Large scale validation of an efficient CRISPR/Cas-based multi gene editing protocol in Escherichia coli

Background The exploitation of the CRISPR/Cas9 machinery coupled to lambda (λ) recombinase-mediated homologous recombination (recombineering) is becoming the method of choice for genome editing in E. coli. First proposed by Jiang and co-workers, the strategy has been subsequently fine-tuned by several authors who demonstrated, by using few selected loci, that the efficiency of mutagenesis (number of mutant colonies over total number of colonies analyzed) can be extremely high (up to 100%). However, from published data it is difficult to appreciate the robustness of the technology, defined as the number of successfully mutated loci over the total number of targeted loci. This information is particularly relevant in high-throughput genome editing, where repetition of experiments to rescue missing mutants would be impractical. This work describes a “brute force” validation activity, which culminated in the definition of a robust, simple and rapid protocol for single or multiple gene deletions. Results We first set up our own version of the CRISPR/Cas9 protocol and then we evaluated the mutagenesis efficiency by changing different parameters including sequence of guide RNAs, length and concentration of donor DNAs, and use of single stranded and double stranded donor DNAs. We then validated the optimized conditions targeting 78 “dispensable” genes. This work led to the definition of a protocol, featuring the use of double stranded synthetic donor DNAs, which guarantees mutagenesis efficiencies consistently higher than 10% and a robustness of 100%. The procedure can be applied also for simultaneous gene deletions. Conclusions This work defines for the first time the robustness of a CRISPR/Cas9-based protocol based on a large sample size. Since the technical solutions here proposed can be applied to other similar procedures, the data could be of general interest for the scientific community working on bacterial genome editing and, in particular, for those involved in synthetic biology projects requiring high throughput procedures. Electronic supplementary material The online version of this article (doi:10.1186/s12934-017-0681-1) contains supplementary material, which is available to authorized users.


Background
Escherichia coli is one of the most extensively studied living organisms on earth and as such has become an instrumental model system for the understanding of a plethora of gene functions and regulations in both prokaryotes and eukaryotes. Moreover, E. coli also plays an invaluable role in modern biological engineering and industrial microbiology: it is a very versatile host for the production of heterologous proteins and their mass-production in industrial fermentation systems. Genetically modified E. coli cells are currently used in a wide range of processes, such as vaccine development, bioremediation, production of biofuels and production of immobilized enzymes. Additional biotechnological applications are expected to be developed thanks to the exploitation of modern techniques of metabolic engineering and synthetic biology.
To fully exploit the use of synthetic biology in E. coli, the availability of efficient genome editing systems, also applicable in high-throughput modalities is necessary. Currently, there are three main approaches for manipulation of chromosomal DNA in E. coli, all utilizing phage recombinase-mediated homologous recombination (recombineering), using either the Rac prophage system [1,2] or the three bacteriophage λ Red proteins Exo, Beta, and Gam [3][4][5]. Classically, gene knockout mutants are created by inserting antibiotic resistance markers (or other selection markers) between double-stranded DNA (ds-DNA) PCR products derived from the upstream and downstream regions of the target gene. Mutant colonies are isolated in the appropriate selective medium after transformation with linear or circular constructs and, when necessary, the selection marker is subsequently eliminated by counter-selection, leaving a "scarless" chromosomal mutation. Court and co-workers elegantly demonstrated that chromosomal gene mutations can be achieved without the need of selection markers and using synthetic single stranded DNAs (ss-DNAs) or ds-DNAs, which anneal to their complementary chromosomal regions during replication and mediate recombination and gene modification [6,7]. While effective, these two approaches are not ideal for high-throughput applications since they are laborious and time consuming (in the case of the first approach) and feature mutagenesis efficiencies often below 1% (in the case of the second approach) [8][9][10]. More recently, a third approach, proposed for the first time by Jiang and co-workers [9], makes use of the CRISPR/Cas9 technology [11][12][13]. Briefly, the strain to be modified is first genetically manipulated to express the Cas9 nuclease and the λ Red machinery, and subsequently the strain is co-transformed with (i) a plasmid (pCRISPR) encoding the guide RNA, which anneals with the chromosomal region to be modified and promotes a site-specific DNA cleavage by the Cas9, and (ii) a donor DNA (PCR-derived or chemically synthesized) partially homologous to the cleaved extremities, which promotes the repair of the double stranded break through λ Red-mediated recombination thereby introducing the desired mutation. The presence of the λ Red machinery plays an important role in the process since in its absence the mutagenesis efficiency was shown to drop quite substantially [10]. With this strategy, Jiang and co-workers reported mutation efficiencies as high as 65%. Subsequently, other authors fine-tuned the Jiang et al. protocol by adding innovative solutions and expanding its application for extensive gene deletions and replacements [14][15][16][17]. Thanks to the contribution of all these authors, CRISPR/Cas9 coupled to recombineering is becoming the most effective approach for genome editing in bacteria and in particular in E. coli.
However, one aspect of the CRISPR/Cas9 technology that still remains to be thoroughly addressed is the definition of its "robustness". In fact, while it has been extensively demonstrated that the efficiency of mutagenesis (number of mutant colonies/total number of colonies analyzed) can be extremely high (close to 100%), the robustness (number of mutated loci/total number targeted loci) has not yet been defined experimentally in a rigorous manner. Knowing the robustness of the specific CRISPR/Cas-based protocol in use is particularly relevant in high-throughput applications, where the repetition of mutagenesis experiments and/or the analysis of large numbers of colonies would be impractical. There is a paucity of papers addressing the "robustness issue" and even in these works the number of targeted loci is limited. For instance, to evaluate the robustness of their protocol, Reisch and Prather [18] using an innovative two plasmid system, one expressing a tightly regulated Cas9 and the other the gRNA, demonstrated that point mutations could be introduced into two dispensable genes with efficiencies close to 100%. On the basis of these data the authors concluded that their system is robust enough to make point mutations (and also larger deletions) in any genomic location carrying an appropriate PAM site. To support their conclusion, the authors also claimed that using the same protocol they successfully inactivated five additional genes. Another recent study describes a CRISPR-based strategy that allowed the integration of entire metabolic pathways in seven distinct loci using ds-DNAs encoding homologous regions to the insertion site of different size with 75-100% efficiencies [19].
In the present paper, we present a "brute force" validation effort, which culminated in the definition of a highly robust, simple and rapid protocol for both single and multiple scarless chromosome manipulations. First, we tested several experimental conditions targeting four dispensable gene loci with the aim of investigating the parameters that mostly influence mutagenesis efficiency. On the basis of this analysis we defined a mutagenesis protocol, which we subsequently validated on a panel of 78 additional genes. The overall approach, which involved the construction of approximately one hundred pCRISPR plasmids, the execution of a few hundred transformation experiments and the analysis of several thousand colony PCR and sequencing, led to the definition of a high fidelity and rapid protocol, which guarantees the generation of single and multiple mutations of dispensable genes with a 100% confidence.

