A change of view: homologous recombination at single-molecule resolution

Homologous recombination (HR) is an important pathway that enables the exchange of genetic information between DNA molecules. Single-molecule (SM) optical approaches are ideal for probing dynamic spatiotemporal processes that cannot be easily studied through traditional experimental approaches. SM optical techniques have uncovered many features of HR and are providing new insights into molecular mechanisms with unprecedented detail. Super-resolution optical microscopy offers the ability to image fluorescent molecules inside fixed or living cells with greater spatial resolution than is possible with conventional microscopy. Its application to visualize recombination is an emerging area of tremendous potential. Homologous recombination (HR) is an important pathway that enables the exchange of genetic information between DNA molecules. Single-molecule (SM) optical approaches are ideal for probing dynamic spatiotemporal processes that cannot be easily studied through traditional experimental approaches. SM optical techniques have uncovered many features of HR and are providing new insights into molecular mechanisms with unprecedented detail. Super-resolution optical microscopy offers the ability to image fluorescent molecules inside fixed or living cells with greater spatial resolution than is possible with conventional microscopy. Its application to visualize recombination is an emerging area of tremendous potential. In this Review, Kanieckiet al. discuss how the use of single-molecule optical microscopy and super-resolution optical microscopy methods has benefited our understanding of homologous recombination by generating detailed insights into the molecules and processes involved. Genetic recombination occurs in all organisms and is vital for genome stability. Indeed, in humans, aberrant recombination can lead to diseases such as cancer. Our understanding of homologous recombination is built upon more than a century of scientific inquiry, but achieving a more complete picture using ensemble biochemical and genetic approaches is hampered by population heterogeneity and transient recombination intermediates. Recent advances in single-molecule and super-resolution microscopy methods help to overcome these limitations and have led to new and refined insights into recombination mechanisms, including a detailed understanding of DNA helicase function and synaptonemal complex structure. The ability to view cellular processes at single-molecule resolution promises to transform our understanding of recombination and related processes.

Homologous recombination (HR) is a conserved pathway for catalysing the exchange of genetic infor mation between DNA molecules and has been shown to contribute to important cellular processes such as double-strand break repair (DSBR), rescue of collapsed or stalled replication forks, horizontal gene transfer and meiosis [1][2][3] . These processes all make vital contributions to genome integrity and/or enhance genetic diver sity. Furthermore, unchecked recombination can lead to gross chromosomal rearrangements, toxic inter mediates that can lead to cell death, and loss of heterozygosity [4][5][6] . Indeed, mutations in the proteins that carry out HR can cause serious deleterious effects, a feature emphasized by their involvement in human diseases, including cancer 4,[7][8][9] .
Early genetic studies of recombination in eukary otes led to major theoretical advances exemplified by the Holliday model of recombination, followed by the Meselson-Radding model and later the DSBR model of recombination 5,10,11 . Most recently, studies of gap repair in Drosophila melanogaster led to the synthesis-dependent strand annealing (SDSA) model, which is currently con sidered a major pathway for both mitotic and meiotic recombination in many organisms 12 . These successive models have been built upon decades of research that have helped to define the molecules and pathways involved in recombination (FIG. 1; TABLE 1).
Despite the wealth of knowledge about recom bination that has been generated by genetic and bio chemical approaches, these methods have limitations.
They cannot readily be used to access information on biochemical processes that might proceed via multi ple pathways, they can overlook transient or otherwise underrepresented intermediates, and they have lim ited access to detailed information on dynamic reac tion intermediates. These limitations are now being addressed by applying single-molecule optical microscopy and super-resolution optical microscopy to in vitro and in vivo studies of DNA recombination; these approaches can help provide detailed spatiotemporal information about heterogeneous biological reaction mechanisms.
In this Review, we discuss how singlemolecule (SM) and superresolution (SR) optical microscopy are contributing to our understanding of HR. We begin by briefly describing the basic concepts behind some of the most common SM optical methods before discussing how they are being applied to study proteins involved in HR, including motor proteins, singlestranded DNA (ssDNA)binding proteins and the RAD51/RecA family of DNA recombinases, named after the eukaryotic DNA repair protein RAD51 and its bacterial homologue RecA. We then outline the different types of SR microscopy that have been used to study HR and highlight exam ples of insights these studies have provided. This Review is not intended to provide a detailed overview of HR, nor do we provide extensive discussion of prior bulk biochemical or genetic studies; for more comprehen sive descriptions of HR mechanisms and discussion of experimental approaches other than optical microscopy, we direct readers to several excellent reviews 2-4,13-18 .
SM approaches for studying recombination Here, we briefly highlight some of the SM methods that have been applied to study processes related to HR; for more extensive information on SM methodologies, we refer readers to other reviews [19][20][21][22][23][24] . Two of the more com mon fluorescence microscopy methods that have been used to visualize in vitro SM reactions are widefield epiillumination and total internal reflection microscopy (TIRFM) (FIG. 2a-d). The main difference between these two approaches is the illumination scheme. With epi illumination, which is often used in combination with laser tweezers to capture tethered molecules of interest (FIG. 2b), all fluorescent molecules present in the sample contribute to the detected signal (FIG. 2a). By comparison, only a very small volume of the sample is illuminated in TIRFM because the illumination scheme generates an evanescent field that penetrates only ~100-300 nm into the sample chamber 25,26 (FIG. 2c,d). Importantly, this approach can reduce the background signal by sev eral orders of magnitude relative to epiillumination (compare FIG. 2a and 2c) 25,26 . Applied to recombination related reactions, epiilluminationbased techniques and TIRFMbased techniques can be used to visualize fluorescently tagged proteins as they move along nucleic acids, to monitor the colocalization of proteins on nucleic acids or to monitor the binding and dissociation Nature Reviews | Genetics b | DSB formation is followed by end-processing reactions in which the 5ʹ single-stranded DNA (ssDNA) ends that flank DSBs are resected to expose 3ʹ ssDNA ends. c | These ends are quickly bound by ssDNA-binding proteinssingle-stranded DNA-binding protein (SSB) in bacteria or replication protein A (RPA) in eukaryotes -that protect the ssDNA from degradation and remove any secondary structure. d | Mediator proteins (such as the Escherichia coli RecFOR proteins, Saccharomyces cerevisiae DNA repair and recombination protein Rad52 or human breast cancer type 2 susceptibility protein (BRCA2); not depicted) then stimulate the replacement of ssDNA-binding proteins with recombinases from the highly conserved RAD51/RecA family of DNA recombinases, which are all ATP-dependent DNA-binding proteins that form extended helical filaments on DNA. These filaments are referred to as presynaptic complexes, and they catalyse all the key DNA transactions that occur during the early stages of recombination. e | The presynaptic filament first performs a homology search to locate a donor double-stranded DNA (dsDNA) molecule with sequence homology to the presynaptic ssDNA. Once a homologous donor has been identified, the presynaptic complex catalyses strand invasion, which results in the formation of a D-loop intermediate whereby the presynaptic ssDNA is paired with the homologous donor DNA. f | The 3ʹ end of the invading ssDNA strand within the D-loop then serves as a primer for DNA polymerases to begin copying information from the donor template to complete repair of the damaged DNA through either the double-strand break repair (DSBR) pathway or the synthesis-dependent strand annealing (SDSA) pathway. During DSBR, both of the 3ʹ overhangs from the processed DSB are paired with the donor DNA, and subsequent ligation of the newly replicated ends results in the formation of a double Holliday junction. Resolution of the double Holliday junction by cleavage of the DNA strands can yield either crossover products (as depicted) or non-crossover products (not shown). During SDSA, only one of the 3ʹ overhangs is paired with the donor DNA, and the second end is thought to remain unpaired. The 3ʹ invading end is used as a primer for DNA replication. The newly synthesized strand is then displaced from the donor DNA and annealed to the complementary 3ʹ overhang from the other end of the processed DSB. Flap removal and ligation restores the broken DNA, producing only non-crossover products. We refer readers to several reviews for more detailed descriptions of each of these reaction steps 3,4,14,15,17 . DMC1, meiotic recombination protein DMC1/LIM15 homologue.

