Accelerated Adaptive Evolution of Gram‐Negative Bacteria With a Broad‐Host‐Range Dual Base Editor

ABSTRACT

Adaptive laboratory evolution (ALE) is a powerful strategy for exploring functional solution spaces, yet its pace is fundamentally constrained by low spontaneous mutation rates. Here, we present a broadly applicable genetic device that accelerates genome diversification in Gram‐negative bacteria through the transient action of a dual base editor (ACBE3). This construct combines cytidine and adenosine deaminase activities with a single‐stranded DNA‐binding module, enabling efficient genome‐wide mutagenesis without reliance on host‐specific replication or repair systems. The expression of ACBE3 increased mutation rates by up to four orders of magnitude in Escherichia coli and Pseudomonas putida , outperforming classical random mutagenesis approaches such as UV irradiation and chemical treatment. Whole‐genome sequencing revealed extensive and uniformly distributed mutations, predominantly consistent with deamination signatures, as well as a broader spectrum of nucleotide changes and small indels. The system displayed strong orthogonality, functioning across diverse bacterial taxa. As a proof of concept, coupling ACBE3‐mediated diversification with selective growth conditions enabled the rapid evolution of Acinetobacter calcoaceticus P320, yielding mutant strains with enhanced growth on n‐dodecane within only a few induction cycles. This approach decouples diversification from selection, enabling the controlled exploration of adaptive landscapes at unprecedented speeds. While high mutational loads impose cellular burdens and require tight regulatory control, the platform offers a versatile route to accelerate microbial evolution at both single‐strain and community levels. Our results thus document dual base editing as a powerful framework for fostering evolution in microbial biotechnology.

We developed a broad‐host‐range dual base editor that transiently induces genome‐wide hypermutation, accelerating adaptive laboratory evolution by up to four orders of magnitude. The platform rapidly generates improved bacterial phenotypes across diverse Gram‐negative species, enabling faster evolutionary engineering for biotechnology.

1. Introduction

Mutation, recombination, and horizontal gene transfer (HGT) confer exceptional genetic plasticity to bacteria, enabling them to evolve and adapt to various environmental changes (Levin and Cornejo 2009). The same mechanisms have been applied in biotechnology through adaptive laboratory evolution (ALE), a strategy that takes advantage of natural evolutionary dynamics to guide microbial populations toward desired phenotypes. By cultivating bacterial populations over many generations under specific environmental constraints, ALE facilitates the enrichment of beneficial mutations—point mutations, insertions, deletions, gene duplications, or regulatory rewiring—which enhance microbial fitness (Barrick and Lenski 2013; Phaneuf et al. 2020; Hirasawa and Maeda 2022). Selection pressures such as nutrient limitation, physicochemical stress, toxic compounds, or recalcitrant substrates drive the exploration of the fitness landscape toward optimized solutions. Because ALE typically operates on single strains under a specific selection, the key player in ALE is spontaneous mutation and natural diversification within a fixed genetic background—rather than acquisition of new functions by HGT. ALE has been applied most intensively to Escherichia coli (Peng et al. 2025) but its versatility makes it applicable across highly diverse physiological backgrounds in environmental, industrial, and pathogenic microbes, including lactic acid bacteria, Corynebacterium glutamicum , Bacillus, Pseudomonas, and many others (Espeso et al. 2020; Hirasawa and Maeda 2022; Wang et al. 2022; Dvořák et al. 2024). Yet, the main drawback of ALE is the long experimental timelines (Lee et al. 2011). Even when automation platforms have emerged to address this issue (Ghiaci et al. 2024), weeks or months of intensive handling are necessary to obtain the desired phenotypes, which has hampered its application in the context of bacterial consortia. Typically, ALE runs on slow genomic diversification driven by spontaneous mutations in bacteria, typically in the range of 1E‐8 mutations per generation/base pairs (He et al. 2025). Therefore, increasing the mutation rates of bacteria could enhance genomic diversification and thus accelerate the evolutionary adaptation of the strain(s) at stake. Multiple strategies have been deployed to boost genome‐level mutagenesis, including exogenous mutagens (UV irradiation, alkylating agents), mutator strains, and plasmid‐borne mutator devices. Chemical mutagenesis and UV are effective (Alcantara‐Diaz 2004; Arshad et al. 2010) but problematic to apply in continuous evolution as they exhibit high and host‐specific toxicity. Mutator strains—via disruption of mismatch repair (MMR) components or conditional expression of mutator MMR alleles—have been established in E. coli , P. putida , and B. subtilis (Caillet‐Fauquet et al. 1984; Nyerges et al. 2014; Aparicio et al. 2020; Tanneur et al. 2025), yielding high mutation rates. However, these approaches require substantial prior knowledge of the MMR systems of the target bacteria. Genome‐wide hypermutagenesis can also be achieved by interfering with the DNA replication fidelity machinery (Badran and Liu 2015) or by fusing the cytidine base editor PmCDA1 to the E. coli RNA polymerase α‐subunit (Eom et al. 2021). However, note that, while powerful, these strategies depend on determinants (e.g., RNA polymerase subunits, dnaQ926) that are strain‐ or species‐specific. Other methods use orthogonal replisomes with dedicated error‐prone DNA polymerases (e.g., T7, φ29) (Diercks et al. 2025; Rehm et al. 2026) to hyper‐diversify DNA; however, they act on defined episomal stretches and are not suitable for genomic diversification.

In this study, we present an orthogonal, broadly applicable genetic tool to increase mutagenesis and genomic variability in a wide range of Gram‐negative bacteria, with potential use also at the community level. Candidate genes encoding activities capable of triggering genome‐scale hypermutagenesis were sought according to the following criteria: (i) direct activity on DNA to avoid host‐specific dependencies; (ii) high catalytic performance to maximize mutagenic output; and (iii) high orthogonality to ensure broad‐host‐range applicability. We selected cytidine deaminases (CDAs) and adenosine deaminases (ADAs) as mutator modules because they (i) catalyse direct deamination of DNA, (ii) exhibit potent activities, and (iii) have been used for genome editing across a wide diversity of organisms.

