Efficient and iterative retron-mediated in vivo recombineering in Escherichia coli

Abstract Recombineering is an important tool in gene editing, enabling fast, precise and highly specific in vivo modification of microbial genomes. Oligonucleotide-mediated recombineering via the in vivo production of single-stranded DNA can overcome the limitations of traditional recombineering methods that rely on the exogenous delivery of editing templates. By modifying a previously reported plasmid-based system for fully in vivo single-stranded DNA recombineering, we demonstrate iterative editing of independent loci by utilizing a temperature-sensitive origin of replication for easy curing of the editing plasmid from recombinant cells. Optimization of the promoters driving the expression of the system’s functional components, combined with targeted counterselection against unedited cells with Cas9 nuclease, enabled editing efficiencies of 90–100%. The addition of a dominant-negative mutL allele to the system allowed single-nucleotide edits that were otherwise unachievable due to mismatch repair. Finally, we tested alternative recombinases and found that efficiency significantly increased for some targets. Requiring only a single cloning step for retargeting, our system provides an easy-to-use method for rapid, efficient construction of desired mutants. Graphical Abstract


Introduction
Genome editing technologies have proven to be invaluable molecular tools, enabling rapid advances in functional genomics, metabolic engineering, therapeutics development, bioremediation and more (1)(2)(3)(4). Traditional methods for genome editing use allelic exchange via the RecA-dependent process of homologous recombination (HR). In HR, circular or linear DNA substrates, encoding the desired modification flanked by targeted regions of homology, are introduced into recombination-proficient (ΔrecBCΔsbcBC or ΔrecD) cells (5,6). Endogenous recombination proteins facilitate crossover between the editing template and the target locus, incorporating the desired mutations into the host chromosome. While this process is useful for targeted mutagenesis, it is sometimes inefficient and always laborious, requiring numerous cloning steps to incorporate sizable regions of homology (0.5-5 kb) and extensive screening to identify recombinant cells (7).
The development of in vivo recombination-mediated engineering (recombineering), which relies on phage-derived proteins to facilitate DNA integration, has greatly improved the speed and efficiency with which targeted mutations can be made. For example, the λ-phage-derived Red system in Escherichia coli yields more than a 100-fold increase in recombinant cells over traditional HR through a process independent of RecA (8,9). Recombineering reduces editing template design constraints by enabling efficient recombination with homology arms of <50 bp on double-stranded DNA (10). In addition, short single-stranded DNA (ssDNA) oligonucleotides (oligos) can be used as suitable recombineering substrates. In fact, ssDNA oligos have produced recombinant cells with just one of the three λ-Red proteins, the ssDNA-annealing protein Beta, further simplifying the requirements for successful recombineering (9). Furthermore, the Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-Cas9 bacterial immune system has been adapted as a programmable counterselection tool to selectively target and kill unedited cells after recombineering. This technique enhances the apparent efficiency by increasing the mutant to wild-type ratio of the population, allowing highly efficient scarless genome editing (11)(12)(13)(14). Additional efforts for the improvement of λ-Red editing efficiency have used engineered mutator strains, where the overexpression of HR-involved enzymes, the deletion of host exonucleases and/or disabling the mismatch repair system significantly improve recombination efficiency but can unintentionally lead to the accumulation of off-target mutations (15)(16)(17).
Farzadfard and Lu (18) described a novel approach for recombineering, termed Synthetic Cellular Recorders Integrating Biological Events (SCRIBE), where the ssDNA is generated in vivo and then incorporated into the chromosome. The in vivo ssDNA is produced by the Ec86 retron of E. coli BL21, composed of the msr and msd RNA elements and the reverse transcriptase (RT). Upon transcription of the msr-msd sequence, inverted repeats flanking the msr-msd RNA form a secondary structure specifically recognized by the RT and the msdRNA is reverse transcribed to ssDNA. Simultaneous expression of bet on the SCRIBE plasmid allows for efficient recombination of the newly synthesized ssDNA into the lagging strand of the targeted locus during DNA replication ( Figure 1A). By placing control of this system under an inducible promoter, the cells record their exposure to the inducer as the fraction of cells with the msd encoded mutation. By simply changing the msd sequence, SCRIBE can be easily retargeted to modify any desired genomic site. The authors report a maximum efficiency for SCRIBE of 10 -4 recombinants per generation (18). While sufficient for making mutations conferring easily selectable phenotypes, this rate of recombination is not robust enough to obviate the need for extensive screening when making nonselectable edits (19). Farzadfard et al. (20) improve upon their original SCRIBE design to achieve highly efficient ssDNA recombineering via transient knockdown of host exonuclease expression through CRISPR interference (CRISPRi), reporting nearly 100% recombination efficiency for one of their chosen targets. Additionally, the authors demonstrate that SCRIBE enables the incorporation of multiple mutations at distinct loci (18,20). However, with no easy method for plasmid curing, doing so requires that each editing plasmid contain a unique selection marker for plasmid maintenance, a strategy limited by the number of orthogonal selection markers and compatible plasmid origins available.
In this work, we sought to enable iterative genome editing by moving the functional components to easily curable plasmids while maintaining the SCRIBE system's high recombination efficiency. We hypothesized that combining an improved SCRIBE system with Cas9 counterselection against wild-type cells would provide an easy-to-use genome editing platform that precluded the need to electroporate ssDNA for recombineering. In addition, the system was designed so that the editing plasmid can be retargeted through a single round of PCR amplification and cloning to reduce construction times and enable the rapid generation of multilocus mutants.

