stimulate growth in a 3′→5′ direction (Figure 4C) (Belan et al., 2021). Roy et al. also showed that Rad55-Rad57 antagonism of anti-recombinase Srs2 might be through promoting faster re-assembly of Rad51 rather than inhibiting the anti-recombinase itself, as previously suggested (Liu J. et al., 2011; Roy et al., 2021).

Excessive recombination, also referred to as hyper-recombination, however, can be genotoxic and must be prevented. Counteracting the effects of RAD51 mediators that promote filament assembly are the negative regulators, or anti-recombinases. Several SF1 helicases are known to display anti-recombinase activity. Bacterial UvrD and PcrA have been implicated in dismantling of RecA filaments in genetic and biochemical experiments (Veaute et al., 2005; Bidnenko et al., 2006; Lestini and Michel, 2007; Petrova et al., 2015). In particular, smFRET studies showed that PcrA strips RecA filaments off DNA through a reeling motion (Park et al., 2010). The yeast SF1 helicase Srs2 has also been shown to prevent recombination by dismantling the Rad51-ssDNA nucleofilament through stimulation of ATP hydrolysis by Rad51 and its dissociation (Krejci et al., 2003; Veaute et al., 2003; Antony et al., 2009). Actions of Srs2 on different HR intermediates have been visualized at the single-molecule level. In one smFRET study, Srs2 cleared Rad51 bound to short ssDNA overhangs and exhibited repetitive motion at the ssDNA/dsDNA junction, proposed to prevent reformation of Rad51 filament (Qiu et al., 2013). Single-stranded DNA curtain experiments showed that Srs2 was capable of processively translocating on naked ssDNA, as well as RPA-coated ssDNA, Rad51-ssDNA, and ssDNA bound by both RPA and Rad52 (De Tullio et al., 2017; Kaniecki et al., 2017). While translocating on protein-bound ssDNA, Srs2 also efficiently removed RPA, Rad51, Rad52, and short heteroduplexes formed with Rad51. Remarkably, this robust anti-recombination function of Srs2 was strongly inhibited by the presence of meiosis-specific recombinase Dmc1 within the presynaptic filament (Crickard et al., 2018).

In addition to SF1 helicases, members of the RecQ subfamily of SF2 helicases have also been implicated in anti-recombination functions (Branzei and Szakal, 2017; Larsen and Hickson, 2013). A DNA curtain study showed that Sgs1, apart from its role in end resection, also acted on presynaptic filaments (Crickard et al., 2019). Sgs1 was observed translocating on RPA-coated ssDNA and, in accordance with its expected anti-recombinase activity, displacing Rad51 while translocating on the Rad51-ssDNA filament. Sgs1-mediated Rad51 removal was found to be independent of Rad51 ATP hydrolysis, in stark contrast to the mechanism employed by Srs2 (Antony et al., 2009; Kaniecki et al., 2017). Though similar to the case of Srs2, Sgs1 action was also inhibited by Dmc1 (Crickard et al., 2019). Functional conservation of RECQ helicases in anti-recombination was recently demonstrated for the human RECQ5 on DNA curtain (Figure 4D, left) (Xue et al., 2021). RECQ5 not only translocated on ssDNA bound by RPA, RAD51, or DMC1, but also removed these proteins in the process (Figure 4D, right). Real-time observation of RAD51 removal by RECQ5 is consistent with previous results from biochemical assays (Hu et al., 2007). Similar to Sgs1, RECQ5 was able to strip ATPase-deficient RAD51 from ssDNA, suggesting a mechanism not coupled to RAD51 ATP hydrolysis. The ability of RECQ5 to translocate and disrupt DMC1 filaments contrasts with the inhibitory effects of Dmc1 on Sgs1 and Srs2, suggesting that it may play a role in meiosis (Xue et al., 2021). Finally, as mentioned above in the context of DNA end resection, BLM showed robust dsDNA unwinding but little interaction with RPA- or active ATP-bound RAD51-coated ssDNA in DNA curtain assays, even though it was considered an anti-recombinase capable disrupting inactive ADP-bound RAD51 filaments (Bugreev et al., 2007; Xue et al., 2019a). The apparent differences in the abilities of RECQ5 and BLM to interact with different HR intermediates may arise from differences in protein domain architecture and reflect division of labor among RECQ helicases in HR.

