Insights into homology search from cryo-EM structures of RecA-DNA recombination intermediates

Abstract

The fundamental reaction in homologous recombination is the exchange of strands between two homologous DNA molecules. This reaction is carried out by the RecA family of ATPases that polymerize on ssDNA to form a presynaptic filament. This filament then binds to dsDNA to form a synaptic filament, a key intermediate that mediates the search for homology and subsequent strand exchange to produce a new heteroduplex. A recent cryo-EM analysis of synaptic filaments has now shed light on this process. The dsDNA strands are separated on binding to the filament. One strand is sequestrated while the other is freed to sample pairing with the ssDNA. Homology, through heteroduplex formation, promotes dsDNA opening. Lack of homology suppresses it, keeping local synapses short so that multiple synapses can form and increasing the probability of encountering homology.

Introduction

Central to homologous recombination is the strand exchange reaction carried out by the RecA family of ATPases, conserved from bacteria to humans. This reaction brings together single-stranded DNA (ssDNA) and double stranded DNA (dsDNA), searches for homology between the two DNA molecules, and carries out the exchange of the ssDNA for the homologous strand of the dsDNA [1,2].

In the absence of ssDNA, RecA exists in an inactive state as a mixture of monomers and oligomers, the latter arranged in a helical filament of 6.2 RecA protomers per turn. Binding to ssDNA and ATP cooperatively induce a conformational change to a narrower and more extended filament, termed the presynaptic filament [2-4]. This conformational change also activates ATP hydrolysis, which cycles the filament between the active and inactive states (Figure 1). Initial insights into the structures of these two states came from helical reconstructions of negative-stain electron microscopy (EM) experiments [5]. Subsequently, a crystal structure reported the high-resolution structure of monomeric RecA, in a crystal-packing arrangement that serendipitously resembled the EM reconstruction of the inactive filament [6]. However, the constraints of crystallizing a polymer precluded the crystallization of a DNA-bound, active-state assembly. This problem was solved with the use of a mini filament that was constructed by fusing five RecA genes and mutating the first and last protomers to prevent further polymerization. This approach yielded high resolution crystal structures of the inactive and active filament states [4].

Figure 1. Schematic of the RecA-catalyzed strand-exchange reaction.

RecA is shown as yellow spheres, with the ADP and ATP-bound forms indicated by the letters D and T, respectively. The primary ssDNA is shown as a dark brown line, and the donor dsDNA as double lines, colored green for the complementary strand and red for the homologous strand. ATP hydrolysis, indicated by the release of a phosphate group (P_i), is activated upon ssDNA binding. After ATP hydrolysis, RecA protomers likely do not fully dissociate, and the active filament reassembles following the exchange of ADP for ATP, driven by the higher cellular concentration of the latter. By contrast, the donor dsDNA likely fully dissociates on ATP hydrolysis, rebinding stochastically at a different register (shown hypothetically as a shifted dsDNA) when the presynaptic filament reassembles, continuing the search for homology. With a fully exchanged postsynaptic filament, ATP hydrolysis results in the release of the new heteroduplex and the displaced homologous strand of the donor. With only a limited stretch of homology, ATP hydrolysis results in postsynaptic filaments that contain D-loops and other joint ssDNA-dsDNA molecules.

The crystal structure showed that the presynaptic filament structure contains three nucleotides per RecA. The ssDNA is arranged in sets of 3 nucleotides, termed nt-triplet, with the Watson-Crick edges of the base groups exposed to the solvent. Each triplet is contacted by three RecA protomers, and each RecA in turn contacts three nucleotides, underscoring the importance of a precise RecA-RecA relationship in the active filament [4]. Within each nt-triplet, the ssDNA conformation is very similar to B-type DNA. In-between triplets, however, the ssDNA backbone has a negative twist and is stretched, giving rise to a global ssDNA conformation that is unwound and stretched relative to B DNA. This separation of adjacent triplets, which prevents them from stacking, is stabilized by RecA loops termed L1 and L2.