Horizontal gene transfer
The transfer of genetic material between the genomes of two organisms that does not occur through parent-progeny transmission. Also referred to as lateral gene transfer.

Loss of heterozygosity
A loss of one of the alleles at a given locus as a result of a genomic change, such as mitotic deletion, gene conversion or chromosome mis-segregation.

Holliday model
An early model of double-strand break repair that proposed a cross-stranded structure was the result of two single-stranded DNA breaks and accounted for both gene conversion and crossing over. While the model was later updated by Meselson and Radding, the crossover structure called the Holliday junction remains a cornerstone of recombination.
kinetics of proteins. In these instances, the information acquired from the observed molecules is usually subject to standard optical resolution limits, which are typically in the order of a few hundred nanometres, although resolution in the order of ~1 to tens of nanometres can be obtained by mathematical treatment of the resulting data to identify the locations of individual fluorescent spots 27 . However, much higher spatial resolution can be obtained by measuring fluorescence resonance energy transfer (FRET) between biomolecules that have been appropriately labelled with donor and acceptor fluoro phores 28,29 . The distance regime accessible to FRET stud ies is in the order of ~10-50 Å, which is comparable to the dimensions of many proteins and makes it an ideal technique for probing biomolecular associations or structural changes that occur within these distances 28,29 .
Single-molecule FRET (smFRET), most commonly used in combination with TIRFM, has become one of the most widely used methods within the field of SM research 19,21,30 (FIG. 2e). However, as with any FRETbased study, its success is dependent on judicious selection of sites for dye attachment, which will vary according to the system being studied and the specific question being addressed. Importantly, all the SM optical imaging methods described here require that the molecules under obser vation are tethered to a surface. For instance, to manipu late molecules using an optical trap, the molecules must be anchored to the surface of a bead (FIG. 2b), whereas with TIRFM, the molecules can be detected only if they are within the evanescent field, which is located at the slide surface or coverslip surface (FIG. 2c,d). These attach ments are often made through biotin-streptavidin or digoxigenin -antibody linkages. Surfaces must also be blocked to prevent nonspecific adsorption of the molecules of interest, which might compromise their biological activities. Beads are often blocked with a non specific protein, such as bovine serum albumin (BSA). Microscope slides can also be blocked with BSA, but another commonly used method is to chemically mod ify the glass surface with a layer of polyethylene glycol (FIG. 2e). With DNA curtains, nonspecific interactions are minimized by coating the surface of the microscope slide with a lipid bilayer 31,32 (FIG. 2f).
SM-based insights into recombination SM optical techniques are ideal for probing dynamic spatiotemporal processes during HR that can not be accessed through traditional experimental approaches. We discuss their application to motor proteins, which were among the first HR proteins to be studied by SM techniques; proteins involved in assembling the presynaptic complex, namely, the ssDNAbinding proteins, which are among the first proteins to arrive at processed doublestrand breaks (DSBs); and the RAD51/RecA family of DNA recom binases, which catalyse the DNA pairing and strand invasion reactions (FIG. 1).

Synthesis-dependent strand annealing
(SDSA). A mode of double-strand break repair that proceeds through the early steps of homologous recombination but does not include second-end capture or the Holliday junction. Instead, after DNA synthesis primed by the initial captured end, the heteroduplex joint is dissociated to re-anneal with the second end of the double-strand break and serve as the template for further gap repair and resolution.

Meiotic recombination
A specialized type of homologous recombination that takes place during meiosis and is required to generate a physical linkage called a chiasma between two (non-sister) homologous chromosomes. Most, but not all, eukaryotes require the meiosis-specific recombinase meiotic recombination protein DMC1/LIM15 homologue (DMC1) to promote meiotic recombination.

Single-molecule optical microscopy
A class of techniques that use optical microscopy to study the biochemical and biophysical properties of biological molecules.