Once the stable presynaptic filament forms, it must locate sequence homology elsewhere in the genome, against a vast background of heterologous sequence. The recombinase then catalyzes a strand exchange reaction to form a heteroduplex containing the ssDNA base paired with the complementary strand and displacing the strand containing the homologous sequence (D-loop). At its core, the homology search process is highly similar to target search by other ubiquitous sequence specific DNA-binding proteins. A theoretical solution to the target search problem has been known for four decades as facilitated diffusion (Berg et al., 1981). Understanding of this process, involving a combination of 1D and 3D diffusion as well as microscopic hopping and intersegmental transfer, has seen significant contributions from single-molecule experiments. Following their earlier work on RecA filament assembly, the Kowalczykowski and Ha groups were the first to shed light on the mechanism behind homology search by RecA presynaptic filaments (Forget and Kowalczykowski, 2012; Ragunathan et al., 2012). In the first study, dsDNA serving as the homologous sequence donor was held between two optically trapped beads in a “dumbbell” configuration (Figure 5A). By systematically varying the distance between the beads, hence the contour length and 3D conformation of the DNA, it was shown that RecA filaments conducted homology search via multiple weak contacts for sampling DNA sequence within a 3D volume, a mechanism the authors termed “intersegmental contact sampling” (Figure 5A) (Forget and Kowalczykowski, 2012). In the second smFRET study, free dsDNA homology donor and surface-immobilized RecA-ssDNA were labeled with donor and acceptor dyes, respectively. By observing the dynamic FRET values while controlling for sequence homology, a sliding model was proposed, in which RecA filaments can diffuse along the dsDNA track while efficiently sampling for homology as short as six nucleotides (Ragunathan et al., 2012). Together these studies demonstrate that homology search by RecA filaments occurs through facilitated diffusion using a combination of 1D sliding and 3D diffusion, expedited via intersegmental contact sampling.

More detailed understanding of minimum sequence homology requirements and kinetics of sampling were uncovered in a pair of papers using single-stranded DNA curtains (Lee et al., 2015; Qi et al., 2015). In these experiments, presynaptic filaments were assembled on long ssDNA tethered to the lipid bilayer surface, while fluorescently labeled duplexes containing varying degrees of homology were free in solution (Figure 5B). The first study revealed that eight nucleotides of homology was the minimum requirement for recognition by Rad51 and stable capture, while subsequent strand exchange occurred in precise three nucleotide steps (Qi et al., 2015). The follow-up work illustrated that the base triplet stepping for homology recognition (Figure 5B) was a conserved feature in the RecA family of recombinases from RecA to Rad51, including the meiosis-specific Dmc1. Dmc1, however, was also unique in its ability to stabilize internal mismatches. Whereas mismatches in RecA or Rad51 filaments could be tolerated but would not contribute to stabilizing recognition complex (Lee et al., 2015).

Adding to the complexity of homology search is the fact that the process in eukaryotic cells also involves the multi-functional SWI2/SNF2 motor proteins Rad54 and Rdh54 (Ceballos and Heyer, 2011). Early single-molecule work had shown that both Rad54 and Rdh54 are highly processive translocases on dsDNA (Amitani et al., 2006; Nimonkar et al., 2007; Prasad et al., 2007). While Rad54 is also known to facilitate Rad51-mediated homologous DNA pairing in vitro and homology search in vivo, its exact mechanism of action remained unknown (Petukhova et al., 1998; Renkawitz et al., 2013). Recently, using double-tethered dsDNA curtains as sequence donor and labeled partial duplex DNA with 3′ Rad51-ssDNA filament, Rad54 was shown to promote targeting to homologous DNA by translocating with the presynaptic filament on dsDNA (Figure 5C) (Crickard et al., 2020). This ATP-dependent behavior adds to the 3D diffusion mechanism of the homology search and serves in reducing dimensionality and increasing search efficiency. Moreover, while driving active translocation, Rad54 induced transient strand opening coupled to RPA binding, potentially allowing the Rad51 presynaptic complex to sample both strands of dsDNA donor for homology (Crickard et al., 2020).

Synapsis is the step in which the two broken DNA ends are brought together to close proximity such that NHEJ machinery may assemble in a stable complex and assess the actions needed to restore the structural integrity of DNA. Detailed mechanistic insights on this critical early step are therefore prerequisite to understanding of the pathway. The dynamic nature of the process involving two DNA ends has made smFRET the single-molecule platform of choice in studying the system. By measuring the fluorescence energy transfer between the donor-labeled surface-immobilized fragment and the acceptor-labeled freely-diffusing fragment, smFRET allows real-time monitoring of intermolecular synapsis. High FRET indicates close proximity of the two DNA ends, while fluctuating FRET values would suggest dynamics in the process of DNA end alignment.

