A dual gene-specific mutator system installs all transition mutations at similar frequencies in vivo

Abstract Targeted in vivo hypermutation accelerates directed evolution of proteins through concurrent DNA diversification and selection. Although systems employing a fusion protein of a nucleobase deaminase and T7 RNA polymerase present gene-specific targeting, their mutational spectra have been limited to exclusive or dominant C:G→T:A mutations. Here we describe eMutaT7transition, a new gene-specific hypermutation system, that installs all transition mutations (C:G→T:A and A:T→G:C) at comparable frequencies. By using two mutator proteins in which two efficient deaminases, PmCDA1 and TadA-8e, are separately fused to T7 RNA polymerase, we obtained similar numbers of C:G→T:A and A:T→G:C substitutions at a sufficiently high frequency (∼6.7 substitutions in 1.3 kb gene during 80-h in vivo mutagenesis). Through eMutaT7transition-mediated TEM-1 evolution for antibiotic resistance, we generated many mutations found in clinical isolates. Overall, with a high mutation frequency and wider mutational spectrum, eMutaT7transition is a potential first-line method for gene-specific in vivo hypermutation.


INTRODUCTION
Directed evolution is a powerful approach that mimics natural evolution to improve biomolecular activity (1,2). Traditional directed evolution relies on in vitro gene diversification such as error-prone PCR or randomized oligonucleotide pools (2). In contrast, continuous directed evolution (CDE) adopts in vivo hypermutation, allowing for simultaneous gene diversification, selection, and replication in cells; this technique significantly enhances the depth and scale of biomolecular evolution (3)(4)(5). As random mutagenesis in the genome is highly deleterious to cells, in vivo hypermutation methods should aim to introduce mutations in a relatively narrow region around the target gene (5).
The deaminase-T7RNAP system was first reported in bacteria (MutaT7) (17) and further extended to mammalian cells (TRACE) (18), yeast (TRIDENT) (21) and plants (22). We previously demonstrated that the mutation frequency of MutaT7 could be enhanced 7-to 20fold with a more efficient cytidine deaminase, Petromyzon marinus cytidine deaminase (PmCDA1) (20). This Pm-CDA1 T7RNAP mutator (previously termed eMutaT7, but here renamed eMutaT7 PmCDA1 ) generated ∼4 mutations per 1 kb per day in Escherichia coli, representing the fastest gene-specific in vivo mutagenesis. The major limitation of eMutaT7 PmCDA1 is a narrow mutational spectrum: it mainly generates C→T mutations on the coding strand and, with the Shoulders group's dual promoter/terminator approach that induces transcription in both directions, introduces C→T and G→A mutations (C:G→T:A) (17,20). Mutations could be expanded to A→G and T→C mutations (A:T→G:C) with engineered tRNA adenosine deaminases, TadA-7.10 (11,19) and yeTadA1.0 (21), but they either had a mutation frequency much lower than eMutaT7 PmCDA1 (19), or presented C:G→T:A as dominant mutations (∼95%) in nonselective conditions when combined with PmCDA1 T7RNAP (21).
Here, we report on eMutaT7 transition , a new dual mutator system that introduces all transition mutations (C:G→T:A and A:T→G:C) at comparable frequencies. The eMutaT7 transition system uses two mutators, eMutaT7 PmCDA1 and eMutaT7  . The latter is the fusion of T7RNAP and a recently evolved E. coli adenosine deaminase, TadA-8e (23), which had much higher mutational activity than the previously evolved TadA-7.10 (11). We optimized the expression of the two mutators and a uracil glycosylase inhibitor, and demonstrated that the frequencies of the C:G→T:A and A:T→G:C mutations were not significantly different. Furthermore, overall mutation frequency was not markedly reduced. eMutaT7 transition also promoted rapid continuous directed evolution of antibiotic resistance with various transition substitutions, suggesting that it is a viable alternative for gene-specific in vivo hypermutation with an improved mutational spectrum.