CDAs and ADAs convert, respectively, cytidine to uridine and adenosine to inosine. Upon DNA replication, these edits produce C → T (cytosine to thymine) and A → G (adenine to guanine) transitions (Figure 1). These activities have been instrumentalized to deploy different genetic tools—either fused to Cas9 (base editors, BEs), T7 polymerase, or RNA polymerase—for genetic editing and hypermutagenesis toolkits in a broad variety of organisms, including bacteria, yeast, mammalian cells, and plants (Nishida et al. 2016; Moore et al. 2018; Banno et al. 2018; Álvarez et al. 2020; Eom et al. 2021; Velázquez et al. 2022). Their cross‐kingdom portability and efficiency make them promising candidates for genome‐scale DNA mutagenesis in diverse bacteria. In this work, we evaluated the mutagenic performance of two CDAs (PmCDA1 and rApoBEC1) and one ADA (ABE9e), and engineered a modular dual base editor (ACBE3) that combines rAPOBEC1, ABE9e, and a Single‐Stranded DNA‐binding protein. While the term base editor (BE) typically refers to protein conjugates of nucleotide deaminases and Cas9, in this work it is applied more broadly to non‐targeted activities over DNA. Our novel dual editor increased mutagenesis by ~1E4‐fold in several Gram‐negative bacteria, generating extensive population‐level genomic variability. We further demonstrated the utility of ACBE3 in accelerating ALE approaches via the rapid evolution of the n‐dodecane degrader Acinetobacter calcoaceticus P320. The value of this approach for accelerated evolution of yeast has been recently documented in Pichia pastoris (Pan et al. 2026). Potential applications at both the single‐strain and community levels are also discussed.

FIGURE 1.

Enzymatic transformations of base editors. Cytidine De‐aminases (CDAs) remove the amino group from the nitrogenous base, rendering uridine. Adenosine Deaminases (ADAs), upon amino group removal, render inosine. During DNA replication, these enzymatic activities translate into C → T and A → G mutations, respectively, as indicated. Genes encoding CDAs used in this work (_r apoBEC1 from rat, pmCDA1 from sea lamprey) and the ADA ABE9e (evolved version of E. coli tadA) are also sketched to the right of the reactions.

2. Materials and Methods

2.1. Strains and Media

Strains used in this work (Table S1) were routinely grown in LB broth (Sambrook et al. 1989) with shaking (170 rpm) at 30°C with the exception of E. coli (37°C). V. natriegens was cultured at 37°C in LBS (modified LB media at 20 g L⁻¹ NaCl). M9 mineral media were prepared according to (Sambrook et al. 1989). DSM457 mineral media (Brunner + trace elements) were prepared according to the DSMZ directions. Citrate 0.2% and n‐dodecane (Sigma‐Aldrich 8.20543) were added to the mineral media as the sole carbon source when stated. The medium was amended with a homemade vitamin cocktail diluted 1000× when specified (Biotin 20 mg L⁻¹, Pyridoxine 10 mg L⁻¹, Riboflavin 50 mg L⁻¹, D‐pantothenic acid 50 mg L⁻¹, P‐aminobenzoic acid 50 mg L⁻¹, Folic acid 20 mg L⁻¹, Thiamine 50 mg L⁻¹, and vitamin B12 50 mg L⁻¹). Solid media were prepared by adding 2% agar to the liquid media. When stated, antibiotics were included at the following concentrations: Chloramphenicol (Cm) at 30 μg mL⁻¹; Streptomycin (Sm) was used at different concentrations depending on the sensitivity of the strain at stake: 500 μg mL⁻¹ for Burkholderia sp. JS872, 200 μg mL⁻¹ for V. natrigens, 100 μg mL⁻¹ for P. putida KT2440, P. aeruginosa PA01, Xanthomonas sp. and Cupriavidus sp. UYMMa02a, 50 μg mL⁻¹ for E. coli , 10 μg mL⁻¹ for A. calcoaceticus P320. Rifampicin (Rif) was also used at different concentrations depending on the strain: 50 μg mL⁻¹ for E. coli and V. natrigens and 100 μg mL⁻¹ for Burkholderia sp. JS872, P. putida KT2440, P. aeruginosa PA01, Xanthomonas sp., Cupriavidus sp. UYMMa02a and A. calcoaceticus P320. Bacterial cultures were induced with dicyclopropyl ketone (DCPK) at 0.05% or with 3‐methyl‐benzoate (3MB) at 1 mM.

2.2. General Procedures and Primers

DNA manipulations were performed using routine laboratory techniques (Sambrook et al. 1989) and manufacturer recommendations. The QIAprep Spin Miniprep Kit (Qiagen) was used for plasmid purification, and bacterial whole‐genome isolations were performed using the DNeasy UltraClean Microbial Kit (Qiagen). Primers were obtained from Sigma‐Aldrich and are listed in Table S2. Q5 polymerase (New England Biolabs) was used for PCR amplification intended for plasmid construction, while diagnostic PCRs were performed with DNA Amplitools Master Mix (Biotools). The Nucleospin Gel and PCR Clean‐up Kit (Macherey‐Nagel) was used to purify the PCR amplicons. Plasmid DNA sequencing was outsourced to Macrogen‐Spain/PlasmidSaurus, and whole‐genome sequencing was performed by Novogene/PlasmidSaurus. The results were analysed using LaserGene and Geneious software. Transformation of E. coli laboratory strains was carried out with chemically competent cells using the CaCl₂ method (Sambrook et al. 1989). Plasmids were delivered to environmental strains via tri‐parental mating (de Lorenzo and Timmis 1994) and trans‐conjugant selection was performed in solid M9 minimal media supplemented with 0.2% w/v citrate and Sm. For V. natriegens trans‐conjugant selection, minimal media was supplemented with Sm and 2% NaCl to support optimal growth.

2.3. Plasmid Construction

Single Base Editor genes _r apoBEC1 (0.7 Kb) and pmCDA1 (0.6 Kb) were amplified, respectively, with primer pairs APOBEC1‐Fw/Rv and pmCDA1‐Fw/Rv from plasmid templates 221‐Ptet‐rAPOBEC1‐T7RNAPol and 221‐Ptet‐pmCDA1‐T7RNAPol (Velázquez et al. 2022) and cloned EcoRI/BamHI in pSEVA429. ABE9e (0.5 Kb) was synthesized by Twist Bioscience, amplified with ABE9e‐Fw/Rv, and cloned with BamHI/SalI in pSEVA429. Using the same DNA templates, double Base Editors were constructed as follows. ACBE1: _r apoBEC1 and ABE9e were amplified with APOBEC1‐Fw/Rv2 and ABE9e‐Link‐APO‐Fw/ABE9e‐Rv2, respectively. Both amplicons were fused by sewing PCR with external primers, and the product (1.2 Kb) was cloned EcoRI/BamHI in pSEVA429. ACBE2: ABE9e and _r apoBEC1 and were amplified, respectively, with ABE9e‐Fw3/Linker‐ABE9e‐Rv3 and Linker‐APO‐Fw/APO‐Rv2, both amplicons fused by sewing PCR with external primers and the product (1.2 Kb) cloned EcoRI/BamHI in pSEVA429. ACBE3: _r apoBEC1 and ABE9e were amplified with APOBEC1‐Fw/Linker‐APO‐Rv and Linker‐ABE9e/ABE9e‐Rv3, respectively, while the P. putida KT2440 ssb gene (PP_0485) was amplified by colony PCR with ssb‐Fw/Rv (0.5 Kb). _r apoBEC1 and ssb amplicons were fused by sewing PCR with external primers, and the product (1.2 Kb) was fused with the ABE9e amplicon by another round of sewing PCR with APOBEC1‐Fw/ABE9e‐Rv3. The final product (1.8 Kb) was cloned EcoRI/SphI in pSEVA428. All clonings were performed in E. coli DH5a competent cells, selected in LB‐Sm, checked by miniprep/restriction, and fully sequenced to ensure sequence accuracy. Plasmids used in this work are listed in Table S1. The DNA sequences and protein translations of all BEs constructed in this study can be found in the Supporting Information.

