The results described in this paper provide additional insight into the cellular processes occurring during recombineering. First, the discovery that ExoVII degrades the ends of dsDNA cassettes solves the mystery of why dsDNA with both 5′ ends phosphorothioated is still recombinogenic despite PT bonds blocking Lambda Exo. Dually-phosphorothioated dsDNA enters the cell, after which ExoVII degrades the 5′ PT bonds of one or both strands. This step may also require the action of a helicase or another endogenous nuclease in order to generate ssDNA ends to which ExoVII can bind. After the action of ExoVII, Lambda Exo degrades the rest of the leading-targeting strand, leaving behind the lagging-targeting ssDNA recombination intermediate 
. Therefore, dually-phosphorothioated dsDNA recombines with a frequency equal to or exceeding that of unmodified dsDNA despite requiring the action of one or more endogenous nucleases in addition to Lambda Exo. This suggests that there is a great deal of interaction between endogenous nucleases and recombineering cassettes. The variable dsDNA recombination frequencies exhibited in this study by nuclease knockout strains support this notion. For example, the nuc4−
strain exhibits decreased recombination frequency for all tested VPT cassettes. The origin of this phenotype is uncertain; it is possible that one of the removed exonucleases has a role in dsDNA recombination, or that the presence of exogenous linear dsDNA is somewhat toxic to cells with these four nucleases removed, thereby selecting against cells which take up the cassettes necessary for recombination. Along similar lines, strain EcNR2.recJ−
appears to be non-recombinogenic, for reasons that are unclear, while EcNR2.xseA−
has slightly enhanced recombination frequency for cassettes without PT bonds blocking the leading-targeting strand.. Taken together, this suggests that modifying endogenous nuclease activity may be a powerful lever for affecting Lambda Red recombination, and possibly a valuable strategy for engineering further improvements of dsDNA recombination frequency.
The VPT series recombination frequency data for EcNR2 () also further supports the recently proposed dsDNA recombineering mechanism involving a full-length ssDNA intermediate 
. Other proposed mechanisms 
have suggested an intermediate in which the 5′ ends of both homology regions have been recessed by Lambda Exo, leaving behind the dsDNA heterology region flanked by 3′ homology ssDNA overhangs. These proposed mechanisms have different implications on the expected recombination frequencies of certain VPT series cassettes. According to the recently proposed mechanism, VPT7 would be expected to have a recombination frequency significantly less than that of unmodified dsDNA, as PT bonds between the 5′ homology and heterology regions of both strands would prevent Lambda Exo from degrading one strand and generating a full-length ssDNA intermediate. In contrast, the placement of PT bonds in VPT7 would be expected to facilitate Lambda Exo’s generation of the 3′ overhang intermediate suggested in the other proposed mechanisms. Thus, if this mechanism were correct, VPT7 would be expected to have equal or greater recombination frequency than unmodified dsDNA. As shown in and Table S2
, VPT7 has 10.2-fold lower recombination frequency than unmodified dsDNA (VPT1) in EcNR2. This result is consistent with those from similar previous experiments 
, supporting the proposed mechanism involving a full-length ssDNA intermediate, and further refuting the previously proposed mechanisms.
This work also lends more insight into recombineering with oligonucleotides. We have shown that removing selected endogenous nucleases improves multiplex recombination frequency, despite the use of a high total concentration (5.2 µM) of oligos. This suggests that the concentration of any given oligo within the cell is not enough to overcome the action of endogenous nucleases, and therefore that oligo entry into the cell is a limiting factor in MAGE. This observation initially appears to contradict the conclusion presented in a recent study by Sawitzke et al
, which showed that the recombination frequency of a given oligonucleotide could be increased by the addition of non-specific carrier oligos which prevent endogenous exonucleases from degrading the recombinogenic oligo. However, this phenomenon was tested only for very low concentrations of the recombinogenic oligo (up to 0.01 µM). Even at this concentration, adding carrier oligos (at 0.1 µM) had less than a 2-fold enhancement of recombination frequency; a more pronounced enhancement was observed only for very low concentrations of recombinogenic oligo (0.001 µM and below). In concentration regimes typical for MAGE (>1 µM total oligos), it has been shown that adding a second oligonucleotide decreases
the recombination frequency of the first oligonucleotide 
. Presumably, at these concentrations (which have been previously been shown to be optimal for recombination frequency 
and were therefore used in this work) any benefit conferred by the saturation of endogenous exonucleases is outweighed by competition for cellular entry. These findings therefore suggest that enhancing oligo uptake is likely to be a fruitful avenue for further improving MAGE.
