A general strategy for engineering noncanonical amino acid dependent bacterial growth

Abstract

Synthetic auxotrophy in which bacterial viability depends on the presence of a synthetic amino acid provides a robust strategy for the containment of genetically modified organisms and the development of safe, live vaccines. However, a simple, general strategy to evolve essential proteins to be dependent on synthetic amino acids is lacking. Using a temperature sensitive selection system, we evolved an Escherichia coli (E. coli) sliding clamp variant with an orthogonal protein-protein interface, which contains a Leu273 to p-benzoylphenyl alanine (pBzF) mutation. The E. coli strain with this variant DNA clamp has a very low escape frequency (< 10⁻¹⁰), and its growth is strictly dependent on the presence of pBzF. This selection strategy can be generally applied to create ncAA dependence of other organisms with DNA clamp homologues.

Graphical Abstract

The use of auxotrophic bacterial strains for the containment of genetically modified organisms or as conditional, live vaccines suffers from the potential availability of the missing nutrient or metabolite from altered metabolism or the host¹. To overcome this challenge, recent efforts have focused on the generation of auxotrophs that are dependent on non-naturally occurring compounds for growth, specifically a non-canonical amino acid (ncAA) that is added exogenously to the growth medium^2–6. This approach is enabled by the fact that ncAAs with unique physicochemical properties can be readily added to the genetic codes of prokaryotic and eukaryotic organisms^7–11. All that is required is an orthogonal translation system consisting of an exogenous tRNA and aminoacyl-tRNA synthetase (aaRS) pair that does not cross react with tRNAs or aaRSs in the host, and that recognize a unique codon, usually a nonsense codon^7–8.

A number of strategies have been reported to construct ncAA dependent auxotrophs with low escape frequencies. For example, structurally distinct ncAAs were substituted at permissive sites in multiple essential enzymes to generate strains dependent on the simultaneous presence of the ncAAs⁵. In another example, the ncAA L-4,4′-biphenylalanine (bipA) was used to generate auxotrophic bacteria in which the core of an essential enzyme was redesigned to pack around bipA⁶. Other approaches that do not require multiple ncAA substitution or computational design include the substitution of an essential Lys residue by N-ε-acetyl-L-lysine with subsequent host enzyme deacetylation to generate Lys³; engineering a metal ion binding site in which His is substituted with the structurally similar analogue, 3-methyl-L-histidine²; and engineering of a dimer interface that depends on the presence of an ncAA⁴. Here we describe a general strategy to develop ncAA auxotrophs by engineering the protein-protein interface of the E. coli sliding clamp to be strictly dependent on the presence of an ncAA for bacterial survival. Using this approach, we developed an E. coli auxotroph that requires pBzF for survival with an escape frequency below 10^-10.

The β-subunit (sliding clamp, SC) of the DNA polymerase III holoenzyme, encoded by the dnaN gene, is a dimeric protein that completely encircles the DNA as nucleotides are added. SC plays a critical role in the processivity of DNA replication and is essential for bacterial survival^12–14. The SC forms a head-to-tail homodimer, with each monomer having identical chain topology¹⁵. The conserved structure of SCs across gram positive and gram negative bacteria make it a readily generalizable strategy for generating other ncAA dependent auxotrophs^16–17.

To construct an orthogonal β-subunit interface, we first disrupted the native E. coli SC (SC^Eco) interface by incorporating the large and structurally distinct ncAA, pBzF, at the dimer interface, which resulted in a loss of cell viability. We then performed a selection with a library of residues around the pBzF side chain to restore its SC function. To this end, we conditionally knocked out the dnaN gene of E. coli in the presence of the plasmid, pKD-ESC which expresses SC^Eco and λ-red recombinase. pKD-ESC also has a temperature sensitive replicon (repA101ts) that limits the growth temperature of the knockout strain, DH10B ΔdnaN::tetR/pKD-ESC (termed EV1) to below 37 °C¹⁸. The EV1 strain was then transformed with the plasmid pUltra-BzF⁴ which contains the tRNA_CUA/aaRS pair that encodes pBzF to afford strain EV1.BzF (Figure 1). To disrupt the dimer interface^19–20, we substituted Leu273 in SC^Eco (encoded by plasmid pBESC-AX) with the larger ncAA pBzF, expecting unfavorable van der Waals interactions with the surrounding hydrophobic residues (Figure S1b, c). Of note, an additional mutation V70A was introduced into this variant (termed SC^Eco-beta) to avoid potential steric hindrance between the benzophenone ring of pBzF and Val70 at the engineered interface (Figure S1). The mutant strain EV1.BzF/pBESC-AX could not survive at 42 °C, whereas the wild type analog, EV1.BzF/pBESC grew well under the same condition (Figure S2).