2.4. Rifampicin Assay

The mutation rate of the strains analysed in this study was estimated by quantifying the emergence of rifampicin‐resistant (Rif^R) colonies (Krašovec et al. 2019). Cultures were grown overnight in LB (wild‐type strains; LBS for V. natriegens ) or in LB‐Sm (LBS‐Sm for V. natriegens ) for plasmid‐bearing strains. Overnight cultures were then back‐diluted to an OD₆₀₀ of ~0.1 in 3 mL of the corresponding medium and incubated for 18 h under induction with 0.05% DCPK or 1 mM 3MB. To determine the total viable counts, appropriate serial dilutions were plated on a rich solid medium (LB/LBS for wild‐type strains; LB/LBS‐Sm for plasmid‐bearing strains). In parallel, dilutions were plated on rifampicin‐containing agar to quantify resistant mutants (LB/LBS‐Rif for wild‐type strains and LB/LBS‐Sm‐Rif for plasmid‐bearing strains). CFU counts were used to calculate the absolute mutant frequency as the ratio of Rif^R CFU/total viable CFU, and the results were reported as Rif^R CFU per 1 E9 viable cells. Three independent biological replicates were performed, and the data are presented as means with standard deviations. To identify the point mutations underlying Rif resistance in E. coli DH5α and P. putida KT2440, the rpoB gene was PCR‐amplified from 20 to 30 independent Rif^R colonies using the primer pairs rpoB‐coli‐Fw/Rv ( E. coli ) and rpoB‐F/R ( P. putida ). PCR products were purified and Sanger‐sequenced using the same primers, and the sequences were aligned with the corresponding wild‐type rpoB reference to identify mutations. The impact on Cell Survival (ICV) was determined for E. coli /pSEVA428‐ACBE3 and P. putida KT2440/pSEVA428‐ACBE3 in independent experiments (three replicas): overnight cultures on LB‐Sm were back‐diluted to OD₆₀₀ ~ 0.1 (20 mL of the same media), incubated until OD₆₀₀ ~ 0.5, then split into two aliquots of 3 mL. One aliquot was added with 3MB 1 mM. Induced and uninduced samples were incubated for 3 h, and culture dilutions were plated in LB‐Sm, and CFUs were counted after overnight incubation. ICV was calculated from the CFU counts as the ratio between treated and control samples and expressed as the percentage of CFUs.

2.5. Ultraviolet (UV) and N‐Methyl‐N′‐Nitro‐N‐Nitrosoguanidine (MNNG) Treatments

E. coli DH5α and P. putida KT2440 were grown overnight in LB medium and subsequently diluted to OD₆₀₀ ~ 0.1 in 20 mL of LB. Cultures were incubated until the mid‐exponential phase (OD₆₀₀ ~ 0.6–0.7). For UV exposure, 1 mL of culture was transferred to a 24‐well plate and exposed at 15 cm of distance to UV light from a 254 nm lamp (VL‐6MC, VilberLourmat, France; 400 μW cm⁻²) during 240 s (UV dose 96 mJ cm⁻²). Immediately after irradiation, 1 mL of LB was added, and the sample was transferred to a 10 mL tube and incubated for 3 h under standard growth conditions for recovery. For MNNG treatment, 2 mL of culture was transferred to a microcentrifuge tube, centrifuged for 1 min at 11,000 rpm, and the pellet was resuspended in 1 mL 1× PBS supplemented with 30 μL MNNG (34 mM stock), yielding a final concentration of ~1 mM. Samples were incubated for 30 min at room temperature, centrifuged for 1 min at 11,000 rpm, resuspended in 2 mL LB, and incubated for 3 h under standard growth conditions for bacterial recovery. Control samples were processed in parallel under identical conditions, omitting UV exposure or MNNG addition. After a 3 h recovery period, the treated and control cultures were serially diluted and plated on LB agar, followed by overnight incubation to determine the number of viable cells. The same cultures were then further incubated for 18 h, and appropriate dilutions were plated on LB and LB‐Rif to estimate the mutagenesis frequency. ICV (Impact on Cell Survival) was calculated from CFU counts as the ratio between treated and control samples and expressed as a percentage of CFUs. The mutagenesis rate was determined as the ratio of Rif^R to total viable cells (absolute mutant frequency) and reported as Rif^R CFU per 1E9 viable cells.

2.6. Diversification of P. putida KT2440 and Genome Analysis

P. putida KT2440 carrying pSEVA428‐ACBE3 was grown overnight in 20 mL LB supplemented with streptomycin and 3MB (LB‐Sm‐3MB). The resulting culture was back‐diluted to an OD₆₀₀ of ~0.1 in 20 mL of the same medium and incubated overnight. This procedure was repeated for ten consecutive cycles, after which the cells were plated on LB agar containing streptomycin and rifampicin (LB‐Sm‐Rif). Five single colonies were randomly selected, grown in LB‐Sm, and genomic DNA (gDNA) was extracted for sequencing by Novogene (Illumina, 150× depth on average). Raw reads were analysed using Geneious software with the P. putida KT2440 genome and pSEVA428‐ACBE3 sequence as references. SNP detection was performed using the following thresholds: minimum variant frequency, 0.9; maximum variant p‐value, 1E−6; and minimum strand bias p‐value, 1E−5. Single nucleotide polymorphisms (SNPs) identified in the genome were classified by SNP type, and the relative abundance of each change was calculated and reported as a percentage of the total SNPs. Plasmid sequence variations were described separately for each analysed clone.

