Homologous recombination was described in Escherichia coli in the mid-1940s (351), and for many years it was thought to be the result of a sexual process, analogous to that found in eukaryotes. When the sensitivity to DNA damage of the first recombination-deficient mutants was noticed, it was realized that recombination in this bacterium may serve the needs of DNA repair as well (105, 107, 266, 267). Subsequently, genetic studies delineated two recombinational pathwaysthe primary, RecBC pathway, serving the needs of "sexual" recombination, and the secondary, RecF pathway, kicking in when the primary pathway is inactive and moonlighting at "postreplication repair" of daughter strand gaps (102, 106, 108). Still later, biochemical characterization of recombinational activities suggested that their primary role is in DNA repair (131, 132). Finally, the realization that disintegrated replication forks are reassembled by recombination justified the "repair" purpose for the RecBC pathway (130, 333) and prompted a revision of our ideas about the relationships of DNA replication and recombination.

The goal of this review is to consolidate genetic data on homologous recombination, physical data on DNA damage and repair, and biochemical data on recombinational enzymes under a different idea in an attempt to highlight new areas for the future in vitro and in vivo experiments. The different idea is that the primary role of the homologous recombination system in E. coli is to repair lesions associated with DNA replication of damaged template DNA (130, 336). Therefore, this review differs from other recent reviews on homologous recombination in E. coli (108, 320, 377) in that its two main emphases are on (i) the evidence for recombinational repair in bacteria and (ii) the interactions of various recombinational repair proteins with each other and with the replication machinery. The recombinational repair machinery is conserved among eubacteria, and so the same two basic pathways are present in such dissimilar species as E. coli and Bacillus subtilis. Therefore, although concentrating on the E. coli recombinational repair paradigm, occasionally I use evidence from other eubacteria.

Damage reversal and one-strand repair. Bacterial genomic DNA, like any macromolecule, is subject to constant chemical and physical assault. Repair of the resulting lesions is essential if DNA is to serve as the template for transcription and its own reduplication. In the course of evolution, a complex enzymatic machinery has evolved to maintain this centrally important molecule in usable form (195). Repair of some DNA modifications simply reverses the damage, returning DNA directly to its original state. For instance, photolyase, using near UV-visible light, splits UV-induced pyrimidine dimers (reviewed in reference 545). Another example is the suicidal Ada protein of E. coli, which transfers a methyl group from the modified base O⁶-methylguanine to itself (reviewed in reference 580).

Two-strand repair. Although the bulk of DNA damage affects one strand of a duplex DNA segment, occasionally both DNA strands are damaged opposite each other, resulting in two-strand damage, a term proposed by Howard-Flanders (266). To repair two-strand damage without the loss of sequence information, a cell needs a higher level of redundancy, an extra homologous sequence whose strands could be used to fix both DNA strands of the damaged sequence. The principle of such two-strand repair is depicted in Fig. 1. An affected duplex homologously pairs and exchanges strands with an intact homologous duplex (Fig. 1B). The resulting joint molecule is "resolved" by symmetric single-strand cuts in homologous strands, yielding two new DNA molecules, each containing a single one-strand lesion (Fig. 1C). Now the damaged strands can be mended by one-strand repair with the complementary strands as templates (Fig. 1D).

View larger version (17K): [in this window] [in a new window]   FIG. 1.   The idea of two-strand repair. (A) A DNA molecule with a two-strand lesion (small open rectangles in the solid duplex) is shown side-by-side with an intact homolog (open duplex). (B) The two sequences have exchanged strands in a homologous region, converting the two-strand lesion into a pair of one-strand lesions. (C) Junction resolution (the strands to cut are shown in panel B) separates the chromosomes from each other. (D) Excision repair removes the one-strand lesions, completing the overall repair reaction. Note that if black and white "parental" DNAs are not identical, the resulting chromosomes may become "recombinant."