Additionally, studying the consequences of different levels of oligo phosphorothioation in the EcNR2, EcNR2.xseA− and nuc5− strains enabled the deconvolution of the countervailing effects of phosphorothioation. Phosphorothioate bonds improve recombination frequency by protecting oligos against nuclease degradation, but also hinder recombination frequency – possibly by reducing the strength of the annealing interaction between the oligo and the lagging strand of the replication fork, or by causing toxicity to the cell. Two PT bonds on both ends of recombineering oligonucleotides was found to be optimal for all three strains tested in this work, but future nuclease knockout strains will need to be optimized in order to determine the ideal number of PT bonds for oligos recombined into that strain.
Beyond providing additional mechanistic insight into Lambda Red recombination, this work also achieves several important improvements of oligo and dsDNA recombineering. Firstly, removing ExoVII was shown to improve the inheritance of mutations at the 3′ ends of oligonucleotides and dsDNA cassettes. This may be particularly useful in oligonucleotide recombineering, as it allows more mutations to be reliably introduced by a single oligo. This could be leveraged for several applications, such as simultaneously modifying several residues near the active site of a protein, recoding 
a larger region with a given oligonucleotide, or modifying several genetic features (e.g. promoter strength, ribosome binding site strength, and the presence or absence of a premature stop codon) with a single oligo. Similarly, the improvement of MAGE recombination frequency via the use of EcNR2.xseA−
will also have substantial utility. These strains facilitate greater modification of a population of cells in a single MAGE cycle (or in multiple cycles), which will be useful for projects that seek the rapid allelic diversification of a population of cells 
. Such advancements are also expected to be useful for improving the Red-mediated diversification of BACs and plasmids. Similarly, the enhanced recombination frequency of these strains also means that fewer cycles will be needed in order to achieve an isogenic recoded population of cells, or to identify a strain with all desired genetic changes among a set of screened clones.
Finally, this work provides guidelines as to the appropriate strain to use for a given recombineering application. It should be noted that the nuc5−
strain was observed to have poor regrowth after electroporation, taking roughly twice as long as EcNR2 or EcNR2.xseA−
to recover to confluence. The pre-electroporation growth rate of nuc5−
was only slightly less than those of EcNR2 and EcNR2.xseA−
(150 minutes vs.
125 minutes to reach mid-log growth phase from a 1
100 dilution of overnight culture), likely due to the ability of mutS
removal to suppress the known cold-sensitive growth phenotype of recJ
quadruple mutants 
. Therefore, while nuc5−
has somewhat better multiplex recombination frequency than EcNR2.xseA−
, its poor regrowth properties cause each recombination cycle to take notably longer. Thus, using EcNR2.xseA−
is likely optimal for applications in which multiple cycles are necessary; the nuc5−
strain may be preferable for applications in which a single cycle is sufficient and fast regrowth is not necessary. While quadruple mutants for recJ, xonA, xseA,
have increased point mutation rates, this phenomenon is epistatic to mutS
. Given that strains used for genome engineering are often mutS−
, removing these nucleases (as in nuc5−
) does not further exacerbate the mutator phenotype. However, the combined removal of xonA
, and exoX
has been shown to increase rates of rearrangement mutations involving repetitive sequences 
, and therefore this strain should not be used for applications in which genomic stability is paramount. Directed evolution and/or the restoration of selected nucleases may facilitate improved growth rates and genomic stability, without substantially compromising recombination frequency. This may be a valuable avenue for future study.
In conclusion, this work helps elucidate how endogenous nucleases act on ssDNA and dsDNA cassettes used for recombineering. In addition to providing further mechanistic insight into Lambda Red recombination, this work also facilitated the creation of nuclease knockout strains with markedly improved recombination frequency and mutation inheritance. These strains will be highly useful as chassis in future recombineering efforts, and may enable new and powerful applications of Lambda Red technology.