Materials
All PCR experiments were conducted with KOD Plus neo DNA polymerase (Toyobo, Japan). T4 polynucleotide kinase and T4 DNA ligases were purchased from Enzynomics (South Korea). Plasmids and DNA fragments were purified with LaboPass TM plasmid DNA purification kit mini and LaboPass™ Gel extraction kit (Cosmogenetech, South Korea). Sequences of all DNA constructs in this study were confirmed by Sanger sequencing (Macrogen, South Korea and Bionics, South Korea). Antibiotics (carbenicillin, chlo-ramphenicol, kanamycin), arabinose, and Isopropyl ␤-D-1-thiogalactopyranoside (IPTG) were purchased from LPS solution (South Korea). Streptomycin was purchased from Sigma Aldrich. Tetracycline was purchased from Bio Basic. Cefotaxime and ceftazidime were purchased from Tokyo chemical industry (Japan). H-p-Chloro-DL-Phe-OH (p-Cl-Phe) was purchased from Bachem (Switzerland).
All plasmids expressing variants of mutators or targets (mutation, deletion, and insertion) were constructed using the site-directed mutagenesis PCR method (25). Plasmids expressing eMutaT7 PmCDA1 and UGI in different conditions (deletion of UGI, an optimized ribosomal binding site (RBS) for UGI, or a constitutive promoter for UGI) were made on pHyo094. Sequence of the optimized RBS region is AACAGAGCGCGCTCTGTTTGAGTACTAGCAAT AAATAAGGAGGATTTTTT (the underlined sequence indicates RBS) (26). Plasmids harboring TadA-8e were made on pDae029. Plasmids expressing PmCDA1 TadA-8e T7RNAP with different linkers were constructed on pDae036.
For evolution of antibiotic resistance, a target plasmid (pGE158) was constructed from pHyo245, which contains the pheS A294G gene between dual promoter/terminator pairs in a low-copy-number plasmid (20): Ampicillin resistance gene in pHyo245 was replaced with tetracycline resistance gene and pheS A294G was replaced with the TEM-1 gene by IVA cloning. Tetracycline resistance gene was amplified from the plasmid pREMCM3 (27) and the TEM-1 gene was obtained from pHyo182 (20).
W3110 ΔalkA Δnfi strain (cDJ085) and W3110 ΔlacZ::KanR-P T7 -gfp-T T7 (cDJ092) were constructed by homologous recombination method (28) The alkA and nfi genes in W3110 were replaced with the streptomycin resistance gene and the kanamycin resistance gene, respectively. The lacZ gene in W3110 was replaced with the kanamycin resistance gene and gfp gene. 30 g/ml of streptomycin or kanamycin was used for selection. Proper gene deletion was confirmed by colony PCR using 2X TOP simple TM DyeMIX-Tenuto (Enzynomics).

PAGE 3 OF 11
Nucleic Acids Research, 2023, Vol. 51, No. 10 e59 In vivo hypermutation Three biological replicates of W3110 or the Δung strain (cHYO057) harboring a mutator plasmid and a target plasmid (pHyo182, pDae117, pDae118, and pDae119 for a single promoter) were grown overnight in LB medium with 35 g/ml chloramphenicol and 50 g/ml carbenicillin (cycle #0). On the following day, the overnight cultures were diluted 100-fold in a fresh LB medium supplemented with 35 g/ml chloramphenicol, 50 g/ml carbenicillin, 0.2% arabinose, and 0.1 mM IPTG in a 96-deep well plate (Bioneer, South Korea) and incubated at 37 • C with shaking (cycle #1). Bacterial cells were diluted every 4 hours and this growth cycle was repeated up to 20 times for accumulation of mutations. At the end of cycle, a fraction of cells were stored at -80 • C with 15% glycerol. To identify mutations in the target gene, cells at cycle #20 were streaked on LB-Agar plates with 35 g/ml chloramphenicol and 50 g/ml carbenicillin. Three or six colonies were randomly chosen for isolation of target plasmids. The target genes in the purified target plasmids were sequenced by Sanger sequencing. Mutations were counted in the region between 147-bp upstream and 138-bp downstream of the pheS A294G gene (total 1269 bp), malE gene (total 1389 bp), and gfp gene (total 1005 bp). Primer 314 and 315 were used for amplification and sequencing of the target gene that has a single promoter system.

PheS A294G suppression assay
Suppression frequency of the pheS A294G toxicity was determined as previously described (20). Cells obtained at the endpoint of each cycle (overnight culture for cycle #0) were diluted to OD 600 ∼0.2. Serial 10-fold dilutions of cells (5 l) using LB broth were placed on YEG-agar plates with or without additives (16 mM p-Cl-Phe, 0.2% arabinose, and 0.1 mM IPTG) and grown overnight at 37 • C. On the following day, the number of colonies on each condition was counted to calculate the suppression frequency. The suppression frequency was calculated as N 1 /N 0 (N 1 : colony forming unit (CFU) in the p-Cl-Phe plates and N 0 : CFU in plates without p-Cl-Phe).