Early models derived from work using purified proteins in bulk biochemical assays, electron microscopy, and x-ray scattering as well as laser microirradiation of cells followed by immunofluorescence imaging suggested that DNA-PKcs is recruited by Ku to DNA breaks and together they are able to bridge the broken DNA ends (DeFazio et al., 2002; Hammel et al., 2010b; Kim et al., 2005; Weterings et al., 2003). Nonetheless, the lack of spatiotemporal resolution precluded these studies from revealing any transient intermediate steps or subcomplexes in the process. Using smFRET with two DNA fragments containing four nucleotide homology at the ends and differentially labeled with donor and acceptor fluorophores, Rothenberg and coworkers observed co-localization of the donor/acceptor pairs after addition of purified NHEJ components except DNA-PKcs (Figure 6A, left) (Reid et al., 2015). Although aggregated joining was observed in the presence of DNA-PKcs, this result cast doubt over the requirement of DNA-PKcs in synapsis. This end joining process mediated by Ku70-Ku80, XRCC4-LIG4, and XLF was revealed to be dynamic, as shown by FRET efficiency distributions (Figure 6A, right). These distributions exhibited widths indicative of the possibility that the DNA ends may be positioned in a side-by-side manner, in addition to end-to-end. Examination of fluctuating FRET trajectories in conjunction with using substrates that varied in end chemistry also supported the notion of DNA ends in “adjacent configuration” during the synaptic process that is highly dynamic (Figure 6A). More mechanistic details were uncovered in a later follow-up smFRET study by the Rothenberg and Lieber laboratories (Zhao et al., 2019). It was shown that the first “flexible” stage of blunt end synapsis (FS), mediated by Ku and XRCC4-LIG4, involves the dsDNA ends being brought into a parallel side-by-side configuration where they can still slide along each other, as evidenced by fluctuating FRET efficiency values. Flexible synapsis, shown to be independent of DNA-PKcs, can then be converted to a close synaptic state (CS) by XLF or PAXX, where the two DNA ends are aligned in close proximity in an end-to-end manner.

The lack of DNA-PKcs requirement in synapsis as monitored by smFRET contradicts existing evidence for its role in NHEJ in vivo (Baumann and West, 1998; Cottarel et al., 2013; Jiang et al., 2015; Zhao et al., 2006). This apparent discrepancy was further investigated by Loparo and coworkers using smFRET and cell-free extract of X. laevis eggs (Figure 6B) (Graham et al., 2016). Xenopus egg extract represents a more physiological system compared to biochemical reconstitution of purified proteins, and has been established for single-molecule imaging studies as well as being capable of Ku- and DNA-PK-dependent DNA end joining (Di Virgilio and Gautier, 2005; Labhart, 1999; Yardimci et al., 2012). In addition to intermolecular synapsis of two separate DNA fragments, a longer 2 kbp DNA fragment with donor/acceptor-labeled blunt ends and an internal biotin for surface immobilization was used in this study to facilitate intramolecular end joining (Figure 6B, left) (Graham et al., 2017). Based on the distance between the donor/acceptor dyes, synapsis was observed to occur through two distinct stages: long-range (LR) where both dyes were present but no FRET, and short-range (SR) where FRET was seen between the dye pair (Figure 6B, right). In contrast to previous single-molecule work, LR synapsis in Xenopus egg extracts required both Ku70-Ku80 and DNA-PKcs, though the kinase activity of the latter is not needed. Transition from LR to SR synapsis would occur after several seconds and require the catalytic activity of DNA-PK, as well as the presence of XRCC4-LIG4 and XLF, though not the catalytic activity of LIG4.

Unlike order of assembly studies, quantifying biophysical observables such as step-wise reaction energetics has mostly been intractable for bulk biochemistry. In particular, a novel DNA substrate featuring two free DNA ends tethered via a leash held by magnetic tweezers has been developed as a unique single-molecule force spectroscopy approach to probe the energetics of NHEJ synapsis with reconstitution of purified proteins (Kostrz et al., 2019; Wang et al., 2018). By cycling between low and high forces on a single tether and monitoring changes in tether length, Strick and coworkers demonstrated that Ku and DNA-PKcs are required to first establish a brief (∼100 ms) stage of synapsis of DNA ends (Wang et al., 2018) (Figure 6C). This initial step is further stabilized by either XRCC4-LIG4 and XLS and/or PAXX, each contributing k _B T-scale energy, leading to long-lived (∼seconds) intermediate stages and stable (∼minute) synaptic complexes. Notably, these results support the two distinct stages of synapsis observed by Graham et al. using smFRET. The subcomplex containing Ku and DNA-PKcs and stabilized by PAXX (∼2 s) appears consistent with the long range synapsis, while the full complex further stabilized by XRCC4-LIG4 and XLF (∼66 s) would correspond to the short range synapsis (Stinson and Loparo, 2021). Most recently, the same technique was applied to demonstrate the dynamic properties of prokayrotic NHEJ synapsis involving just the Ku heterodimer and Ligase D (Oz et al., 2021). Although debates remain regarding whether DNA-PKcs is required for synapsis, as the results appear to be dependent on the system employed, recent single-molecule work have unambiguously shown the process to be a dynamic process with distinct stages.