Results
Components of our CRISPR/Cas9 genome editing protocol Figure 1 schematically represents our CRISPR/Cas9based E. coli genome editing protocol. Briefly, the procedure makes use of three main elements: the pCasRed plasmid, the pCRISPR-SacB-gDNA plasmid, and the synthetic, mutation-inducing oligonucleotide [donor DNA (dDNA)]. The pCasRed plasmid carries the chloramphenicol resistance gene (Cm R ) and encodes the Cas9 nuclease, the λ Red (Exo, Beta, Gam) cassette and the tracrRNA. The cas9 gene and the tracrRNA coding sequence are under the control of constitutive promoters while the λ Red gene cassette is transcribed from the pBAD arabinose-inducible promoter [20]. The pCRISPR-SacB-gDNA plasmid derives from a pCRISPR plasmid [9], where the kanamycin resistance gene (Km R ) is fused to the sacB gene encoding the Bacillus subtilis levansucrase. SacB is toxic in E. coli if grown in media containing 5% sucrose [21][22][23][24] and, as previously demonstrated by Hale and co-workers, plasmid-borne sucrose toxicity can be exploited to cure high copy number plasmids [25]. We used this strategy to remove the pCRISPR-SacB-gDNA Fig. 1 Overview of CRISPR/Cas9 genome editing strategy in Escherichia coli. The strain to be mutagenized [E. coli BL21(DE3)∆ompA] is first transformed with the pCasRed plasmid expressing the λ Red (Exo, Beta, Gam) machinery, the Cas9 endonuclease, and tracrRNA. Subsequently, the strain is co-transformed with pCRISPR-SacB-gDNA, and a synthetic, mutation-inducing oligonucleotide [donor DNA (dDNA)]. The pCRISPR-SacB-gDNA plasmid encodes the gRNA that specifies the site of cleavage and the endonuclease Cas9 recognizes the gRNA together with the tracrRNA, which anneals to gRNA forming a three-component complex. After the base pairing of gRNA to the target site, the Cas9 mediates the chromosomal DNA double strand break (upper panel). The double strand break is repaired by λ Red-mediated homologous recombination taking place between the extremities of the cleaved chromosomal DNA and the donor DNA (lower panel). For the sequence of constitutive promoters P1 and P2 see ADDGENE #4287 [9]; for sequence of constitutive promoter P3 see ADDGENE #42875 [9] and for P4 constitutive promoter sequence see ADDGENE #13036 [24]. For the arabinose-inducible promoter pBAD see pKOBEG plasmid [20] plasmid after the gene mutation has been introduced. In addition to the sacB gene, pCRISPR-SacB-gDNA plasmid carries the synthetic DNA fragment (gDNA), transcribed from a constitutive promoter [9], encoding the RNA guide necessary to drive the Cas9-dependent double stranded break to the desired site within the bacterial genome.
In the following sections we provide details of the experimental data that led us to define a protocol, which, under the conditions described here, guarantees the inactivation of any dispensable gene and, after construction of the pCRISPR plasmids and dDNAs (see "Methods"), allows the sequential mutation of genes at a pace of one mutation every other day. Finally, we also describe a highly efficient protocol for the simultaneous introduction of multiple mutations.

Definition of the robustness of gDNA selection method
gDNAs within a target gene are usually selected among those 30 nucleotide sequences followed by an NGG (PAM) trinucleotide, which do not share homologies with other regions in the chromosome (to avoid offtarget cleavage). Routine bioinformatics tools (BLAST) are generally used to identify such sequences. However, what has not been fully investigated is the robustness of bioinformatics in gDNA selection. In other words, do all predicted gDNAs promote an efficient CRISPR/Cas9mediated cleavage at the selected target site? To test this, we selected several gDNAs and we analyzed their capacity to promote Cas9 cleavage. Theoretically, if a gDNA efficiently drives Cas9 cleavage no colonies should be isolated since non-homologous end-joining repair (NHEJ) works poorly in E. coli [26,27]. Practically, colonies are still recoverable and represent "escapers" in which the gDNA from pCRISPR-SacB-gDNA is lost, or mutations in the PAM region and/or in the seed region (8-12 nucleotides upstream from the PAM region) [9] have occurred. Alternatively, as suggested by Cui and Bikard [28], if Cas9-mediated cleavage is not efficient, homologous recombination between the cleaved chromosome and the un-cleaved sister chromosome could take place. Finally, gDNA promoter mutations and mutations to the cas9 gene itself could occur, leading to a basal level of colonies lacking of dsDNA breaks. However, whatever mechanism is involved, the number of "escaper" colonies should be orders of magnitude lower compared to the colonies obtained by transforming the strain with the "empty" pCRISPR-SacB plasmid. Therefore, the drop in transformation efficiency between pCRISPR-SacB and pCRISPR-SacB-gDNA indicates the quality of the gDNA.
To establish the effectiveness of gDNAs in our protocol, we created sets of pCRISPR-gDNA plasmids (Additional file 1: Table S1) by cloning synthetic gDNAs (Additional file 1: Table S2) between the two tracrRNAcomplementary repeat regions ( Fig. 1) and we compared the transformation efficiencies of the recombinant plasmids with the efficiency of the "empty" pCRISPR vector (Fig. 2). In particular, we selected three target genes, ompF, lpp and fecA, and we designed four sets of different gDNAs, two of them targeting different sites at the 5′ and 3′ regions of ompF gene, respectively, and the other two sets targeting lpp and fecA at different positions (Fig. 2). Overall, 20 pCRISPR-gDNA plasmids were generated and used to transform E. coli BL21(DE3)ΔompA(pCasRed). The BL21(DE3)ΔompA was used since we observed that the strain maintained good transformation efficiencies even in the presence of pCasRed plasmid. Colonies were selected on LB plates supplemented with Cm and Km. While the "empty" pCRISPR vector routinely gave a transformation efficiency of 0.5-2 × 10 6 CFUs/μg of plasmid DNA, most of pCRISPR-gDNA plasmids had transformation efficiencies in the range of 0.5-2 × 10 3 CFUs/μg of plasmid DNA (Fig. 2). However, two out of the 20 pCRISPR-gDNA plasmids (pCRISPR-ompF_5'C and pCRISPR-lpp_C) gave a transformation efficiency close to the one observed with the empty vector.
We randomly analyzed ten clones from the transformations with pCRISPR-ompF_5′G and pCRISPR-ompF_5′C (transformation efficiency of 1 × 10 3 and 1 × 10 5 CFUs/μg, respectively) using PCR and sequence analyses of both the chromosomal regions targeted by the corresponding gDNAs and the gDNA regions of pCRISPR-ompF_5′G and pCRISPR-ompF_5′C. While the chromosome target sites had wild type sequences in both sets of transformant colonies, the colonies transformed with pCRISPR-ompF_5′G carried rearranged plasmids missing the gDNA ompF_5′G region, probably due to homologous recombination occurring between the two repeats flanking the gDNA. By contrast, all the colonies analyzed from the transformation with pCRISPR-ompF_5′C carried a wild type gDNA sequence (Additional file 2: Figure S1) [9].
From these experiments we conclude that in silico selection allows the identification of gDNAs, which in most cases drive the Cas9-mediated cleavage with high efficiency. However, approximately 10% of selected gDNAs did not cause bacterial killing. "Good and bad" gDNAs can easily be discriminated by colony counting after transformation with the pCRISPR plasmids carrying the selected gDNAs. Therefore, before modifying any target gene we routinely run the colony testing and if transformation efficiencies higher than 10 3 CFUs/μg of plasmid DNA is obtained we change the gDNA sequence of the plasmid.