The presynaptic filament then binds to donor dsDNA forming a complex, referred to as the synaptic filament, for which there was no structural information until recently. This is a key intermediate in the reaction because it is when the filament queries the dsDNA for homology to the ssDNA, and, on encountering homology, catalyzes the exchange of homologous strands between the ssDNA and dsDNA [1-3]. Strand exchange gives rise to a postsynaptic filament that contains a new heteroduplex and the displaced homologous strand from the donor duplex, which are released on ATP hydrolysis[7]. Homology that is limited in length results in joint molecules where the ssDNA and dsDNA remain attached through D-loop structure(s). It is thought that these undergo additional ATPase cycles of dsDNA binding, homology search and strand exchange until the correct register of the two DNA molecules is encountered (Figure 1).

RecA consists of a flexibly-linked ~30-residue N-terminal α helix (αN), followed by a ~240 residue ATPase core, and a 64-residue globular domain C-terminal domain (CTD) [4,6]. The αN helix mediates oligomerization in both the inactive and active states by binding to the helicase core of another RecA in the 5’ direction (of ssDNA). In the inactive state, the ATP-binding site is exposed, while in the active filaments it packs with the 5’-adjacent RecA, which provides a 2-lysine motif that activates the ATP hydrolysis and also forms a second polymerization interface via the ATP [4]. The repositioning of the helicase core is accommodated by a conformational change in the linker between αN and the helicase core. The L1 and L2 loops and nearby elements that interact with the ssDNA reside in the helicase core and form the primary ssDNA binding site (also referred to as S1; Figure 2a) . Their engagement with the ssDNA cooperates with ATP binding to bring about the conformational change from the inactive to the active state of the filament [4].

Figure 2. Examples of intermediates identified by the cryo-EM analysis of a strand-exchange reaction.

(a) Top, semi-transparent rendering of the molecular surface of RecA (yellow) with a stick representation of a 3 bp heteroduplex and a 5 nt segment of the complementary strand [ref]. The DNA segments are colored as in Figure 1, and they are also indicated schematically on the right. The primary DNA binding site is marked by the L1 and L2 loops (showing through the semi-transparent RecA surface), while the secondary site involves the β6 strand and the L2 loop. The αN oligomerization motif and the CTD are indicated by dashed circles. Bottom rendering includes a donor duplex (blue) bound to the CTD. This duplex is connected to the complementary strand of a heteroduplex triplet centered on the RecA that is 2 protomers away, in the 3’ direction of the primary ssDNA, from that to which the duplex binds. The other strand of the duplex is connected to the 5 nt homologous strand on the same protomer as the duplex.

(b) Schematics of mini filaments with strand exchange intermediates of non-homologous dsDNA. Top row shows hypothetical transient states during strand separation after the initial binding of the dsDNA to a random RecA protomer, designated as “D” in this example, with the labels starting with the letter “A” at the 3’ end of the 9-RecA mini filament. DNA is depicted as in (a), except for the dotted blue lines that indicate the flexible disposition of the 5’-most duplex and of the separated strand. Strand separation terminates when the second duplex binds to a CTD (labeled) to give rise to the structures at the bottom row observed in the cryo-EM analysis. The relative abundance of each structure is derived from the approximately 20 % probability of strand separation terminating at each RecA step.

(c) Schematics of mini filaments with strand exchange intermediates of partially-homologous dsDNA. The Watson-Crick pairing of the complementary strand with the ssDNA promotes strand separation through the pairing energy as well as by keeping the complementary strand away from the homologous strand. Thus, dsDNA opening proceeds unencumbered across the region of homology. Beyond that, strand separation tapers off as with the non-homologous dsDNA in (b).