Homologous recombination versus recombinational repair. Since the machinery for the two-strand repair is complex and not copious and since the repair incidents are rather infrequent, this type of repair is more accessible to genetic than to biochemical study. The principal genetic assay for two-strand repair is to monitor the formation of new chromosomes resulting from alternative resolution of joint molecules. A joint molecule (Fig. 2A) can be redrawn to show that the DNA junctions are able to isomerize (Fig. 2B). This isomerization of the junctions creates two possible ways of resolving each junction (shown by numbers beside the arrows [Fig. 2B]). If the resolution is random, in 50% of the cases the participating chromosomes will exchange shoulders, forming two "recombinant" chromosomes (Fig. 2C). If the parental chromosomes were genetically marked, progeny carrying recombinant chromosomes would be detected genetically as having traits that initially resided on separate parental chromosomes.

View larger version (18K): [in this window] [in a new window]   FIG. 2.   The four ways to resolve a joint molecule with a double junction. Lowercase letters w, x, y, and z designate unique sites which serve as markers on the homologous chromosomes. A junction is resolved by two symmetrical single-strand cuts (small black arrows in panel B) across each other. Each such diagonal pair of cuts is numbered either 1 or 2 for each junction. If junctions freely isomerize and are resolved independently of each other, four outcomes of the resolution are expected. In two of the outcomes, the chromosome arms will be exchanged, resulting in recombinant chromosomes. (A) A joint molecule with two junctions as shown in Fig. 1B. (B) The same joint molecule isomerized to show both junctions in the open planar configuration (498). (C) The four resolution outcomes, numbered according to the resolution options realized at the left and the right junctions.

The two mechanisms of two-strand damage. Two-strand lesions appear in DNA in two distinct ways. DNA synthesis in a region increases recombination in this region (447), suggesting that one source of two-strand lesions is DNA replication. The fact that replication of DNA containing one-strand lesions stimulates recombination between this DNA and an intact homolog (360) suggests that DNA replication causes two-strand lesions when it runs into unrepaired one-strand lesions. There are at least two mechanisms of replication-dependent conversion of one-strand damage into two-strand damage. In vivo, a noncoding lesion (for example, an abasic site) is an absolute block to DNA replication in growing cells (347); similarly, in vitro, a noncoding lesion in template DNA blocks the progress of the major E. coli DNA polymerases (59). In the chromosome, replication is likely to reinitiate downstream of a noncoding lesion (see "Elongation phase of DNA replication in E. coli" below), leaving behind an unfillable single-strand gap (Fig. 3) (see "Origin of daughter strand gaps and mechanism of their repair: early studies" below). Such an unfillable gap is called a daughter strand gap, since it appears in one of the two daughter branches after the replication fork passage (538, 734). Another type of one-strand lesion, a single-stranded interruption in template DNA, is proposed to cause a disintegration (collapse) of a replication fork (see "Evidence for replication fork disintegration" below) (234, 597). As a result, a double-strand end is detached from the full-length duplex molecule (Fig. 3). Finally, inhibited replication forks are broken, similarly releasing one of the replicating branches as a free double-stranded end (263, 334).

View larger version (20K): [in this window] [in a new window]   FIG. 3.   The two major types of replication-induced two-strand lesions. A replication fork moves from left to right along the template DNA with unrepaired one-strand lesions. The left template contains a noncoding lesion (T=T, thymine dimer), the right template has a single-strand interruption. Additional explanations are given in the figure and in the text.

The two recombinational repair pathways of E. coli. The two types of replication-induced two-strand lesions are repaired in E. coli by two separate pathways, both dependent on the recA gene but named after the critical genes that distinguish between them (Fig. 4). Daughter strand gaps are repaired by the RecF pathway (see "Repair of daughter strand gaps" below), while disintegrated replication forks are repaired by the RecBC pathway (see Double-strand and repair" below). The three common phases (see "Two-strand repair" above) of the two repair reactions are (Fig. 4) (i) presynapsis, during which the damaged DNA is prepared for homology search, followed closely by synapsis, during which homologous pairing and strand exchange with the intact sister duplex occur; (ii) DNA replication restart; and (iii) postsynapsis, during which the recombination intermediates are resolved.

View larger version (20K): [in this window] [in a new window]   FIG. 4.   The two pathways of recombinational repair in E. coli. On the left, the RecF (daughter strand gap repair) pathway is shown; on the right, the RecBC (double-strand end repair) pathway is shown. Additional explanations are given in the text.