Strains, plasmids and culture conditions
Bacterial strains used in this work are listed in Supplementary  Table S1. Plasmid construction methods and primers are described in Supplementary Table S2 (along with Addgene #s for available plasmids) and the Primers_GeneFrags.fa file, respectively. The pFF745 plasmid was a gift from Timothy Lu (Addgene #61450). Cloning steps for retargeting the editing plasmid are illustrated in Supplementary Figure S2. Briefly, primers with overhangs encoding the new target sequence were used to PCR amplify the vector using Q5 High-Fidelity 2x Master Mix (New England Biolabs, NEB), followed by DpnI digestion to remove template DNA, agarose gel electrophoresis to verify amplicon size and column-based DNA cleanup. DNA fragments were then assembled using the 2x NEBuilder HiFi DNA Assembly Master Mix (NEB) with the following modifications: 1 μl Master Mix + 0.5 μl each DNA fragment, incubated at 50 ∘ C for 15 min. Chemically competent E. coli NEB5α cells were used for cloning. Antibiotics were used at the following concentrations unless otherwise noted: spectinomycin (50 μg/ml), kanamycin (50 μg/ml), chloramphenicol (40 μg/ml), rifampicin (20 μg/ml), tetracycline (20 μg/ml) and carbenicillin (100 μg/ml). The tetA gene was PCR amplified from pUC18-mini-Tn7T-Gm-TetAR using primers 2490 and 2491 and integrated into E. coli MG1655 using λ-Red recombination with pKD46 (21), creating the MG1655 yhiS::tetAR strain.

Recombineering assays and recombination efficiency determination
Since the number of recombinant cells should increase with time, all experiments were performed in freshly transformed cells. Approximately 100 ng of editing plasmid was transformed into competent E. coli cells via either heat shock at 42 ∘ C or electroporation in a 0.1-cm gap cuvette with a single 1.8-kV pulse. Cells were recovered in 1 ml SOB at 30 ∘ C for 2 h and then outgrown overnight at 30 ∘ C in 5 ml Luria Bertani (LB) broth + appropriate antibiotic(s) to select transformants. The total number of viable cells was determined by spotting 10 μl 10-fold serially diluted overnight cultures on LB plates + appropriate antibiotic(s). For the rpoB and tetR assays, 10 μL of 10-fold serially diluted overnight cultures were spotted on LB plates with rifampicin or tetracycline to determine the number of recombinant cells. For the ackA assays, 1 ml of cells from overnight cultures were pelleted, washed and resuspended with sterile phosphate buffered saline after which washed cells were spotted on M9 chloroacetate plates (M9 minimal media + 10 mM sodium chloroacetate + 2% glycerol + 0.1% SOB) to determine the number of ackA mutant cells. Recombination efficiency was determined by dividing the number of recombinant cells by the total number of viable transformants. Efficiencies reported are the mean and standard error for three independent replicates. For the rpoB experiments, spontaneous rifampicin resistance was assessed for the wild-type strain as described in Section 2.6 below and the spontaneous mutation rate was subtracted from all samples plated on rifampicin. All statistical analyses were performed using the GraphPad Prism 5 program.

