NT-CRISPR, combining natural transformation and CRISPR-Cas9 counterselection for markerless and scarless genome editing in Vibrio natriegens

Design of and rationale behind NT-CRISPR

For NT-CRISPR, we envisioned a one-plasmid design to achieve a convenient and fast workflow. One necessary component of the NT-CRISPR system is the master regulator for natural competence tfoX. Similar to a previous approach²⁰, we use the tfoX gene from Vibrio cholerae under the control of the isopropyl ß-D-1-thiogalactopyranoside (IPTG) inducible P_tac promoter. In addition, our design requires the components of the CRISPR-Cas9 system, namely cas9 and gRNA, both driven by the anhydrotetracycline (ATc) inducible P_tet promoter (Fig. 1a). In our construct, we used improved variants of P_tac and P_tet²⁹. We found that simultaneously maintaining the coding sequences for Cas9 and a gRNA targeting a genomic sequence to be a major challenge in V. natriegens. A possible reason was trace production of Cas9 and gRNA through leaky promoter activity which was probably sufficient to introduce DNA DSBs and consequently causes cell death. This problem persisted even when a weak ribosome binding site was used and the strongest available SsrA-derived protein degradation tag (M0050)³⁰ was fused to Cas9 to reduce cellular abundance of this protein. Previously, it had been shown that by these two measures, the Cas9 and gRNA coding sequences could be established together in E. coli cells¹⁴. However, when V. natriegens cells were transformed with the plasmid carrying all CRISPR-Cas9 components, we obtained only very few colonies carrying a wide range of deleterious mutations in the plasmid, rendering the CRISPR-Cas9 system non-functional. In contrast, the same plasmid with a non-binding control gRNA could be easily introduced to this host.

The final solution to the strong toxicity of Cas9 and gRNA produced at basal levels was the expression of the anti-CRISPR protein-encoding gene acrIIA4 under control of the constitutive promoter J23106. AcrIIA4 shows a high affinity to Cas9 and the Cas9-gRNA complex, efficiently inhibiting Cas9-gRNA from binding to the DNA target and consequently preventing the introduction of DSB^31,32. Anti-CRISPR proteins have been used in eukaryotes to reduce off-target effects^33,34 or to achieve spatiotemporal control over CRISPR-Cas9 activity, for example through an optogenetic approach³⁵ or by selective expression in certain cell types³⁶. Moreover, anti-CRISPR proteins were used in the prokaryotic species Clostridium acetobutylicum to enable cells to carry cas9 and gRNA simultaneously without toxic effects³⁷. In a similar fashion, we used constitutive expression of acrIIA4 to compensate for the basal production of Cas9 and gRNA. After adding acrIIA4 to our design, we were finally able to transform V. natriegens with a plasmid carrying both cas9 and a gene coding for a chromosome targeting gRNA. The full plasmid design is shown in Fig. 1a.

The first step after transformation of the NT-CRISPR plasmid (Fig. 1a) is the natural transformation, which requires the addition of IPTG to induce production of the master regulator TfoX. For a simple deletion, a transforming DNA (tDNA), with sequences homologous to the regions flanking the target sequence to be deleted, is added to the cells (Fig. 1b). Only a minority of cells in the population will be modified, while the vast majority of cells will still have the wild type sequence. The last step in the workflow is the CRISPR-Cas9 mediated counterselection (Fig. 1c). Therefore, ATc is added to induce expression of both cas9 and the gRNA gene to overcome the AcrIIA4 threshold causing DSB and cell death in cells with the wild type sequence. In contrast, modified cells are immune to CRISPR-Cas9 counterselection because the target sequence is no longer present in the genome (Fig. 1c). As a result, the counterselection step enriches successfully edited cells.

