Continuous Directed Evolution of a Feedback-Resistant Arabidopsis Arogenate Dehydratase in Plantized Escherichia coli

Abstract

Continuous directed evolution (CDE) is a powerful tool for enzyme engineering due to the depth and scale of evolutionary search that it enables. If suitably controlled and calibrated, CDE could be widely applied in plant breeding and biotechnology to improve plant enzymes ex planta. We tested this concept by evolving Arabidopsis arogenate dehydratase (AtADT2) for resistance to feedback inhibition. We used an Escherichia coli platform with a phenylalanine biosynthesis pathway reconfigured (“plantized”) to mimic the plant pathway, a T7RNA polymerase-base deaminase hypermutation system (eMutaT7), and 4-fluorophenylalanine as selective agent. Selection schemes were prevalidated using a known feedback-resistant AtADT2 variant. We obtained variants that had 4-fluorophenylalanine resistance at least matching the known variant and that carried mutations in the ACT domain responsible for feedback inhibition. We conclude that ex planta CDE of plant enzymes in a microbial platform is a viable way to tailor characteristics that involve interaction with small molecules.

Introduction

Classical directed evolution has greatly advanced protein engineering^1,2 but is limited by dependence on labor-intensive cycles of in vitro generation of sequence diversity and screening of variants.^3,4 Classical directed evolution campaigns can thus only handle relatively few evolutionary cycles and independent evolving populations, and tend to incrementally climb local adaptive peaks rather than explore the fitness landscape more widely because they lack the mutational power needed to cross fitness valleys.³⁻⁵ Continuous directed evolution (CDE) systems overcome these limitations of depth and scale by hypermutating the target gene in vivo and, for enzymes, by coupling activity of the target gene to growth of the platform cell.^3,4 CDE systems thus require only serial subculturing because they work by selecting for growth alone, and in principle they can run indefinitely.³⁻⁵ So far, CDE systems are largely confined to Escherichia coli and yeast,⁴ and while progress in CDE in mammalian cells⁶ and plants⁷ continues, long generation times and small library sizes are enduring obstacles.

Tailoring enzyme properties by CDE has many potential uses in plant breeding and plant biotechnology, including developing new herbicide-resistance genes, extending enzyme life, and relieving feedback inhibition of flux-determining enzymes.⁷⁻¹⁰ As CDE in a plant platform is far less powerful than in a microbial one, a preferable strategy is to run CDE on a plant enzyme in E. coli or yeast, then return the improved gene to the plant by genome editing.⁹ This strategy is still at the concept stage. Further, for the E. coli systems, the differences between conditions in plant and prokaryote cells (e.g., metabolic pathway architecture, metabolite levels, redox poise, protein-folding, and degradation systems⁸) and the uncertain durability of the hypermutation machinery in long evolution campaigns^4,5,8 are potential roadblocks. Nor is it clear that E. coli CDE systems work in minimal media, which is critical because many selection schemes require auxotrophs.⁵

We therefore undertook a proof-of-concept study in E. coli to evolve a plant arogenate dehydratase (ADT) (EC 4.2.1.91) for resistance to feedback inhibition by phenylalanine. We used eMutaT7,¹¹ one of several CDE systems whose mutagenesis machinery is a T7RNA polymerase (T7RNAP)-nucleobase deaminase fusion that is directed to the target gene by a T7 promoter.⁴ The eMutaT7 system uses cytidine deaminase (CD), which makes mainly C → T mutations (Figure 1A). ADT was chosen as the target because (i) it requires rewiring (“plantizing”) of the E. coli aromatic amino acid synthesis network to reproduce that in plants (Figure 1B), making the study more demanding and informative, and (ii) a known feedback-resistant ADT variant¹² is available to prevalidate and calibrate selection schemes. Having confirmed that eMutaT7 works in minimal medium, we showed that short evolution campaigns deliver feedback-resistant ADT variants. This opens the way to custom applications of ex planta CDE using eMutaT7 and related systems to evolve plant enzymes to meet various needs.

Figure 1.