Promoter optimization
The P lacO promoter of the Ec86 retron cassette in pTLlSc-rpoB was replaced with P J23101 , and P VanCC was inserted upstream of bet as described in Supplementary Table S2. Expression of the Ec86 retron cassette and beta recombinase by P J23101 and P VanCC , respectively, was further optimized by PCR amplifying pTJVSc-rpoB with primers 2326 & 2358 and 2325 & 2359, followed by assembly of the two resulting amplicons using the 2x NEBuilder HiFi DNA Assembly Master Mix (NEB). The assembled pTJVSc-rpoB promoter library was transformed into E. coli NEB5α cells, and recombination efficiency was determined. In addition, the remainder of the pTJVSc-rpoB culture was diluted 1:100 into 5 ml LB broth + rifampicin and grown overnight again at 30 ∘ C. The following morning, the plasmid library was extracted, retransformed into E. coli NEB5α cells, and the process was repeated three more times. After the fourth round of selection, the pTJVSc-rpoB plasmid was extracted from four randomly chosen transformants and the P J23101 and P VanCC promoters were sequenced with primers 746 and 2252, respectively, to assess the convergence of the population on a single optimal promoter sequence, yielding pTJV1Sc-rpoB as indicated in Supplementary Table S2.

Cas9 counterselection
For pCas9CR4 for CRISPR/Cas9 directed counterselection against unedited wild-type cells, transformants with editing plasmid were outgrown overnight. The following morning cultures were diluted 1:10 into LB + appropriate antibiotics with or without anhydrotetracycline (aTc; 0.1 μg/ml) to induce Cas9 expression and then grown overnight at 30 ∘ C. Serial dilutions were spotted onto LB + rifampicin for the rpoB assays, LB + tetracycline for the tetA reversion assays, or M9 chloroacetate for the ackA assays, then incubated overnight at 37 ∘ C. The efficiency of Cas9 counterselection was determined by dividing the number of colonies from the aTc-induced cultures that grew with selection by the total number of viable cells plated on a nonselective plate. Results are reported as the mean with standard error for three independent replicates.

Efficiency improvement with negative mutator alleles
The mutL gene was cloned into pCas9CR4 as described in Supplementary Table S2. The cymR repressor and P cymRC promoter (22) were inserted upstream of mutL to allow for inducible expression by the addition of cumate, yielding pCas9CyMutL. Escherichia coli MG1655 cells harboring pCas9CyMutL were then transformed with pTJV1Sc-rpoB1 or pTJV1Sc-tetA and recovered in 1 ml SOB at 30 ∘ C for 2 h, after which cells were transferred to 5 ml LB + chloramphenicol + spectinomycin with or without cumate (100 μM) and outgrown overnight at 30 ∘ C. Cultures were diluted 1:10 into LB + chloramphenicol + spectinomycin + cumate with or without aTc (0.1 μg/ml) and grown overnight at 30 ∘ C. Recombination efficiency and efficiency of Cas9 counterselection were determined the following day, and the results are reported as the mean with standard error for three independent replicates.

Off-target mutation frequency analysis
Escherichia coli MG1655 was separately transformed with pTV1β-rpoB and pCas9CyMutL and plated on LB + spectinomycin and LB + chloramphenicol, respectively, for the selection of transformants. Single colonies from each transformation were used to inoculate 5 ml LB broth + appropriate antibiotic in triplicate and grown overnight at 30 ∘ C, with pCas9CyMutL grown with or without cumate (100 μM). Cultures were then diluted 1:10 into the same media and grown for an additional overnight at 30 ∘ C, after which 10 μL of 10-fold serial dilutions of each were spotted onto LB only and LB + rifampicin to assess the spontaneous mutation frequency.
For genome sequencing, a single colony from each transformation and wild-type MG1655 were grown in LB broth + appropriate antibiotic overnight at 30 ∘ C, with pCas9CyMutL grown with and without cumate (100 μM) for induction of mutL expression. Cultures were then diluted 1:100, grown overnight and repeated for a total of two setbacks. Genomic DNA was extracted after the first setback for wild-type and after the second setback for the other samples using a cetyltrimethyl ammonium bromide (CTAB)/phenol-chloroform extraction protocol (23).
Samples were prepared for whole-genome sequencing using the NEBNext Ultra II FS DNA Library Prep Kit for Illumina (NEB). Sequencing was performed using the NovaSeq6000 sequencing platform (Novogene Co. Sacramento, CA). Sequencing data were quality filtered and adapters were trimmed using the Trim Galore script (24). Mutations were identified using the Breseq mutational analysis pipeline (25) set to polymorphism mode with the default parameters and a minimum coverage cutoff of 20× reads. The E. coli MG1655 reference genome (National Center for Biotechnology Information (NCBI) accession: NC_000913) was used as the reference sequence. The wild-type MG1655 parent strain was used as a control to assess differences between the reference genome and our lab strain and sequence variations identified were subtracted from those found in our experimental samples.