Inducible cell killing by CRISPR-Cas9

As described above, it is crucial for the NT-CRISPR method to tightly control CRISPR-Cas9 and induce cell death only upon induction. We reasoned that the ratio between acrIIA4 and cas9 expression strength is important. Expression of acrIIA4 that is too weak might not be sufficient to compensate for the leaky cas9 and gRNA gene expression, still leading to premature cell death. On the other hand, acrIIA4 expression that is too strong, could prevent efficient counterselection despite a full induction of the CRISPR-Cas9 system. To test a range of acrIIA4 and cas9 expression strengths, we created twelve plasmids representing the combinatorics of four different ribosome binding sites (RBS) for acrIIA4 with three distinct RBS for cas9. The RBS used differ in their expression strengths which were recently quantified in V. natriegens with fluorescent reporter experiments³⁰. These variants were tested for their ability to trigger cell death upon addition of ATc as the inducer of Cas9 and gRNA production. We used a gRNA targeting the non-essential wbfF gene, which is involved in capsule polysaccharide biosynthesis³⁸. Combinations of different RBS for cas9 and acrIIA4 were tested to identify translation rates and consequently a protein stoichiometry which leads to efficient inducible cell killing. As a simple experiment, we cultivated cells in microplates and continuously measured OD₆₀₀ to track growth of the cultures. To induce expression of cas9 and the gRNA gene, ATc was added in a final concentration of 200 ng/mL to exponentially growing cells 1 h after the start of the batch cultivation. We observed a reduction in OD₆₀₀, indicating cell death about 1.5 h after ATc-mediated induction of cas9 and the gRNA gene with ATc for all strains carrying plasmids using the moderately strong RBS B0032³⁰ for acrIIA4 (Supplementary Fig. S1a). While the choice of the RBS for acrIIA4 and consequently its expression strength, apparently has a strong impact on the inducibility of the CRISPR-Cas9 system, we found no difference between the three tested RBS for cas9, neither in this growth experiment in liquid culture (Supplementary Fig. S1a) nor in a separate agar plate-based assay which yielded counterselection efficiencies similar to those obtained applying the actual NT-CRISPR editing workflow (Supplementary Fig. S1b). Of the three variants that showed inducible cell killing in the growth experiment, we proceeded with plasmid pST_116, which carries the strongest of the three tested RBS for cas9 B0033³⁰.

Dependence of gene deletion efficiencies on tDNA amount and homologous sequence length

We chose wbfF as a first target to demonstrate gene deletion by our NT-CRISPR-based approach in V. natriegens since deletion mutants of this gene show an almost transparent colony morphology²⁰. This phenotype allows for distinction between wild type and wbfF mutant colonies to conveniently calculate gene deletion efficiencies. We reproduced this phenotype by our approach using tDNA comprising 3 kb homologous sequences flanking the designed deletion of the wbfF coding sequence (1734 bp) (Fig. 2a). Subsequently, we characterized the genome editing efficiency dependent on the amount of tDNA supplied (1 ng, 10 ng or 100 ng) and on the induction state of the CRISPR-Cas9 system (Fig. 2b). We obtained almost 100% of all colonies with the transparent ∆wbfF morphology for all three DNA amounts tested when CRISPR-Cas9 was induced. This suggests a remarkably high editing efficiency of NT-CRISPR, even when just 1 ng of tDNA was used. However, in absence of counterselection we observed a decrease in the fraction of positive colonies with decreasing tDNA amounts (Fig. 2b). This is in accordance with a previous study describing natural transformation for genome engineering in V. natriegens²⁰. This trend also became apparent through the absolute number of positive colonies (CFU/µL) which decreased with lower amounts of tDNA for both the induced and uninduced samples (Supplementary Fig. S2a, b). With 100 ng of added tDNA and without counterselection, we still obtained ~20% of positive colonies (Fig. 2b), further supporting the potential of natural transformation for genome engineering in V. natriegens.