2.7. Accelerated Evolution of A. calcoaceticus P320

A. calcoaceticus P320/pSEVA428‐ACBE3 was first inoculated from a glycerol stock in 20 mL of Brunner‐Citrate‐Sm. After overnight growth, cultures were back‐diluted to OD₆₀₀ ~ 0.1 in the same media and induced with 3MB 1 mM for 5 h (Step I—diversification). The cultures were then washed to remove residual citrate by spinning down the cells at 4000 rpm/10 min and resuspending the pellet in 10 mL Brunner media (2X). The final suspension was back‐diluted to OD₆₀₀ ~ 0.1 in 20 mL of fresh Brunner media added with vitamins, trace elements, 100 μL of n‐dodecane (4 g L⁻¹), and Sm 5 μg mL⁻¹ (lower Sm concentration was used during growth in n‐dodecane because excessive cellular burden was observed at 10 μg mL⁻¹). Cultures were incubated for 3 days at 30°C with shaking (Step II‐ Selection), washed as before, and inoculated again in Brunner‐Citrate‐Sm‐3MB to start a new evolution cycle. Control A. calcoaceticus P320 wild‐type strain was subjected to the same procedure but omitting the antibiotic and the inducer in the culture medias. After each evolution cycle (Step I and II), dilutions were plated in solid Brunner media (without antibiotics) with 100 μL of n‐dodecane on top of the media. The plates were incubated at 30°C for 4 days and inspected visually. One large colony emerging from the plasmid‐containing strain was picked and isolated by streaking twice on the same medium without antibiotic. Then, a single colony (P320^evo) was grown on LB media and a glycerol stock was stored at −80°C. Growth dynamics of evolved and wild‐type bacteria were determined by culturing the strains overnight in Brunner‐Citrate, washing 2× the cultures twice as explained above, back‐diluting the suspension in 20 mL of Brunner‐Citrate or Brunner/n‐dodecane at the stated concentrations, and monitoring OD₆₀₀. Genome sequencing was outsourced to PlasmidSaurus, and SNPs were analysed using the company on‐line application.

2.8. n‐Dodecane Quantification Assay

Improvement of degradation performance of A. calcoaceticus P320 evolved and wild type strains was determined as follows. The test was conducted in 100 mL glass vials, sealed with PTFE silicon caps, using 10 mL of Brunner Mineral Medium (MM) supplemented by 1–5‐10 g/L dodecane (Sigma‐Aldrich). The strains were grown overnight in LB broth, centrifuged (4000 rpm, 5 min), and resuspended in physiological solution (0.9% NaCl). Both strains were inoculated at 10⁶ cell/mL final concentration and incubated at 30°C under shaking. Negative control, represented by abiotic non‐inoculated vials, was established. After 3 days (1 g L⁻¹ C12), 4 days (5 g L⁻¹ C12) or 7 days (10 g L⁻¹ C12), 5 replicate vials were analysed per each strain. To each vial, 10 mL of hexane (equal volume of the aqueous phase) were added, followed by shaking and sonication for 10 min. A volume of 4 mL of the organic phase was transferred to glass vials, dehydrated with anhydrous Na₂SO₄, and stored at −20°C. The n‐dodecane was quantified by gas chromatography using a Shimatzu GC‐2030 gas chromatographer equipped with a flame ionization detector (250°C) and a Polyethylene Glycol (PEG) Mega‐Wax (30 m × 0.25 mm) column, injecting an extract volume of 1 μL. A calibration curve was constructed using a n‐dodecane standard (> 99% purity, Sigma‐Aldrich) for GC quantification. Degradation percentage was calculated as follows: (initial n‐dodecane concentration − residual n‐dodecane concentration)/initial n‐dodecane concentration × 100. Outliers have been removed when detected by applying the ‘outliers’ package in R. Significance of the statistical differences observed comparing the parental wt strain and P320^evo was calculated by t‐test using Excel.

3. Results and Discussion

3.1. Selection and Mutagenic Efficacy of Base Editors

Most CDAs currently used for genome editing (Figure 1) are based on pmCDA1 from the sea lamprey Petromyzon marinus (Rogozin et al. 2007) and _r apoBEC1 from Rattus norvegicus (Harris et al. 2002). To our knowledge, the only active ADAs reported so far are a family of derivatives of the E. coli gene tadA, including tadA7, tadA8, and tadA9 (Gaudelli et al. 2017; Richter et al. 2020; Yan et al. 2021; Ranzau et al. 2023). As mentioned before, these BEs have been used for genome editing in a variety of organisms, including bacteria, yeasts, and mammal/plant cells, either guided by Cas9 to modify specific DNA targets or fused to T7‐RNA‐Polymerase to hyper‐diversify local DNA regions downstream of a T7 promoter. On this basis, we aimed to test the ability of different CDAs and ADAs to promote broad genomic mutagenesis when expressed as single proteins. The CDAs pmCDA1 and _r apoBEC1 were selected and the tadA9 variant described by Yan et al. (2021) and later instrumentalized as ABE9e by Shelake et al. (2023) was the selected ADA, here designated ABE9e (Figure 1). The three selected base editors (BEs) were PCR‐amplified and cloned into the expression vector pSEVA429 (Sm^R; RK2 origin of replication; alkS/P _alkB expression system inducible by dicyclopropyl ketone [DCPK]). Following the transformation of the resulting constructs (pSEVA429‐pmCDA1, pSEVA429‐ _r apoBEC1, and pSEVA429‐ABE9e; Figure 2A) and the empty vector into E. coli DH5α and Pseudomonas putida KT2440, we employed a Rifampicin Assay as a proxy for mutagenesis. Numerous mutations in rpoB (encoding the β subunit of RNA polymerase) are known to confer resistance to rifampicin (Severinov et al. 1994; Jatsenko et al. 2009), enabling a quantitative readout of the mutagenic rate (Rif^R mutants per viable cell), together with a qualitative analysis of the specific genomic changes underlying the resistant phenotype (Figure 2B). Briefly, plasmid‐bearing strains were grown overnight in LB supplemented with streptomycin (LB‐Sm), inoculated into fresh LB‐Sm‐DCPK at OD₆₀₀ ~ 0.1, and incubated for 18 h. Appropriate dilutions were then plated on LB‐Sm (for viable counts) and LB‐Sm‐Rif (to score rpoB mutants). Control strains lacking plasmids were processed identically but without streptomycin in the culture. After recording the CFU counts, the mutagenic activity was calculated as the ratio of Rif^R mutants to viable cells. The rpoB gene from 20 to 30 Rif^R colonies was PCR‐amplified, sequenced, and analysed. The results plotted in Figure 2C show that the controls (no plasmid or empty pSEVA429) exhibited basal mutagenesis of approximately 1E‐8 Rif^R mutants per viable cell in both E. coli DH5α and P. putida KT2440. The expression of the cytidine deaminases PmCDA1 and rApoBEC1 elevated these values by approximately two orders of magnitude in E. coli and four orders in P. putida . The adenosine deaminase ABE9e increased mutagenesis by one order of magnitude in E. coli and two orders in P. putida . Mutation profiling indicated high accuracy for rApoBEC1 in both strains (100% of rpoB changes were the expected C → T transitions from cytidine deamination), whereas PmCDA1 performed comparably well in KT2440 but yielded 87% C → T changes in E. coli . ABE9e exhibited high specificity in KT2440, with 92% of A → G transitions corresponding to adenosine deamination, while demonstrating lower accuracy (68%) in E. coli . Collectively, these results support the capacity of single cytidine or adenosine deaminases to drive potent and accurate genome‐wide mutagenesis in two bacterial‐model organisms. Therefore, we fused ABE9e with a CDA into a single polypeptide to generate a dual base editor that combined both activities. _r apoBEC1 was selected as the CDA module owing to its higher efficiency and accuracy and because this CDA has been reported to impose low toxicity in bacterial cells (Harris et al. 2002).