Improvement of recombineering with reverse-transcribed ssDNA
To enable an efficient and iterative system for genome editing with ssDNA reverse-transcribed in vivo, we moved the SCRIBE system's functional components to the pKDsgRNA plasmid that has the temperature-sensitive variant of the pSC101 origin to allow for easy curing with growth at 37-42 ∘ C (21,26,27). This plasmid, pTLlSc, possesses the msr-msd coding sequence and Ec86 RT for the synthesis of ssDNA, the bet gene to facilitate incorporation of the ssDNA and the S. pyogenes single-guide RNA (sgRNA) to enable Cas9 targeting. The judicious design of this plasmid Figure 1. Optimization of the origin of replication and promoter elements enables an efficient and curable system for in vivo recombineering . (A) In vivo retron-generated ssDNA serves as an editing substrate for beta recombination at the lagging strand of the replication fork. (B) Rifampicin resistance conferred by the P564L mutation of the rpoB gene was used to measure the recombination efficiency. The pSC101 origin of replication allows for curing of the editing plasmid by growth at 42 ∘ C after desired mutation(s) are made. (C) Modifications made to promoters and origin of replication for enhanced recombination efficiency and curability. (D) Recombination efficiency of each editing plasmid reported as the average of three independent replicates. Error bars represent SEM, and statistical significance relative to pFF745-rpoB is denoted by asterisks (ns = not significant; **P < 0.01; ***P < 0.001; one-way analysis of variance and Tukey's test of log-transformed values). enables easy one-step cloning to retarget both the msd and sgRNA (Supplementary Figure S2). To assess the recombination efficiency using an easily selectable mutation, the msr-msd was retargeted to introduce three consecutive nucleotide substitutions and create the P564L point mutation of rpoB that confers rifampicin resistance ( Figure 1B). Here, we define recombination efficiency as the number of cells obtained on selective media divided by the total number of viable cells. First, we compared the efficiency of pFF745, described by Farzadfard et al., which possesses a high-copy pUC origin targeted to rpoB, with our temperature-sensitive plasmid pTLlSc-rpoB. As expected, recombination efficiency was about 100-fold lower with the temperature-sensitive plasmid, presumably because of decreased expression of the retron elements and bet compared to the high-copy pUC origin on pFF745 ( Figure 1D). We sought to improve this efficiency by increasing the transcription of the msr-msd, RT and bet through promoter replacements. The strong constitutive promoter P J23101 (28) replaced the P lacO promoter driving expression of the msr-msd RNA and RT. The promoter of bet was changed to P VanCC (22) because it is both strong and nonhomologous to P J23101 . Replacing these promoters individually did not increase recombination efficiency, but when combined, efficiency increased nearly 10-fold ( Figure 1D).
To further increase recombination efficiency, we used a selection strategy where a library of variant plasmids was created by cloning degeneracies into both the P J23101 and P VanCC promoters. We hypothesized that variants with increased efficiency would overtake the population after repeated rounds of outgrowth, selection, plasmid extraction and re-transformation (Supplementary Figure S1A). After four passages, the efficiency of the evolved pool was slightly higher than the original construct (Supplementary Figure S1B), and sequencing of the promoter from randomly chosen transformants revealed a convergence of the pool toward a single promoter sequence that had a recombination efficiency of nearly 10-fold greater than the parent plasmid (Supplementary Figure S1C-D). Moreover, the efficiency of the improved construct, pTJV1Sc-rpoB, was similar to the original SCRIBE plasmid, pFF745-rpoB ( Figure 1D).