EMutaT7 continuous directed evolution system and the plantized E. coli platform strain. (A) eMutaT7 hypermutates a plasmid-borne target gene using a T7RNAP-cytidine deaminase (CD) fusion that is encoded by a second plasmid. (B) Native E. coli (left), Arabidopsis (right), and plantized E. coli (center) phenylalanine synthesis pathways. The native pathways proceed from prephenate via 3-phenylpyruvate in E. coli and via arogenate in Arabidopsis. PheA, bifunctional chorismate mutase/prephenate dehydratase; TyrB, tyrosine aminotransferase; TyrA, chorismate mutase activity of TyrA; AtPAT, Arabidopsis prephenate aminotransferase; AtADT2, Arabidopsis arogenate dehydratase 2. (C) Growth assays in MOPS minimal medium comparing E. coli ΔpheA strains expressing AtPAT alone or with AtADT2. Data are means ± s.e.m. of three independent replicates.

Results and Discussion

Function in Minimal Medium

As eMutaT7 has so far been used only in rich media^4,11 and might be metabolically burdensome enough to prevent growth in minimal media,⁵ we first tested its operation in MOPS minimal medium. In this and all subsequent experiments, an arabinose concentration of 0.2% was used to induce maximal mutator expression.¹¹ A Δung strain was used as platform; the target was the gene encoding PheS A294G, a tRNA ligase that sensitizes E. coli to p-chlorophenylalanine. A campaign of three serial passages in minimal medium (without selection) gave multiple p-chlorophenylalanine-resistant mutants that all contained C → T nonsense mutations in the pheS coding sequence (Table S1). This result confirms that eMutaT7 allows E. coli growth and efficient mutagenesis in minimal medium.

Plantizing the E. coli Platform

E. coli and plants such as Arabidopsis synthesize phenylalanine from prephenate, but by different routes (Figure 1B).¹³E. coli converts prephenate to phenylalanine via 3-phenylpyruvate using the prephenate dehydrogenase activity of PheA and the aminotransferase TyrB, whereas Arabidopsis does so via arogenate using prephenate aminotransferase (AtPAT) and arogenate dehydratase (AtADT2), which phenylalanine feedback-inhibits.¹³ To plantize the E. coli pathway, we deleted pheA and inserted an AtPAT cDNA flanked by the constitutive PJ23101 promoter/B0032 RBS and the rrnB T1 terminator into the flu locus. The AtADT2 gene, expressed from the eMutaT7 pHyo182 vector, completes the plantized pathway. The chorismate mutase activity of E. coli TyrA suffices to replace that of PheA in this pathway. Growth tests on MOPS minimal medium confirmed operation of the plantized pathway and its dependence on AtADT2 (Figure 1C).

Prevalidating the Selection Scheme

We adopted a classical analogue selection scheme¹⁴ using 4-fluorophenylalanine (4FPA), which—like phenylalanine—inhibits plant ADT activity¹⁵ and also inhibits E. coli growth.¹⁶ AtADT resistance to inhibition by 4FPA is predicted to lead to an expanded free phenylalanine pool that competes out 4FPA’s inhibitory effect. To confirm that this is so, we compared growth of the plantized strain expressing wild-type AtADT2 or its feedback-resistant mutant.¹² In line with prediction, growth of the wild-type AtADT2 strain was blocked by a high concentration (5 mM) of 4FPA (Figure 2A), whereas the feedback-resistant AtADT2 strain grew well after a lag (Figure 2B). As 4FPA concentration was reduced below 5 mM, the growth difference between the strains diminished progressively (Figure S1). These results validate the selection scheme. As recommended,⁴ we also ran a pilot test to check that the selection scheme does not yield frequent cheater mutations, e.g., ones not in the target ORF. To this end, we modified the mutator vector to express T7RNAP alone instead of the T7RNAP-CD fusion and compared the frequency of 4FPA-resistant mutant recovery from cultures harboring each vector. The T7RNAP-CD strain gave resistant colonies at a frequency of 6.0 × 10^–7 whereas the T7RNAP strain frequency was only 2.6 × 10^–8 (Table S2). This 23-fold difference further validates the selection scheme.

Figure 2.

Use of the known feedback-resistant AtADT2 variant to prevalidate the planned select-ion scheme. (A) 4FPA-sensitivity of the plantized eMutaT7 strain expressing wild-type AtADT2. (B) Reduced 4FPA sensitivity of this strain expressing the feedback-resistant variant (AtADT2-R). Data are means ± s.e.m. of three independent replicates.