Definition of the optimal conditions to be used in gene knockout
Next, we investigated the efficiency of gene inactivation using synthetic oligonucleotides (donor DNA, dDNA) under different experimental conditions. We analyzed three parameters: length and concentration of the mutagenic dDNA and extent of gene deletion. These parameters were first tested targeting the 5′ end of the ompF gene using the pCRISPR-ompF_5′G plasmid (Fig. 2) and single stranded donor DNAs annealing to the lagging strand (Lg-ss-dDNAs) (Additional file 1: Table S3). In fact, similar to the homologous recombination using synthetic oligonucleotides [5], Lg-ss-dDNAs have been reported to be more efficient in promoting double strand break repair than those targeting the leading strand (Ld-ss-dDNAs) [10,18]. Donor DNAs were designed with homology arms of equal length upstream and downstream from the deletion. Subsequently, the best conditions were validated targeting three additional gene loci, the 3′ end of the ompF gene, and the fecA and lpp genes, using either single stranded donor DNA (Ldss-dDNA) or double stranded donor DNA (ds-dDNA). All parameters were tested in at least three independent experiments and mutation efficiencies were determined by analyzing a total of at least 20 colonies. Table 1 summarizes the data obtained from all experiments. As far as the use of ss-dDNA is concerned, for short gene deletions (around 30 bp), 10 μg of both 70 and  Table S1. The tables report the transformation efficiencies (CFU/μg) of each pCRISPR-gDNA in BL21(DE3)∆ompA(pCasRed) 120 bp oligonucleotides were very effective, resulting in efficiencies of mutagenesis >50% for all four genes. Deletions longer than 30 bp and up to approximately 500 bp, could be obtained with relatively good efficiencies (>10%) as long as 10 μg of 120 bp ss-dDNAs were used. However, efficiencies dropped quite substantially (<5%) for deletions longer than 500 bp. No major appreciable difference in deletion efficiencies at all four gene loci was observed when the Ld-ss-dDNA was used in place of the Lg-ss-dDNA. As far as ds-dDNA is concerned, the use of 120 bp ds oligonucleotides was not only as efficient as ss-dDNAs for short deletions but also allowed the generation of the long full gene deletions that failed using ss-dDNAs.
From the experimental data obtained on four gene loci, we conclude that, in order to create relatively short deletions, ss-dDNAs of 70-120 nucleotides targeting either the lagging or the leading strands perform well (100% robustness and >50% efficiency). However, 120 bp-ds-dDNAs showed high robustness and efficiency when extended deletions (of up to 2.200 bp) were required.

Validation of 30 bp gene deletions for mutant library production
To further validate the conclusions of the experiments described above, we attempted the mutation of 78 additional genes that are classified as "dispensable" according to the Keio library [29] (see Additional file 1: Table S4 for the list of genes). First, we constructed the 78 pCRISPR-SacB-gDNAs plasmids (Additional file 1: Table S1) encoding the gRNAs targeting the genes next to their 5′ ends to avoid the expression of truncated but nonetheless functional proteins (Additional file 1: Table S2). Subsequently, we verified the effectiveness of the cloned gDNAs in guiding Cas9 cleavage by transformation and colony counting (see above). In line with the statistics previously obtained, approximately 10% of selected gDNAs (7 out of 78, highlighted with an asterisk in Additional file 1: Table S2) did not promote Cas9 cleavage, and therefore other functional gDNAs were selected (Additional file 1: Table S2).
After selecting the 78 pCRISPR-SacB-gDNAs plasmids with transformation efficiencies three logs lower than pCRISPR-SacB, we next designed 78 70 bp-ss-dDNAs to delete 30 bp and introduce a premature stop codon in each gene, downstream the deleted region (Additional file 1: Table S3). Of these 78 dDNAs, 46 targeted the leading strand and 32 the lagging strand. Each pCRISPR-SacB-gDNA plasmid, with its corresponding dDNA, was used to transform E. coli BL21(DE3)ΔompA(pCasRed) strain, giving transformation efficiencies usually between 2 and 5 × 10 2 CFUs/μg of plasmid DNA. From each transformation ten colonies were randomly selected and analyzed by PCR to identify those carrying the 30 bp deletion and to establish the "success rate" of mutagenesis. We arbitrarily applied the "ten colony rule" (i.e., mutagenesis efficiencies higher that 10%) because we considered such number still compatible with manual high-throughput procedures. Figure 3a summarizes the results of our screening. Fortyeight of the selected 78 genes were successfully mutated, with efficiencies ≥10% (at least one mutant/10 colonies). More specifically, 20 of the 46 Ld-ss-dDNAs created the desired mutations, while Lg-ss-dDNAs mutated 28 out of 32 genes. The inability to obtain 30 mutations, at least with a sufficiently high frequency to fulfill the "ten colony rule", was partially unexpected, since the missing mutations had been previously obtained in E. coli K-12 BW25113 [29]. Therefore, either such mutations are not compatible with the genetic background of our recipient strain [our strain is an E. coli BL21(DE3) derivative carrying the deletion of the ompA gene and pCasRed plasmid] or the experimental conditions previously defined were not sufficiently robust.
To discriminate between the two possibilities, we first asked the question whether the use of dDNAs targeting the opposite strands could rescue the missing mutants. Figure 3b reports the results of this experiment. The use of Lg-ss-dDNA rescued 14 of the 26 mutants that failed to be isolated with the Ld-ss-dDNA. In contrast, when the mutation of the five genes that were not mutated with the Lg-ss-dDNA was attempted only one mutant was rescued using Ld-ss-dDNA. Overall, the data indicate that 42 out of a total of 58 mutagenesis attempts with the Lg-ss-dDNA were successful, corresponding to a success rate of 72%. By contrast, the use of the Ld-ss-dDNA was successful in 22 of the 51 mutagenesis experiments (success rate 43%).
These data demonstrate that, differently from our preliminary validation experiment but in line with previous studies [10,18], Lg-ss-dDNA performed better than Ldss-dDNA both in terms of success rate (72% rate versus 43%) and efficiency of mutagenesis (average bar heights with Ld-ss-dDNA: 10%; average bar heights with Lg-ss-dDNA: 31%). However, a non-negligible fraction of genes [15 out of 78 (19%)] failed to be mutagenized, regardless the type of the dDNA used.
Considering our experimental data that showed how ds-dDNAs successfully guided long deletions not obtained with ss-dDNAs, we thus re-attempted the deletion of all 78 genes using the same pCRISPR-SacB-gDNAs, but making the 70 base dDNAs double stranded. As shown in Fig. 3c, all genes that were successfully mutated using either the Lg-ss-dDNAs or the Ld-ss-dDNAs (63 genes in total) were also inactivated with the ds-dDNA. In addition, seven genes that were not deleted with the ss-dDNAs were mutagenized, bringing the success rate to 90% (70/78). Importantly, when ds-dDNAs were used the mutagenesis efficiency was also improved compared to the efficiencies obtained with ss-dDNAs (average bar heights with ds-dDNA: 48.9%).
As pointed out above, the inability to obtain some mutations classified as "dispensable" according to the Keio library [29] could be due to the fact that such mutations are not compatible with the genetic background of our recipient strain [our strain is an E. coli BL21(DE3) derivative carrying the deletion of the ompA gene and pCasRed plasmid]. To investigate the possible role of the ΔompA deletion in preventing the inactivation of the eight genes, we tested whether they could be achieved in BL21(DE3) wild type strain. To this aim, pCasRed plasmid was introduced into BL21(DE3) and BL21(DE3) (pCasRed) was subsequently subjected to the gene deletion experiments by co-transforming the strain with the eight pCRISPR/ds-dDNAs couples. As a control, the deletion of the degP and yncD genes [two genes that were successfully inactivated in BL21(DE3)ΔompA was also attempted]. As shown in Fig. 3 and Additional file 1: Table  S6, all eight genes were successfully inactivated, with good efficiencies, thus confirming the incompatibility of these mutations with the ΔompA gene inactivation. It is noteworthy that all the eight genes encode membraneassociated proteins whose inactivation in the absence of one of the major E. coli outer membrane proteins might impair the membrane function and/or structural integrity. An important outcome of the mutagenesis experiments carried out in BL21(DE3) strain is that when ds-dDNA is used, our CRISPR/Cas9 mutagenesis protocol can efficiently inactivate any "dispensable" gene and therefore the protocol has a robustness of 100%.