Super-resolution optical microscopy
A class of microscopy techniques used to enhance the spatial resolution of an optical microscope.
Motor proteins involved in DNA recombination. DNA motor proteins play a wide range of roles in all pro cesses related to nucleic acid metabolism [33][34][35] , includ ing HR [36][37][38][39][40][41][42][43][44] (TABLE 1). In general, these proteins use chemomechanical energy derived from ATP hydroly sis to unwind nucleic acid duplexes or to displace pro teins from nucleic acids [36][37][38][39][40][41][42][43][44]  enzyme subunit RecD 34 . RecBCD enzyme subunit RecC holds the complex together, helps to separate the DNA strands and coordinates the response to crossover hotspot instigator sites (chi sites), which together result in 3ʹ ssDNA overhangs that serve as substrates for assembly of the RecA presynaptic complex 45,47 (FIG. 1). Despite this wealth of knowledge, many aspects of RecBCD activity and regu lation remained unknown owing to the diffi culty of analysing heterogeneous populations of mole cules by traditional ensemble approaches. Specifically, it was not known how quickly RecBCD was able to move along DNA, how far it could go before falling off the DNA and what happened to the protein complex when it encountered a chi site. To overcome these problems, single molecules of RecBCD acting on DNA were visual ized using widefield epifluorescence microscopy 48 . Fluorescently labelled DNA molecules were attached to polystyrene beads and captured within an optical trap (FIG. 2b), RecBCD was loaded onto the DNA, reactions were initiated by moving the DNA into a stream of buffer containing ATP, and timelapse images were collected. Analysis of the data showed that RecBCD could travel along the DNA at velocities often exceeding 1,000 bp s −1 (at 37 ˚C) and for distances of tens of kilobases 48 . Remarkably, RecBCD was later shown to briefly pause at chi sites before resuming translocation at roughly half of its initial velocity 49 . This study also revealed that when powered by RecD, the lead motor before chi recognition, the RecBCD complex travels at a velocity of ~600 bp s −1 (at room temperature), whereas the complex moves at ~300 bp s −1 when powered by RecB 49 . Chi recognition causes RecD to stop translocating, at which point RecB takes over as the lead motor. Thus, the pause at chi sites is a consequence of the differential velocities of RecD and RecB and reflects the time it takes for the slower RecB motor to catch up to RecD. Chi recognition also alters RecB nuclease activity such that it selectively degrades only one ssDNA strand and allows productive DSB processing, although the precise underlying mecha nism remains uncertain 45,47 . Together, these SM imaging studies have led to a remarkably detailed understanding of RecBCD regulation that would not have been possible using only traditional approaches 45 .
The level of detail that can be obtained from SM methods is exemplified by a series of smFRET studies that documented the properties of four structurally homologous helicase proteins: ATP dependent DNA helicase Rep and DNA helicase II (UvrD) from E. coli, ATPdependent DNA helicase PcrA from Staphylococcus aureus and ATPdependent helicase Srs2 from Saccharomyces cerevisiae [50][51][52] . Genetic and biochemical studies have shown that S. cerevisiae Srs2 removes DNA repair protein Rad51 from ssDNA, thus acting as a negative regulator of genetic recombi nation 39,42,53,54 , and similar properties have been attrib uted to Rep, PcrA, UvrD and many other HRrelated helicases 37,42,[55][56][57][58] . In these experiments, smFRET sig nals were detected with TIRFM and used to measure the movement of the helicases along DNA (FIG. 2e). In the first study, a dsDNA fragment with a 3ʹ ssDNA overhang was tagged with a fluorescence acceptor dye at the ssDNA-dsDNA junction. The fluorescence donor dye was coupled to Rep through sitespecific labelling of a cysteine residue. This setup allowed the distance between Rep and the ssDNA-dsDNA junction to be measured using smFRET 51 (FIG. 2e). As expected, addi tion of the helicase Rep resulted in an increase in the smFRET signal as the helicase moved along the ssDNA towards the ssDNA-dsDNA junction (FIG. 2e). This study also detected unusual cyclical changes in the smFRET signals that were interpreted as repeated transloca tion events, which suggested that the motor domains of the helicases dissociated from the DNA upon reach ing the ssDNA-dsDNA junction (FIG. 2e). However, some part of the protein retained in contact with the ssDNA to enable the process to be repeated for many cycles before the helicase dissociated into free solution. Notably, this type of cycling behaviour can be observed only by SM imaging. Similar smFRET studies have now shown that Srs2 and PcrA remove Rad51 and RecA filaments Figure 2 | Examples of single-molecule imaging methods. a | Simplified illustration of the illumination geometry used for typical epifluorescence microscopy. The excitation light passes through the entire sample volume, so all fluorophores present within this volume will contribute to the signal detected by the camera. Therefore, with this method, it is essential to minimize the concentration of free fluorescently tagged molecules present in the sample. b | Schematic illustration of an optical trapping setup that is often used in conjunction with epifluorescence microscopy. In this case, an infrared laser (shown in red) is used to trap a polystyrene bead. The bead is attached to one end of a fluorescently labelled DNA molecule (shown in blue), typically through a biotinstreptavidin linkage, and the trap can be used to position the bead in the sample chamber for optimal viewing. In the illustration, RecBCD is bound to the end of the DNA, and RecBCD helicase and/or nuclease activity would be revealed as a reduction in the length of the double-stranded DNA (dsDNA) in real time. c | Illumination geometry used for typical objective-type total internal reflection microscopy (TIRFM). Here, a laser beam is passed through the objective lens and reflected off the interface between the coverslip and the aqueous buffer holding the sample. This generates an evanescent field, which penetrates ~100-300 nm into the sample chamber. The key feature of TIRFM is that the only fluorophores that will emit photons are those that are within the evanescent field. d | Schematic showing the illumination geometry for prism-type TIRFM. In this case, the laser is directed through a prism placed on top of the microscope slide glass. e | Schematic illustration of how single-molecule fluorescence resonance energy transfer (smFRET) works and how it can be applied to studying protein movement on DNA. Here, a hypothetical motor protein is labelled with a 'green' fluorescent donor dye (Cy3), and the dsDNA-single-stranded DNA (ssDNA) junction is labelled with a 'red' fluorescent acceptor dye (Cy5). The green signal is excited with a laser of an appropriate wavelength, and only a green signal is emitted when the protein is far away from the junction (that is, >50 Å). The red signal becomes visible as the protein approaches the junction because FRET occurs between the green donor dye and red acceptor dye, causing the red dye to emit photons. The resulting data can be expressed in graphical form as separate red and green signals, which should be anti-correlated, or as a FRET value, which is essentially a ratio of the two signals. Although compatible with other types of fluorescence microscopy, smFRET is often used with TIRFM, which usually requires that the molecules of interest be tethered to a surface to confine them to the evanescent field. In the example shown, the DNA is attached to a microscope slide through a biotin-streptavidin linkage, and the surface is covalently modified with polyethylene glycol (PEG) to prevent proteins or DNA from adhering to the glass. f | Schematic illustration of DNA curtains. Here, the individual strands of DNA are anchored at one end to a lipid bilayer through a biotin-streptavidin linkage and then aligned along the leading edge of a nanofabricated chromium barrier by application of buffer flow. The other end of the DNA can be attached nonspecifically to a series of anchor points so that the molecules are confined within the evanescent field of a total internal reflection microscope. In the example shown, ssDNA is coated with fluorescently labelled replication protein A (RPA) (a GFP-RPA fusion protein) (in green), which allows visualization of the DNA. Part b is adapted from REF. 150, Macmillan Publishers Limited. a.u., arbitrary units.

Evanescent field
An oscillating electromagnetic field that is spatially concentrated at the interface between two materials with different refractive indices, for instance, glass and water. Also known as an evanescent wave. In the case of total internal reflection microscopy, the evanescent field is confined near the interface between the aqueous buffer and the glass microscope slide (or coverslip).

Single-molecule FRET
(smFRET). A technique that allows the direct measurement of distances between macromolecules of interest within a range of ~10-50 Å. Also referred to as single-pair FRET (spFRET). smFRET is commonly used with total internal reflection microscopy but can be adapted for use in many types of single-molecule imaging systems.

Presynaptic complex
A nucleoprotein complex that comprises the presynaptic single-stranded DNA and the associated protein cofactors necessary for promoting homologous recombination (HR). Key among these factors are the RAD51/RecA family of DNA recombinases, which act together with a number of associated factors to promote HR.

Crossover hotspot instigator sites
(chi sites). Specific cis-acting sites consisting of an eight-nucleotide DNA sequence (5ʹ-GCTGGTGG-3ʹ) that is over-represented in the Escherichia coli genome and helps regulate the properties of the RecBCD complex by controlling the activities of the RecBCD enzyme subunits RecB and RecD ATP-dependent motor proteins.

Translocation
General term used to indicate the ATP-dependent movement of a motor protein (such as a helicase or a polymerase) along a nucleic acid.