Assays for cell viability and off-target mutagenesis
Cell viability and off-target mutagenesis were assayed as previously described (20). Overnight cultures of the cells harboring the plasmid expressing eMutaT7 TadA-8e , no mutator, or MP6 were diluted 100-fold in LB supplemented with 35 g/ml chloramphenicol and grown to a log phase (OD 600 = 0.2-0.5) at 37 • C. Cells were diluted to OD 600 ∼0.2 and serial 10-fold dilutions of cells (5 l) using LB broth were placed on LB-agar supplemented with 35 g/ml chloramphenicol and 0.2% arabinose. After overnight growth at 37 • C, the number of colonies on the plates were counted to calculate CFU/ml. To evaluate the off-target mutagenesis via rifampicin resistance, cells taken at cycle #0 and cycle #20 were grown to log phase in LB supplemented with 35 g/ml chloramphenicol and 50 g/ml carbenicillin, and subjected to viability assay on plates with or without rifampicin (50 g/ml).

Fluctuation analysis
Fluctuation analysis was performed as previously described (29). Cells harboring a mutator plasmid (pDae079, eMutaT7 transition ) and a target plasmid (pHyo182 for a T7 promoter; pDae120 for constitutive promoter (BBa J23100)) were grown overnight in LB medium with 35 g/ml chloramphenicol and 50 g/ml carbenicillin. The cultures were diluted 1:10 6 with induction media containing 35 g/ml chloramphenicol, 50 g/ml carbenicillin, 0.2% arabinose and 0.1 mM IPTG, and divided into 32 wells (50 L each) in a 96-deep well plate. This plate was sealed and incubated for 6 hours (pHyo182) or 16 hours (pDae120) at 37 • C with shaking. To assess the total cell counts, 8 cultures were resuspended and plated on a YEG-agar plate at the required dilutions. The remaining 24 cultures were resuspended using a pipette, and placed on YEG-agar plates with additional ingredients (16 mM p-Cl-Phe, 0.2% arabinose, and 0.1 mM IPTG). Colonies on YEG-agar plates with or without additives were counted after an overnight incubation.
The Ma-Sandri-Sarkar (MSS) maximum likelihood method was used to compute the loss-of-function mutation rate (30), and the 95% confidence intervals (CIs) were calculated as previously described (29). The FAL-COR webtool (https://lianglab.brocku.ca/FALCOR) was used with both of these methods (31). The calculated loss-of-function mutation rates serve simply as comparative estimates for per-base-pair mutation rates in our study.

High-throughput sequencing and data analysis
Cells taken at cycle 0 and cycle 20 were sequenced as previously described (20). Cells taken at cycle 0 (n = 1) and cycle 20 (n = 3) were grown in 15 ml of LB broth without arabinose and IPTG, and the plasmids were extracted with Plasmid DNA Miniprep Kit. The 3288 bp DNA fragments containing the pheS A294G gene were amplified using primer 512 and 513 covering from 999 bp upstream from T7 promoter and to 1020 bp downstream from T7 terminator. The 2 × 151 paired-end sequencing library was constructed using TruSeq Nano DNA Kit and were sequenced using NovaseqTM (Illumina; operated by Macrogen).
The quality of the sequencing data was checked with FastQC (v0.11.8). Raw reads were trimmed to remove adapter sequences and low-quality end sequences using Trimmomatic (v0.38) (32). Processed data were aligned to the reference sequence (3288 bp) using Burrows-Wheeler Aligner (BWA v0.7.17) with MEM mode and BAM files generated by mapping were sorted using SAMtools (v1.9) (33,34). Sorted BAM files were subject to SAMtools mpileup to obtain a pileup output with maximum depth option, which was set as total number of trimmed reads, and output tag list option consisting of DP, DP4 and AD. Alleles for each locus were called using BCFtools (v1.9), which was a set of utilities of SAMtools package, with multiallelic-caller option. Allele count for each allele and ratio (each allele count/total allele count) were calculated based on AD information of VCF files.