CRISPR/Cas9 counterselection against wild-type cells
We next investigated whether our improved in vivo recombineering system could be combined with CRISPR/Cas9 counterselection against wild-type cells for a highly efficient genome editing system. The pCas9CR4 plasmid has cas9 under the control of an inducible P tet promoter that was engineered to enable co-maintenance with a genome targeting sgRNA until induction with anhydrotetracycline (aTc) (27). We first transformed pCas9CR4 into the E. coli strains NEB5α, BL21-DE3 and MG1655, and subsequently transformed the pTJV1Sc-rpoB editing vector, which possessed an sgRNA targeted to wild-type rpoB. After recovery, the cells were transferred to LB broth with spectinomycin and chloramphenicol to select for both plasmids and grown overnight. The following morning, the cultures were then passed into media with or without aTc and grown an additional overnight for induction of Cas9 expression and killing of unedited cells. Cells were then spotted onto plates with and without rifampicin to directly select for the rpoB mutation and assess total viable cells, respectively (Figure 2A). The recombination efficiency was with pTJV1Sc-rpoB and the RpoB P564L frequencies were determined by plating on LB + rifampicin. (C) Escherichia coli MG1655::tetAR was transformed with pTJV1Sc-rpoB, pTJV1Sc-tetA and pTJV1Sc-ackA. Frequencies of the RpoB P564L , TetA *70Y and AckA E54* mutations were determined by plating cells on LB + rifampicin, tetracycline and chloroacetate, respectively. (D) Sequential mutation of distinct loci was performed by transforming E. coli MG1655 with pTJV1Sc-rpoB, determining recombination efficiency by plating on LB + rifampicin, curing pTJV1Sc-rpoB by growing cells at 42 ∘ C, then transforming once more with pTJV1Sc-ackA. Double mutant frequency was assessed by plating on LB + rifampicin + sodium chloroacetate. Averages are based on three independent replicates. Error bars represent SEM. Statistical significance is denoted by asterisks (** = P-value < 0.01; *** = P-value <0.001; two-tailed Student's t-test of log-transformed values). between 1 × 10 -5 and 3 × 10 -3 without induction of Cas9 expression for all three strains. However, when cas9 was induced, the average efficiency was ∼5% for BL21 and 60-80% for NEB5α and MG1655, indicating a substantial reduction in wild-type cells ( Figure 2B).
To demonstrate that our system could mutagenize different targets, we examined the recombination efficiency of tetA and ackA. We inserted the tetA encoded tetracycline efflux pump that possessed a premature stop codon of about 200 bp from the start codon into MG1655 using traditional λ-Red recombineering (21). Our experiments then mutated the stop codon back to a sense codon, creating the TetA *70Y mutation to enable the fulllength translation of tetA and confer tetracycline resistance to the cells. We obtained tetracycline-resistant colonies with efficiencies of 5.4 × 10 -4 and 1.8 × 10 -5 with and without Cas9 counterselection, respectively, much lower than those observed for rpoB ( Figure 2C). Next, the acetate kinase gene, ackA, was mutated with three nucleotide point mutations to create a premature stop codon (AckA E54* ) so that the cells are unable to metabolize acetate and are thus resistant to the toxic acetate analog chloroacetate. A recombination efficiency of 5.6% without Cas9 counterselection and over 100% with induction of Cas9 was obtained, as slightly more colonies were observed on the selection plate than on the nonselective plate on average ( Figure 2C). These experiments show that targeted counterselection of unedited cells with Cas9 can successfully enrich for recombinants across distinct loci.
The pCas9CR4 plasmid used in these experiments was designed for tight repression of cas9 and an ssrA degradation tag decreases Cas9 stability, enabling co-maintenance of both genome targeting sgRNA template and cas9 on two plasmids. Although induction of cas9 is required for cell death by double-strand break (DSB), we questioned whether some transient level of Cas9 expression also enriches for recombinants, thus we performed experiments to compare the number of mutants obtained with the SCRIBE system in the presence and absence of pCas9CR4. More recombinant cells were produced for all three targets with cells harboring pCas9CR4 than cells without, even without induction of cas9 expression ( Figure 2C). In fact, we were unable to obtain tetracycline-resistant cells when using the pTJV1Sc-tetA plasmid alone. These results suggest that even low levels of cas9 expression in the uninduced state can increase recombination efficiency possibly by slowing the growth of wild-type cells that must repair DSB or because this low-level DSB stimulates recombination.
The proposed mechanism of ssDNA recombineering asserts that allelic replacement occurs at the replication fork, where the supplied ssDNA replaces an Okazaki fragment on the lagging strand. This mechanism results in a lagging strand bias, where recombination efficiencies for oligos targeting the lagging strand are higher than those targeting the leading strand. To assess whether in vivo produced ssDNA proceeds through the same mechanism, we targeted the leading strand of rpoB for editing (Supplementary Figure S3A). Surprisingly, this resulted in a similar efficiency as targeting the lagging strand (Supplementary Figure S3B). We wondered whether this was a function of the in vivo generation of ssDNA or if the same would be true of the rpoB target using traditional λ-Red recombineering. We constructed a control plasmid in which the msr-msd and RT coding sequences were removed (pTV1β-rpoB), resulting in constitutive expression of Beta only and sgRNA targeting rpoB. We then performed a traditional recombineering experiment by transforming exogenous ssDNA into cells with this plasmid and pCas9CR4 and observed that the lagging oligo was much more efficient, consistent with previous observations (7,9,29) and the proposed mechanism of ssDNA recombineering (Supplementary Figure S3C). To evaluate whether the same phenomenon is true of other targets using SCRIBE, we targeted the leading strand of tetA with the in vivo system (Supplementary Figure S3A). In this case, over 1000-fold fewer recombinant cells were obtained than when targeting the lagging strand (Supplementary Figure S3D). While we suspect that recombineering with in vivo ssDNA occurs at the lagging strand of the replication fork, our data suggest that the constant availability of ssDNA may be sufficient to overcome the lagging strand bias in some cases or that an alternative mechanism may influence recombination rates at some locations.