Processivity
The tendency of a helicase (or any other nucleic acid motor protein) to continue to move along a nucleic acid molecule rather than dissociating into free solution.
from ssDNA, respectively, and have suggested that the repetitive shuttling of these helicases prevents aberrant recombination by clearing these recombinases from ssDNA 51,52,55 .
The part played by conformational changes and oligomerization in the regulation of helicases has been difficult to measure with ensemble techniques because of the transient nature of conformational changes and the population heterogeneity associated with oligomer ization 33,59,60 . Structural studies had predicted that the helicases Rep, UvrD and PcrA might undergo large con formational changes -all three proteins were thought to transition between 'open' and 'closed' helicase confor mations via an ~150˚ rotation of the 2B domain relative to the rest of the protein [61][62][63] . Recent SM studies have investigated how structure and function affect helicase activity using elegant combinations of smFRET to probe protein conformation and forcebased measurements to probe DNA unwinding 59,60 . In the first of these studies, Rep and PcrA protein variants were generated that con tained pairs of engineered cysteines that could be used to 'lock' the proteins into the closed conformation 59 . These proteins were then probed for activity in an smFRET assay in which the proteins were immobilized on a slide surface and a DNA substrate containing an 18 bp duplex region and a 3ʹ 20nucleotide ssDNA tail was added 59 . A Cy3 donor was attached to the 3ʹ ssDNA end of the sub strate, and a Cy5 acceptor was attached to the opposite end of the duplex. Partial unwinding of the duplex led to an increased smFRET signal as the donor and acceptor dyes approached one another, and complete unwinding resulted in a loss of smFRET owing to displacement of a DNA strand 59 . Interestingly, this study found that Rep or PcrA locked in the closed conformation exhibited greater processivity than the wildtype enzymes -for instance, 95% of the Rep molecules locked into the closed conformation could completely unwind a 4 kb region of duplex DNA, whereas only 7% of the wildtype Rep could fully unwind the same region of duplex DNA. This result suggested that reorientation of the 2B domain into the open conformation restricts the ability of Rep and PcrA to unwind long stretches of dsDNA 59 .
In a different study, smFRET analysis of UvrD labelled with donor and acceptor dye pairs on the 2B and 1A domains revealed the existence of two distinct activ ities, which were termed 'frustrated' and 'long distance' unwinding 60 . Frustrated activity was characterized by repetitive cycles of unwinding and rezipping over dis tances shorter than 20 bp, whereas longdistance activity was less repetitive and allowed for unwinding of tracts of dsDNA longer than 20 bp. Counting the number of fluo rescent UvrD molecules present on each DNA revealed that the frustrated and longdistance behaviours cor related with one and two UvrD proteins, respectively. This study also established an smFRET assay that was able to report on whether a particular molecule of UvrD was in the open or closed state 60 . These findings led to a model in which UvrD (and related helicases) can switch directions by transiently sampling the open con figuration: the 2B domain anchors the helicase to the DNA so that the motor domains can dissociate from one strand and then rebind the second strand to move in the opposite direction 60 . The frequency of opentoclosed transitions dictates how much DNA a given enzyme can unwind before switching directions and rewinding the same substrate; this frequency is influenced by the number of molecules present, which explains why two UvrD proteins acting in concert exhibit longdistance unwinding events 60 .
Single-stranded binding proteins. All recombination reactions involve an ssDNA intermediate, which is rapidly coated by essential proteins -singlestranded DNAbinding protein (SSB) in bacteria and replication protein A (RPA) in eukaryotes 64,65 . Biochemical and genetic studies have shown that SSB and RPA protect the ssDNA from nucleases, eliminate secondary struc ture to enable recombinase loading, serve as DNA dam age signalling complexes and promote recruitment of downstream factors 64,66-71 (FIG. 1). Although the need for these proteins in all aspects of DNA metabolism is well established, we still have a fairly poor understanding of their ssDNAbinding properties, in particular their dynamic behaviour on ssDNA and the mechanisms that allow them to be replaced by downstream factors during the repair of DNA damage. The development of ssDNA curtains (FIG. 2f) afforded the opportunity to study the properties of RPA on long ssDNA substrates 31,72 . These experiments have revealed that GFPtagged RPA can bind tightly to ssDNA and that essentially no protein dissociation occurs within a 2hour observation win dow when free RPA was absent from solution [73][74][75] . Introducing free RPA resulted in rapid turnover of the DNAbound proteins, suggesting RPA was replaced through a concentrationdependent mechanism referred to as facilitated dissociation 73 . Facilitated dissociation is a mass action process that can take place when a protein is capable of interacting with its ligand through multiple contacts (for example, the three ssDNAbinding domains of RPA). Transient disruption of any single contact is not sufficient to allow the protein to release its ligand, and in the absence of competing interactions, the disrupted contact can rapidly reform. However, when free proteins are present, they can rapidly engage any sites within the ligand that are transiently released by the bound pro teins, which in turn reduces the probability of rebinding. The advantage of facilitated dissociation is that it offers the potential for extremely tight binding interactions but still allows rapid displacement of the bound proteins 44,45 . Facilitated dissociation has also been reported for sev eral other DNAbinding proteins 76,77 , suggesting that this process is more common in biological systems than is currently appreciated 78,79 .
Other aspects of RPA and SSB behaviour on ssDNA have been studied using SM approaches. For instance, smFRET studies have shown that SSB can diffuse along ssDNA, and this process may be involved in the displacement of SSB from ssDNA during the assem bly of the RecA presynaptic complex 80 . SSB can even be pushed along ssDNA through the action of the E. coli helicases Rep and UvrD, which suggests a role for these helicases in assisting the reorganization or removal

Nucleoprotein filaments
The helical filament structures formed by members of the RAD51/RecA family of recombinases as they bind to single-stranded DNA. They contain one protein monomer for every three DNA bases and six protein monomers per helical turn. These nucleoprotein filaments are a key component of the presynaptic complex.
of SSB from DNA repair intermediates 81 . A TIRFM study looking at timeresolved widefield images of SSB bound to ssDNA revealed that SSB can form reversible long range contacts that can lead to extensive condensation of SSBbound ssDNA intermediates 82 . Finally, DNA cur tain studies have shown that RPA can downregulate the processivity of the DNAendprocessing nuclease exo nuclease 1 (EXO1), which suggests an unforeseen role for RPA in regulating EXO1 activities 83 . These studies highlight some of the new discoveries that have been made through the use of SM optical microscopy, in par ticular, that SSB and RPA are highly dynamic complexes that are capable of moving, exchanging and forming higherorder structures.