Statistical analysis
For high-throughput sequencing data ( Figure 4 and Supplementary Figure S6), Mann-Whitney test (unpaired Wilcoxon test) was used to assess the significance of the substitution frequency caused by the eMutaT7 transition system. Calculation was conducted using Stata (USA). Statistical significance was determined with P values. P < 0.05 was considered significant for this experiment. For other data, statistical analyses comparing groups in pairs were performed using two-sided Mann-Whitney test (Figures 1-3, Supplementary Figures S1C, S3B, and S5B) without assuming that the data follow normal distribution or twotailed Student's t-test (Supplementary Figure S1B) assuming that the data follow normal distribution. Calculation was conducted using GraphPad prism 5. P < 0.05 was considered significant.

TEM-1 evolution and identification of the evolved mutants
TEM-1 evolution experiments were performed as previously described (20). Strains were grown in LB medium supplemented with 6 g/ml tetracycline, 35 g/ml chlo-ramphenicol, 0.2% arabinose, and 0.1 mM IPTG. Cells were grown without selection pressure at the initial cycle. Then, multiple cultures were grown with different concentrations of an antibiotic (cefotaxime and ceftazidime) at the same time and the culture grown at the highest antibiotic concentration (OD 600 > 1) were used for the next round of evolution. After the final cycle, the target plasmids were purified and re-inserted into fresh W3110 cells harboring the T7RNAP-expressing plasmid (pHyo183) for validation of antibiotic resistance. Twelve colonies were randomly selected for MIC measurement and those with high MIC values (five colonies with 400-1600 g/ml MIC for CTX, three colonies with 4000 g/ml MIC for CAZ) were subjected to the target gene sequencing by Sanger method.

MIC determination
MIC values were measured as previously described (20). Cells were grown overnight in LB medium supplemented with 6 g/ml tetracycline, 35 g/ml chloramphenicol. They were diluted 10 000-fold into fresh LB broth with increasing concentrations of antibiotics (2-fold) in 96-deep well plates, and grown at 37 • C with shaking (290 rpm) overnight. Final cell density (OD 600 ) was measured by M200 microplate reader (TECAN, Switzerland).

RESULTS AND DISCUSSION
eMutaT7 TadA-8e promotes rapid gene-specific in vivo hypermutation To date, TadA-8e is the most efficient TadA variant, presenting a rate constant (k app ) 590 times higher than that of the previous TadA-7.10, and has been successfully used for genome editing (23). To evaluate their efficiency in gene-specific in vivo hypermutation, we fused TadA-7.10 and TadA-8e to the N-terminus of T7RNAP, creating eMutaT7 TadA-7.10 and eMutaT7 TadA-8e , respectively (Figures 1A, 1 and 2). As in the previous characterization of eMutaT7 PmCDA1 (20), we expressed the mutator and induced hypermutation in the target gene, pheS A294G, which was inserted between T7 promoter and T7 terminator in a low-copy-number plasmid. We determined mutational suppression of the pheS A294G toxicity by counting viable cells in the presence of p-chloro-phenylalanine (p-Cl-Phe), which is toxic to cells containing intact pheS A294G. We performed 20 rounds of in vivo hypermutation (4 h growth and 100-fold dilution to a new medium for a single round) without p-Cl-Phe and then sampled cells at different time points for the cell viability assay. We found that the suppression frequencies of eMutaT7 TadA-8e were several orders of magnitude higher than the eMutaT7 TadA-7.10 frequencies after 8 h, indicating that eMutaT7 TadA-8e induces genespecific hypermutation much faster than eMutaT7 TadA-7.10 ( Figure 1B).
To examine whether eMutaT7 TadA-8e generates mutations in the target gene, we randomly selected three clones from cells that had undergone 20 rounds of hypermutation and sequenced the target gene by Sanger method. We also included as negative controls cells that had an empty vector, expressed TadA-8e without T7RNAP, or contained the eMutaT7 TadA-8e plasmid without induction ( Figures 1A, 3-5). Notably, we found ∼6.7 substitutions per clone in the eMutaT7 TadA-8e -expressing cells, while eMutaT7 TadA-7.10expressing cells and negative controls did not exhibit mutations ( Figure 1C and Supplementary Figure S1A). This mutation frequency is definitely much higher than that of eMutaT7 TadA-7.10 and only 2.4-fold lower than that of eMutaT7 PmCDA1 (20). Interestingly, we identified nine A→G (45%) and eleven T→C (55%) mutations on the coding strand, indicating that eMutaT7 TadA-8e causes mutations on both DNA strands ( Figure 1C and Supplementary Figure S1A). We observed that eMutaT7 TadA-8e neither noticeably reduced cell viability (Supplementary Figure S1B) nor induced rifampicin resistance (Supplementary Figure S1C). This result suggests that eMutaT7 TadA-8e does not generate significant off-target mutations in the genome.