Frequency of two-strand lesions. Under conditions of laboratory growth, two-strand lesions are too infrequent to be detectable in wild-type (WT) cells directly by physical techniques, although they are detectable in recombinational repair mutants (434). After massive DNA damage, daughter strand gaps are detected as single-stranded regions of several hundreds of nucleotides in the chromosomal DNA (278, 710) or as interruptions in the newly synthesized DNA (538, 714), double-strand breaks are detected as an immediate chromosome fragmentation (61, 323, 683), and disintegrated replication forks are detected as replication-induced chromosome fragmentation (60, 715).

Recombinational repair capacity of E. coli cells. WT E. coli cells grown in a nutritionally poor medium are able to survive 53 to 71 cross-links per chromosome (595). It can be calculated on the basis of the data with excision repair-deficient strains (714) that E. coli cells are still viable after repairing 100 to 200 daughter strand gaps per chromosome. E. coli cells should also be able to tolerate multiple disintegration of replication forks, because recombinational repair should reattach the resulting double-stranded ends to the circular domains of the chromosome. The only two-strand DNA lesion that has proved to be deadly for E. coli is a double-strand break. E. coli survives only two or three double-strand breaks in its chromosome (325, 683), which suggests that whenever a double-strand break occurs in an unreplicated portion of the chromosome, it cannot be repaired. Whether E. coli is an exception among bacteria in its inability to repair multiple double-strand breaks remains to be determined. There is a eubacterium, Deinococcus radiodurans, which can repair >100 double-strand breaks per chromosome (437), but this extreme resistance to DNA damage stands out in the bacterial world.

The two main E. coli reservoirs in nature are (i) the animal gut, where the microbe is dividing and concentrated; and (ii) the natural water of lakes and ponds, where the microbe is starving and diluted (560). In the gut, that is, in the environment rich in nutrients and protected from the elements, E. coli is likely to replicate its DNA for many generations without much need to repair it. However, when E. coli finds itself in the water, where DNA replication stops and DNA repair is anemic while the possibilities for damage of DNA are significant, the E. coli genome must accumulate a tremendous amount of DNA damage. Unfortunately, the gut is a discontinuous niche, since the animal the gut belongs to will eventually die; therefore, to survive in the long run, E. coli has to exit the old gut and recolonize a young one. When the battered E. coli from the water eventually makes it to a new gut and starts replicating, it finds its DNA riddled with unrepaired lesions.

The sporadic occurrence of massive DNA damage separated by long periods of undisturbed growth calls for a modest standby repair system, capable of rapid induction in response to increased DNA repair needs. Such an arrangement is indeed found in E. coli; the rapid increase in its DNA repair capacity is called the SOS response (506).

Repair instead of DNA damage checkpoints: the prokaryotic strategy. The bulk of two-strand DNA lesions in enterobacteria are probably generated as a result of DNA replication on template DNA containing one-strand lesions. An easy way to prevent this aggravation would be to stop DNA synthesis altogether when one-strand lesions are sensed. This is exactly what eukaryotic cells dothey employ checkpoint mechanisms to delay chromosomal replication when their DNA is damaged (reviewed in references 295 and 401). Since prokaryotes would also benefit from such a strategy, it was argued that E. coli might have a system to delay DNA synthesis when its chromosome is damaged (73).

Organization of the SOS regulon. DNA lesions inhibit DNA replication. Inhibition of DNA replication in E. coli induces the SOS response: an increased expression of some 20 genes aimed at restoring the capacity of the chromosome to replicate (Table 1). The resulting enhancement of the ability of the cell to repair and tolerate DNA damage is achieved in several independent ways. The capacity of the cell for excision repair (see "Damage reversal and one-strand repair" above) is enhanced by overproduction of the UvrD helicase and the UvrA and UvrC subunits of the UvrABC excinuclease. Induced amounts of DNA polymerase II increase the capacity of the cell for DNA synthesis across abasic sites (58, 484, 660). Up to a 50-fold increase in the amount of RecA protein (292, 543) and a similar increase in the expression of RecN protein (494) enhance the recombinational repair. The SOS induction makes possible repeated disengagement of replisomes stalled at the lesions in template DNA to allow resumption of the synthesis downstream, a phenomenon known as replisome reactivation (see "Replisome reactivation and model for RecFOR catalysis of RecA polymerization at daughter strand gaps" below). When recombinational repair cannot fix certain DNA lesions, the UmuD'C complex catalyzes translesion DNA synthesis (see "Backup repair of daughter strand gaps: translesion DNA synthesis" below). Overproduction of SfiA protein inhibits cell division (41), providing extra time for completion of recombinational repair. If all these measures fail to restore DNA replication, the lingering SOS induction awakens colicinogenic plasmids and dormant prophages, whose expression lyses the cell. The lysis of doomed cells benefits the viable cells of the same clone when resources are limited, since inviable bacterial cells can multiply for several generations, wasting precious nutrients. The lysis by induction of a prophage or colicinogenic plasmid is therefore an example of "bacterial apoptosis," which could have evolved to increase the number of viable cells in a clone.