Dynamics of the presynaptic complex.
The presynaptic complex is a crucial recombination intermediate formed when RPA is replaced by RAD51 (or meiotic recombi nation protein DMC1/LIM15 homologue (DMC1)) in eukaryotes or when SSB is replaced by RecA in bacteria (FIG. 1). Decades of biochemical and genetic studies have led to a detailed understanding of biochemical proper ties of RecA and Rad51 nucleoprotein filaments 15,17,18,56,70,84 . However, we still do not fully understand how these fila ments are assembled or how they search the genome for homologous sequences, discriminate between homo logous and nonhomologous sequences and promote the strandpairing reactions necessary for genetic recombi nation. We also lack an understanding of the precise complement, stoichiometry and spatial organization of the HR accessory factors that must interact with the pre synaptic complex to promote its stability and activities. SM imaging methods offer the potential to help fill these gaps in our understanding.
An early SM investigation into presynaptic complex dynamics used smFRET to observe assembly of RecA fil aments on short ssDNA overhangs 85 (FIG. 3a). This study suggested that ~5 monomers of RecA are required to form a stable nucleation point for filament assembly and that subsequent extension occurs in single monomer units. Most surprisingly, these experiments suggested that filament extension occurs in both the 5ʹ-3ʹ and 3ʹ-5ʹ directions. However, monomer binding occurred more rapidly at the 3ʹ end of the growing filaments, while monomer dissociation occurred at similar rates at either end, leading to more rapid net extension in the 5ʹ-3ʹ direction 85 .
Bidirectional growth of RecA filaments was also observed in studies using fluorescently tagged RecA and a three channel laminar flow system to control the assembly of RecA filaments on beadtethered dsDNA molecules 23,86,87 (FIG. 3b). The centre channel was used for observation by epifluorescence microscopy, whereas an adjacent channel was used to capture the bead-DNA complex in an optical trap. The captured DNA was then 'dipped' into the third flow stream containing fluorescent RecA and moved back into the central channel for obser vation. This dipping allowed the authors to precisely control the timing with which the RecA and the DNA were mixed together and to control the amount of RecA that bound to the DNA based upon the length of time that the DNA remained in the RecA channel. The result ing snapshots revealed that filament assembly occurs through a ratelimiting nucleation step involving the initial binding of ~4-5 RecA proteins. Nucleation was followed by a rapid growth phase in which the filaments could extend in both the 5ʹ-3ʹ and 3ʹ-5ʹ directions.
Building upon this work, another study used TIRFM to look at long molecules (~48,000 nucleotides) of SSB coated ssDNA that were anchored to a flowcell surface 88 . Use of a clever twocolour labelling strategy, involving RecA labelled with either Cy3 or fluorescein, allowed the authors to distinguish between nucleation and growth within the same filaments (FIG. 3c), which provided a clear demonstration that growth was bidirectional but occurred roughly twice as quickly in the 5ʹ-3ʹ direction compared with the 3ʹ-5ʹ direction 88 . Next, the role of the RecFOR proteins, which help promote RecA assembly on ssDNA 89,90 , was investigated by SM observation 88 . RecFOR caused the tethered SSB-ssDNA to non specifically absorb to the slide surface, so these meas urements relied upon epifluorescence observations of a beadtethered DNA, which enabled the DNA to be held away from the slide surface 88 . These experiments showed that in the presence of RecFOR, the frequency of RecA nucleation increased, as did the rate of filament growth, suggesting that RecA was more readily able to replace SSB when RecFOR was present 88 . Together with previous bulk biochemical studies 15 , these observations support a model in which RecFOR promotes RecA presynaptic complex assembly by weakening contacts between SSB and ssDNA, perhaps by partially unwrapping SSB from the ssDNA 88 .
DNA curtains have also been used to study the assembly and stability of the presynaptic complex, using a GFP-RPA fusion protein as a readout for pro tein occupancy on the ssDNA [72][73][74]91 ; the absence of GFP fluorescence indicates that the unlabelled (or dark) protein of interest has displaced GFP-RPA from the ssDNA (FIG. 3d). One notable advantage of this labelling strategy is that it avoids the need to prepare fluores cently tagged recombinases, which have proved diffi cult to generate without substantially compromising biochemical activities. This DNA curtain approach has been applied to E. coli RecA, S. cerevisiae and human RAD51 and DMC1 and Caenorhabditis elegans RAD51 (REFS 72-75,91-95). Collectively, these studies have con firmed that human RAD51 exhibits bidirectional, ATP dependent filament assembly 75 ; that the yeast mediator protein DNA repair and recombination protein Rad52 can bind extensively to an assembled presynaptic complex, which highlights potential roles for Rad52 after presynaptic complex assembly 91 ; and that the C. elegans RAD51like protein 1 (RFS1)-RFS1 inter acting partner (RIP1) complex (a paralogue of RAD51) stabilizes filaments by preventing RAD51 dissociation even in the absence of ATP, which highlights the dra matic effect that the RAD51 paralogues can have on presynaptic complex stability 95 .
When RAD51 or RecA bind to DNA, they stretch it by ~50% relative to the contour length of BDNA 84,96 . To explore the possibility of a relationship between

Homology search
The process during which the presynaptic complex searches the genome for a double-stranded DNA sequence that is homologous to the single-stranded DNA sequence present at the processed ends of a double-strand break.

Strand invasion
A reaction catalysed by the RAD51/RecA family of recombinases, which allows for Watson-Crick pairing interactions between a single-stranded DNA molecule and the complementary strand within a homologous double-stranded DNA, resulting in displacement of the non-complementary strand.

D-Loop
The paired intermediate generated by a successful strand invasion reaction wherein the 3ʹ end of the invading single-stranded DNA strand is now available as a primer for DNA synthesis.
DNA tension and filament stability, an epifluor escence microscope equipped with dual optical traps 97-100 was used to visualize fluorescent RAD51 filaments while simultaneously stretching the dsDNA; these studies demonstrated that increased tension on the DNA prevented protein dissociation and thereby increased filament stability 97 . Surprisingly, when ten sion was released, filament disassembly occurred in bursts interspersed by long pauses. Analysis of the data revealed a clear relationship between extension of the presynaptic complex and the dissociation of RAD51 97 . Moreover, the patterns of protein disso ciation that were observed led to a model in which RAD51 could hydrolyse ATP anywhere within the stretched filament, but dissociation of the resulting ADPbound proteins could occur only at the filament ends; ATPbound proteins were unable to dissociate from filaments regardless of their position 97 . Thus, the bursts of RAD51 dissociation corresponded to filaments with ADPbound proteins at their ends, whereas pauses occurred when ATPbound proteins were at the filament ends. Together, the studies highlighted above are beginning to yield new insights into the behaviours of the pre synaptic complex, but much remains to be discovered. Future investigations of the presynaptic complex with SM approaches might help to address remaining ques tions about protein composition, protein distributions and reaction mechanisms.
The homology search and strand invasion. The pre synaptic filament has a number of vital roles in genetic recombination. First, it must locate a homologous dsDNA sequence present elsewhere in the genome in a process that is referred to as the homology search. Then, it must promote strand invasion by catalysing Watson-Crick pairing interactions between the presynaptic ssDNA and the complementary strand within the homo logous dsDNA sequence while also displacing the non complementary strand to generate a D-loop structure 15,17 . Both the homology search and strand invasion are highly dynamic processes that are anticipated to proceed through a heterogeneous and asynchronous set of inter mediates, which makes it difficult to study the molecular details with biochemical and genetic methods.
The homology search has been the subject of a num ber of indepth reviews 86,101-103 , and thus here we only briefly highlight the key areas where progress has been made using a combination of bulk biochemical stud ies and SM studies. For instance, these types of studies are beginning to present a more unified picture of the basic principles that guide sequence alignment during the homology search. They have shown that the process involves a combination of intersegmental transfer 104,105 , shortdistance 1dimensional sliding 106 and a length dependent recognition mechanism that reduces search complexity by allowing the search process to focus on tracts of microhomology longer than eight nucleotides, which have a high probability of being the correct tar get 94, [107][108][109][110] . Other studies suggest that differential exten sion of the presynaptic ssDNA relative to the incoming dsDNA gives rise to differential levels of tension between the two DNA molecules, which promotes more rapid turnover of nonhomologous pairing interactions 111 . However, many unanswered questions remain, includ ing how recombination accessory proteins, such as the motor proteins DNA repair and recombination protein Rad54 and DNA repair and recombination pro tein Rdh54 (REFS 3,4,15,16), influence the search process and how the search takes place within chromatin.
Strand invasion has been more difficult to study using SM optical studies than other aspects of recombination, largely owing to the rapid reaction rates and the small length scales involved in this process. Nevertheless, an assay that combines TIRFM and smFRET has been used to visualize the capture of homologous dsDNA fragments by a RecA presynaptic complex anchored to a slide sur face 112 . Using this approach, strand exchange led to a high FRET signal between a Cy5 acceptor positioned on the presynaptic complex and a Cy3 donor placed on the incoming dsDNA. However, there was a considerable delay, on the order of ~120 milliseconds, between the initial binding of the dsDNA fragment and the appear ance of the FRET signal, which was attributed to the time required to complete strand invasion. Subsequent ana lysis of the delay time distributions suggested that strand exchange took place in 3nucleotide steps 112 . Similar stepping behaviours during strand invasion have been observed using ssDNA curtain assays 92-94 . In these stud ies, Dloop intermediates were generated by incubating RecA, RAD51 or DMC1 presynaptic complexes with ◀ fluorescently tagged dsDNA fragments containing an internal microhomology of 8-15 nucleotides in length, which was complementary to the ssDNA. The resulting lifetime changes occurred in 3nucleotide increments, consistent with the notion that strand exchange inter mediates are stabilized by the RAD51/RecA family mem bers in 3nucleotide steps 92-94 . Moreover, the energetic signature of the stepping behaviour was conserved for RecA, RAD51 and DMC1, indicating that this behaviour appears to be broadly conserved among members of the RAD51/RecA family of DNA recombinases. However, a caveat of both approaches is that they do not directly 'see' individual strand exchange steps as they are taking place. Rather, the existence of the 3nucleotide steps is inferred from either a modeldependent analysis of kinetic data 112 or from the stability of the resulting strand exchange products 92-94 . Direct observation of the steps themselves remains a goal of future technical developments.