Deletion of genes associated with hypoxanthine repair does not significantly increase eMutaT7 TadA-8e activity
In the eMutaT7 PmCDA1 system, deletion of a gene encoding a uracil-DNA glycosylase (UNG) enhanced the mutation frequency (20). UNG removes uracil (deaminated cytosine) and initiates the base excision repair pathway (35). Likewise, we hypothesized that the deletion of genes encoding hypoxanthine (deaminated adenine)-removing enzymes  would further increase the eMutaT7 TadA-8e -mediated mutation frequency. We prepared a strain in which two genes involved in hypoxanthine repair, nfi (36,37) and alkA (38), are deleted and analyzed eMutaT7 TadA-8e -mediated hypermutation ( Figure 2A). Twenty rounds of targeted hypermutation revealed that the mutation frequency in the Δnfi ΔalkA strain did not increase significantly from the wild-type level (11 and 7.2 substitutions per clone on average, respectively) ( Figure 2B and Supplementary Figure S2). Because a DNA repair enzyme often reduces the mutation rate by more than an order of magnitude (39)(40)(41)(42) and the construction of a gene deletion strain requires additional experimental steps, we concluded that the Δnfi ΔalkA strain has no obvious advantage over the wild-type strain for eMutaT7  . No significant increase of mutations in the Δnfi strain was also previously observed (19).

Optimized expression of uracil glycosylase inhibitor increases eMutaT7 PmCDA1 activity
Although we co-expressed a UNG inhibitor (UGI) with eMutaT7 PmCDA1 from the plasmid pHyo094, we did not obtain an efficiency level that matched the Δung strain (20). Proper UGI expression can greatly expand eMutaT7 PmCDA1 utility by avoiding the ung deletion. To enhance UGI activity, we initially tested a new constitutive promoter for ugi or a triply fused protein, UGI PmCDA1 T7RNAP. However, both were less efficient than the Δung strain (Supplementary Figure S3). Next, we optimized the ribosomal bind-ing site (RBS) of ugi (26) (Figure 2C), and obtained a suppression frequency indistinguishable from that of the Δung strain ( Figure 2D). Thus, we were able to avoid the ung deletion for efficient eMutaT7 PmCDA1 -mediated mutagenesis.

Dual expression system introduces all transition mutations at comparable frequencies
We examined whether the two deaminases could simultaneously install both C:G→T:A and A:T→G:C mutations at similar frequencies. Initially, we tested two triplefused proteins, PmCDA1 TadA-8e T7RNAP and TadA-8e PmCDA1 T7RNAP, in which two deaminases were attached to the N-terminus of T7RNAP in different orders ( Figure 3A, 2 and 3). Sequencing of clones after 20 rounds of in vivo hypermutation revealed that Pm-CDA1 TadA-8e T7RNAP installed more A:T→G:C mutations (84%) than C:G→T:A (16%), whereas TadA-8e PmCDA1 T7RNAP generated more C:G→T:A (96%) than A:T→G:C (4%) ( Figure 3B and Supplementary Figure S4A). This result indicates that the deaminase closer to T7RNAP is more active. Shorter or longer linker lengths between enzymes did not significantly reduce the gap (Supplementary Figure S5).
Next, we tested the expression of two mutators, eMutaT7 PmCDA1 and eMutaT7 TadA-8e , from a single plasmid ( Figure 3C, 4-7). The pDae079 plasmid, in which the eMutaT7 TadA-8e gene is located in front of the eMutaT7 PmCDA1 gene, yielded the same amounts . The ratios of A:T→G:C and C:G→T:A frequencies adjusted by subtracting frequencies at cycle 0 as well as the substitution frequencies were shown above. Y-axes above and below 0.0001% are in log-and linear-scale, respectively. P values were obtained with two-sided Mann-Whitney tests and presented as -log 10 (P value). of A:T→G:C (50%) and C:G→T:A (50%) mutations (P = 0.87; Figure 3D and Supplementary Figure S4B). In contrast, the pDae080 plasmid, which reversed the order of the two mutators, disproportionately generated C:G→T:A (85%) over A:T→G:C (15%) (P = 0.012; Figure  3D and Supplementary Figure S4B). As expected, weaker UGI expression without the optimized RBS significantly reduced C:G→T:A mutations in the wild-type strain (P = 0.0046; Figure 3D and Supplementary Figure S4B) but produced comparable numbers of mutations in the Δung strain (A:T→G:C, 38%; C:G→T:A, 62%; P = 0.37; Figure 3D and Supplementary Figure S4B). We thus selected pDae079 for eMutaT7 transition , which on average installed 5.7 transition mutations in the 1269-bp gene during 80-hour in vivo hypermutation.