Figure 1. Strategy used to engineer ncAA dependent bacteria.

(a) Directed evolution of the E. coli sliding clamp dimer interface with a library that contains randomly mutated residues (cyan) surrounding the 273 residue (magenta). The engineered dimer interface has unique interactions of pBzF273 with surrounding residues (lower panel). (b) A temperature sensitive replicon in plasmid pKD-ESC (blue circle) was used to conditionally knock out the essential gene, dnaN. The selection was conducted at 42 °C with SC^Eco-beta library that is encoded in pBK-SC-Lib (green circle). The orthogonal translational machinery (encoded in pUltra-BzF, gray circle) was used to site specifically incorporate pBzF at the dimer interface of mutant sliding clamp. The evolved SC^Eco-beta variant that is encoded in pBK-SC-Hit (gold circle) enables pBzF dependent growth of EV2.BzF.hit (highlighted in red).

To restore the variant SC^Eco-beta subunit interface and, as a result, SC function in DNA replication, we carried out saturation mutagenesis on the residues surrounding the pBzF273 side chain of the mutant SC^Eco-beta in order to repack the largely hydrophobic dimer interface. Plasmid pBK-SC-Lib encoded a dnaN gene library with the following mutations: Leu273TAG, Val70Ala, and NNK (N = A, C, G or T; K = G or T) for the surrounding residues, Phe′75, Ile′78, Met′97, Phe′106, Leu′108, and Ile272 (where the prime residues correspond to the other subunit). The transformant EV1.BzF/pBK-SC-Lib was plated on permissive LB-agar media containing 1 mM pBzF and isopropyl 1-thio-β-D-galactopyranoside (IPTG, 1 mM) at 42 °C and grown overnight. Surviving colonies were amplified in the presence of pBzF and regrown by replica spotting on permissive (LB-agar, 1 mM pBzF) and non-permissive (LB-agar, no ncAA) media at 42 °C.

Using this selection strategy, we identified ten SC^Eco-beta variants which showed a pBzF dependent growth phenotype when introduced into the EV1.BzF strain, and after curing of the complementing pKD-ESC plasmid at 42 °C (Figure 1b and Table S1). One variant, h2, had a second TAG mutation at residue 108 in addition to the Leu273TAG mutation (Table S1). Because we hypothesized that two TAG codons would need to be mutated for this variant to escape pBzF growth dependence (the single base pair mutation frequency is ~10⁻¹⁰ per generation)²¹, which should result in a very low escape frequency, we first characterized this variant. However, the engineered strain, EV2.BzF/pBK-SC-h2 (EV2.BzF.h2) had a higher than expected escape frequency (4 × 10⁻¹⁰ escapees per colony forming unit (c.f.u.)) (Table S5). All the escape mutants had a mutation in the tyrT anticodon GTA to CTA to give supF amber suppressor, suggesting that tyrosine incorporation at residues 108 and 273 of h2 is the escape mechanism of EV2.BzF.h2. A complementation experiment with a plasmid library containing saturation mutations (NNKs) at residues 108 and 273 of h2 surprisingly showed that multiple combinations of hydrophobic amino acids at these residues resulted in the EV2.BzF.h2 growth in non-permissive media (Figure S3).

Next, we investigated clone h1 in which the mutations are mostly composed of small hydrophobic amino acids (Figure 2a). The engineered E. coli, EV2.BzF.h1 showed pBzF dependent growth in permissive media, whereas no growth was observed in non-permissive media over 2 weeks (Figure S4a). Unfortunately, EV2.BzF.h1 exhibited a comparable escape frequency (1.3 × 10⁻¹⁰ escapees per c.f.u) to EV2.BzF.h2 (Table S5). All isolated escapees had a TAG to TGG mutation at position 273, which encodes tryptophan. In a complementation experiment, we found that residue 273 in h1 can tolerate mutation to Phe, Trp, Met, or Val with 84%, 8%, 5%, and 3% frequency among all the growing colonies, respectively (Figure S4b, c).

Figure 2. Structure of h1 and h5 dimer interface.