SR approaches for studying recombination
Until recently, optical microscopy suffered from a fun damental limit in spatial resolution owing to the wave nature of visible light. This phenomenon, first described as the Abbe limit in 1873 and quantified by the Rayleigh criterion, means the maximum achievable resolution for discrete fluorescent sources is equal to onehalf of the emission wavelength of the fluorophores under obser vation (~250 nm for GFP, which has an emission wave length of 509 nm) [113][114][115][116] . Unfortunately, this value remains two orders of magnitude above the size regime that is relevant to many molecular interactions and subcellular structures -in other words, the smallest distance that can be resolved using fluorescent labels and traditional imaging techniques is usually much greater than the dis tances relevant to many biological processes. However, new imaging techniques are now beginning to leverage advances in fluorophore photophysics and image ana lysis to enable measurements within spatial regimes that were previously inaccessible [114][115][116][117][118][119] . These new imaging techniques include stochastic optical recon struction microscopy (STORM) 120-122 , photoactivated localization microscopy (PALM) [123][124][125] , stimulated emis sion depletion (STED) microscopy [126][127][128] and structured illumination microscopy (SIM) [129][130][131] .
STORM and PALM both use the same general principles for generating SR images, namely, they take advantage of the fact that single fluorophores can be localized with great precision by using relatively simple equations to calculate the centroid position of a fluo rescent spot representing a single fluorescent dye [113][114][115] . However, these localization procedures work only if the fluorescent dyes are relatively far apart from one another. Thus, a key feature of these methods is that not all the fluorophores present in a sample are visual ized at once. Instead, a series of images are collected in which only a small subset of the fluorophores present on a labelled subcellular structure are activated at any given time, and a complete composite picture is then generated by integrating all the fluorophore position information derived from the image series [113][114][115] (FIG. 4A). In both PALM and STORM, the fluorophores present in the sample are driven between 'light' or 'on' (that is, visible) and 'dark' or 'off ' (that is, not visible) states using variations in excitation wavelength, which enables only a small number of the total fluorophores present to be excited at a given time. The main difference between the two methods is that PALM uses engineered versions of GFP that can be switched on and off, whereas STORM typically relies upon the use of antibodies coupled to organic fluorophores [113][114][115] .
STED shares one common feature with PALM and STORM; namely, it relies upon the excitation of only a small subset of the fluorophores present in a given image at any given time to generate an SR image. However, in STED, this is accomplished by using a focused doughnut shaped light beam to switch fluorophores between 'on' and 'off ' states at defined spatial coordinates -only those fluorophores present in the very centre of the doughnut are 'on' , and the remaining fluorophores are switched 'off '   (FIG. 4B). The dough nutshaped beam is then scanned across the sample to generate a complete image, and the resolution of the resulting image is defined by the size of the doughnut centre. SIM also uses patterned light for sample illumi nation, but in this case, the entire sample is illuminated with a series of striped light patterns [113][114][115] (FIG. 4C). The recorded images are then computationally processed to produce a mathematically reconstructed image that has about twice the resolution of normal fluorescence microscopy.
STORM, PALM and STED excel at imaging fixed cells or chromosome spreads, but they are currently more difficult to use for livecell imaging, especially for biological processes that occur rapidly, because these methods require the timeconsuming collection of many successive images to build a composite picture of the structures of interest [113][114][115] . SIM offers great promise for livecell imaging because images can be cap tured at higher acquisition rates, currently on the order of several seconds, albeit with lower spatial resolution than that offered by STORM, PALM and STED [113][114][115] .
SR-based insights into recombination SR approaches are being actively applied by a number of laboratories to study the nucleoprotein structures that participate in meiosis and meiotic recombination. Although there are currently only a few such studies in the published literature, we anticipate this number to increase in the coming years as their potential for studying HR becomes more widely appreciated.