Iterative mutagenesis of two targets
Above all, the high efficiency of mutagenesis found at our target sites confirms a robust system for genome editing, precluding the need to screen large numbers of colonies to identify mutants. To demonstrate that our system could iteratively construct genome modifications, cells that were mutated at the rpoB locus were grown at 42 ∘ C to cure the cells of pTJV1Sc-rpoB. Subsequent transformation with pTJV1Sc-ackA enabled mutation of the ackA gene and produced cells resistant to both rifampicin and chloroacetate ( Figure 2D). Amplification and Sanger sequencing of the rpoB and ackA loci showed that both intended mutations were made, confirming that iterative use of the system could edit multiple loci independently (Supplementary Figure S4).

Co-Expression of dominant-negative mutL
Recombineering with ssDNA results in transient production of heteroduplex DNA that the methyl-directed mismatch repair (MMR) system can repair, thus decreasing the observed rate of mutagenesis. Several strategies are known to decrease the rate of MMR and enable more efficient mutagenesis (15,17,(30)(31)(32). For example, single bp mismatches are repaired with different efficiencies, while C:C mismatches evade repair entirely and prevent repair of mismatches within 3 bp (17,31). The introduction of several adjacent mismatches also prevents repair by MMR (15). However, these strategies present design constraints that cannot always be followed. An alternate strategy is the recombinant expression of negative mutS or mutL variants that are dominant over wild-type enzymes and prevent efficient repair (32). Accordingly, we cloned mutL with the dominant negative E32K mutation into the pCas9CR4 plasmid under the control of a cumate inducible promoter (22) to maintain a two-plasmid system with independent control of cas9 and mutL ( Figure 3A). To test this system, we introduced a single bp mismatch of A:C in rpoB with and without induction of the negative mutL. In cells harboring pCas9CyMutL with mutL in the off-state, the average frequency of rifampicin-resistant mutants was 4.9 × 10 -6 without Cas9induction and 4.5 × 10 -4 with Cas9 counterselection ( Figure 3B). In contrast, the induction of mutL enabled rpoB mutations with an average recombination efficiency of 0.2% and 18% without and with Cas9 counterselection ( Figure 3B). We also re-examined the tetA reversion assays, which introduced a single bp T:T mismatch that is less efficiently repaired than the A:C mismatch used for rpoB. Induction of the mutator allele increased the recombination efficiency by about 100-fold both with and without Cas9 counterselection, achieving a maximum efficiency of 0.3% on average ( Figure 3C). When generating a single A:G mismatch in the ackA gene, mutL induction improved recombination by 1000-and 100-fold with and without Cas9 counterselection, respectively, resulting in an average maximum efficiency of 56% ( Figure 3D). These results further highlight the versatility of our system in generating precise mutations across a diverse range of loci.
To further illustrate the efficiency enhancement these tools offer over previous methods, we performed the same experiments using the no-SCAR method described previously (12). Singlestranded editing oligos targeting the lagging strand were used to introduce the same single bp mismatches as before in cells harboring pCas9CyMutL and pKDsgRNA, a plasmid encoding the λ-Red proteins Exo, Beta and Gam under the control of the arabinoseinducible P BAD promoter, as well as an sgRNA targeting the wild type gene sequences. When targeting rpoB, induction of mutL before transformation, during recovery, or both achieved similarly high recombination rates as the retron system, yielding average efficiencies >2% with direct selection on rifampicin and 100% with Cas9 counterselection, a 20-fold increase in recombinants than without mutL (Supplementary Figure S5A). Regardless of whether mutL was induced, Cas9 counterselection effectively removed unedited cells with 95% of cells plated on aTc being rifampicin-resistant even without mutL expression. On the contrary, when targeting ackA, mutL induction improved the recombination efficiency less than 10-fold, yielding a maximum efficiency of 4.8 × 10 -4 when plated on chloroacetate (Supplementary Figure S5B). However, sequencing the ackA gene of several putative mutant colonies on the chloroacetate selection media revealed none had acquired the intended mutation. Furthermore, Cas9 