Levels of SOS induction. The SOS response is by no means a desperate attempt to stay alive, as its name inaccurately implies (506), but, rather, an orderly and measured reaction of the cell to DNA synthesis inhibition. General information on the E. coli genes induced during the SOS response is summarized in Table 1. The strength of SOS boxes in the operator regions of the SOS genes correlates well with the in vitro LexA repressor affinities for the corresponding promoters and is likely to determine the timing of expression of a given gene during the SOS induction. According to thus inferred order of expression during the SOS induction, the genes of the SOS regulon could be loosely grouped into three categories. The first genes to be induced are mostly those responsible for one-strand repair (uvrA, uvrB, and uvrD) or damage tolerance (polB) (Fig. 5). The LexA repressor itself is also induced immediately. The DinI gene product, which delays activation of translesion DNA synthesis (753), is likely to be synthesized at this stage, too. Increase in expression of the immediately induced genes is usually less than 10 times that of their constitutive expression. If the increased expression of the one-strand repair genes does not help to regain normal rates of DNA synthesis, the genes of recombinational repair, recA and recN, are induced (Fig. 5). The maximal induction of these genes is higher, 20- to 50-fold over their regular levels. When DNA damage is massive, so that even the enhanced recombinational repair cannot overcome the inhibition of DNA replication, the third group of genes, represented by sfiA and umuDC, is called into action. Since these genes are expressed at very low levels during regular DNA synthesis, their SOS induction could be more than 100-fold. Expression of the umuDC operon inhibits recombinational repair and makes possible translesion DNA synthesis (see "Backup repair of daughter strand gaps: translesion DNA synthesis" below), while SfiA protein delays cell division. As DNA replication rates return to normal, the three expression groups of the SOS genes are likely to become repressed in the reverse order (Fig. 5). Alternatively, if a cell cannot repair its DNA damage and is doomed to generate a dead lineage, prophages and colicin plasmids are induced to lyse it (Fig. 5).

View larger version (21K): [in this window] [in a new window]   FIG. 5.   An idealized induction kinetics of the four groups of LexA-controlled genes (also, see reference 611). The graph illustrates the well-regulated nature of the SOS response. Both the x axis (time) and y axis (the level of the SOS induction) are in arbitrary units; therefore, the heights of the three curves are not to be compared.

The entire 4.7-Mbp circular chromosome of E. coli is traversed by a single replication bubble emanating from the unique replication origin. Both replication forks of the replication bubble are active; they meet in a chromosome region called the terminus, which is situated across from the origin. The terminus is delineated by termination sites arranged so as to form a replication fork trapreplication forks can enter the terminus, but they cannot exit it (253). To replicate the whole chromosome within a less-than-1-h bacterial cell cycle, replication forks have to proceed at about 650 bp/s (68). However, even the higher speed of almost 800 bp/s is insufficient when the cell cycle of E. coli is squeezed into 24 min in a rich medium. To prevent underreplication, E. coli starts a new round of DNA replication well before the completion of the ongoing round. Thus, in cells growing in a rich medium, there are one to three replication bubbles (two to six replication forks) (244).

Conceptually, replication of the E. coli chromosome is subdivided into three major phases: initiation, elongation, and termination (reviewed in references 25 and 406). Termination is the least understood phase (253) and is not immediately relevant to the needs of recombinational repair, although inhibition of DNA replication associated with termination sometimes causes disintegration of replication forks with their subsequent recombinational repair (263, 573). Elongation is the phase at which the two-strand lesion formation occurs and the recombinational repair machinery meets the replication machinery. Initiation of chromosomal DNA replication is helpful in defining interactions of the key replication proteins. An initiation strategy of multicopy plasmids is relevant because it utilizes the host reinitiation mechanism after the completion of recombinational repair.