XRCC4-like factor (XLF, or Cernunnos) is identified as an interactor of XRCC4 and regulator of ligation (Ahnesorg et al., 2006; Buck et al., 2006). X-ray crystallography and electron microscopy studies have shown that XLF and XRCC4 can form filaments in crystals (Andres et al., 2012; Hammel et al., 2010a; Mahaney et al., 2013; Ropars et al., 2011). Filamentous structures of XRCC4, XLF, and LIG4 have also been observed at DSB sites using super-resolution fluorescence microscopy (Reid et al., 2015). The mode of interaction between XLF-XRCC4 complexes and DNA remained elusive until the collaborative work from the Modesti, Peterman, and Wuite groups. In a single-molecule tour de force, dual- and quadruple optical traps were combined with wide-field fluorescence imaging to demonstrate that XRCC4-XLF complexes robustly bridged two independent DNA fragments (Brouwer et al., 2016). These complexes acted like sleeves that were able to withstand high applied forces and capable of sliding along DNA molecules (Figure 7A). Specific contributions from XLF in synapsis in the presence of other NHEJ core proteins were elucidated using smFRET (Figure 7B, left). Mutagenesis in XLF and XRCC4 showed that close alignment of donor/acceptor dye labeled DNA ends in the xenopus egg extracts system required interactions between these two proteins (Graham et al., 2018). Moreover, binding of a single dye-labeled XLF dimer was sufficient to mediate this short-range synapsis, which is shown to also be dependent on interactions of both XLF head domains with XRCC4 (Figure 7B, right). These findings call into question the requirement and relevance of XLF-XRCC4 filaments, as observed in bulk, in NHEJ. More corroborating evidence incompatible with the XLF filament hypothesis emerged in a subsequent smFRET study by Rothenberg and Lieber laboratories using reconstituted human NHEJ proteins. XLF was found to drive DNA ends into close proximity in a manner that is not strongly dependent on XLF concentrations, suggesting that only one to a few XLF dimers are needed at the DNA ends (Zhao et al., 2019).

Many DNA ends at DSB sites, regardless of their origins, require end processing before repair. The arsenal of NHEJ end-processing enzymes include nucleases, polymerases, kinases, phosphatases, and phosphodiesterases (Chang et al., 2017). The effects of chemically diverse DNA ends have on the dynamics of how they come together during synapsis is a question uniquely suited for single-molecule studies. Pairing efficiency as monitored by smFRET was shown to be strongly affected by phosphorylation status of the 5′ end of compatible DNA ends with four nucleotide overhangs in a minimal reconstituted system (Reid et al., 2017). Two distinct kinetic regimes, transient (<5 s) and persistent (>30 s), were found to exist for end pairing during the process, and that their energetics are modulated by the 5′ phosphate, through recognition by LIG4. In the absence of other end processing factors in this single-molecule work, a model involving an iterative process was proposed, where incompatible ends within a synaptic complex would fall apart to provide access by the processing enzymes and thus generating new compatible ends for synapsis and ligation (Reid et al., 2017). Subsequent smFRET work further examined the ability of LIG4 to sense complex ends in the minimal reconstituted system. At DNA ends with overhangs containing varying degrees of complementarity, LIG4 was shown to promote alignment of complementary ends in pre-catalytic positions, but allow dynamic sampling of alignments for terminal mismatches or ends with embedded complementarity that requires nucleolytic end process before ligation (Conlin et al., 2017). While the above-mentioned work demonstrated the participation of LIG4 in the alignment of DNA ends, in order to understand how end processing is coordinated with alignment of DNA, simultaneous observation of end processing enzymes and synapsis would be required. Taking advantage of the xenopus egg extracts system, Stinson et al. recently expanded on the requirement of LIG4-mediated close alignment of the DNA ends and showed that end processing is coordinated to take place within this synaptic complex (Stinson et al., 2020). Synaptic complex formation was monitored through smFRET with donor/acceptor dye-labeled DNA ends as before. To observe pol λ activity, the first incoming nucleotide was labeled with a fluorescent quencher, which once incorporated leads to quenching of the donor fluorophore. To observe Tdp1 activity, one of the 3′ adducts is conjugated to the donor fluorophore, which once processed by Tdp1 will be lost. It was observed that donor signal loss was preceded by high FRET, indicating close alignment of DNA end (Stinson et al., 2020). These data clearly demonstrated that end processing by pol λ and Tdp1 occurs within the short-range synaptic complex. This level of coordination between end processing and ligation during synapsis has thus been proposed as a regulatory mechanism to minimize errors and maximize fidelity of NHEJ (Stinson et al., 2020; Stinson and Loparo, 2021). Finally, attesting to the flexibility of NHEJ, it was recently reported in a smFRET study that pol μ, another X family polymerase participating in NHEJ alongside pol λ, alone can mediate synapsis of 3′ overhangs with at least 1 nt homology, in the absence of Ku (Zhao et al., 2020b).