Pilot eMutaT7 Trial

We first ran eMutaT7 in the noncontinuous mode used to develop it¹¹ to confirm that feedback-resistant mutants could be recovered by building a mutant library and then screening it (Figure 3). Triplicate populations were mutagenized during three serial passages on nonselective LB medium, after which they were plated on selective MOPS minimal medium containing 5 mM 4FPA. Nine resistant colonies were isolated and restreaked on selective medium to eliminate residual nonresistant cells, and the AtADT2 ORFs were sequenced. Each ORF had one or other of six single or double mutations (Figure 3 and Table S3). Multiple mutants carried T190I or R213W mutations in AtADT2’s ACT feedback-regulation domain (Figure 3) in which the known feedback-insensitivity mutation is located.^12,17 All mutations were C → T transitions (Table S3), as expected.⁴ In only one case was there a mutation in the promoter controlling AtADT2 expression. This trial confirmed that, in our application, the eMutaT7 mutational machinery performed as reported.¹²

Figure 3.

Trial of eMutaT7 in discontinuous mode. After accumulating a library of mutants in nonselective conditions in rich medium, cells were plated on selective minimal medium, and the AtADT2 gene from resistant colonies was sequenced. Red asterisks are nonsynonymous mutations. Black asterisks are synonymous mutations. The ACT domain is shaded gray.

eMuta in CDE Mode

We next ran eMutaT7 in continuous mode and under selection,^4,12 starting with ten independent populations and raising the 4FPA concentration stepwise from 0.5 to 30 mM during a total of 26 passages (Figure 4A). Each concentration step took five passages, the number needed for the growth of most populations to improve substantially (Figure 4B); five passages also allowed resistance mutations to reach population frequencies high enough to detect via bulk sequencing of AtADT2 amplicons. Bulk sequencing of populations can monitor types and frequencies of mutations, i.e., assess a campaign’s progress and help decide when to end it. It cannot, however, show whether mutations in a population reside in the same or different genes, or distinguish beneficial from neutral mutations. Populations that showed no growth were replaced by dividing populations that did grow or, for selection steps up to 1 mM 4FPA, by populations maintained on minimal medium. Populations were plated as above on MOPS minimal medium plus 5 mM 4FPA at the 5 and 10 mM 4FPA steps during the campaign and when it ended at 30 mM. Bulk sequencing of amplicons from these plates showed three patterns in AtADT2 ORF mutations (Table S4).

Figure 4.

Operation of eMutaT7 in continuous mode and characterization of the mutants obtained. (A) Evolutionary campaign plan. 4FPA concentration was raised in five steps from 0.5 to 30 mM, with five passages at each step. (B) Data from the second (1 mM 4FPA) and final (30 mM 4FPA) steps illustrating how growth of most populations improves between first and last passages at these steps. Each data point is a separate population. (C) Expression levels of wild-type AtADT2 and of three variants; AtADT2 protein levels (percent of soluble protein) are given under the gel. (D) Growth in MOPS minimal medium alone or plus 5 mM 4FPA of E. coli strains expressing AtADT2 variants. Data are means ± s.e.m. of three independent replicates. EV, empty vector. (E) Structure model of the AtADT2 phenylalanine-binding pocket of the ACT domain showing positions of the benchmark S222N (blue), A208V (green), and R213W (violet) mutations that confer insensitivity to 4FPA.

First, A208V or R213W mutations, both in the ACT domain, had swept two or more independent populations by 5 mM 4FPA and persisted until the 30 mM step. Selection of these mutations in independent populations, and of R213W also in noncontinuous mode (Figure 3 and Table S3), points to convergent evolution of fitness-enhancing mutations. Also pointing to convergent evolution, A208V and R213W both replaced other mutations over time. Second, one population (C6) was dominated at the 5 and 10 mM steps by a G201* nonsense mutation that truncates the ORF upstream of the regulatory ACT domain. As a G → T transversion, this mutation was probably not made by eMutaT7,^4,12 i.e., was spontaneous. By the 30 mM step, the nonsense mutation had been displaced by a double mutant with the above A208V ACT domain mutation plus D240N, also in the ACT domain. The displacement suggests that the A208V plus D240N substitutions confer greater 4FPA resistance than the truncation. Third, while single mutants dominated all populations at the 5 mM step, by the 10 mM step one population was dominated by a double mutant, and there were three double mutants by the 30 mM step. Further, one of the mutations added between the 10 and 30 mM steps, A17V, was among those recovered in noncontinuous mode (Figure 3 and Table S3). These patterns fit with sustained, target-specific hypermutagenesis by eMutaT7 throughout ∼150 generations (26 passages × ∼6 generations per passage). Sequencing the pHyo094 mutator plasmid at the end of the campaign confirmed the integrity of the T7RNAP-CD fusion, showing that mutations did not degrade or inactivate the mutational mechanism, despite the potential for this to happen.⁸