In a second experiment, we investigated the impact of different lengths of the homologous sequences of the tDNA on the success of genome editing with NT-CRISPR (Fig. 2c). When CRISPR-Cas9 was induced, we again obtained high editing efficiencies of at least 90% for all tested fragment lengths, with exception of the samples with 50 bp homologous flanks. More than 98% editing efficiency was achieved for 3000 bp and 1000 bp homologous flanks. However, a dependency of the efficiency of natural transformation on the length of the tDNAs is suggested by the fact that a lower absolute number of colonies with ∆wbfF mutant morphology (given as CFU/µL) was obtained for tDNAs with shortened homologous flanks. This was observed both with and without CRISPR-Cas9-based counterselection (Supplementary Fig. S2c, d). Decreases in putative ∆wbfF CFU/µl by ~40 fold and ~4 fold were observed when reducing the flanks from 3000 to 1000 bp and 1000 to 500 bp, respectively (Supplementary Fig. S2c). Further shortening the flanks down to 100 bp did not result in considerable additional drops in the number of putative ∆wbfF colonies (given as CFU/µl) (Supplementary Fig. S2c). Note that a decreasing length of tDNAs correlates with an increasing number of molecules for 100 ng mass of tDNA. An almost constant number of colonies obtained for fragments with homologous flanks of 500 bp, 200 bp and 100 bp suggests that a possible reduction in uptake or recombination efficiency of shorter fragments is partly compensated by a higher concentration of tDNA molecules. Another result with potential practical implications is the high editing efficiency of ~90% obtained by NT-CRISPR with an asymmetric tDNA fragment with 3000 bp and 50 bp homologous sequence surrounding the deletion (Fig. 2c). This fragment can be easily generated in one PCR by adding the short 50 bp arm as an overhang to one of the PCR primers used to amplify the long arm. Applying such an asymmetric DNA fragment for natural transformation-mediated locus exchange by homologous recombination was successfully established for V. cholerae, but required the deletion of at least two ssDNA exonucleases to achieve reasonable editing efficiencies²⁰. If deletion of the orthologous ssDNA exonucleases in V. natriegens could even further increase the editing efficiency with short tDNA fragments remains unknown. Nonetheless, due to the highly efficient counterselection, asymmetrical tDNA fragments with one short arm might represent a convenient approach to introduce single gene deletions.

Collectively, the initial characterizations suggest that combining natural transformation with efficient, targeted CRISPR-Cas9-mediated counterselection considerably increases genome editing frequencies in V. natriegens. Moreover, even low amounts of tDNA like 1 ng and homologous flanks as short as 100 bp are sufficient to produce more than 90% correct colonies.

NT-CRISPR proof-of-concept for generating deletions of different lengths, point mutations and integrations in a set of target genes

Inspired by these promising results, we sought to apply CRISPR-NT for various types of genome edits and to expand the set of chromosomal target sequences for our proof-of-concept study in V. natriegens. We assembled NT-CRISPR plasmids with 15 different gRNAs targeting a range of different sequences, both on chromosome 1 and 2. The reason behind selecting these genes is described below. In order to quantify the “killing efficiency” of the respective gRNAs, we performed the NT-CRISPR protocol without addition of tDNA. Samples were either induced with ATc to induce CRISPR-Cas9-based counterselection or remained uninduced. The ratios of the resulting CFU/µL allows for calculation of killing efficiency in the NT-CRISPR workflow. The shown gRNAs resulted in killing efficiencies within a narrow range between 99.999% (malQ) and 99.964% (vnp1) (Fig. 2d). The value of the strongest tested gRNA that targeted malQ translates to the survival of just one out of 100,000 cells. A non-targeting gRNA was included as a control and resulted in a slight reduction of cells upon Cas9 and gRNA induction, possibly due to unspecific toxicity or protein production burden. An empty plasmid control ruled out negative effects of the inducer ATc at the applied concentration.

In the NT-CRISPR method, editing through natural transformation and CRISPR-Cas9-based counterselection occur in two temporally distinct steps. Consequently, with knowledge of the killing efficiency, it is possible to calculate the apparent editing efficiency from the initial editing efficiency before counterselection. The apparent editing efficiency is defined as the fraction of positive cells after counterselection. With a highly efficient gRNA even an initial editing efficiency of just 10⁻⁶ (one edited cell out of one million) was computed to result in 10⁻¹ or 10% of all remaining cells being correct after counterselection (Fig. 2e). With the weakest shown gRNA, an initial editing efficiency of 10⁻³ (0.1%), which is far below the values obtained for simple deletions without counterselection (Fig. 2b), was calculated to be sufficient for achieving apparent editing efficiencies of almost 100% through robust counterselection. Different gRNAs for the same gene should not influence the initial natural transformation-dependent editing step but might yield different counterselection efficiencies. Since we observed a very narrow range of killing efficiencies for the different gRNAs shown (Fig. 2d), we do not expect screening of gRNAs for enrichment of certain genetic modifications to be necessary, as long as the used gRNA is in principle functional and yields a killing efficiency approximately within the reported range.