Models of h1 and h5 were generated using the Rosetta protein design software. pBzF273 (magenta) is located at the interface with surrounding residues [(a) Ala′70, Leu′75, Gly′78, Leu′97, Met′106, Cys′108, and Val272 in h1; (b) Ala′70, Met′75, Phe′78, Leu′97, Thr′106, Ala′108, and Val272 in h5; cyan]. H-bond is shown with a red dashed line. Hydrogens are masked for clarity.

We then investigated all other hits from the selection to identify the SC^Eco-beta variant that gave the lowest mutational escape frequency with a canonical amino acid substitution at residue 273. Because EV2.BzF.h1 had a reasonably low escape frequency, we used this strain as the host for complementation by the other hits (h3-h10) in which an NNK mutation was introduced at site 273. In the case of clone 5, the only canonical amino acid capable of functional replacement of pBzF273 and conferring cell growth was phenylalanine (Figure 3a, b). Because the amber codon (UAG) is two mutations away from either of the phenylalanine codons (UUU and UUC), EV2.BzF.h5 should have lower escape frequency than EV2.BzF.h1 or EV2.BzF.h2. Indeed, EV2.BzF.h5 exhibited a pBzF dependent growth profile (Figure 3c) with non-detectable escapees (no escapees from the culture of 2 × 10¹⁰ in non-permissive solid media for 2 weeks at 37 °C) (Table S5).

Figure 3. Characterization of EV2.BzF.h5 dependence on pBzF for survival.

(a) EV2.BzF.h1 strain was transformed with a focused library of h5 in pEvol-h5-NNK plasmid for complementation. The growth was monitored both in permissive and non-permissive media that contains 0.02% _L-arabinose. The wild type and blank plasmid were used for positive and negative controls, respectively. (b) Frequency of amino acid mutation at residue 273 of h5 in pEvol-h5-NNK plasmids isolated from survived EV2.BzF.h1 clones in non-permissive media. (c) Growth curve of EV2.BzF.wt (square) or EV2.BzF.h5 (triangle) cells in permissive (LB, 1 mM pBzF, opaque) and non-permissive (LB, no ncAA, hollow) media. Error bars represent the standard deviation of three technical replicates.

The doubling times of EV2.BzF.h1, EV2.BzF.h2, and EV2.BzF.h5 are slightly longer than the EV2.BzF.wt that expresses wild type SC^Eco (Table S5), perhaps due to lower levels of expression of the SC^Eco-beta variants compared to the wild type SC^Eco in the E. coli cytosol. Indeed, smaller amounts of the h5 protein were observed in the soluble fraction compared to wild type when each was overexpressed in E. coli (Figure S5). We were also able to confirm the dimeric nature of variant h5 by exploiting the photoreactivity of the benzophenone side chain of pBzF²². The soluble fractions of the lysate from E. coli cells that expressed c-terminal his-tagged SC^Eco and corresponding variant proteins were irradiated with long wavelength UV light, and crude samples were analyzed by denaturing SDS gel electrophoresis and western blotting. Clone h5 showed clear dimer bands upon irradiation, whereas wild type and non-irradiated h5 did not form dimer bands (Figure S5).

To understand the nature of the structural modifications, we modeled the engineered interfaces of h1 and h5 using the Rosetta software²³ (Figure 2, see Supporting Information for a detailed description). Modeling revealed that the mutations at residues 70, 75, and 97 observed in the h1 and h5 variants allow the bulky “second ring” of pBzF (i.e., the phenyl ring farthest from the protein backbone) to adopt an orientation that would be difficult to achieve in the wild type protein. The models also suggest that mutations at position 106 facilitated an energetically favorable rotamer of the pBzF. Notably, Thr′106 in h5 also appears to introduce a hydrogen bonding interaction to the carbonyl of the pBzF273 residue.

In conclusion, by evolving the sliding clamp to be dependent on pBzF for function, we generated a mutant E. coli auxotrophic for pBzF with an undetectable escape rate for a single engineered essential protein, it is noteworthy that the strategy described here should be generally useful to engineer ncAA auxotrophs by creating ncAA dependent essential proteins. Given that the SC dimer interface is well conserved across gram negative pathogens including Pseudomonas aeruginosa (P. aeruginosa), Acinetobacter baumannii (A. baumannii), and Enterobacter cloacae (E. cloacae)¹⁷, the hit variants evolved in this study should be readily transferred to SCs of those strains for the development of conditional live vaccines (Figure S6). As such, we are now expanding this strategy to develop conditional live vaccines for P. aeruginosa and A. baumannii infections which are a significant risk for hospitalized patients, and those with impaired lung or immune system function. We are also applying a similar strategy to identify a second engineered ncAA dependent auxotroph.