Creation of multi-gene mutants by stepwise approach
In order to generate multi-gene mutants in a stepwise modality, the pCRISPR-gDNA plasmid used to knockout a given gene has to be removed to allow the next round of transformation with a pCRISPR-gDNA plasmid targeting another gene locus. The pCRISPR plasmid is a high copy number plasmid (>400 copies/cell in exponential growth phase) and therefore it cannot be easily removed by simply growing bacteria in liquid media deprived of kanamycin, the pCRISPR antibiotic resistance marker. Elegant solutions have been proposed to get rid of high copy number plasmids [10,18,30]. We adopted the strategy proposed by Hale and co-workers [25] by introducing the suicide sacB gene downstream of the Km cassette into the pCRISPR plasmid, thereby generating the pCRISPR-SacB plasmid. E. coli strains carrying any pCRISPR-SacB plasmid derivative can survive in the presence of sucrose only if they rapidly lose the plasmid [21][22][23][24]. The efficiency of pCRISPR-SacB plasmid elimination in sucrose containing media can accelerate the entire process of multiple stepwise gene inactivation. Once a mutant colony is selected by colony PCR, it can be inoculated in sucrose-containing LB and the overnight culture can be directly used to make competent cells for the subsequent mutagenesis step. We validated the procedure by creating a BL21(DE3)ΔompA(pCasRed) strain carrying the inactivation of both fecA and lpp genes. First of all, the gDNA fecA_B and gDNA lpp_B were inserted into pCRISPR-SacB plasmid generating plasmids pCRISPR-SacB-fecA_B and pCRISPR-SacB-lpp_B plasmids (Additional file 1: Table S1), respectively. Next, BL21(DE3) ΔompA(pCasRed) strain was co-transformed with pCRISPR-SacB-fecA_B plasmid and the 120 base oligo fecA_B-120-∆30 to create the 30 bp fecA gene deletion.
Seven out of the ten colonies selected on Km/Cm LB plate carried the expected mutation (Fig. 4). One mutant clone was grown overnight in LB supplemented with Cm and 5% sucrose and competent cells, directly prepared from the overnight culture, were co-transformed with pCRISPR-SacB-lpp_B plasmid and the double stranded 120 base lpp_B-120-∆240 oligo designed to eliminate the entire lpp gene. BL21(DE3)ompA/ΔfecA/Δlpp colonies were obtained with an efficiency of 60% (Fig. 4). By the time the two pCRISPR-SacB-gDNA plasmids and the two dDNAs were available, four working days were sufficient to generate the BL21(DE3)ΔompA derivative strain carrying the ΔfecA and Δlpp double gene deletions.

