Although homologous recombination has been reported in E. coli, the number of reported recombinants was two to three orders of magnitude lower (≈1 per 106 cells) than the native homologous recombination efficiencies of other microbes, such as Saccharomyces cerevisiae. However, when bacteriophage-based recombination proteins, such as Rec E/T or Red αβγ, are expressed in E. coli, recombination efficiencies reach up to about 1 per 103–4 cells without any further optimization [1, 3, 15]. In the Rec E/T system, RecE is an exonuclease with 5' to 3' activity and RecT is a ssDNA binding protein that stabilizes the ssDNA intermediate involved in annealing to the newly introduced complementary DNA strand during replication [16, 17]. The Lambda Red system consists of three proteins: Exo (also known as a), Beta, and Gam. Exo is an exonuclease, which degrades DNA in the 5' to 3' direction (similar to RecE); Beta is a ssDNA binding protein, which binds to ssDNA (similar to RecT), stabilizing it, and protecting it from further action by exonucleases; and finally Gam inhibits RecBCD and SbcCD activity in the host, thereby protecting the exogenous DNA from being degraded by natural mechanisms [18, 19]. Comparing the two phage systems, RecE and Exo serve the same function within the cell. The same is true for RecT and Beta, however, RecE does not function with Beta, and Exo does not function with RecT, so they are not interchangeable . Recombineering with dsDNA requires the presence of all three Lambda Red proteins, but ssDNA recombineering only requires the Beta protein [17, 21]. A recent study by Fu et al.  reported that the Rec E/T system could be used for direct DNA cloning; Rec E/T is highly efficient at linear–linear homologous recombination and is more efficient than the Lambda Red proteins. Despite this, the majority of current research efforts employ the Lambda Red proteins, which is the main focus of the rest of this review.
2.1 Current mechanism
Understanding of the mechanism of recombineering has evolved significantly over the last few years. Initially, it was thought that recombineering occurred by strand invasion [23, 24]; however, in a recA– background, recombineering remains highly efficient and therefore a RecA–-independent mechanism was proposed, wherein the newly introduced DNA is incorporated by annealing at replication forks . Stahl et al.  performed a detailed study to determine which mechanism was more likely, and recombination products from crosses of Lambda phage showed characteristics more consistent with annealing. Later studies demonstrating enhanced targeting of dsDNA recombineering to the lagging strand provided further support for the replication-fork annealing model [26, 27]. The most recent mechanism (Fig. 1), proposed by Mosberg et al. , hypothesized that replication-fork annealing occurred through a fully ss intermediate. When dsDNA is used, the first step in recombination is the degradation of one complete strand by Lambda phage Exo. It is believed that Exo and Beta act synergistically, with Exo aiding in the binding of Beta to the ss intermediate ; thus overcoming any potential issues with secondary structures of a fully ss intermediate. Beta binds to the ssDNA to protect it from further degradation and catalyzes its placement and annealing to the lagging strand of the replication fork, acting as an Okazaki fragment [28, 30, 31]. The homology regions of the oligo bind to complementary regions and, during the next round of replication, the insertion (or deletion, mutation, etc.) is incorporated into the newly synthesized DNA. This proposed mechanism is supported by experimental evidence, such as higher efficiencies for oligos targeting the lagging strand [28, 32, 33] and only Beta is needed for ssDNA recombineering [17, 21]. Since the mechanism of annealing to the chromosome is suspected to be the same for both ss- and dsDNA, the main advantage of using dsDNA recombineering is that larger inserts can be created with the polymerase chain reaction (PCR) (possibly including selectable markers) without any further processing to obtain ssDNA. This picture of the mechanism of recombineering is not complete, because recombination does occur on the leading strand, but how that occurs is not yet known.
Figure 1. Current proposed mechanism of Lambda Red recombineering in E. coli. Upon induction of Lambda Red proteins, the Gam protein blocks activity of RecBCD (ExoV) and SbcCD, which degrade exogenous DNA. The Exo protein degrades one DNA strand in the 5' to 3' direction and recruits Beta to bind the exposed ssDNA to protect it from further degradation. Beta also promotes annealing to the lagging strand at the replication fork during DNA replication. The ssDNA displaces an Okazaki fragment at the replication fork and becomes integrated into the newly synthesized DNA.
Download figure to PowerPoint
2.2 Improvements in efficiency
With a better understanding of the recombineering mechanism, there have been numerous recent studies focused on improving the efficiency of Lambda Red based recombineering. In general, the most efficient methods now entail (i) optimizing the oligo used for recombineering to enhance replication-fork annealing and limit proofreading; (ii) mechanisms to modify the replisome itself; and (iii) multiplexing recombineering targets and automation.