Organization of the synaptonemal complex.
Meiosis is characterized by a single round of replication fol lowed by two rounds of cell division to yield haploid gametes from a diploid precursor [132][133][134][135][136] (FIG. 5a). Meiotic recombination is required for chromosome segre gation, and it also generates new combinations of parental alleles. During meiosis, the endonuclease meiotic recombination protein SPO11 initiates programmed DSBs, which are then repaired by HR between homo logues. This prefer ence for repair using the homologue, instead of the sister, is referred to as 'homologue bias' , but its underlying mechanisms remain unknown [132][133][134][135][136] . Importantly, the earliest stages of meiotic recombina tion coincide with the formation of a meiosisspecific chromo some structure referred to as the synaptone mal complex [132][133][134][135][136] (FIG. 5b). The synaptonemal complex comprises two lateral elements, which serve as attach ment points for homologous chromosomes, and the DNA, which is arranged as a linear array of loops teth ered to the lateral elements [132][133][134][135][136] (FIG. 5b). The lateral elements are held together by a central axis region that Nature Reviews | Genetics The precise location of these spatially resolved fluorophores can be defined by fitting each observed 'spot' to a Gaussian (or similar) function. These cumulative data for all the 'spot' locations are then used to reconstruct the final SR image (part Ad). As shown in part B, stimulated emission depletion (STED) microscopy uses two superimposed laser beams for sample illumination, depicted in red and blue (part Ba). One laser (blue) is used to excite (that is, turn 'on') the fluorophores conjugated to the subcellular structure of interest, and a second doughnut-shaped laser (red) is used to push the fluorophores back into an unexcited state (that is, 'off'). Superimposition of these two beams yields a very small region within the centre that can be used to excite fluorophores in the sample (part Bb), which can be scanned across a sample to yield an SR image (part Bc). In structured illumination microscopy (SIM), shown in part C, a subset of fluorophores on the labelled sample is illuminated with a pattern of light, depicted as a blue 'striped' pattern (part Ca), and their positions are recorded. This pattern is then shifted through a series of different orientations, yielding images with encoded high-resolution information, and mathematical deconvolution of the resulting imaging data yields a high-resolution image (part Cb).
can be further subdivided into a single central element, which runs midway between the two lateral elements, and transversal filaments, which link the central and lateral elements [132][133][134][135][136] (FIG. 5b). This structural organiza tion of the synaptonemal complex is broadly conserved among eukaryotes, although the protein components tend to be more divergent [132][133][134][135][136] . Much of our under standing of the synaptonemal complex has come from genetic, electron microscopy and conventional optical microscopy studies [132][133][134][135][136] . However, we still have a lim ited understanding of its 3dimensional organization, a limited knowledge of the detailed spatial distribution and organization of proteins within it, and a very poor Nature Reviews | Genetics  Figure 5 | Super-resolution images of the synaptonemal complex. a | A simplified overview of meiosis, which highlights DNA replication and segregation events involving a diploid cell with a single pair of homologous chromosomes (shown in red and black). Following an initial round of DNA replication, the resulting sister chromatids are linked together by cohesin (not depicted). Homologues are then aligned and linked along their lengths through formation of the synaptonemal complex, which comprises several meiosis-specific proteins [133][134][135][136] . Meiotic recombination takes place within the synaptonemal complex and results in crossover formation between homologues. The sites of crossover are called chiasma, which provide a physical linkage between homologues that is necessary to ensure attachment to the spindle apparatus (depicted as green arrows), allowing homologues to be separated to opposite poles of the cell. A second round of spindle formation allows segregation of the sister chromatids to yield haploid cells. b | Diagram of the synaptonemal complex highlighting the tethered-loop architecture with pairs of sister chromatids attached to the lateral elements, which are in turn held together by central element proteins. c | Confocal and stochastic optical reconstruction microscopy (STORM) images showing the relative locations of synaptonemal complex protein 3 (SYCP3) (magenta) and the SYCP1 N-terminus and C-terminus (green) within the synaptonemal complex 137 . SYCP3 co-localizes with the lateral elements, as does the C-terminus of SYCP1, whereas the N-terminus of SYCP1 is positioned towards the central element. Notably, these differences in protein spatial distributions cannot be distinguished in the confocal image. d | Widefield epifluorescence and STORM images showing the locations of SYCP3 (green) and H3K4me3-modified nucleosomes (magenta) 138 . In these images, the position of SYCP3 indicates the location of the lateral elements, and H3K4me3 highlights what are anticipated to be transcriptionally active chromatin loops emanating outwards from the lateral elements. e | Illustration highlighting the proposed two-layered architecture of the synaptonemal complexes based on structured illumination microscopy images 139 . The key feature of this model is that the pairs of sister chromatid loops emanating outwards may be maintained in separate subcellular volumes, which could help minimize inter-sister recombination during meiosis and may help ensure homologue bias. Part c is adapted with permission from REF. 137 understanding of how its unique structural organiza tion contributes to the homologue bias that is a defining characteristic of meiotic recombination [132][133][134][135][136] .
Questions relating to the structure and function of the synaptonemal complex can potentially be addressed using new SR imaging methods. For instance, the lateral and central elements cannot be resolved by conventional fluorescence microscopy, but they have been resolved by STORM 137,138 . In one initial study, fluorescently tagged antibodies targeting structurespecific proteins (syn aptonemal complex protein 2 (SYCP2; also known as SCP2) and SYCP3 (also known as SCP3) in the lateral elements and SYCP1 (also known as SCP1) in the trans versal filaments) were used to generate STORM images of the synaptonemal complex in mammalian chromosome spreads 137 (FIG. 5c). Antibodies that targeted either the Nterminal domain or the Cterminal domain of SYCP1 enabled each domain to be visualized independently and provided confirmation that the Cterminus of SYCP1 localizes with the lateral elements, whereas its Nterminus localizes with the central elements 137 (FIG. 5c). In another study, STORM was used with fluorescently tagged anti bodies directed against modified histones to look at the distributions of epigenetic markers within the synap tonemal complex, revealing distinctive patterns of his tone modifications 138 (FIG. 5d). For instance, the repressive histone modification H3K27me3 was found in periodic clusters positioned at 500 nm intervals, whereas the cen tromeric histone modification H3K9me3 was located at one end of the synaptonemal complex, consistent with the known location of mouse centromeres. In addition, H3K4me3, a histone modification associated with tran scriptionally active chromatin, was found to be arranged in a radial 'hairlike' pattern that emanated outwards from the synaptonemal complex and that was suggestive of the chromatin loops expected to project outwards from the synaptonemal complex (FIG. 5b,d). Together, these studies have yielded remarkably detailed 3 dimensional pictures of the synaptonemal complex, and a sideby side comparison of images collected by conventional micro scopy with those collected by STORM provides an intuitive understanding of the degree to which spatial resolution is improved with SR techniques 137,138 (FIG. 5c,d).
Two very recent studies have also exploited SR imag ing to further probe protein organization in the synap tonemal complex 139,140 . In the first of these studies, researchers labelled several protein components of the C. elegans synaptonemal complex using either fluores cent antibodies or GFPfusion proteins and then visu alized these proteins in intact gonads by either STORM or PALM 140 . This strategy took advantage of the fact that chromosomes in C. elegans gonads tend to be positioned towards the nuclear envelope, thus allowing them to be illuminated by a TIRFMlike technique 140 . For this study, the labelled proteins included meiosisspecific HORMA domain proteins (HIM3 and its paralogues HTP 1, HTP2 and HTP3) and the structural maintenance of chromosomes (SMC) cohesin complexes, including the meiosisspecific kleisin subunits meiotic recombination protein REC8, COH3 and COH4 (REF. 140). The result ing SR images revealed the organization of these protein components in exquisite detail and suggested a model in which HORMA domain proteins tether cohesin bound sister chromatids to the central regions, with REC8containing cohesin attaching to each sister chro matid and COH3 cohesin forming a boundary between sisters 140 . In the second study, researchers used SIM to visualize the protein components of the D. melanogaster synaptonemal complex 139 . In this case, the samples were fixed and expanded fourfold using a specialized tissue expansion technique originally developed for elec tron microscopy 139 . Several protein components of the synap tonemal complex were labelled using antibodies, including C(2)M, which localizes to the outer edge of the central region, and C(3)G, Corolla and CONA, which are all components that lie near the middle of the D. melanogaster central element 139 . A key finding from this study, which was consistent with the protein organiza tion observed in C. elegans 140 , was that the synaptonemal complex seemed to be organized into two distinct lay ers, suggesting that each sister chromatid is maintained within a spatially distinct volume 139 (FIG. 5e). Although speculative, this model could help to explain why mei otic recombination is biased towards pairing interactions between homologues instead of sisters 139,140 .

DMC1 and the presynaptic complex.
Most eukaryotes have evolved a meiosisspecific recombinase, DMC1, and both it and RAD51 are required for progression through meiosis 132,141,142 . It is not understood why this is the case, nor is it understood how RAD51 and DMC1 are organized with the presynaptic complex during meiosis or how this organization might contribute to the unique requirements of meiotic recombination. Conventional microscopy studies had suggested the existence of par tially offset RAD51 and DMC1 cofoci [141][142][143][144] , which led to the proposal of an asymmetric loading model in which RAD51only filaments formed on one end of a meiotic DSB and DMC1only filaments formed on the other end of the same DSB [143][144][145] . However, these studies lacked the spatial resolution necessary to confirm the model, and the model itself was at odds with a number of bio chemical and genetic studies 146 . To help resolve these issues, fluorescently tagged antibodies directed against RAD51 and DMC1 were used with STORM to examine the organization of the presynaptic filaments at both ends of a meiotic DSB 146 . The results showed that the two ends of the DSB were separated from one another by ~400 nm, and it was suggested that this distance reflected structural constraints imposed by the tethered loop architecture of the synaptonemal complex 146 . Notably, RAD51 and DMC1 cooccupied both ends of the mei otic DSBs, arguing against the asymmetric loading model. These filaments containing both RAD51 and DMC1 were quite small and typically comprised just ~30 recombinase monomers, which is sufficient to cover ~100 nucleotides of ssDNA and is surprising given that the ssDNA overhangs were on the order of ~800 nucleotides in length 141,146 . In addition to revealing new insights into the presynaptic complex, this important finding illustrates how poorly we understand the pro tein stoichiometry and composition of the presynaptic complex and raises questions as to what other proteins are present on the ssDNA, how they are arranged and why they are there 141 .