Simultaneous creation of multi-gene deletions
Finally, we tested the possibility of exploiting the CRISPR/ Cas technology to simultaneously create mutations at two separated genome loci. To test this strategy, we created pCRISPR plasmids carrying two gDNAs designed to guide the Cas-mediated genome cleavage at two distant positions within the E. coli BL21(DE3)ΔompA genome. The two gDNAs were intercalated by the short repeated regions in the configuration "Repeat-gDNA 1 -Repeat-gDNA 2 -Repeat" (Additional file 1: Table S1; Fig. 5). We mimicked the same organization of the CRISPR arrays found in bacteria. The pCRISPR(gDNA) 2 plasmid thus obtained was co-transformed with two mutagenic oligonucleotides designed to repair and mutagenize the genome at the two cleaved sites.
First, we tested this approach by creating pCRISPR-ompF_5′G-ompF_3′I plasmid to delete 30 bp at both the 5′ and 3′ ends of the ompF gene (Fig. 5a). To verify that the plasmid carrying two guide RNAs could drive the simultaneous Cas-mediated cleavage at two chromosomal sites, E. coli BL21(DE3)ΔompA(pCasRed) strain was transformed with pCRISPR-ompF_5′G-ompF_3′I plasmid in the presence of either ompF_5′G-120-∆30 or ompF_3′I-120-∆30 donor DNAs. If pCRISPR-ompF_5′G-ompF_3′I mediates the Cas9 cleavage of the ompF gene at both sites, co-transformation (See figure on previous page.) Fig. 3 Validation of 30 bp deletions on 78 genes. E. coli BL21(DE3)ΔompA(pCasRed) or E. coli BL21(DE3)(pCasRed) were transformed with different mixtures of pCRISPR-SacB-gDNAs and 70-base dDNAs (either ssDNA or dsDNA) to mediate 30 bp deletion at one of the 78 selected gene loci (y axis). Ten colonies from each transformation were analyzed by PCR to identify those carrying the deletion. X axis indicates the percentage of mutants identified out of the total number of colonies analyzed. Bar height indicates the mutation frequency, while the presence of flat colored squares above gene names in each graph indicates that no mutants were identified out of ten colonies analyzed. Absence of bars or flat colored squares above gene names indicate that the gene mutation was not attempted for those specific genes in the experiment indicated in each bar graph. Red bars/squares indicate mutation experiments using Ld-ss-dDNAs; Blue bars/squares indicate mutation experiments using Lg-ss-dDNAs; Green bars/squares indicate mutation experiments using ds-dDNAs; bars with green downward diagonals indicate mutation experiments with ds-dDNAs in BL21(DE3)(pCasRed). a Bar graph representing mutation success rate using 46 Ld-ss-dDNAs (red bars/squares) and 32 Lg-ss-dDNAs (blue bars/squares). b Gene mutations that failed using the Ld-ss-DNAs (26 genes out of 46) and the Lg-ss-DNAs (4 genes out of 32) were re-attempted using ss-dDNAs targeting the opposite strands. The chart represents the mutation success rate of this second round of experiments. c The bar graph represents the mutation success rate in BL21(DE3)ΔompA(pCasRed) (green bars) and in BL21(DE3)(pCasRed) (green downward diagonal bars) using ds-dDNAs with only one dDNA should not generate mutants. Indeed, co-transformation of pCRISPR-ompF_5′G-ompF_3′I with ompF_3′I-120-∆30 donor DNA gave no mutants (from 20 colonies tested) and the cotransformation of pCRISPR-ompF_5′G-ompF_3′I with ompF_5′G-120-∆30 dDNA gave only two mutants out of 20 screened colonies (Table 2). Next, we co-transformed BL21(DE3)ΔompA(pCasRed) with pCRISPR-ompF_5′G-ompF_3′I in the presence of both ompF_5′G-120-∆30 and ompF_3′I-120-∆30 dDNAs and we analyzed 20 colonies by PCR. Twelve out of 20 colonies carried the deletion at both sites while only one of the 20 colonies analyzed carried a single mutation at the 5′ end of the ompF gene ( Fig. 5; Table 2).

Fig. 4
Representation of the stepwise approach used to isolate strains carrying multiple mutations. Day 1: E. coli BL21(DE3)∆ompA(pCasRed) was co-transformed with 1 μg/ml of pCRISPR-SacB-fecA_B and 10 μg/ml of donor fecA-120-∆30nt and transformant colonies were selected on LB agar plates supplemented with Cm (25 μg/ml) and Km (50 μg/ml). Day 2: Ten colonies were randomly selected and screened by PCR using primers designed to generate DNA fragments from mutated colonies of 200 bp. PCR products were analyzed on 2% agarose gels. One mutant clone was subsequently inoculated into 5 ml of LB supplemented with 5% sucrose and 25 μg/ml Cm. Day 3: The overnight culture was used to prepare competent cells, which were subsequently co-transformed with 1 μg/ml of pCRISPR-SacB-lpp_B and 10 μg of double strand donor DNA lpp-120-Δ237. Day 4: Ten colonies were randomly selected and screened by PCR using primers designed to generate DNA fragments from mutated colonies of 400 bp. PCR products were analyzed on 2% agarose gels To confirm the applicability of the simultaneous double gene knockout, we also tested the concomitant inactivation of lpp and fecA genes. For this purpose, pCRISPR-lpp_B-fecA_B was constructed (Fig. 5b) and used to co-transform E. coli BL21(DE3)ΔompA(pCasRed) with lpp_B-120-∆30 and fecA_B-120-∆30 donor DNAs (Table 2). Here, nine out of 29 colonies were mutated in both genes, while single gene mutations were obtained at frequencies of 0% for fecA and 3% for lpp ( Fig. 5b; Table 2).