Sawitzke et al.  investigated the effect of increasing oligo concentration and found that saturation occurred at 3000 oligos per cell. In the same study, the optimal length was also investigated. The optimal length was a 60- or 70-mer, but essentially oligos of 40–70 bases gave the same recombineering efficiencies, with recombineering efficiency decreasing rapidly for oligos of less than 25 nucleotides in length . Wang et al.  investigated the efficiency of recombineering with varying sizes of insertions, deletions, and mismatches. For all cases, efficiency dropped off with increasing length of nonhomologous sequence, such as mismatch, insertion, or deletion; this indicates that minor rewriting can be achieved at much higher efficiencies than large changes. They also reported 90 bp as the optimal oligo length. Phosphothionating the four nucleotides at the 5' end of the oligo protects them from endogenous exonucleases and increases efficiency two-fold . Finally, designing oligos to target the lagging strand increases efficiency about 30-fold [32, 33], which also supports the current mechanism of recombineering, which states that the oligo acts as an Okazaki fragment on the lagging strand of DNA replication.
Several strategies for modifying the cellular machinery involved in DNA replication have been shown to be effective, improving efficiency up to several hundred-fold. The first major strategy was to inactivate the mismatch repair (MMR) system by knocking out MutS, which is responsible for correcting mistakes made during replication. Constantino and Court  reported that, by removing mutS, oligo-mediated recombination increased 400-fold for a GB7G mismatch and 100-fold for most other mismatches. However, the basal rate of mutation increases 100-fold in a ΔmutS background, which can have undesired secondary effects, in particular, when attempting to map engineered mutations onto selectable phenotypes. One approach to circumvent the DNA MMR system is to use chemically modified bases that are not recognized by the MMR proteins. Wang et al.  used 2'-fluorouridine, 5-methyldeoxycytidine, 2,6-diaminopurine, or isodeoxyguanosine instead of the natural bases and found that allelic replacement efficiencies increased 20-fold in a strain with 100-fold lower background mutation rate. Mutations were also introduced at higher efficiencies in cells with a functional MMR system by introducing four or more adjacent mismatches or introducing mismatches at four or more consecutive wobble positions near the mutation site . A more recent study introduced a known mutation into the DNA primase to reduce the frequency of priming for Okazaki fragments , which resulted in longer Okazaki fragments and greater accessibility to ssDNA on the lagging strand. This mutation, dnaGQ576A, resulted in 63% greater allelic replacement per clone than the wild type. A follow-up study removed five endogenous exonucleases (RecJ, ExoI, ExoVII, ExoX, and Lambda Exo); these mutations alone increased recombineering efficiency 46% and when combined with the dnaGQ576A mutation resulted in 111% more clones per cycle than the wild type; however, these mutations came at the expense of cellular growth rate [36, 37].
Multiplexing and automation were also used to increase the efficiency of recombineering. Wang et al.  built a device that automated all the steps in recombineering; thus enabling many more rounds of recombineering in one day. Multiplexing oligos also enabled the efficient creation of combinatorial libraries. This technique, multiplex automated genome engineering (MAGE), was used to create combinatorial libraries containing about 3 billion mutants in a few days, several of which were capable of producing five-fold more lycopene (mutations were directed at genes in the 1-deoxy-D-xylulose-5-phosphate (DXP) pathway) . MAGE was also used to massively rewrite the E. coli chromosome, changing all 314 TAG stop codons to TAA stop codons. In this study, Isaacs et al.  speculated that there was a subpopulation of cells that were highly recombinogenic, and thus, contained many more mutations than other cells. This discovery led to the development of the coselection MAGE approach [38, 39]. The strategy here is that, since DNA is sufficiently unwound at the replication fork so that up to 500 000 bp are exposed, then if one oligo is integrated into the chromosome, it is likely that a second (or third, fourth, etc.) oligo designed to be integrated close by would also be integrated. Therefore, coselection markers, which are easily screened or selected for (e.g. antibiotic resistance or amino acid auxotrophy), are chosen around the chromosome and oligos designed to turn them on/off are mixed in with the oligo mixture used during recombineering. The use of coselection greatly reduced the number of colonies that had to be screened and resulted in the identification of strains with as many as 12 mutations . Sawitzke et al.  also reported increased recombineering efficiencies (≈100-fold) in cells co-electroporated with a selectable plasmid. As such, these techniques, along with access to cheap and accurate oligo libraries (up to several hundred thousand), have set the stage for massive genome engineering at a scale of dozens to hundreds of modifications in parallel and on laboratory timescales.