Deletions

To demonstrate the applicability of NT-CRISPR for deletions, we selected single genes and groups of genes as targets whose loss might lead to an increased plasmid transformation efficiency. So far, the tremendous potential of V. natriegens as a fast-growing strain for the selection and propagation of in vitro assembled plasmids is still hampered due to transformation efficiencies that are much lower than the ones observed for highly engineered E. coli strains³⁰. We targeted the two prophage regions vnp1 and vnp2, as their removal leads to increased survival under osmotic stress conditions³⁹ which might be experienced by the cells during preparation of chemically competent cells. In addition, we chose galE, encoding an enzyme which provides precursors for the synthesis of lipopolysaccharides, because an increased plasmid transformation efficiency was reported for galE mutants of some Gram-negative bacteria, presumably due to the loss of cell surface structures which might impair plasmid uptake^40,41. Along the same line, we reasoned that removal of other surface structures, namely the flagella and pili, could also contribute to increased transformation efficiency. Lastly, we included the gene coding for the extracellular nuclease Xds⁴², which could degrade the incoming plasmid DNA similarly to the Dns nuclease. Deletion of dns was already confirmed to drastically increase the plasmid transformation efficiency of chemically competent cells^5,30.

The target sequences for deletion described above span a wide range in size from 1.0 kb (galE) up to the prophage region of vnp2 with 39 kb and are located on either chromosome 1 or chromosome 2 (Fig. 3a). To quantify the editing efficiency of each of these deletions, we analyzed 50 colonies by colony PCR. For all targets, except for the two prophage regions, all screened colonies were correct including the deletion of a flagellar gene cluster of 31 kb. Also, the prophage regions were deleted with 47 and 49 colonies out of 50 being correct, a remarkably high efficiency considering the size of 36 kb and 39 kb, respectively (Fig. 3b). We sequenced the target regions for four colonies per target and found the desired sequence in all clones, with a single exception. In case of vnp1, one clone missed four bp directly adjacent to the deleted sequence (Supplementary Fig. S3a). It was shown previously that the prophages are activated spontaneously at low frequencies³⁹. It is tempting to speculate that this deviation from the desired sequence is the result of a spontaneous loss of the prophage region, which would still confer resistance to the CRISPR-based counterselection, rather than the targeted deletion by natural transformation.

As described above, the targets for deletion were selected because we hoped that their deletion would improve plasmid transformation efficiency of V. natriegens. We tested all single deletions but did not see any significant increase in transformation efficiency (Supplementary Fig. S4).

One feature of CRISPR-based systems is its inherent modularity. Simultaneous targeting of multiple sequences is possible by simply including multiple gRNAs (Fig. 3c). Assembly of a plasmid harboring all required components to target three different loci is achieved by firstly constructing the gRNA expression cassette and secondly by integrating them into a plasmid already carrying all remaining components (Supplementary Fig. S5a). We tested this approach by simultaneously deleting three of the genomic regions which could be efficiently removed individually. We obtained striking results for the simultaneous deletion of the pili operon, xds and galE, with 24 out of 50 tested colonies carrying all three deletions (Fig. 3d). As an even bigger challenge, we successfully deleted the flagellar gene cluster and both prophage regions simultaneously in two out of 50 colonies (Fig. 3e). The deleted regions sum up to 106 kb, equaling ~2% of the V. natriegens genome. We obtained these results using just 10 ng of each tDNA. For the simultaneous deletion of the pili operon, xds and galE, increasing the amount of tDNA to 100 ng for each target resulted in 44 out of 50 colonies (88%) carrying all three deletions (Supplementary Fig. S6). The killing efficiency with three gRNAs was found to be similar or slightly higher than that with the individual gRNAs (Supplementary Fig. S7), suggesting that undesired recombination events within the sequences identical in all gRNA expression cassettes (P_tet and gRNA scaffold) did not occur at perturbing frequencies. The current design allows construction of NT-CRISPR plasmids with two to five gRNAs (Supplementary Fig. S5b), even though we note that successful deletion of more than three loci was not demonstrated within the scope of this study.