FIGURE 2.

Rif assays of the single base editors _r apoBEC1, pmCDA1 and ABE9e. (A) Plasmid constructs scheme: the three BEs were cloned in pSEVA429 (SmR, RK2 ori, alkS/P _alkB expression system‐DCPK inducible). (B) Rifampicin Assay: upon overnight expression of the base editors in E. coli and P. putida KT2440, cultures were plated in LB‐Sm (viable CFUs) and in LB‐SmRif (rpoB mutants). Control strains with empty pSEVA429 were analysed in the same way. Plasmid‐free strains were grown overnight and plated in LB (viable CFUs) and in LB‐Rif (rpoB mutants). The rpoB gene of 20–30 Rif^R colonies from each experiment was PCR amplified and sequenced. (C) Mutagenic output was calculated and plotted as Rif^R CFUs per 1E9 viable cells in all conditions tested. Absolute mutagenesis (Rif^R mutants per viable cell) is also shown over the bars. Data represent the mean and standard deviation of three biological replicas. Analysis of rpoB SNPs is shown below, depicting the percentage of mutations arising from cytidine de‐aminations (C → T), adenosine de‐aminations (A → G), and other mutations.

3.2. Mutagenic Efficacy of Engineered Dual Deaminases

As illustrated in Figure 3A, we constructed several configurations of chimeric dual base editors: 5′‐ _r apoBEC1 + ABE9e‐3′ (Adenosine–Cytidine Base Editor 1, ACBE1) and 5′‐ABE9e +  _r apoBEC1‐3′ (Adenosine–Cytidine Base Editor 2, ACBE2), incorporating an 18‐aa flexible linker between the coding sequences, both cloned into pSEVA429. Given that nucleotide deaminases act on single‐stranded DNA (ssDNA) (Branton et al. 2020), the third configuration (ACBE3) included the coding sequence of the Single‐Stranded Binding protein from P. putida KT2440 (ssb ^KT, PP_0485). SSB proteins are highly conserved in bacteria and bind ssDNA non‐specifically, tightly coating exposed DNA strands to prevent secondary structure formation and degradation, thereby facilitating efficient DNA replication, recombination, and repair (Marceau 2021). We reasoned that the non‐specific DNA‐binding properties of SSBs could help tether the chimeric editor to the bacterial chromosome, enhancing its ability to modify the DNA. ACBE3 comprised 5′‐ _r apoBEC1 + ssb ^KT + ABE9e‐3′, with flexible linkers between the base editors and ssb ^KT (Figure 3A). This fusion was successfully cloned into pSEVA428 (Sm^R; RK2 origin; xylS/P _m expression system inducible by 3‐methylbenzoate, 3MB) as multiple attempts to clone it into pSEVA429 failed. Although the reason is unclear, the xylS/P _m system exhibits stronger repression than alkS/P _alkB (Calles et al. 2019), which could mitigate the toxicity of the ACBE3 product during the cloning process. Reciprocal fusion (5′‐ABE9e + ssb ^KT +  _r apoBEC1‐3′) was not obtained in either pSEVA429 or pSEVA428, potentially reflecting excessive cellular burden even under uninduced conditions. pSEVA429‐ACBE1, pSEVA429‐ACBE2, and pSEVA428‐ACBE3 were tested in E. coli DH5α and P. putida KT2440 using the same Rif assay described above. Figure 3B shows the quantitative/qualitative results upon induction with DCPK (pSEVA429‐ACBE1 and pSEVA429‐ACBE2) or 3MB (pSEVA428‐ACBE3): ACBE1 increased mutagenesis by two logs in E. coli and three logs in P. putida , with a marked predominance of cytidine deamination (~95%) over adenosine deamination (~4%). ACBE2 yielded similar results in E. coli and produced a four‐log increase in P. putida , still with a noticeable imbalance between the two activities. In P. putida , ACBE3 did not substantially improve the total activity or deamination ratios relative to the other fusions. However, in E. coli, it increased the number of Rif^R mutants by four logs, improving mutagenic potency by nearly 100‐fold over the ACBE1/ACBE2 fusions and achieving a more balanced A → G (23%)/C → T (77%) ratios.

FIGURE 3.

Rif assays of the adenosine–cytidine double base editors ACBE1, ACBE2, and ACBE3. (A) Plasmid constructs scheme: ACBE1 and ACBE2 were cloned in pSEVA429 (SmR, RK2 ori, alkS/P _alkB expression system‐DCPK inducible) while ACBE3 was cloned in pSEVA428 (SmR, RK2 ori, xylS/P _m expression system‐3MB inducible). Flexible linkers between functional modules are shown in red. The ssb gene of P. putida KT2440 (single stranded binding protein). (B) The same Rifampicin Assay and SNP analysis described before was applied upon overnight induction with DCPK (pSEVA429‐ACBE1, pSEVA429‐ACBE2) or with 3MB (pSEVA428‐ACBE3). Mutagenic output was calculated and plotted as Rif^R CFUs per 1E9 viable cells in all conditions tested. Absolute mutagenesis (Rif^R mutants per viable cell) is also shown over the bars. Data represent the mean and standard deviation of three biological replicas. Analysis of rpoB SNPs is shown below depicting the percentage of mutations arising from cytidine de‐aminations (C → T), adenosine de‐aminations (A → G) and other mutations.

We did not fail to notice that uninduced experiments also exhibited considerable mutagenic activity, typically ~10‐fold lower than fully induced conditions (~1E−5 mutants per cell). These values remained high relative to the baseline for E. coli and P. putida (1E−8 mutants per cell), indicating that leaky expression of BEs is sufficient to promote substantial mutagenesis. This is especially notable for ACBE3 (Figure S2), showing narrow dynamic ranges between induced (3.1E−4 and 1.7E−4 mutants per cell for E. coli and P. putida , respectively) and uninduced states (3.5E−5 and 7.3E−5 mutants per cell for E. coli and P. putida , respectively). Given that the xylS/P _m system was reported to be tightly repressed in the absence of an inducer, these results suggest that the chimeric enzyme is highly active and that even low expression levels can drive a mutagenic regime in bacterial cells. This may also reflect the metabolic burden of a highly active base editor and the detrimental effects of widespread DNA lesions in the genome—issues addressed in the next section. Although direct comparisons among the three variants may be confounded by the different expression systems used (alkS/P _alkB for ACBE1/ACBE2 vs. xylS/P _m for ACBE3), all chimeric dual BEs performed robustly in both model strains. We selected ACBE3 based on its ability to catalyse both enzymatic changes at the genomic level and its overall superior mutagenic potency.