High-throughput sequencing demonstrates that eMutaT7 transition introduces all transition mutations at similar frequencies
To further dissect the eMutaT7 transition -mediated in vivo hypermutation, we used next-generation sequencing (NGS) to analyze the sequences of ∼3.3 kb DNA fragments around the target region from mixed pools of cells taken at cycle 0 (n = 1) or cycle 20 (n = 3). We found that, among all substitution types, all four transition substitutions were significantly accumulated at cycle 20 ( Supplementary Figure S6); the adjusted average substitution frequencies (fre-quency differences between cycle 0 and cycle 20) were 0.28% for A→G, 0.22% for T→C, 0.046% for G→A, and 0.41% for C→T, respectively. We further dissected the 3.3 kb DNA into three regions--upstream, target gene, and downstream. Among them, the target gene showed the highest level of adjusted transition substitution frequencies (0.52%, 0.54% and 0.54%, respectively) than the upstream (0.023%, 0.023% and 0.021%) and the downstream (0.089% 0.090% and 0.088%) regions ( Figure 4A and B). This result supports the gene-specific mutagenesis of eMutaT7 transition . As previously observed with eMutaT7 PmCDA1 (20), the downstream region showed higher leakages of gene targeting than the upstream region. Given the very low rifampicin resistance frequencies of eMutaT7 PmCDA1 (20) and eMutaT7 TadA-8e (Supplementary Figure S1C), however, we believe that eMutaT7 transition does not generate high level of off-target mutations in the genome.
The average number of eMutaT7 transition -mediated substitution in the target gene was 6.7 (1269 bp × 0.53%) in NGS analysis, closely recapitulating the result from Sanger sequencing (5.7 substitutions; Figure 3D, 5). The high-throughput sequencing data also confirmed that eMutaT7 transition generates comparable amounts of A:T→G:C and C:G→T:A substitutions, whose average ratio was 1:0.93 ( Figure 4B). Taken together, NGS analysis corroborated that eMutaT7 transition rapidly introduces all transition mutations on the target gene at comparable frequencies.

Additional analyses demonstrate high mutational activity and target tolerance of eMutaT7 transition
To further estimate the mutational activity of eMutaT7 transition , we performed fluctuation analysis for the loss-of-function of pheS A294G (29). We used two target plasmids in which the target gene is controlled by either T7 promoter or an unrelated constitutive promoter. The conversion of the pheS A294G loss-of-function colony counts to loss-of-function mutation rate with the FAL-COR webtool (31) resulted in 3.6 × 10 −5 loss-of-function mutations per generation with T7 promoter and 2.4 × 10 −9 loss-of-function mutations per generation without T7 promoter, indicating 15 000-fold increase of the mutation rate with the proper targeting of eMutaT7 transition ( Figure 5A and Supplementary Figure S7). This result suggests that eMutaT7 transition indeed has a high mutational activity.
We also tested target tolerance of eMutaT7 transition by using different target genes in various genetic contexts.
We initially performed the 20-cycle in vivo hypermutation of two additional genes (malE and gfp) encoding maltose binding protein (MBP) and green fluorescent protein (GFP), respectively, as well as pheS A294G with or without eMutaT7 transition . We found that these three genes contained comparable numbers of transition mutations (average 6.3, 7.2 and 5.0 substitutions in pheS A294G, malE and gfp, respectively), whereas the pheS A294G gene without eMutaT7 transition displayed no mutation (Figure 5B and Supplementary Figure S8A). This result suggests that the presence of pheS A294G itself does not induce hypermutation and that the high mutational activity of eMutaT7 transition is not limited to our model gene, pheS A294G. We also tested the conditions in which T7controlled gfp is inserted in the chromosome ( Figure 5C), malE and gfp are located in a single target plasmid (Figure 5D), or malE and gfp are positioned in a target plasmid and chromosome, respectively ( Figure 5E). We found that three conditions lead to total 7.0, 4.5 and 6.0 mutations, Although majority of these results showed comparable numbers of A:T→G:C and C:G→T:A substitutions (Supplementary Figure S8A-C), one experiment almost exclusively showed A:T→G:C (Supplementary Figure S8D), indicating that only eMutaT7 TadA-8e was active during hypermutation. Because we used the same DNA sequence of T7RNAP for two mutator genes, the deletional recombination of these two mutator genes might generate the eMutaT7 TadA-8e -only mutator plasmid. Indeed, we found that the mutator plasmid obtained from the cycle #20 of this sample was shorter than the original eMutaT7 transition plasmid (Supplementary Figure S8E), suggesting that eMutaT7 transition needs to be improved for longer in vivo hypermutation experiments.