counterselection did not substantially enrich for mutants, with <5% of the colonies on the aTc-induced plate able to grow with chloroacetate when patched (Supplementary Figure S5B), indicating most of the colonies on the counterselection plate escaped Cas9 killing. These results are consistent with previous studies examining the rate of Cas9 escape in bacteria, which report frequencies ranging from ∼10 −3 to 10 −4 in multiple species (11,27,33,34). Overall, the combined mutL and retron-based systems result in consistently high mutagenesis rates for single bp edits and can outperform traditional oligo recombineering methods for certain target loci. The MutL E32K gene was cloned into the pCas9CR4 plasmid under the control of the P CymRC promoter for inducible expression. When cumate is added, MutL E32K is expressed and competes with the native MutL for binding to MutS during mismatch repair. The E32K mutation prevents MutH binding, inhibiting removal and repair of the mismatched base. Cells harboring pCas9CyMutL were transformed with (B) pTJV1Sc-rpoB1, (C) pTJV1Sc-tetA or (D) pTJV1Sc-ackA1 encoding single point mutations for generating the RpoB P564L , TetA *70Y and AckA E54* mutations, respectively, and grown in liquid culture with or without cumate. The unmodified pCas9CR4 in both wild-type E. coli and a mutS mutant were used as controls. Recombination efficiency with and without Cas9 counterselection was assessed by plating on LB + rifampicin, LB + tetracycline or M9 + chloroacetate. Data reported are the average of three independent replicates. Error bars represent SEM and statistical significance relative to pCas9CR4 is denoted by asterisks (ns = not significant; * = P-value < 0.05; ** = P-value < 0.01; *** = P-value <0.001; one-way analysis of variance and Tukey's test of log-transformed values).
Hindrance of MMR increases the frequency of background genome mutations (35,36); however, the inducibility of our system minimizes the amount of time in which MMR is inhibited. Similarly, uncontrolled expression of the λ-Red genes has also been shown to increase the rate of spontaneous mutations (37). As such, constitutive expression of bet in our system could conceivably produce unintended recombination events in the cell. To assess the background rate of mutagenesis in cells with our SCRIBE system, we measured spontaneous rifampicin resistance (38,39). Cells were grown overnight, diluted 1:10, grown overnight and then spotted on rifampicin to assess the mutation frequency. The wild-type cells spontaneously acquired rifampicin resistance at a frequency of 6 × 10 -7 on average. Cells constitutively expressing bet and cells with the uninduced pCas9CyMutL did not differ significantly from the wild-type (Supplementary Figure S6A).
However, induction of the mutator mutL resulted in an average mutation frequency of about 1.3 × 10 -5 , an ∼20-fold increase over the wild-type (Supplementary Figure S6A).
While these results indicate that the unexpressed mutL and constitutive bet did not increase the rate of point mutations, we wanted to assess further whether other genomic mutations and rearrangements that would not manifest as rifampicin resistance were present. Accordingly, we performed whole-genome sequencing on the strains grown for two overnights to identify polymorphisms within the entire population. A complete list of the mutations identified is in Supplemental File 2. Our analysis revealed 41 total mutations in the population constitutively expressing bet, constituting a mutation rate of 5.9 × 10 -11 mutations per cell per generation (Supplementary Figure S6B), with one new IS5 mobile element insertion into the rclA gene. The uninduced pCas9CyMutL cells displayed a slightly higher mutation rate of 2.4 × 10 -10 ; however, both rates were similar to the wild-type control, which exhibited a mutation rate of 1.0 × 10 -10 . In contrast, 478 mutations were found in the population with mutL induced for a mutation rate of 1.59 × 10 -9 mutations per cell per generation (Supplementary Figure S6B), as well as a deletion of the ∼1200 bp insH21 mobile element. Even still, this low frequency of background mutations indicates selecting a mutant with unintended mutations would be a rare occurrence. Overall, these results demonstrate that maintenance of plasmids with the mutL mutator allele or constitutive expression of bet do not increase the frequency of background mutations to a level that would cause concern from a genetic engineering standpoint.