3.3. Benchmarking Activity of Dual Base Editor ACBE3

Classical approaches to increase bacterial mutagenesis have largely relied on mutagenic treatments, namely UV irradiation and the application of alkylating agents such as N‐methyl‐N′‐nitro‐N‐nitrosoguanidine (MNNG) (Alcantara‐Diaz 2004; Arshad et al. 2010). To compare these approaches with ACBE3 activity and report the impact of such treatments on cell viability (ICV), E. coli and P. putida KT2440 were exposed to UV light (UV dose 96 mJ cm⁻²) and incubated with MNNG, measuring the mutagenic activity (Rif^R Assays) and the ICV after treatments expressed as the percentage of CFUs observed between the treated and control cultures (Figure 4A—see Materials and Methods for details). This simple analysis measures the number of colonies emerging from each experimental treatment and may reflect either lethality, toxicity, or just delays in the growth rate. A comparison of the mutagenic performance/survival of the same strains expressing the ACBE3 base editor is shown in Figure 4B. UV light did not increase the mutagenesis but had a high impact on cell viability, 2% for E. coli and 18% for P. putida cells. Previous studies in E. coli showed that single doses of UV irradiation had a moderate impact on mutagenesis, while applying iterative cycles of UV exposure and growth increased the numbers by 10 to 50‐fold compared with non‐treated controls (Shibai et al. 2017; Adebali et al. 2025; Ghosh et al. 2026). These studies also highlighted that viability and mutability were negatively related. Most likely, the mutation frequency reported here is hampered by the toxic effect of the ultraviolet doses applied, which attests to the delicate balance between low mutagenesis and high lethality imposed by UV treatments and the inherent limitations to apply it on ALE approaches. In contrast, MNNG increased mutagenesis by ~100‐fold in both organisms, with uneven ICV rates of 26% for E. coli and 69% for P. putida . While MNNG displayed, as expected, a high mutagenic load, the efficiency of ACBE3 overperformed the alkylating agent 140‐fold in E. coli and 18‐fold in P. putida , also showing a differential ICV (25% and 79%, respectively) comparable to that of MNNG. Toxicity mediated by the expression of BEs probably reflects the metabolic burden associated with DNA lesions produced by nucleotide deamination. The differences between P. putida and E. coli , also detected for MNNG and UV light, could be attributed to the higher endurance of P. putida KT2440 to different environmental stresses (Nikel et al. 2021). These results attest to the high potency of the dual base editor to promote mutagenesis, outperforming classical approaches without higher impacts on cell survival.

FIGURE 4.

Mutagenic efficiency of ACBE3 versus UV light and MNNG. (A) Rif and Impact on Cell Viability (ICV) assays were performed on E. coli DH5α and P. putida KT2440 after the treatment with UV light (254 nm, 4 min) and the alkylating agent N‐methyl‐N′‐nitro‐N‐nitrosoguanidine (MNNG 1 mM, 30 min). (B) The results were compared the with the mutagenic output of ACBE3 and the basal rates of mutagenesis calculated before. Mutagenic output was plotted as Rif^R CFUs per 1E9 viable cells in all conditions tested. Absolute mutagenesis (Rif^R mutants per viable cell) is also shown over the bars (black). ICV (red numbers over the bars) was expressed as the percentage of CFUs between treated and untreated bacterial cultures. Data represent the mean and standard deviation of three biological replicas.

3.4. Diversification at the Genomic Level

The Rif^R assay described above is a useful proxy for bacterial mutagenesis across diverse species, but it presents quantitative and qualitative limitations for assessing the deamination‐based strategy proposed here because: (i) the DNA changes in rpoB that yield Rif^R mutants, while diverse, are finite in number; (ii) rpoB displays an intrinsic bias in the availability of C/A targets whose deamination produces Rif^R mutations; and (iii) this bias can be further influenced by strain‐specific differences in genomic G:C content. To overcome these constraints, we characterized in greater detail the genome‐wide changes arising from the activity of the dual base editor ACBE3. To this end, P. putida KT2440 harbouring pSEVA428‐ACBE3 was subjected to cycles of continuous mutagenesis by overnight growth on LB‐Sm‐3MB and refreshing the culture in the same medium for ten consecutive passages. After plating, single colonies were isolated, genomic DNA was extracted, and whole‐genome sequencing was performed. The analysis of five clones is presented in Figure 5A, which reports single‐nucleotide polymorphisms (SNPs) and indels detected in both the bacterial chromosome and resident plasmid. Figure 5B displays a schematic distribution of the mutations along the P. putida chromosome for the five clones. On average, each clone exhibited > 200 genomic mutations uniformly distributed across the bacterial chromosome. No shared mutations were identified among the five clones analysed. As expected for a dual base editor, the predominant transitions C → T and A → G (equivalently G → A and T → C when detected on the complementary strand) accounted for ~60% of all mutations, with cytidine deamination events being more frequent (~37%) than adenosine deaminations (~23%). Notably, these genome‐wide data contrast with the Rif^R assay results, which overestimated the difference between the two enzymatic modules. Genomic analysis indicates that, although _rAPOBEC1 is more active than ABE9e, both modules operate at broadly comparable rates in vivo. In addition to the expected transitions, a spectrum of transversions was detected at lower frequencies: C → G (~6%), A → C (~4%), G → T (~4%), A → T (~2%), T → A (~0.3%), G → C (~1%), and C → A (~0.2%, observed in two of five clones). However, T → G was not detected in clones Col1–Col5. Indels (1–3 nt) were also common, comprising ~8% deletions and ~14% insertions. This phenomenon has been observed as a byproduct of CDAs activity (C → U) in conjunction with uracil N‐glycosylase (ung) cleavage over uracil, leading to the creation of abasic sites that ultimately result in insertions or deletions at the target site (Rees and Liu 2018). Mutagenic activity was also evident in the resident plasmid: in clones Col4/Col5, a premature stop codon disrupted the xylS ORF, whereas in Col1/Col2/Col3, either silent or missense mutations were detected in the same gene. A missense mutation occurred within the _r apoBEC1 ORF (Col3), and silent substitutions were found in trfA (encoding the plasmid replication initiator) in Col2 and Col3. Additionally, a SNP was detected in the T1 transcriptional terminator (Col2). All plasmid‐borne mutations corresponded to deamination‐type changes in the DNA sequence. These results indicate that, as expected, mutagenesis also affects the plasmid during diversification, with xylS being a frequent target. XylS is the transcriptional regulator that controls the P _m promoter, activating transcription upon binding the 3‐methylbenzoate (3MB) inducer. In the absence of inducer, P _m is not recognized by RNA polymerase, and transcription is triggered only when XylS–3MB binds to the promoter sequence (Domínguez‐Cuevas et al. 2008). Some mutations detected in xylS are predicted to yield a defective regulatory system, effectively limiting expression to basal levels—which are nonetheless sufficient to sustain high mutagenic activity, as mentioned previously. Interestingly, no mutations were observed in the base editor coding sequence in any of the analysed samples. This pattern suggests that ACBE3 mutagenesis tends to self‐limit its own expression, likely by imposing selective pressure to reduce the expression of the regulatory circuitry rather than inactivating the editor itself—an adaptive response that may alleviate the metabolic burden associated with high mutational loads under conditions of strong expression. Collectively, these data indicate that ACBE3 induces stochastic, genome‐wide mutations that exhibit a clear bias—though not limited to—its specific deamination activities, thereby facilitating rapid genomic diversification and variability at the population level in bacterial cells. Our observations also underscore the need for tight control of ACBE3 expression, either by employing stringent expression systems or by applying shorter mutagenic pulses to balance the editing efficacy with cellular fitness.