Point mutations

In addition to deletions, we tested NT-CRISPR for the introduction of point mutations. The three genes, malQ, araA and glpK, involved in catabolism of the alternative carbon sources maltose, arabinose and glycerol, respectively, were chosen as proof-of-concept targets for the introduction of point mutations. The introduction of a premature stop codon by a single point mutation can be easily identified because this prevents V. natriegens from growing on minimal medium with the respective carbon source. In case of malQ, we randomly selected 50 colonies yielded by the NT-CRISPR approach and tested them for their ability to grow on M9 minimal medium with maltose as the sole carbon source. None of the tested colonies could grow indicating successful genome modification (Fig. 4a). Subsequently, the target sequence of four colonies was sequenced and the introduction of the desired point mutation was confirmed for all tested colonies. In case of araA and glpK, a first attempt to introduce a point mutation was not successful and resulted in very few colonies, none of them carrying the desired edit (Supplementary Fig. S8a, b). High editing efficiency of 100%, based on the inability of cells from the tested colony to grow on minimal medium with the respective carbon source, could be achieved by introducing a second silent point mutation, leading to a C–C mismatch (G>C mutation) (Fig. 4a). It was shown previously that V. natriegens has an active methyl-directed mismatch repair (MMR) pathway, preventing the efficient introduction of point mutations by natural transformation²⁰. It is known that C–C mismatches inhibit MMR in a wide range of other organisms^43,44,45.Our data suggest that this is also the case in V. natriegens. The approach to introduce a C–C mismatch in addition to the desired mutation can serve as a convenient solution when MMR hinders successful introduction of certain point mutations. Again, we confirmed the point mutations as well as the additional C–C mismatch by sequencing the target regions of glpK and araA (Supplementary Fig. S3b, c).

Integrations

Lastly, we wanted to test if NT-CRISPR is also applicable for integrations into the genome. The approach presented here could be a powerful tool to fuse fluorescent reporter genes to any gene of interest to study their expression or the localization of their gene product. We selected genes that are expected to differ in expression during different growth phases in a batch culture to follow their expression dynamics by measuring the reporter signal. We fused the coding region of the red fluorescent protein mScarlet-I⁴⁶ to the 3′ end of four genes which are known to be differentially regulated during transition into and in the stationary phase of E. coli. As candidates for genes with high expression level in the exponential growth phase, we chose two ribosomal protein-encoding genes. We picked rpsB and rplS because fusion of fluorescent reporter proteins to the respective ribosomal proteins was possible in E. coli without noticeable detrimental effects on growth or ribosome assembly⁴⁷. We chose hisG, involved in histidine biosynthesis⁴⁸, and rpoS, encoding the stress sigma factor σ^38 49, as candidates for genes upregulated in stationary phase. A control strain was generated by first creating an mScarlet-I transcription unit with a strong constitutive promoter (J23111) and a strong RBS (B0030). Then this construct was integrated into an intergenic region with neighboring genes in convergent orientation (int9).