Little bundles of RecA.
In all the studies cited above, the cells and subcellular structures had to be fixed to ensure good imaging. An advantage of SIM is that it can be readily applied to living cells, which is essential for understanding how subcellular structures evolve in time during the completion of a biological process. Livecell SIM has recently been applied to study RecA during DNA recombination in E. coli. During replica tion of the circular E. coli genome, the resulting sister chromatids move away from one another, posing the questions of whether and how recombination could occur after the homologous sequences have moved to distant cellular compartments. When induced DSBs were tracked in living cells at different time points after replication initiation, it was found that distant sister chromatids could pair even after reaching opposite ends of the cell 147 . Furthermore, fluorescent RecA co localized with the DSB and then polymerized into long filamentous structures that were termed RecA 'bundles' (FIG. 6a,b). The bundles often spanned the entire length of the cell, did not have appreciable turnover as meas ured with fluorescence recovery after photobleaching (FRAP) and resulted in rapid sister pairing ~50 minutes after DSB induction. SIM images revealed that the bun dles were positioned between the inner membrane and the nucleoid and that they comprised a thick immobile central region flanked on either end by thinner, more mobile projections (FIG. 6b-d). In future work, it will be important to determine how these bundles of RecA con tribute to recombination and whether similar dynamic structures might exist in other organisms.
Together, these early SR studies provide detailed new insights into the nucleoprotein complexes that par ticipate in HR and serve as important foundations for future research. In principle, similar approaches could be applied to many other subcellular structures, such as DNA repair foci. It has been known for many years that repair foci demarcate the intracellular locations of recombination intermediates 142,148 , but we know very little about the precise organization and dynamic properties of the proteins, protein complexes and nucleic acid struc tures that form these foci 148 . A recent study using STED microscopy to study repair foci induced by ionizing radi ation 149 demonstrates that SR imaging has the potential to provide a detailed picture of how the nucleoprotein complexes involved in recombination are organized in living cells.

Conclusions and future perspectives
In this Review, we highlight how SM and SR optical microscopy techniques have contributed to our under standing of HR. In vitro SM studies combine high spatial and temporal resolution with the power of reductionist biochemistry to yield detailed new insights into reaction mechanisms. SR microscopy is a rapidly evolving area of research that offers exceptional spatial resolution, which provides new insight into how recombination proteins and related nucleic acid structures are organized in living cells. It is, of course, important to recognize that these microscopy techniques will never operate in isolation, and they will always be guided by both genetic stud ies and bulk biochemical analysis. Nevertheless, future developments in SM and SR optical microscopy tech niques, and in the systems under study, have the potential to make even greater contributions to the HR field.
Improving SR and SM technology. SM and SR studies are technologydriven, and advances in these technologies offer the potential for greater spatial and temporal resolu tion, which in turn will lead to more detailed biophysical insights. For instance, faster cameras with greater spatial resolution and higher sensitivity and brighter, more stable fluorophores would enable higherquality images to be collected over longer periods of time than is currently possible. These advances would also help overcome some of the technical problems that currently limit most SR studies to fixed samples, namely, slow image acquisition rates, challenges in labelling proteins in living cells and phototoxicity associated with the high laser illumination intensities 113,114,118 . Thus, SR imaging might eventually be used to look at HRrelated reaction mechanisms in living cells as they are occurring in real time.
Increasing the complexity of SM studies. As with any reductionist biochemical approach, many of the SM studies of recombination have focused on understand ing the detailed properties of one or two proteins in order to infer their particular contributions to the over all recombination pathway. Therefore, a major chal lenge faced in the SM field will be to gradually increase the complexity of the systems under investigation. For instance, most of the SM studies of recombinase fila ments described above have been confined to analy sis of only the recombinases (such as RecA or Rad51) and therefore reflect only the basal properties of these proteins 23,98 . However, it is clear from many years of genetic, cell biology and biochemical studies that the presynaptic complex contains many other proteins in addition to the recombinases, although the precise spatial organization of these proteins within the pre synaptic complex remains illdefined 3,4,15,16,148 . Moving forward, it will be essential to understand whether and how the presence of these additional recombination accessory proteins influences filament assembly and dynamics, the homology search and strand invasion. It will also be important to determine how these pro cesses take place within the context of chromatin, how chromatinremodelling factors work together with recombination machinery and how recombination is coordinated with DNA replication.
Determining biological function with SR. The struc tures that have been observed by SR imaging will require continued contributions from genetic, biochem ical and biophysical studies to more fully define their mechanistic and biological implications. For instance, although SR studies of the synaptonemal complex have provided beautiful optical images of these structures and new insights into their organization, the impact of this organization on function has yet to be fully revealed [133][134][135][136][137][138][139][140] .While the potential of SR imaging is far from being fully realized, we anticipate continued and rapid developments in this powerful technology that will lead to broader applications 114,118 . For instance, tra ditional approaches have been unable to determine how chromosome organization contributes to homologue bias during meiotic recombination, but this problem may eventually prove amenable to SR imaging.
In conclusion, as SM and SR studies continue to mature, they are likely to provide further insights into HR mechanisms. In the groundbreaking paper describ ing the DSBR model, the authors conclude that while genetic studies have been invaluable in the elucidation of recombination pathways, "…we suggest that bio chemical experiments will be necessary to determine the actual mechanism of initiation of … recombination" (REF. 5). We extend this sentiment by stating that the bio chemical information must span a range of spatial and temporal scales and must also be integrated with an indepth understanding of the biology of recombination and the consequences of its outcomes. The new optical microscopy tools available for SM and SR imaging of HR can help make this a reality, and the coming years should prove extremely fruitful for those investigating HR mechanisms.

Author biographies
Kyle Kaniecki is a PhD candidate in the Department of Genetics and Development at Columbia University, New York, New York, USA. He studies helicasemediated regulation of DNA repair at the singlemole cule level with Eric Greene. Previously, he studied cystic fibrosis geno typetophenotype correlation at Johns Hopkins University, Baltimore, Maryland, USA, with Garry Cutting and neuronal patterning at the University of WisconsinMadison, USA, with Yevgenya Grinblat. He also has experience in drug and diagnostic development in both aca demia and industry.
Luisina De Tullio is a postdoctoral research scientist at Columbia University, New York, New York, USA, with Eric Greene while con currently pursuing a career in academia as an assistant researcher of Consejo Nacional de Investigaciones Científicas y Técnicas

Key points
• Homologous recombination (HR) is an important pathway that enables the exchange of genetic information between DNA molecules. • Singlemolecule (SM) optical approaches are ideal for probing dynamic spatiotemporal processes that cannot be easily studied through traditional experimental approaches. • SM optical techniques have uncovered many features of HR and are providing new insights into molecular mechanisms with unprecedented detail. • Superresolution optical microscopy offers the ability to image flu orescent molecules inside fixed or living cells with greater spatial resolution than is possible with conventional microscopy. Its appli cation to visualize recombination is an emerging area of tremen dous potential.