FIGURE 5.

Genome diversification of P. putida KT2440. (A) P. putida KT2440/pSEVA428‐ACBE3 was subjected to ten cycles of mutagenesis. After plating, five single colonies were isolated and whole genome sequencing was performed. SNPs analysis for the genome and the plasmid is shown in the table, including small indels. For plasmid mutations, the gene/sequence affected, the SNP and the amino acid change are depicted. (B) Distribution of detected SNPs and indels (red bars) along the genome (blue line) is depicted for the five clones analysed.

3.5. Host Range of Dual Base Editor

As mentioned, base editors have been successfully applied in a wide variety of organisms for genome editing. To check the applicability of ACBE3 beyond the platform strains E. coli and P. putida KT2440, a set of Gram‐negative species from different bacterial families was selected, and the mutagenic activity was tested. This group of wild‐type strains included common soil dwellers often involved in bioremediation (Burkholderia, Cupriavidus, Acinetobacter), human and plant pathogenic bacteria ( P. aeruginosa, Xanthomonas), and the halophile Vibrio natriegens . These six strains were transformed with the plasmid pSEVA428‐ACBE3, and Rifampicin assays were performed as explained above. Control strains without the plasmid were also assayed to determine the basal mutagenic activity of each bacterial background. Figure 6 shows the results combined with the mutagenic activities for E. coli and P. putida KT2440: in all tested organisms, the expression of ACBE3 base editor raised the basal mutagenic activity by 3–4 logs, with the exception of V. natriegens, which showed a 100‐fold increase, confirming that the device deploys not only a high activity but also a remarkable autonomy in respect to the host's machinery across different bacterial chassis. These results open the possibility of using a single device to accelerate the evolution of a suite of Gram‐negative bacteria in ALE approaches and also to tackle the possibility of manipulating not only single species but also complex bacterial communities.

FIGURE 6.

Host range of the dual base editor ACBE3. Rif assays were performed on different bacterial strains harbouring the pSEVA428‐ACBE3 under full induction and compared with the basal mutagenic activity of the wild‐type strains. Mutagenic rates were calculated and plotted as Rif^R CFUs per 1E9 viable cells in all conditions tested. Absolute mutagenic rates (Rif^R mutants per viable cell) are also shown over the bars. Data represent the mean and standard deviation of three biological replicas.

3.6. Accelerated Evolution of the Alkane Degrader Acinetobacter calcoaceticus P320

As an example of applicability of this approach, we used the base editor ACBE3 to improve the degradation capabilities of Acinetobacter calcoaceticus P320, a wild‐type environmental bacterium that can degrade and grow on n‐dodecane as the sole carbon source. This strain was previously isolated from the rhizosphere of a Dactylis glomerata L. plant collected from the historically PCB (poly‐chlorinated‐biphenyl) contaminated site of Brescia‐Caffaro and characterized for the potential exploitation for rhizoremediation interventions (Vergani et al. 2017). After transforming A. calcoaceticus P320 with pSEVA428‐ACBE3, the base editor activity was confirmed using the Rif assay. As illustrated in Figure 6, induction initiated a hypermutagenic state similar to that seen in other strains analysed in this study, leading to an increase of 3.5 log units compared to the basal mutagenesis. Subsequently, a cyclic two‐step evolution strategy was employed (Figure 7A): in the first step (Step I: diversification), cells were cultured in minimal medium with citrate as the carbon source, and base editor expression was induced using 3MB. The cell cultures were washed and used to inoculate fresh minimal medium supplemented with trace elements, vitamins, and n‐dodecane as the sole carbon source (Step II: selection). The cultures were incubated for 3 days, washed, and reinoculated into citrate minimal medium under 3MB induction to initiate the next cycle. During the more stringent growth phase on n‐dodecane, vitamins were included; however, they were omitted during growth on citrate to prevent mutations in the genetic determinants involved in vitamin biosynthesis. To determine whether this approach accelerated bacterial evolution, the same protocol was applied in parallel to the strain carrying the mutagenic device and the wild‐type strain. This scheme separates genome diversification (Step I), which aims to generate population‐level genetic variability, from phenotypic enrichment under selective conditions (Step II). We hypothesized that mutants with an enhanced ability to grow on n‐dodecane would emerge during Step II after multiple cycles. Because the evolutionary paths required to produce an improved phenotype may vary, such events could occur at different frequencies but are expected to be rare within the mutant pool. Therefore, relying solely on macroscopic culture parameters to assess growth improvements (e.g., reduced lag phase, faster growth, or higher final optical density) would likely necessitate many generations—typically more than a single cycle—and improvements may only become apparent in subsequent rounds. To detect the evolved variants after each cycle, aliquots from Step II were plated onto a minimal agar medium with n‐dodecane as the sole carbon source. Notably, by Cycle II, plate inspection revealed the emergence of large, fast‐growing colonies against the background of small colonies in the A. calcoaceticus P320/428‐ACBE3 population (Figure 7A), whereas the wild‐type control produced a uniform field of small colonies.

FIGURE 7.

Accelerated evolution of Acinetobacter calcoaceticus P320. (A) Evolution scheme. (B) Growth curves of wild‐type and P320^evo strains in mineral media with n‐dodecane 5 g L⁻¹ and citrate as sole carbon sources. Data represent the mean and standard deviation of three biological replicas. (C) Degradation analysis of n‐dodecane (5 g L⁻¹) in wild‐type and P320^evo strains. Data represent the mean and standard deviation of five biological replicas. (D) Genomic mutations detected in A. calcoaceticus P320^evo strain.