Counterselection for successful integration can be performed through a gRNA, which overlaps the integration site so that successful modification disrupts the gRNA binding sequence and thereby confers resistance against CRISPR-based counterselection. In case of rpoS and rplS, no suitable PAM sequence was available at the desired integration site. As a workaround, we introduced a silent point mutation 300 bp from the integration site. This point mutation was used for the counterselection, expecting the integration of mScarlet-I when the silent point mutation was present. For each integration, 50 randomly selected colonies were screened by colony PCR. When a gRNA was available for direct counterselection at the integration site, we reliably obtained high editing efficiencies with almost all tested colonies being correct (Fig. 4b). In contrast, for rpoS and rplS, editing efficiencies were drastically lower with just 25 and four out of 50 colonies carrying the desired mScarlet-I integration, respectively. Sequencing DNA from four colonies each, which were negative in the PCR screening, revealed that all clones carried the selected silent point mutation, suggesting that not the full-length tDNA fragment was incorporated through homologous recombination. It remains to be investigated if introduction of the silent point mutation closer to the actual integration site than the 300 bp tested here, might lead to a higher editing efficiency. We additionally note that the size of the integrated sequence is relatively short with ~700 bp and ~1300 bp for the translational fusions with mScarlet-I and the integration of the constitutive expression cassette, respectively. Within the scope of this project, we did not evaluate if larger sequences, e.g., sequences encoding full metabolic pathways, can be integrated with similar efficiencies. The generated constructs were tested for their growth behavior and mScarlet-I signal. No growth difference was observed between all tested strains (Fig. 4c). This is also the case for the two strains carrying mScarlet-I fusions to the essential ribosomal genes rpsB and rplS²⁸, which is in accordance with similar experiments in E. coli⁴⁷. The mScarlet-I signal of the tested strains showed the expected growth phase dependency. The rpsB-mScarlet-I and rplS-mScarlet-I strains displayed almost identical courses of fluorescent signal, suggesting a very similar regulation of both ribosomal protein genes. Both strains showed a steady increase in reporter-mediated fluorescence signal in the time window between 2 and 6 h, before reaching a plateau. In contrast, the rpoS-mScarlet-I strain showed an increasing signal after 4 h with an even steeper rise after 6 h. A constant increase from the onset of the stationary phase (~2 h) and throughout the stationary phase was observed for the strain carrying the constitutively expressed mScarlet-I cassette (Fig. 4d). The data are shown as relative mScarlet-I/OD₆₀₀ to account for strong differences in the absolute signal strength (Supplementary Fig. S9). Unfortunately, the signal mediated by hisG-mScarlet-I was hardly measurable and did not allow for quantitative characterization (Supplementary Fig. S10a, b). The results presented here serve as an example of how NT-CRISPR can be used for the construction of reporter strains to study the regulation of important physiological processes.

Overcoming PAM dependency with engineered Cas9

The use of Cas9 is limited by the availability of the PAM sequence, NGG (N = any nucleotide) in case of the commonly used Cas9 from Streptococcus pyogenes. While there are always plenty of possible gRNAs available for larger deletions, introduction of specific point mutations or genomic integrations into a desired locus might be restricted when no PAM can be found nearby, thus requiring inefficient workaround solutions (see generation of rpoS-mScarlet-I and rplS-mScarlet-I above). In recent years, substantial progress has been made in developing new Cas9 variants with a wider PAM spectrum^50,51,52. We tested the near-PAMless Cas9 variant SpG Cas9⁵¹ with the PAM requirement being NGN. In theory, each G and C nucleotide in the genome can serve as a PAM, thereby largely expanding the number of available gRNAs.

We introduced the described mutations into the cas9 sequence in the NT-CRISPR plasmid and tested it with gRNAs targeting wbfF with all possible PAM doublet pairs (NGG, NGA, NGC and NGT). The wbfF gRNA with a NGG PAM showed similar killing efficiencies when used with Cas9 and SpG Cas9, confirming the general compatibility of SpG Cas9 with NT-CRISPR. The tested gRNAs with alternative PAM sequences showed a wide range of killing efficiencies from 99.993% (NGT) and no significant effect for (NGC) (Fig. 5a). We note that other determinants, apart from the PAM sequence, can influence the killing efficiency and the limited number of tested gRNAs does not allow for formulation of general claims about the applicability of SpG Cas9 with alternative PAMs in the framework of NT-CRISPR in V. natriegens. However, it appears as if the killing efficiencies were far lower than the obtained values for the many gRNAs tested before with NGG PAMs in our study (cf. Figure 2d), except for the tested gRNA using NGT as a PAM. Based on the high killing efficiency which we observed with a gRNA using NGT as a PAM, we designed two additional gRNAs overlapping the 3′ end of rpoS and rplS and also measured high killing efficiencies for these (Fig. 5a). Thereafter, we used SpG Cas9 together with these two gRNAs for the integration of mScarlet-I at the 3′ end of rpoS and rplS. For 50 randomly selected colonies, successful mScarlet-I integration was confirmed for 44 and 50 colonies for rpoS and rplS, respectively, compared to 25 and 4 for the indirect approach described above (Fig. 5b). In conclusion, the killing efficiency with SpG Cas9 and alternative PAM sequences tends to be lower than Cas9 with NGG PAMs. Nevertheless, our results suggest that applying SpG Cas9 together with gRNAs using NGT as a PAM could be a suitable strategy for the integration of sequences when no gRNA with NGG is available at the desired target sequence.