eMutaT7 transition evolves TEM-1 with various transition mutations
We previously demonstrated that eMutaT7 PmCDA1 promoted rapid continuous directed evolution of TEM-1 for resistance against third-generation cephalosporin antibiotics, cefotaxime (CTX) and ceftazidime (CAZ) (20). Here, we tested eMutaT7 transition in the same way. We used the dual promoter/terminator approach to install both C→T and G→A mutations (17,20). By sequentially increasing antibiotic concentrations during multiple rounds of in vivo hypermutation, we elevated minimum inhibitory concentrations (MICs) from 0.05 to 400-1600 g/ml in 80 h for CTX ( Figure 6A) and from 0.4 to 4000 g/ml in 48 h for CAZ ( Figure 6B).
In conclusion, this study described a new mutator system that combines eMutaT7 PmCDA1 and eMutaT7 TadA-8e , called eMutaT7 transition . This new system has advantages over previous deaminase-T7RNAP mutators. First, eMutaT7 transition expands the mutational spectrum to all transition substitutions (C:G→T:A and A:T→G:C). eMutaT7 PmCDA1 can mediate 8.4% of all amino acid e59 Nucleic Acids Research, 2023, Vol. 51, No. 10 PAGE 10 OF 11 changes (32 out of total 380 changes), but eMutaT7 transition expands them to 19% (74 changes). Accordingly, we observed in TEM-1 evolution experiments several A:T→G:C substitutions that have been previously identified in clinical or laboratorial isolates. Although transition substitutions nominally compose only a small fraction of all amino acid changes, they generally appear more frequently in natural variants, explaining approximately two-third of single nucleotide polymorphisms in several species (52)(53)(54). Second, all transition substitutions are produced at similar frequencies. This outcome was made possible by the use of two efficient deaminases, PmCDA1 and TadA-8e, along with appropriate expression of the two mutators and a DNA glycosylase inhibitor. In contrast, TRIDENT generated considerably more C:G→T:A substitutions (∼95%) in yeast (21).
Future research should aim to include transversion mutations in the mutational spectrum without significantly sacrificing substitution frequencies. Additionally, eMutaT7 transition would be improved to suppress the deletional recombination for longer in vivo hypermutation experiments; either the different DNA sequences of T7RNAP for two mutators or the recently reported TadA variants that can mutate both cytidine and adenine simultaneously (55,56) may enhance its property. With its good substitution frequencies and wider mutational spectrum, we believe that eMutaT7 transition or its improved variants can become the method of choice in synthetic biology studies requiring evolutionary approach, particularly in evolution or engineering of enzymes, metabolic pathways, or gene circuits.

DATA AVAILABILITY
Illumina sequencing data have been deposited in the ArrayExpress database at EMBL-EBI (www.ebi.ac.uk/ arrayexpress) under accession number E-MTAB-12258. Other data that support the findings of this work can be found in the paper and in the Supplementary Data files. Protein and primer sequences are listed in supplementary tables. All E. coli strains and plasmids described in this work are available upon request. The pDae029 (eMutaT7 TadA-8e ), pDae069 (eMutaT7 PmCDA1 ), and pDae079 (eMutaT7 transition ) have been deposited and are available through Addgene (#187620 for pDae029; #187621 for pDae069; #187622 for pDae079).