A single large colony was picked, streaked twice on the same medium without antibiotic and a single, fast‐growing clone (P320^evo) was finally isolated. To rule out contamination, 16S rRNA gene sequencing was performed for both P320^evo and the parental wild‐type strain, yielding identical sequences. We measured the growth kinetics in a liquid minimal medium using citrate and n‐dodecane 1, 5, and 10 g L⁻¹ as the sole carbon sources (Figures 7B and S1). In n‐dodecane 5 g L⁻¹, the wild‐type strain began to grow after a brief lag phase of approximately 1 day, achieving maximal growth after approximately 6 days, with a final OD₆₀₀ of 1.2. In contrast, P320^evo started growing shortly after inoculation, exhibited faster exponential‐phase dynamics, and reached a final OD₆₀₀ of 2.4, indicating a 75% increase in biomass compared to that of the wild type. In citrate, although P320^evo exhibited marginally superior growth, the variations in growth dynamics and ultimate biomass were considerably less pronounced. The n‐dodecane degradation assays showed that P320^evo exhibits significantly higher degradation percentage compared to the wild‐type strain at all tested concentrations (Figures 7C and S1). In particular, when n‐dodecane was supplemented at 5 g L⁻¹, the degradation increased by 23.9%, raising in 3 days from 29.8% in the wild‐type strain to 53.7% in P320^evo. A similar increase was detected in the degradation assay established with 1 g L⁻¹ n‐dodecane, where degradation percentage shifted from 45.7% in the wild‐type strain to 68.7% in P320^evo. On the opposite, the improvement of degradation performance was limited, even if still significant, after 7 days at the highest concentration tested (10 g L⁻¹), moving from 18.9% in the wild‐type strain to 26.4% in P320^evo (Figure S1), most likely reflecting saturation effects. Whole‐genome sequencing of the wild‐type and evolved strain revealed a circular chromosome of 4.2 Mb and an additional plasmid of 74 Kb in both strains. Notably, P320^evo did not show the pSEVA428‐ACBE3 sequence. A SNP search using the wild‐type sequence as a reference identified only four mutations in P320^evo (Figure 7D): three single point mutations were found in the pas (C → T), rpsL (A → G), and rpoC (C → T) genes, along with a 12‐nt deletion in the dmeF gene. The reported SNPs, consistent with the deamination activity of the dual base editor ACBE3, were missense mutations predicted to cause single amino acid changes in the protein sequence, whereas the deletion in dmeF removed four histidine codons from a large stretch of poly‐his (35 residues). The pas gene (automatic annotation) encodes a polypeptide of 1168 aa, featuring a PAS domain at the amino terminus, a common feature of signalling proteins such as sensor histidine kinases, cyclic‐di‐GMP synthases/hydrolases, and methyl‐accepting chemotaxis proteins (Gu et al. 2000). On the other hand, dmeF orthologs have been reported to be involved in nickel and cobalt efflux systems in Rhizobium (Rubio‐Sanz et al. 2013). The potential impact of these two mutations on the improved phenotype of P320^evo is unclear and most likely represents random changes during the evolution process. However, point mutations in rpsL have been associated with spontaneous resistance to Sm in different bacteria (Hosaka et al. 2004). Streptomycin was used during the evolution procedure to maintain the plasmid pSEBA428‐ACBE3, but the isolated evolved strain P320^evo, while still resistant to the antibiotic, lacked the plasmid. Therefore, it is plausible that this particular clone acquired Sm resistance during the mutagenic procedure and simultaneously lost the hypermutagenic device, which in turn might have limited the mutagenic load. Single mutations in rpoC, which encodes the β subunit of RNA polymerase, have been linked to growth enhancement in E. coli (Conrad et al. 2010; Cheng et al. 2014), P. putida (Espeso et al. 2020), and other bacteria (Xu et al. 2018). Notably, Cohen and Hershberg (2022) discovered that mutations associated with specific traits, such as antibiotic resistance and adaptation to acidic conditions, are localized within defined functional domains of the rpoC gene in E. coli . This phenomenon is likely similar in A. calcoaceticus P320^evo, where a single alteration in rpoC might be specifically linked to growth improvement on n‐dodecane but not on other carbon sources. Although this is beyond the scope of the current study, future research on comparative genomics and the characterization of various mutants could shed light in the adaptive pathways involved in the observed phenotype. Overall, our findings demonstrate the selection of advantageous mutations in response to specific constraints under laboratory conditions within a remarkably short period (2 weeks), resulting in enhanced bacterial phenotypes, such as faster growth, increased biomass and enhanced n‐dodecane degradation performance. This study highlights how rapid evolution through genome diversification can swiftly equip bacteria with improved capabilities in response to specific selective pressures.

3.7. Conclusion

The dual base editor developed here provides a powerful and versatile platform for accelerating microbial evolution by dramatically increasing genome‐wide mutagenesis. Its high catalytic activity and mutational output enable the rapid exploration of adaptive landscapes, generating extensive genetic diversity that extends beyond canonical deamination signatures and facilitates the emergence of beneficial phenotypes within only a few evolutionary cycles. The broad‐host‐range functionality of the system further opens the door to its application not only in single strains but also, potentially, at the level of microbial communities, where controlled diversification could be harnessed to steer the collective functions. At the same time, this very potency poses challenges. The elevated basal activity observed even in the absence of induction, together with the associated cellular burden, may complicate precise control over mutagenesis and trigger self‐limiting dynamics through selection against high‐expression states. These limitations require improved regulatory architectures that allow tighter and more predictable control of editor activity. Future developments will therefore focus on implementing highly stringent and orthogonal expression systems—potentially including temperature‐sensitive (Aparicio et al. 2018) or multi‐layer regulatory circuits (Velázquez and de Lorenzo 2024)—to enable well‐defined pulses of diversification. Such refinements would allow mutagenesis to be deployed on demand, minimizing fitness costs while preserving the exceptional evolutionary acceleration afforded by the system. Nevertheless, even in its current form, the platform represents a substantial advance over existing approaches, enabling rapid, scalable, and broadly applicable evolutionary engineering that was previously difficult to achieve.

Author Contributions

Tomas Aparicio: conceptualization, investigation, methodology, validation, writing – original draft, formal analysis, writing – review and editing. Gabriele Capriglia: investigation, writing – original draft. Francesca Mapelli: investigation, writing – original draft. Sara Borin: conceptualization, writing – original draft. Víctor de Lorenzo: conceptualization, formal analysis, funding acquisition, writing – original draft, writing – review and editing, project administration, supervision, resources.

Funding

This work was supported by HORIZON EUROPE Excellent Science, 101060625.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Growth curves and n‐dodecane degradation analysis of wild‐type and P320_evo strains.

Figure S2: Comparison of mutagenic efficiency of ACBE3 with and without 3MB induction.

Table S1: Bacterial strains and plasmids used in this work.

Table S2: Oligonucleotides used in this study.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.