To validate the evolutionary campaigns, we compared the growth of strains expressing variants A208V or R213W with a strain expressing the known feedback-resistant mutant that was isolated from a mutagenized Arabidopsis population and carries the S222N mutation¹² (first wrongly annotated as S222A¹²). We used a challenge of 5 mM 4FPA, which blocks growth of the strain expressing wild-type AtADT2 (Figure 2A). On minimal medium alone, the three strains expressing variants grew similarly to the strain expressing wild-type AtADT2, but adding 5 mM 4FPA led to marked differences (Figure 4C): relative to the benchmark S222N variant clones, those with the R213W variant grew similarly and those with the A208V variant grew faster. As expected, clones with the wild-type enzyme and clones harboring empty vector did not grow. Variation in growth between replicate clones is expected because their AtADT2 genes are in the eMutaT7 system, and thus subject to ongoing hypermutation and hence to reduction or loss of AtADT2 expression.⁸ The S222N benchmark variant was more highly expressed in soluble form than wild-type AtADT2, as were the A208V and R213W variants to a lesser extent (Figure 4D). This is unlikely to relate to the observed resistance phenotypes because overexpression of feedback-inhibited enzymes cannot per se confer feedback-resistance,¹⁸⁻²⁰ and the benchmark S222N variant gives feedback-resistance in planta without being overexpressed.¹² Structure modeling based on the Chlorobaculum tepidum dehydratase (PDB code: 2QMX) suggested that, like the S222N mutation, the A208V and R213W mutations are located in the ACT domain ligand-binding pocket (Figure 4E). Based on the model, V208, W213, and N222 are all 11–12 Å from the phenylalanine ligand. The A208V and R213W mutations are thus positioned to disrupt phenylalanine binding, like S222N.

Conclusions

CDE experiments require careful setup, checks that the selection system is fit for purpose, and calibration of the dynamic range of selection.⁴ CDE of metabolic enzymes in E. coli additionally often requires the CDE system to work in minimal medium (which not all do⁵) and engineering of the platform strain.²¹ This study went through these steps for a realistic case of a plant enzyme and so provides a heuristic for future work of a similar kind. We established the following crucial points. First, eMutaT7—and, by extension, probably other MutaT7 systems⁴—can be used successfully in minimal medium. Second, prevalidation of the selection scheme with an enzyme having the desired property (i.e., a facsimile of the variants to be sought) is well worth the effort because it defines appropriate selection conditions at the outset. The same applies to checks for the likely frequency of cheaters. Third, bulk sequencing of the target gene in whole populations, which is fast and cheap, gives interim information on a campaign’s progress, and helps decide when to halt it, even if selected mutations are not yet fixed. Similarly, bulk sequencing of the mutator plasmid can detect failure, or incipient failure, of the mutational machinery and thereby dictate termination of a campaign instead of its futile prolongation.

The next major step in validating the directed evolution–genome editing (DE–GE)⁹ pathway to plant metabolic engineering is to return CDE-evolved AtADT2 variants such as A208V and R213W to the original plant, Arabidopsis. This will establish whether the variants confer the predicted phenotype (i.e., accumulation of free phenylalanine and downstream products as seen with other introduced feedback-insensitive alleles^12,18) and, more generally, how well the variants function in the plant host. As said at the outset, the differences between conditions in plant and prokaryote cells mean that it cannot be taken for granted that enzymes evolved in E. coli will necessarily function efficiently in a plant.

Finally, and more generally, evolving plant enzymes in microbes has several advantages over doing this in plants. First and foremost, the short generation times and small sizes of microbes enable evolutionary campaigns that are enormously faster and more powerful.^5,9 Another advantage of microbes is that tools to diversify a specific target gene are far better developed than in plants.⁹ A potential limitation on using microbes, particularly for plant secondary product pathways, is designing a suitable host strain and selection scheme to couple the activity of the target enzyme to growth. Installing a biosensor reporter system that responds to the enzyme’s product may be a way around this.²² Although our study targeted feedback inhibition, eMutaT7 and other CDE systems could potentially be used to improve other plant enzyme characteristics, including substrate preference⁹ and in vivo working life.²³

Methods

Media and Conditions

MOPS minimal medium (0.2% glycerol unless otherwise specified) was as described²⁴ with added micronutrients.²⁵ Antibiotics (μg mL^–1) were carbenicillin (Carb) 100; chloramphenicol (Cmp) 33; kanamycin (Kan) 50; and apramycin (Apra) 50. Cells were washed in MOPS medium minus MgSO₄, FeSO₄, or micronutrients (basal MOPS). Growth was at 37 °C, shaking at 250 rpm.