Discussion
The cutting edge CRISPR/Cas-technology was elegantly applied to bacteria for the very first time by Jiang and co-workers in 2013 [9]. Subsequently, a number of other groups, using modified versions of the protocol proposed coli BL21(DE3)∆ompA(pCasRed) was co-transformed with either 100 ng pCRISPR-ompF_5′G-ompF_3′I plasmid and the two dDNAs ompF_5′G-120-∆30 and ompF_3′I-120-∆30 (10 μg each) (a) or with 100 ng pCRISPR-lpp_B-fecA_B and lpp_B-120-∆30 and fecA_B-120-∆30 dDNAs (10 μg each). b Transformant colonies were selected on LB agar plates supplemented with 25 μg/ml Cm and 50 μg/ml Km. Colony PCR was carried out using two different couple of primers to screen each genomic locus (indicated at the bottom of each gel) on a randomly selected number of colonies and the PCR products separated on 2% agarose gels. Asterisks indicate those colonies in which deletion occurred at both gene loci. The primer sequences used for PCR experiments are reported in Additional file 1: Table S5   Table 2 Efficiency of simultaneous two-loci mutagenesis (ompF 5′/ompF 3′ regions and fecA/lpp genes) using pCRISPR plasmids carrying REPEAT-gDNA 1 -REPEAT-gDNA 2  Thanks to these discoveries, the CRISPR/Cas methodology became the most effective strategy to create gene knockouts and knock-ins in prokaryotes (for a recent review see Choi and Lee [14]). Two are the major breakthroughs of CRISPR/Cas-technology. First, the efficiency of mutagenesis can be as high as 100%. As reported by several authors [9,10], such efficiency cannot be reached by any other genome editing strategy, including the procedure based on the λ Red/synthetic oligonucleotide-mediated homologous recombination described by Court and co-workers.
For instance, when we tested the efficiency of ompF 30 bp deletion transforming BL21(DE3)ΔompA(pCasRed) with the synthetic dDNA in the absence of the corresponding pCRISPR-gDNA, we did not manage to isolate a single mutant colony out of 40 colonies analyzed. Second, even large deletions can be achieved at high frequencies using synthetic oligonucleotides as donor DNAs, thus avoiding the need to generate linear or circular DNA constructs to drive homologous recombination. However, one missing information from published data was the robustness of the CRISPR/Cas protocols. In other words, how confident could one be that any desired mutation can be consistently achieved at high frequency? This information becomes particularly relevant when CRISPR/Cas genome editing is applied in high-throughput formats where the screening of several colonies and the repetition of mutagenesis experiments to rescue missing mutants could be impractical.
To address this issue, we first set up our own version of CRISPR/Cas9-based genome editing and we defined the best conditions by analyzing the mutagenesis efficiency at four gene loci using different parameters, including (i) sequence of the guide RNA (ii) length and concentration of the dDNA, and (iii) use of ss-dDNA and ds-dDNAs. Subsequently, we challenged the robustness of the protocol by attempting 30 bp deletion in 78 additional genes selected on the basis of their being classified as "dispensable" according to the Keio collection [29]. In this way, and for the first time, we could assign a value to the success rate of a CRISPR/Cas9-based genome editing procedure based on a statistically meaningful number of data (81 target loci in total).
The relevant outcome from our work is that we could establish with a high degree of confidence that our CRISPR/Cas9 protocol features an efficiency of mutagenesis consistently higher than 10% and a robustness of 100%. All 81 genes were successfully mutated with high efficiencies, 73 of them in the E. coli BL21(DE3)ΔompA background, and the remain eight genes, whose inactivation was incompatible with the concomitant presence of ompA mutation, were mutated in E. coli BL21(DE3).
Considering that the main components of our CRISPR/ Cas9 genome editing protocol (pCRISPR-SacB-gDNA and pCasRed) are variations-of-the-theme of what already described by others, we believe that our data can be extrapolated to other procedures and therefore could be of general interest. We confirm, thanks to a massive amount of validation data, that CRISPR/Cas genome editing can be very effective and reliable provided that a few experimental solutions are applied. Among those is the use of double stranded synthetic oligonucleotides as donor DNA, and the pre-testing of gDNAs before attempting gene deletion.
Court and co-workers were the first to demonstrate that "λ Red recombineering" in E. coli can be carried out with both single and double stranded synthetic oligonucleotides, and that Lg-ss-dDNAs perform better than Ld-ss-dDNAs in promoting the mutagenesis event [31]. Like other authors [10,18], we confirmed the superiority of the Lg-ss-dDNA even when "λ Red recombineering" is combined with CRISPR/Cas9, a situation whereby the mechanisms that leads to the introduction of the sitespecific mutation must be coupled to the mechanisms of double-stranded DNA repair. However, we also found that the successful mutation of all 81 genes could be achieved, with efficiencies consistently higher than 10%, only if ds-dDNA is used (Fig. 3). Furthermore, we showed that ds-dDNA clearly outperformed ss-Lg-dDNA (and ss-Ld-dDNA) when extended deletions were attempted (Table 1).
We did not investigate on the mechanisms of homologous recombination taking place between ds-dDNA and the two chromosomal extremities generated by the Cas9 cleavage. However, it is reasonable to envisage, in addition to other described mechanisms, the contribution of a single exonuclease activity (possibly the 5′ exonuclease provided by the λ Red machinery) acting on both the ds-dDNA and the extremities of the cleaved chromosomal DNA. The exonuclease activity would release single stranded termini, stabilized by Beta, which would become available for base pairing. Such mechanism would not be effective with ss-dDNAs and therefore this could explain the higher efficiency of mutagenesis of ds-dDNA with respect to ss-dDNA. Alternatively, another possible mechanism is that the double strand break is repaired by dsDNA template (and not by ssDNA template) following a Rec-dependent homologous recombination. Such mechanism can occur even in the absence of the lambda red machinery [28,32]. To test this possibility, the mutagenesis of proX and yiaD, two genes successfully mutagenized with ds-dDNA only, and with ssDNAs and dsDNA, respectively (Fig. 3), was re-attempted in BL21(DE3)∆ompA(pCas9) in which the λ Red machinery was removed. The strain was transformed with either pCRISPR-SacB-proX or pCRISPR-SacB-yiaD in the presence of the corresponding ds-dDNAs, and subsequently 100 colonies from each transformation were PCR analyzed. No mutant colonies were identified (data not shown). Therefore, under the experimental conditions we used, the efficiency of dsDNA homologous recombination, if occurring, was at least lower than 1%.
The second factor to consider to achieve high performance is the addition of a rapid experimental step aimed at demonstrating the effectiveness of gDNAs. Theoretically, any sequence in the genome next to the "NGG" (PAM) trinucleotide should represent a potential Cas9 cleavage site. Indeed, most of the gDNAs that we designed and cloned into pCRISPR plasmid produced gRNAs that promoted Cas-mediated cleavage with high specificity. However, 10% of the pCRISPR-gDNAs did not work, as judged by their inability to substantially reduce bacterial viability after transformation. We tried to identify the cause root of their failure, such as searching for the presence of sequences forming internal stem-loop structures or partially complementary to other chromosomal regions not carrying the NGG trinucleotides, but we did not find any plausible explanation so far. Whatever reason might be, the use of inefficient gDNAs would affect the overall robustness of the procedure. Therefore, we routinely transform the strain to be mutagenized with all pCRISPR-gDNAs before starting mutagenesis. A drop in transformation efficiency of at least three orders of magnitude is indicative of the quality of the gDNAs.
One important aspect to consider in setting up a CRISPR/Cas9-based mutagenesis protocol is the possibility to eliminate the whole machinery once genome editing is terminated. As far as the Cas9/λ Red genes are concerned, since they reside on a low copy number plasmid expressing the Cm R , they can be easily removed by growing the strain in the absence of the antibiotic. We experimentally confirmed the effectiveness of losing pCasRed by growing strains overnight in the absence of Cm (see "Methods"). To eliminate the high copy number plasmid encoding the gRNA, a few strategies have been devised, the most popular one being the use of a temperature sensitive origin of replication [10]. To get rid of our pCRISPR-SacB-gDNA we envisaged to exploit the toxic effect of the sacB gene when E. coli is grown in media containing high concentrations of sucrose. SacB fused to an antibiotic resistance marker (usually chloramphenicol) has been used in chromosomal gene knock-in/knock-out experiments [21][22][23][24] and, more recently, for plasmid curing [25]. Indeed, we found that by simply inoculating the mutant colonies in LB supplemented with 5% sucrose, bacterial cells rapidly lose the plasmid and, if necessary, the culture is ready to be used for an additional round of mutagenesis. Multiple mutations can be introduced in the same recipient strain at a pace of one mutation every two working days (Fig. 5).
The pCRISPR-SacB-gDNA plasmid can accommodate more than one gRNA coding sequences potentially allowing the simultaneous modification of more than one gene. We demonstrated that by cloning two gDNAs in the configuration repeat-gDNA 1 -repeat-gDNA 2repeat, double gene inactivation was achieved with very high efficiencies. Therefore, not only bacterial cells could successfully take up all three elements simultaneously [pCRISPR-(gDNA) 2 -SacB and two dDNAs] but also the gRNA was efficiently transcribed and processed to promote the Cas9-mediated cleavage at both chromosomal sites. It will be interesting to investigate the limits of the system and see how many loci can be simultaneously modified.