Initiation of chromosomal DNA replication in E. coli. For a replication fork to start, the DNA duplex must be open. At the origin of chromosomal DNA replication, this opening is effected by binding of the initiator protein, DnaA. DnaA recognizes and binds to a degenerate nonanucleotide (T/C)(T/C)(A/T/C)T(A/C)C(A/G)(A/C/T)(A/C) (562). At the origin of chromosomal DNA replication, four DnaA recognition sites are found in a cluster. Binding of 10 to 20 DnaA monomers to this cluster of DnaA binding sites leads to an opening of DNA duplex nearby.

Nucleoid segregation and the problem of accessibility. Nucleoid segregation sets the "window of opportunity" for recombinational repair. In rapidly growing E. coli cells, nuclear bodies, or nucleoids, are seen in the electron microscope as dimers (688, 741), and even after being freed from their "cells," many purified nucleoids have a doublet appearance (240, 458, 490), indicating that the separation of nascent nucleoids is concomitant with DNA replication. It has been proposed that replicated daughter branches of the parental chromosome do not stay entangled but from the very beginning form separate bodies, growing out from the replication point in opposite directions (381, 740). Spatial separation of the origin-proximal markers after origin duplication was visualized in live cells (215).

recA and peculiarities of recA null mutants. recA happened to be the first E. coli recombinational repair gene to be discovered (107). recA is not a part of any operon, is surrounded by genes unrelated to DNA metabolism, and has its own promoter and terminator (260, 546). Normally, recA expression maintains 1,000 to 10,000 RecA monomers per cell (292, 445, 543, 558). RecA production is induced by DNA-damaging treatments such as UV irradiation or nalidixic acid, resulting in up to a 50-fold increase in the amount of the protein (see "Organization of the SOS regulon" above) (292, 543).

The variety of phenotypes of recA mutant cells stems from a single deficiency, the inability to form an active RecA filament. The in vitro properties and activities of RecA filament still bewilder and fascinate those who study them. For in-depth treatment of the enchanting RecA biochemistry, see the excellent reviews by Kowalczykowski (319) and Roca and Cox (524).

RecA without DNA. In high-concentration solutions, RecA aggregates to form oligomers, filaments, and bundles (70, 71, 246, 737). One of the major species in these aggregates consists of rings of six to eight monomers (70, 71, 246). These rings are characterized by electron microscopy for RecA from Thermus aquaticus, due to their greater stability (758). Surprisingly, they resemble in gross details both the hexameric rings of helicases like DnaB or RuvB and the F₁-ATPase (168, 760).

Filament formation by RecA around ssDNA. In vitro, in the presence of physiological concentrations of Mg²⁺ and ATP, RecA assembles around ssDNA into a helical filament (Fig. 6), an entity proficient in all known RecA activities (in the absence of the nucleotide cofactor or in the presence of ADP, RecA forms similar filaments but with different parameters; since such filaments are inactive in RecA-promoted reactions, they are not discussed in this review). At physiological pH, RecA filament does not readily assemble on duplex DNA; however, a filament assembled on ssDNA extends into a contiguous double-stranded region (362, 572). Every RecA monomer within a filament binds a single ATP molecule (307). RecA filament assembled on ssDNA slowly hydrolyses ATP at a rate of about 30 ATP molecules per min per monomer (69); duplex DNA-bound filament has an even lower ATPase activity (362, 502).

View larger version (85K): [in this window] [in a new window]   FIG. 6.   RecA filament in vitro: photographs and molecular model. (A) A relaxed circular duplex DNA completely covered by RecA filament. Naked duplex DNA of the same length, lying mostly inside the RecA filament, illustrates that DNA within RecA filament is stretched 1.5 times. Reprinted from reference 625 with permission of the publisher. (B) A RecA-covered circular ssDNA molecule; the arrow points to a segment complexed by SSB. Below, another ssDNA circle of the same length can be seen, but since it is entirely complexed with SSB, it appears very different, i.e., much smaller and kinky. Note that while RecA filament stretches ssDNA, SSB compacts it. Reprinted from reference 246 with permission of the publisher. (C) A crystal structure-based molecular model of an 18-monomer segment of RecA filament with three symmetry-related monomers in gray. Each of these monomers is enlarged on the right to show residues conserved among eubacterial RecA proteins (see reference 524 for details). In such a filament, the 5' end of DNA1 would be at the top. Reprinted from reference 524 with permission of the publisher.