Alternative recombinases enhance efficiency for some targets
Recently, Wannier et al. identified recT from a Collinsella stercoris phage (CspRecT) with improved recombination efficiency compared to Beta for oligo mediated recombineering in E. coli (40). To examine the effect of alternative recombinases in our system, we replaced bet with CspRecT, as well as EcRecT from the E. coli BL21-DE3 chromosome ( Figure 4A). We then compared the efficiency of these plasmids across three different targets. Surprisingly, our results showed inconsistency for which DNA-binding protein promoted the highest number of recombinants between the different targets. For the rpoB mutation, EcRecT was more efficient than both Beta and CspRecT, with 92% efficiency when using Cas9 counterselection ( Figure 4B). On the contrary, the ackA mutation was more efficient with Beta, where 80% of cells were mutants compared to 4.5% for CspRecT and no ackA mutants with EcRecT ( Figure 4C). Given the improvement in efficiency seen for tetA when expressing MutL E32K , we used pCas9CyMutL in combination with the alternative recombinases. We obtained over 100% efficiency for the tetA reversion with CspRecT, both with and without Cas9 counterselection. Although slightly more colonies were observed on the tetracycline plate than on the nonselective plate for two of the replicates (Figure 4D), the CspRecT clearly outperformed Beta and EcRecT where relatively low levels of mutants were found. Thus, in circumstances where a mutation is difficult to obtain, it is advisable to try these alternative recombinases, while the mechanism causing these discrepancies warrants further investigation.

Discussion
In moving the SCRIBE machinery onto the temperature-sensitive pSC101 origin of replication and combining it with Cas9 counterselection, we successfully constructed an in vivo system for ssDNA recombineering that is highly efficient and iterative.
Targeted counterselection using CRISPR/Cas9 enabled efficiencies near 100% after a single night of outgrowth post-transformation, reducing the number of colonies needing to be screened for mutant identification. Moreover, both the ssDNA template and sgRNA sequence can be retargeted for modification of any genomic locus with an appropriate PAM in a single cloning step using sequence overlap cloning methods. Numerous loci can be mutated without the need for orthogonal selection markers or extensive screening procedures, increasing the speed and ease of generating desirable mutants. In addition, the pCas9CR4 and pCas9CyMutL plasmids can be cured using pKDsgRNA-p15 (Addgene #62656) as previously described (12), yielding plasmid-free mutants for downstream applications.
Recombineering is an important tool in the biological sciences, and continued enhancement of recombination efficiency across diverse target loci will decrease strain construction times. Recently, Lopez et al. showed that increasing the abundance of ssDNA produced by retrons significantly improves recombination efficiencies (41), consistent with the enhanced efficiencies we observed when increasing the msr-msd/RT expression and bet through promoter modifications. A complementary strategy is to reduce the rate of ssDNA degradation by host nucleases, as the High-efficiency SCRIBE (HiSCRIBE) system by Farzadfard et al. (20) describes. This system improves upon their original SCRIBE design by coupling a strong ribosome-binding site for bet on the SCRIBE plasmid with CRISPRi-enabled transcriptional interference of the endogenous exonucleases. Recombination efficiencies of nearly 100% for galK and ∼25% for a kanamycin resistance gene target were obtained with these improvements. However, employing the nuclease-deficient dCas9 for CRISPRi prevents the use of the nuclease-active Cas9 for counterselection. Although the authors show that CRISPR/Cas9 counterselection can select for recombinant cells at high efficiencies in an exonuclease knockout strain, this strategy prevents easy system portability since a mutant strain must be used.
Lim et al. (42) also describe a CRISPR/retron-based editing system enabling trackable editing and functional screening across multiple loci in a population. However, this system requires the sequential transformation and maintenance of three plasmids. Moreover, the construction of their retron/sgRNA plasmid requires multiple enzymatic digestions for cloning, contrary to our twoplasmid system, which simultaneously retargets the retron and sgRNA in a single two-fragment cloning step. Most importantly, both the HiSCRIBE and the CRISPR/retron system of Lim et al. (20,42) would require multiple orthogonal selection markers to mutate multiple loci within a single cell, as no easy method of plasmid curing is included in either design, preventing iterative mutagenesis. Although the recombination efficiencies achieved with our system in the absence of counterselection are less than those reported for HiSCRIBE, Cas9-induced cell death of unedited cells in the population enabled apparent efficiencies on par with HiSCRIBE. The ease with which our system can be retargeted and used iteratively to incorporate multiple edits combined with its high efficiency in mutant selection provides a quick and straightforward method for constructing mutants relevant to both research and biotechnological applications. Methods enabling multiplexed editing of numerous genomic loci within populations and continuous evolution of targeted loci, using retron-mediated recombineering, have been recently described (20,(42)(43)(44) and rely on efficient editing for generating a library of variants. Our system would be well suited for constructing such libraries, with the added advantage of being curable, thus enabling editing of distant loci within the same cell through repeated rounds of library recombineering.

Supplementary data
Supplementary data are available at SYNBIO online.

Data availability
Raw sequence reads for the off-target mutation frequency experiments are available under the BioProject ID PRJNA813865 in the NCBI databases.

Funding
This work was supported by the Department of Microbiology and Cell Science, Institute of Food and Agricultural Sciences, University of Florida.