In the crystal structure of a partial postsynaptic mini filament, which contained a heteroduplex but lacked the displaced strand, the strand-exchanged heteroduplex is arranged in base pair triplets, each very similar to B DNA, with adjacent triplets separated as in the presynaptic filament [4]. Together with the structure of the presynaptic filament, the postsynaptic structure suggested that part of catalysis of strand exchange involves RecA holding the otherwise flexible ssDNA substrate in a B-DNA conformation that can sample only Watson-Crick pairing with a strand released from the donor dsDNA [4], with the 3-nt step confirmed by single-molecule imaging studies [8]. Other aspects of strand exchange, including how the double-stranded nature of dsDNA is destabilized to release a strand and how RecA searches for homology were incompletely understood due to the lack of structural information on synaptic filaments [3,8-11]. Mutagenesis had identified a secondary DNA-binding site on the RecA helicase domain that is necessary for strand exchange but not for ssDNA binding and active filament formation. This site, which is centered on the β6 beta strand, was thus thought to bind to both the donor dsDNA substrate and its displaced homologous strand product of the reaction [12-15].

Cryo-EM analyses of synaptic mini filaments

These and other aspects of strand exchange have now been revealed with cryo-EM analyses of synaptic filaments and D-loop intermediates [16]. The study used reactions containing a mini filament of nine RecA protomers, 27 nt primary ssDNA, the nonhydrolyzable ATP analog ATPλS, and dsDNA that was either entirely non-homologous or had a 10 nt region of homology in the middle. Unlike X-ray crystallography, cryo-EM affords the ability to computationally classify a wide range of compositionally and conformationally distinct particles, akin to biochemical fractionation. The data sets, each consisting of approximately 1 to 1.5 million particles, were thus subjected to iterative computational classification. For each data set, an ensemble of over fifty synaptic mini filament structures were characterized, providing a snapshot of the most common complexes present at the synaptic stage of the reaction.

This analysis revealed that the initial binding site of donor dsDNA is the RecA CTD domain (Figure 2a), which is at the periphery of the filament [16]. The role of the CTD was previously unclear, although early NMR chemical shift perturbation studies with an isolated CTD fragment had suggested it has dsDNA affinity [17,18]. The binding of the dsDNA to the CTD directs the DNA towards the filament interior, where it would clash with the RecA L2 loop. It is proposed that this configuration results in the L2 loop inserting into the dsDNA, separating the two strands and initiating the opening of the duplex [16]. Although no structures were captured of the L2 loop in the process of opening the dsDNA, the post-opening structures recovered have key L2 loop hydrophobic residues capping the last base-pair at the opening. Similar loops or beta hairpins inserting into and opening dsDNA is a common motif in DNA-damage recognition proteins. Like the proposed role of the L2 loop, the Nucleotide Excision Repair factors XPC and XPE open up two adjacent base pairs by inserting and L2-sized hairpin and loop, respectively [19,20].

Once duplex opening is initiated by the L2 loop insertion, it is propagated through one strand of the dsDNA interacting with the secondary DNA-binding site, termed S2 site (Figure 2a). This frees the other strand of the dsDNA to sample Watson-Crick base pairing with the primary ssDNA at the S1 site (Figure 2b shows a schematic for this process). In the reaction with nonhomologous dsDNA, dsDNA opening can stochastically terminate at each RecA step, with the as-yet unopened dsDNA portion stacking with an L2 loop and binding to another CTD (Figure 2b). This gives rise to a D loop-like DNA structure of two CTD-bound duplexes linked with a single strand at the S2 site, while the other strand (the one to become complementary strand if there was homology) is disordered [16].

The particle classification analysis recovered nearly all possible combinations of S2-site connected paired duplexes on the 9-RecA mini filament; the number of particles fractionating in each class then gave a measure of the abundance of the particular D loop-like structure (see Figure 2b for examples of such intermediates). The abundance of a structure decreased with increasing distance between the two duplexes (i.e., length of S2-bound strand). This indicated that in the absence of homology, which was the case in this reaction, dsDNA opening has a substantial, ~20 % probability of terminating at each RecA step (Figure 2b) [16]. The significance of this may be to locally limit the length of ssDNA sampled for pairing if homology is not encountered. This in turn could allow the formation of multiple synapses, far apart on the donor dsDNA, increasing the probability of the filament encountering the correct register of the two sequences. This is consistent with the ‘intersegmental contact sampling’ model, proposed based on single-molecule imaging, where multiple RecA-DNA contacts facilitate the search for homology [11].