Plasmid Constructs

AtADT2 and prephenate aminotransferase (AtPAT) were PCR-amplified from Arabidopsis cDNA generated using Superscript III (Invitrogen, Waltham, MA) and TA-cloned into pGEM-T Easy (Promega, Madison, WI). Both sequences were truncated to remove the plastid targeting peptide (Table S5); AtADT2 residue numbers in the text are for the truncated protein. The eMutaT7 target (pHyo182) and mutator (pHyo094) vectors were from Park and Kim.¹¹ The AtADT2 sequence and pHyo182 backbone were PCR-amplified and Gibson-assembled (NEB, Ipswich, MA). The S222N mutant of AtADT2(12) was recreated by site-directed mutagenesis (Q5 Kit, NEB, Ipswich, MA). The empty vector was created by PCR-amplifying the pHyo182 backbone outside the start and stop codons of the AtADT2 ORF and ligating the ends. The cytidine deaminase domain was deleted from the control pHyo094 vector by site-directed mutagenesis as above. All constructs were sequence-verified.

Strain Engineering

The uracil-DNA glycosylase ung and pheA genes of E. coli MG1655 were deleted by recombineering;²⁶ the Kan cassette from the corresponding Keio collection mutant was exchanged for the target gene and then removed via recombination of FRT sites. An expression cassette containing the pJ23101 promoter and B0032 ribosome binding site (iGEM collection) and the rrnB T1 terminator was cloned into the pGETSfluURA3 backbone²⁷ via Gibson assembly. The AtPAT gene was then inserted into this cassette by Gibson assembly. The resulting expression cassette was PCR-amplified and recombineered into the flu locus as above, replacing Kan with Apra.

pheS Mutagenesis in Minimal Medium

Single colonies of Δung cells harboring pHyo182-PheS and pHyo094 were inoculated in 3 mL of MOPS medium containing Carb and Cmp. Overnight cultures were diluted 100-fold into the same medium plus 0.2% arabinose and 0.1 mM IPTG. These cultures were grown for ∼14 h and subcultured twice more. Cells were pelleted (4000g, 5 min), resuspended in 200 μL of basal MOPS, diluted to OD₆₀₀ 0.2, serially diluted 10-fold, spotted (5 μL aliquots) on MOPS plates (Carb, Cmp, 0.2% arabinose, 0.1 mM IPTG, ±10 mM p-chlorophenylalanine), and incubated overnight. Single colonies were restreaked on LB (Carb and Cmp) plates and grown overnight for sequencing of target gene amplicons.

Complementation of the ΔpheA Strain with AtPAT and AtADT2

Single colonies of the Δung::FRT ΔpheA::FRT Δflu::AtPAT strain harboring pHyo094 and pHyo182 alone or containing AtADT2 were inoculated into 3 mL of MOPS medium (Carb, Cmp, 1 mM phenylalanine) and incubated overnight. Cells were pelleted as above, washed with 3 × 1 mL of basal MOPS, resuspended in 200 μL of basal MOPS, and inoculated into 3 mL of MOPS medium (Carb, Cmp, 0.002% arabinose, 1 mM IPTG) at an OD₆₀₀ of 0.02, after which growth was monitored at OD₆₀₀.

4FPA Sensitivity Calibration

Single colonies of the Δung::FRT ΔpheA::FRT Δflu::AtPAT strain harboring pHyo094 and pHyo182 alone or containing AtADT2 wild type or the feedback-resistant variant¹² were inoculated into 3 mL of MOPS medium (Carb, Cmp, 1 mM phenylalanine) and incubated overnight. Cells were pelleted, washed, resuspended, and inoculated into MOPS medium as above (complementation section) except that various concentrations of 4FPA (p-fluoro-dl-phenylalanine, Sigma-Aldrich, catalog number F5251) were added. Growth was monitored as above.