Bioinformatics
The "Keio collection" [29], a collection of systematic single-gene knockout mutants of E. coli, was used as a basis to evaluate both the biological functions and the non-essentiality of the selected genes listed in Additional file 1: Table S5. To properly select the gDNA and to avoid off-target effects of Cas9, we used the following procedure. First, we identified the PAM sequences in the first half of the target genes. Next, we selected the 30 bp upstream the PAM sequences and we blasted them [33] against E. coli BL21(DE3) genome. Those sequences with no complementary nucleotides within the ten nucleotide seed-region and with a homology lower than 15% in the remaining part of the guide were selected.

Bacterial strains and culture conditions
Escherichia coli DH5α strain was routinely grown in LB broth (SIGMA) at 37 °C and used for cloning experiments. E. coli BL21(DE3)∆ompA strain generated as previously reported [34], was grown in LB broth at 37 or 30 °C when required, and was employed for genome editing experiments. Stock preparations of strains were prepared in LB + 20% glycerol and stored at −80 °C. Each bacterial manipulation was started using an overnight culture from a frozen/glycerol stock. When required, kanamycin and chloramphenicol were added to final concentrations of 50 or 25 μg/ml, respectively.

Construction of plasmids
Information about all primers, plasmids and E. coli strains used in this study are provided in Additional file 1: Tables S1, S5.
The pCasRed plasmid carries a chloramphenicol resistance gene (Cm R ) and encodes the Cas9 nuclease, the λ Red (Exo, Beta, Gam) cassette and the tracrRNA. The Cas9 gene and the tracrRNA coding sequence are under the control of constitutive promoters while the λ Red gene cassette is transcribed using an arabinose-inducible promoter.
The pCas9 plasmid (ADDGENE #4287 [9]) was used as template for the construction of pCasRed plasmid as follows. The λ Red cassette was PCR amplified from pKOBEG plasmid [20] and was cloned into pCas9 plasmid using the polymerase incomplete primer extension (PIPE) cloning method [35]. Briefly, the vector pCas9 was linearized by V-PIPE PCR amplification with primers Pipe1 pCAS9-F/ Pipe1 pCAS9-R. The insert, the λ Red cassette, was I-PIPE PCR amplified with primers redF/redR, which contain 5′ sequences complementary to the two distinct ends of the amplified vector. In this manner, annealing occurred directionally by mixing the PCR products, V-PCR and I-PCR, and by transforming E. coli HK-100 strain, pCasRed plasmid was generated. The resulting construct was analyzed by DNA sequencing and used to transform E. coli BL21(DE3)∆ompA electrocompetent cells.
The pCRISPR-SacB plasmid, derived from pCRISPR plasmid (ADDGENE #42875, [9]), where a kanamycin resistance gene (Km R ) is fused to the sacB gene encoding the Bacillus subtilis levansucrase, carries the synthetic DNA fragment (gDNA) coding for the guide RNA necessary to drive the Cas9-dependent double stranded break at the desired site of the bacterial genome. The construction of the pCRISPR-SacB plasmid was carried out by the PIPE method in two steps. In a first step, the kanamycin resistance cassette of pCRISPR plasmid was replaced by a "cat-sac cassette" containing the chloramphenicol acetyltransferase gene, along with the sacB gene, from pKM154 plasmid (ADDGENE #13035) [24]. The vector pCRISPR was linearized by V-PIPE PCR amplification with primers pipeCRISPR-F and pipeCRISPR-R to exclude the kanamycin gene. The insert, the cat-sac cassette, was I-PIPE PCR amplified with cat/sac-pipeF and cat/sac-pipeR primers, which contain 5′ sequence complementary to the two distinct ends of the amplified vector. In this manner, annealing occurred directionally by mixing the PCR products, V-PCR and I-PCR, and after transformation pCRISPR-CatSacB plasmid was isolated. In a second step, the chloramphenicol resistance cassette of the pCRISPR-CatSacB plasmid was then replaced with the kanamycin gene from the original pCRISPR plasmid. The pCRISPR-CatSacB plasmid was linearized by using V-crSac F and V-crSac R primers, excluding the chloramphenicol acetyltransferase gene. The kanamycin gene was amplified from pCRISPR plasmid using primers I-kanaF and I-kanaR. The PCR products were then mixed together and used to transform E. coli, generating the pCRISPR-SacB plasmid. The resulting construct was analyzed by DNA sequencing.
Plasmids expressing the gRNAs, were constructed by phosphorylation and annealing of oligonucleotides (gDNAs) listed in Additional file 1: Table S2, followed by ligation into pCRISPR (or pCRISPR-SacB) digested with BsaI (New England BioLabs), generating the plasmids listed in Additional file 1: Table S1. The resulting constructs were used to transform E. coli DH5α strain (Invitrogen) and the plasmids prepared by QIAprep Spin Miniprep Kit, Qiagen (QIAGEN kit) were analyzed by DNA sequencing.
To prepare E. coli BL21(DE3)∆ompA(pCasRed) a 5 ml overnight culture (LB medium) inoculated from a single colony of BL21(DE3)∆ompA obtained from an LBagar plate was grown at 37 °C under vigorous agitation. The overnight culture was diluted 100-fold and grown at 37 °C (200 r.p.m.) until the optical density at 600 nm (OD 600 ) reached 0.6-0.8 (~3 h). Then the cells were harvested at 4000 r.p.m. for 20 min at 4 °C and washed three times with cold MilliQ water. After a final wash in 10% glycerol the cells were aliquoted and stored at −80 °C. 50 μl of competent cells were then electroporated using 1 mm Gene Pulser cuvette (Bio-Rad) at 1.8 kV with 1 ng of pCasRed plasmid. Competent cells of E. coli BL21(DE3)∆ompA(pCasRed) were prepared by growing a single colony in LB medium with 25 μg/ml chloramphenicol at 37 °C under shaking. The overnight culture was diluted to an OD 600 of 0.1 and grown at 37 °C under shaking (200 r.p.m.) to an OD 600nm of 0.2 and then l-arabinose was added to a final concentration of 0.2% for λ Red induction. After induction, the culture was grown to an OD 600 of 0.7 and then cells were washed and aliquoted as described above.