The bound duplexes were observed to have two distinct angles and modes of stacking with the L2 loop. The analysis of binding modes and their abundance at each RecA position further suggested that the duplex opening propagates in the 3’ to 5’ position (relative to the primary ssDNA) [16].

The data from the reaction containing the 10-nt homology dsDNA was analyzed similarly (Figure 2c) [16]. The resulting ensemble of synaptic structures exhibited two major differences from that with the non-homologous DNA. First, patchy density for the complementary strand base pairing with the primary ssDNA was observed in classes of S2-connected paired duplexes that spanned the central 10-nt homology region. For each of these classes, further sub-classification for the strength of this complementary-strand density revealed one class with strong density, essentially a bone-fide D-loop structure with a central heteroduplex, and another class with no density. This latter class was similar to the structures seen with non-complementary dsDNA, and presumably corresponds to where the dsDNA opened up in a segment outside the region of homology. Pairing was evident even in classes where one or both duplexes were beyond the region of homology, with the portion(s) of the complementary strand that extended beyond the region of homology being disordered (Figure 2c). The second difference involved the abundance of S2-paired duplexes as a function of increasing RecA–RecA distance. By contrast to the non-homologous DNA reaction, their abundance was constant across the region of homology as long as there was heteroduplex. This indicated that pairing suppresses the stochastic termination of dsDNA opening, allowing for the sampling of Watson-Crick hydrogen bonds with the primary ssDNA for as long as there is successful pairing. Mechanistically, this is likely a result of the pairing energy stabilizing the strand-separated state of the donor dsDNA (Figure 2c). In addition, heteroduplex pairing would keep the complementary strand away from the homologous strand, thus stabilizing the sequestration of the latter at the S2 site.

High resolution structure of a D-loop intermediate

The aforementioned D-loop structures had residual heterogeneity, because dsDNA binding events that were offset by one or two nucleotides relative to the homology region could not be further sub-classified (the duplexes would be in the same positions, while the differences in complementary strand density relative to the duplexes were too minor for classification). Thus, to obtain a single, well-defined D-loop for high resolution structure determination, the 10-nt region of homology of the dsDNA was embedded in a bubble of mismatched base pairs [16]. The 2.8 Å-resolution structure revealed the details of the protein-DNA contacts and provided additional insights into the S2 site and the opening of the dsDNA (Figure 3a).

Figure 3. High resolution structure of a 10 nucleotide D-loop.

(a) Cartoon representation of a D-loop on the 9-RecA mini filament. The DNA is colored as in Figure 2, and the CTD domains to which the duplexes bind are denoted by superscript letters. The ATPγS molecules at each RecA-RecA interface are shown in sticks and the magnesium ion as green spheres. The DNA sequence is shown below the structure.

(b) Close-up view of the S2 ssDNA, colored as in (a), except that the β6 and L2 structural elements of the S2 site are colored in orange. The β6 and L2 elements are labeled with the superscript denoting the RecA protomer to which they belong. S2 ssDNA nucleotides are numbered according to (a). Green-dotted lines indicate hydrogen bond contacts. Reproduced with permission from Yang et al., [16].

The S2 site to which the homologous strand of the dsDNA binds consists of portions of the helicase core β sheet as well as the L2 loop (Figure 3b). The homologous-strand conformation is characterized by extensive stacking interactions of bases as well as sugar groups, both within the DNA and between the DNA and RecA, with the Watson-Crick edges of its bases generally sequestered. Each S2 site sequesters five nucleotides of the homologous strand, in contrast to the three-nucleotide per RecA repeat of the primary ssDNA and heteroduplex binding sites (Figure 3b) [16]. It is possible that this evolved to increase the efficiency of the homology search, since the opening of a longer stretch of dsDNA would produce a longer complementary strand for homology sampling with the relatively shorter primary ssDNA.