Two DNA-binding sites in RecA filament. Soon after the beginning of biochemical characterization of RecA protein, it was realized that, to promote homologous pairing, RecA must have at least two DNA-binding sites: the primary site accommodating DNA1, around which the filament was assembled, and the secondary site for DNA2, to be compared with DNA1 (269). Since then, the idea of at least two DNA-binding sites within RecA filament has been substantiated with a variety of evidence.

Cleavage of LexA repressor by RecA filament. RecA filament holding a single DNA strand promotes autocleavage of the SOS response repressor, LexA (259, 366), as well as autocleavage of phage repressors (172, 522, 559) and of the UmuD protein (77). It is said that in these reactions RecA plays a role of coprotease, because there are conditions under which LexA, phage  repressor, and UmuD cleave themselves in the absence of RecA (77, 364). The LexA-binding site lies deep within the filament groove and overlaps with the secondary DNA-binding site (759), explaining why, when both DNA-binding sites are occupied by ssDNA, LexA cleavage is inhibited (152, 515, 642). RecA filament assembled on duplex DNA promotes LexA cleavage at 5 to 20% of the rate observed with a filament assembled on ssDNA (152, 642). This feature of the SOS repressor cleavage makes biological senseif all the single-stranded regions associated with DNA lesions are made double stranded (supposedly by pairing with intact homologous sequences), there is no reason to boost the repair capacity of the cell any further.

Detection by RecA filament of homology to ssDNA bound in the primary site. Although RecA forms a filament around ssDNA1 in a sequence-independent manner, the filament itself is "a sequence-specific DNA-binding entity, with the specificity determined by the bound DNA" (524). The amount of nonhomologous DNA in vivo is overwhelming, since even an identical sequence, shifted a single nucleotide out of register, becomes perfectly heterologous to DNA1. Heterologous DNA is not neutral in homology searches: preincubation of presynaptic filaments with heterologous DNA inhibits subsequent homologous pairing (214). The problem of the complexity of natural DNA is compounded by the enormous intracellular DNA packing densities, measured in E. coli at 20 to 100 mg/ml (57), and the restricted accessibility due to the nucleoid segregation (see "Nucleoid segregation and the problem of accessibility" above). Under these conditions, RecA has to find homology to the damaged DNA within minutes, as illustrated by the extreme DNA damage sensitivity of a partially active recA allele, proficient in homologous recombination in vivo and capable of "slow" recombinational reactions in vitro (273). If not repaired quickly, the damaged DNA could be degraded or segregated from its intact sister or could bring about even greater damage if the upcoming replication fork runs into it. On the other hand, if recombinational repair mends DNA damage mostly at replication forks, the affected and the intact homologous DNA segments should initially be in close proximity with each other.

Strand exchange between DNA1 and DNA2 catalyzed by RecA filament. When homology is found (i.e., when a homologous duplex DNA2 is aligned with ssDNA1), RecA filament catalyzes the exchange of strands between the two DNA molecules. In this process, ssDNA1 forms hydrogen bonds with the complementary strand of DNA2 while the identical strand of the duplex is displaced (135) (Fig. 7). The outgoing strand of DNA2 is accommodated in the secondary ssDNA binding site at the periphery of the RecA filament to be extracted from there later by SSB (346, 420).

View larger version (24K): [in this window] [in a new window]   FIG. 7.   The distinguishable phases of RecA-promoted reactions, as they are thought to happen in vivo. Black lines indicate a damaged duplex; white lines indicate an intact duplex; the open rectangle with rounded corners indicates a RecA filament; the irregularity in one of the black DNA strands indicates a noncoding lesion. The left side represents daughter strand gap repair; the right side represents double-strand end repair. Explanations are given in the figure.