Gene knockout using CRISPR-Cas9
The genes encoding ompF, lpp and fecA were used as targets to establish the proof of concept for genome editing via CRISPR-Cas9 system in E. coli. All mutagenesis oligonucleotides (donor DNA or dDNAs) (Additional file 1: Table S3) (Sigma-Aldrich) were HPLC purified grade. The dDNAs were designed to delete a region ranging from 30 (∆30) to 2325 (∆2325) nucleotides from target genes, removing the protospacer and PAM regions, thus disrupting the Cas9 cleavage site and at the same time adding an in-frame stop codon downstream the deleted region.
For the leading and lagging strand design, the oligonucleotides annealing to the 3′ > 5′ strand moving clockwise from OriC up to ter were Ls-ss-dDNAs, while oligonucleotides annealing to the same strand but moving counterclockwise from OriC up to ter were Lg-ss-dDNAs. The opposite was true for the oligonucleotides annealing to the 5′ > 3′ strand moving clockwise from OriC up to ter and for those annealing to the same strand moving counterclockwise from OriC to ter.
The ds-dDNAs were generated by annealing 10 µg of both forward and reverse oligonucleotides in a total volume of 20 µl at 95 °C for 5 min and allowing the reaction mixture to cool down at room temperature. The annealing reaction was verified by loading 500 ng of each single stranded oligonucleotides and 1 μg of total DNA in the annealing reaction and by visualizing the bands using ATLAS ClearSight DNA Stain (BIOATLAS). The bands corresponding to the single stranded oligonucleotides disappeared in the annealing reaction sample. For CRISPR/Cas9-mediated gene knockouts 50 μl of E. coli BL21(DE3)∆ompA(pCasRed) or BL21(DE3)(pCasRed) competent cells, corresponding to 10 9 competent cells, were electroporated using 1 mm Gene Pulser cuvette (Bio-rad) at 1.8 kV with 100 ng of pCRISPR-gDNA and different quantities of dDNA ranging from 1 to 100 μg. As control, 100 ng of an empty pCRISPR plasmid was used. Cells were then immediately re-suspended in 1 ml of LB medium and allowed to recover at 30 °C for 3 h under agitation before being plated on LB agar with 25 μg/ml chloramphenicol and 50 μg/ml kanamycin and incubated at 37 °C overnight. Mutants were screened by colony PCR using GoTaq master mix (Promega-M7123). Briefly, cells were picked from individual colony using a pipette tip, directly resuspended in PCR reaction mix and DNA amplification was carried out according to the standard cycling GoTaq protocol. The deletion in ompF gene was analyzed by using primers: seqs-ompF 1F/ ompR (∆30), seqs-ompF 1F/ompR2 (∆100), ompF 1F/ ompF 4R2 (∆500, ∆1089) (Additional file 1: Table S5). The deletion of lpp was analyzed by using primers: seqslpp F/seqs-lpp R (∆30, ∆237) (Additional file 1: Table S5). The deletion in fecA gene was analyzed by using primers: FecA_F1/FecA_R1 (∆30), dFecA-seqF/dFecA-seqR (∆500, ∆2325) (Additional file 1: Table S5).

Plasmid curing and creation of multiple mutations by stepwise approach
To cure mutant strains from pCRISPR plasmid derivatives after each round of mutation, we adopted the strategy proposed by Hale and co-workers [25] by introducing the suicide sacB gene downstream of the Km cassette into the pCRISPR plasmid, generating the pCRISPR-SacB plasmid. E. coli strains carrying any pCRISPR-SacB plasmid derivative can survive in the presence of sucrose only if they rapidly lose the plasmid [21][22][23][24]. To test the effectiveness of this strategy, we transformed the BL21(DE3)∆ompA(pCasRed) strain with pCRISPR-SacB and one transformant colony was grown in LB medium supplemented either with 25 μg/ml Cm alone (the antibiotic resistance marker of pCas9-λ Red plasmid) or with 25 μg/ml Cm and 5% sucrose. After overnight growth, 100 μl of each culture were plated on LB agar plates supplemented with either Km and Cm, or Cm only. Not a single colony from the sucrose-containing culture could be isolated on the Km/Cm containing plate (Fig. 6), indicating that all bacteria had lost the pCRISPR-SacB plasmid. By contrast, confluent growth was observed on the Km/Cm plate seeded with bacteria grown in sucrosedeprived medium. The loss of pCRISPR plasmid from BL21(DE3)∆ompA(pCasRed)(pCRISPR-SacB) strain grown in the presence of sucrose was confirmed by plasmid extraction (Fig. 6).
For pCRISPR plasmid curing after each mutation round, the mutant colony, carrying both the pCasRed and pCRISPR-SacB plasmids, was inoculated in LB medium containing 5% sucrose and 25 μg/ml chloramphenicol, and grown overnight at 37 °C under shaking conditions. The overnight culture was directly used to prepare competent cells as described above. Briefly, at day 1, 50 μl of E. coli BL21(DE3)∆ompA(pCasRed) competent cells were co-transformed with 100 ng of pCRISPR-SacB-fecA_B plasmid and 10 μg of fecA_ B-120-∆30 dDNA. At day 2, transformant colonies were PCR screened and one mutant clone was inoculated in 5 ml LB supplemented with 5% sucrose and 25 μg/ml chloramphenicol. The overnight culture was used to prepare competent cells as described above. The new Fig. 6 pCRISPR-SacB-gDNA plasmid curing using 5% sucrose containing medium. A single colony from E. coli BL21(DE3)ΔompA strain carrying both the pCasRed (Cm resistance) and pCRISPR-SacB-gDNA (Km resistance) was grown at 37 °C in LB-medium containing 5% sucrose and 25 μg/ml Cm. After 14 h growth, 100 μl of culture were plated on LB-agar plates containing either Cm (25 μg/ml) + Km (50 μg/ml) or Cm (25 μg/ml) alone. The loss of pCRISPR-SacB-gDNA plasmid was verified by 1.5% agarose gel analysis of plasmids extracted from bacteria directly collected from the Cmcontaining agar plate. As control, the same colony was grown in the absence of 5% sucrose and plasmid extraction was carried out from bacteria collected from LB-agar plate containing 25 μg/ml Cm recipient strain BL21(DE3)∆ompA/∆fecA(pCasRed) was co-transformed with 100 ng of pCRISPR-SacB-lpp_B plasmid and 10 μg of ds-dDNA lpp_B-120-∆237 and transformant colonies were selected on LB-Agar 50 μg/ ml kanamycin and 25 μg/ml chloramphenicol. The day after transformant colonies were PCR screened to identify the E. coli BL21(DE3)∆ompA/∆fecA/∆lpp(pCasRed) mutant.
The curing of the pCasRed plasmid was carried out after overnight growth at 37 °C under shaking in the absence in the culture medium of chloramphenicol. The day after the culture was plated for single colony on LBagar plate and colonies were analyzed for chloramphenicol resistance. No chloramphenicol resistance colonies were recovered.

Authors' contributions
FZ designed the experimental set-up and experiments were performed by FZ, IZ, DF, EK. FZ, IZ, DF, EK, CI, LFF and GG analyzed and interpreted the acquired data. The manuscript was primarily prepared by GG and FZ, and includes corrections and intellectual contributions from all other authors. All authors read and approved the final manuscript.