The high-resolution D-loop structure also provided a mechanistic explanation for the apparent 3’ to 5’ preference of dsDNA opening. The 3’ duplex is bound by the S2 site starting with the first nucleotide immediately after DNA opening by the L2 loop, likely cooperating with the L2 loop insertion in dsDNA opening. This mechanism is not available to the 5’ duplex, which has a spacer of several unbound nucleotides before the S2 contacts (Figure 3b). The same pattern of S2 contacts was also observed in the structure of another, 12 nt D-loop [16]. The conserved nature of this feature supports the notion that the first duplex to bind to a CTD becomes the 3’ duplex, while the binding of the 5’ duplex occurs as the opening of the dsDNA terminates.

It is important to note that these reactions were performed in the apparent absence of ATP hydrolysis (the maps had clear density for the γ phosphate group of the slowly-hydrolyzable ATPγS used in the reactions, and similar results were obtained with the non-hydrolyzable AMPPNP analog; N.P. Pavletich personal communication). While ATP hydrolysis is not an inherent requirement for the exchange of short DNA molecules (except for the release of products), it is essential with longer, physiologically relevant substrates [21-24]. This requirement for ATP hydrolysis is incompletely understood. Biochemical data have indicated it plays a role in bypassing heterology, and in the extension of initial joint molecules and branch migration [25,26]. This presumably involves the release of the portions of the donor duplex that have not exchanged, followed by their resampling in a new round of the reaction. As ADP exchanges for ATP, the primary ssDNA likely does not diffuse away because RecA protomers are topologically wrapped around the ssDNA through the αN interface. The unexchanged portions of the opened-up donor dsDNA likely pair again, allowing the reconstituted dsDNA to rebind stochastically at a different register (shown hypothetically as a shifted dsDNA in Figure 1) to continue the search for homology. With long DNA substrates, about ~100 ATP molecules are hydrolyzed per base pair exchanged in vitro [27].

Many of the findings from these RecA studies likely also apply to Rad51, the eukaryotic RecA homolog. Cryo-EM studies show that Rad51 binds to primary ssDNA and heteroduplex similarly to RecA [28,29], and mutagenesis has identified secondary site residues at positions analogous to those of RecA [30]. Although Rad51 lacks the RecA CTD, it contains an N-terminal domain (NTD) that occupies an analogous position at the filament periphery [28,29,31], except that it is oriented with its solvent exposed surface pointing to the 5’ end of the filament instead of the 3’ end that the RecA CTD points to [16]. As such, if the Rad51 NTD is involved in the initial binding of the dsDNA, the resulting synaptic structures may differ in the arrangement of the duplexes and polarity preference of dsDNA opening.

Conclusions

RecA and recombination have been the focus of a large body of literature, ranging from biochemistry to single-molecule imaging and structural biology. The crystal structures of the presynaptic and partial postsynaptic filaments had set up the basic framework for understanding the mechanism of the strand exchange reaction, but the questions of how the donor dsDNA is processed and how the homology search occurs had been poorly understood. The recent structural analysis of the synaptic filament has now shed light on these questions. This was made possible by the ability of cryo-EM to identify and quantify distinct conformational and compositional states in the ensemble of recombination intermediates, painting a detailed picture of the reaction akin to a partition function.

The data point to the initial binding of the dsDNA to a CTD as the event that also initiates strand separation, likely mediated by the L2 loop inserting into the duplex. In the absence of local homology, strand separation has a substantial probability of stochastically terminating at each RecA step. Homology suppresses this process, allowing strand separation to extend unencumbered. This presumably limits the futile strand separation in the absence of homology, allowing for the formation of multiple synapses to sample homology elsewhere along the dsDNA. One of the remaining questions is how these initial joint molecules, often having multiple local synapses, get processed through multiple rounds of ATP hydrolysis to eventually lead to the exchange of long stretches of DNA.