Cheater Frequency Test

Single colonies of the Δung::FRT ΔpheA::FRT Δflu::AtPAT strain harboring pHyo182-AtADT2 and pHyo094 (T7RNAP-CD fusion or T7RNAP alone) were inoculated into 3 mL of LB medium (Carb and Cmp) and incubated overnight. Cultures were diluted 100-fold into LB (Carb, Cmp, 0.1 mM IPTG, 0.2% arabinose), grown for 4 h, and subcultured twice more. Cells were pelleted, washed, resuspended, inoculated into MOPS, and grown as above. Cultures were then diluted to OD₆₀₀ 1, and 100 μL aliquots were plated on MOPS medium (Carb, Cmp, 1 mM IPTG, 0.002% arabinose) containing 5 mM 4FPA; colonies were counted after 4 days. Aliquots were also diluted 10⁵-fold and plated on LB (Carb and Cmp) for total viable cell counts.

Discontinuous Mutagenesis

Procedures were as for cheater frequency tests except that the plasmids were pHyo182-AtADT2 and pHyo094. Colonies that grew in the presence of 5 mM 4FPA were restreaked on MOPS plates containing 5 mM 4FPA to minimize false positives; AtADT2 amplicons from resistant colonies were sequenced.

Continuous Mutagenesis with 4FPA Selection

Ten single colonies of the Δung::FRT ΔpheA::FRT Δflu::AtPAT strain harboring pHyo182-AtADT2 and pHyo094 were inoculated into 3 mL of MOPS (Carb, Cmp, 1 mM phenylalanine), incubated overnight, washed, and resuspended as above. Cells were inoculated into 3 mL of MOPS medium (Carb, Cmp, 1 mM IPTG, 0.002% arabinose) ± 0.5 mM 4FPA at starting OD₆₀₀ 0.02. Cultures were grown for 48 h, monitoring OD₆₀₀, and cells were pelleted, resuspended, and subcultured. 4FPA concentration was increased stepwise: 0.5, 1, 5, 10, and 30 mM (Figure 4A). Cultures that did not grow in 0.5 or 1 mM 4FPA were replaced with cells from paired MOPS cultures without 4FPA. Each 4FPA concentration was maintained for five passages. After passages 15, 20, and 26, cells were pelleted, washed, diluted, and plated as above on MOPS medium (Carb, Cmp, 1 mM IPTG, 0.002% arabinose) with 5 mM 4FPA; AtADT2 amplicons from resistant colonies were sequenced. AtADT2 mutants were tested for 4FPA resistance using the above growth protocol, inoculated into 5 mM 4FPA, and grown for several days.

DNA Procedures

Target gene sequencing in clones (pheS mutagenesis and discontinuous AtADT2 evolution) or populations (continuous AtADT2 evolution) served as templates. Cells were resuspended in water and amplified by colony PCR using Phusion polymerase (Thermo Fisher Scientific, Waltham, MA). Mutant clones were restreaked on LB plates (Carb and Cmp), and mutant populations were maintained on MOPS plates (Carb, Cmp, 1 mM IPTG, 0.002% arabinose) containing 5 mM 4FPA. The T7 promoter plus AtADT2 ORF were amplified using flanking primers and sequenced using internal primers. To sequence pHyo094, mutant populations were plated on LB (Carb and Cmp) and then inoculated into LB (Carb and Cmp). Plasmids were isolated and sequenced using internal primers.

Protein Expression

Procedures were as in continuous mutagenesis except that cells were harvested at OD₆₀₀ ∼ 1, resuspended in 50 mM NaH₂PO4, 300 mM NaCl, 5 mM DTT, pH 8.0, and sonicated (8 × 15 s pulse/45 s on ice). Lysates were cleared by centrifuging twice (22 000g, 5 min). Protein was quantified by Bradford assay and 10 μg of total protein analyzed by SDS-PAGE (15% acrylamide) and ImageJ software.

Homology Modeling

The truncated AtADT2 sequence was modeled against 2QMX structure using Phyre2. Models were aligned and analyzed with ChimeraX.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.2c00511.

• Figure S1: Growth curves of plantized E. coli with wild-type AtADT2 or AtADT2-R (PDF)

• Table S1: mutational summary of p-chlorophenylalanine-resistant mutants (XLSX)

• Table S2: comparison of resistant colony frequencies of T7-cytidine deaminase and T7-only control (XLSX)

• Table S3: mutational summary of 4FPA-resistant colonies from noncontinuous evolution (XLSX)

• Table S4: mutational summary of 4FPA-resistant populations from continuous evolution (XLSX)

• Table S5: nucleotide sequences used in this study (XLSX)

Author Contributions

B.J.L. conceived the project with support from A.D.H. B.J.L. and A.D.H. designed experiments; B.J.L. performed the experiments. Both authors analyzed the data and wrote the manuscript.

The authors declare no competing financial interest.