E. coli K-12, strain AG1 [recA1 endA1 gyrA96 thi-1 hsdR17 (rKmKþ) supE44 relA1] harboring ASKA+ and ASKA-plasmid libraries, W3110, and W3110 ts 29-13 were obtained from the National BioResource Project (NIG, Japan): E. coli (https://shigen.nig.ac.jp/ecoli/strain/). BL21-AI [F⁻ ompT hsdS_B (r_B ⁻ m_B ⁻)gal dcm araB::T7RNAP-tetA] was purchased commercially (ThermoFisher Scientific, MA, United States).

LB medium (1% BactoTriptone, 0.5% Bacto Yeast Extract, 1% NaCl) was used to culture all E. coli strains. Antibiotics (chloramphenicol 50 mg/L, kanamycin 25 mg/L, and tetracycline 1.25 mg/L) were added as needed. Liquid culture was performed in an air incubator with shaking (180 rpm). To prepare solid media, 2% agar was added. Bacteria were incubated at 37°C unless otherwise noted. Bacterial transformation was performed by electroporation using a Gene Pulser II™ electroporator (Bio-Rad Laboratories, CA, United States).

The procedure for preliminary toxicity and GFP-fluorescence evaluation for all ORFs using ASKA+ plasmid library has been described previously (Kitagawa et al., 2005). For quantitative assays, overnight cultures of E. coli harboring the ASKA-plasmid encoding the tested ORFs were used. The cultures diluted 10 to 10³-fold were inoculated 250 μL onto solid medium containing 1 mM IPTG and chloramphenicol, a selection marker for ASKA-, and surviving colonies were counted. The total number of bacteria was determined by inoculating 10⁶-fold diluted cultures onto a solid medium without IPTG. The toxicity due to overexpression was assessed by the survivor frequency (survivors/total CFU).

The ASKA-clone encoding tyrS was isolated from AG1 carrying this plasmid using the Qiagen Plasmid Midi (Qiagen Inc., CA, United States) and transformed into BL21-AI. The tyrS in the genome of the resulting BL21-AI [tyrS ⁺ ; ASKA-(tyrS)] strain was substituted with a knock-out cassette containing the neomycin resistance gene using λRed recombinase (Datsenko and Wanner, 2000). Among individual colonies resistant to both chloramphenicol and kanamycin, clones with successful substitution were identified by PCR using the genomic DNA as a template. Finally, the deletion of tyrS was directly confirmed by sequencing the tyrS locus using the Illumina sequencing, as detailed below.

Plasmid loss was tested for BL21-AI [tyrS ^- ; ASKA-(tyrS)] and the control parent strain, BL21-AI [tyrS ⁺ ; ASKA-(tyrS)]. Every 24 h, the bacteria were diluted 10³-fold (1 μL/1 mL fresh LB medium without selection marker antibiotics) and incubated again at 25°C. This dilution and reculturing was repeated 12 times, approximately 120 generations. On inoculation into a new medium, an aliquot of the culture was withdrawn, diluted 10⁶-fold, and inoculated (250 μL) onto solid medium. Sixty-four colonies were selected and inoculated on solid medium containing chloramphenicol to determine plasmid maintenance.

To verify genetic stability, bacteria were cultured in the presence of chloramphenicol and kanamycin which prevent contamination by non-tested bacteria during the 1 month long serial cultures. Dilution and reculturing procedures were identical to those used in the plasmid loss test. On inoculation for reculturing, an aliquot of the cultures was withdrawn, diluted 10-fold, and inoculated (250 μL) onto solid medium containing 5 mM IPTG. Surviving colonies were counted after overnight incubation. The total number of bacteria was also measured by inoculating 10⁶-fold diluted cultures on IPTG-free solid medium.

The procedure for genome sequencing has been described previously (Kato, 2023). Briefly, total DNA, including genomic and plasmid DNAs, was extracted using the DNeasy Blood and Tissue Kit (Qiagen). After checking the integrity of the extracted DNA, sequencing libraries were prepared using a TruSeq DNA PCR-free Sample Preparation Kit (Illumina, Inc., San Diego, CA, United States). The purified libraries were subjected to paired-end (2 × 150 bp) sequencing by Macrogen using the NovaSeq6000 platform. Raw reads were trimmed to remove adapter sequences, mapped to the E. coli BL21-AI (NZ_CP047231.1) and ASKA-(tyrS) reference sequence using BWA-MEM v0.7.17. Variants (SNPs and small INDELs) were identified using Samtools. Variant functional annotation was performed with SnpEff v.5.0e (https://pcingola.github.io/SnpEff/).

Bacterial samples were homogenized using a Multi-beads Shocker^® (Yasuikikai, Osaka, Japan) at 1,500 rpm for 2 min. Total RNA was extracted using the RNeasy Mini Kit and the QIAcube Connect Extraction System (Qiagen). The concentration of the total RNA samples was determined using the Synergy LX multimode reader (BioTek, VT, United States) and the QuantiFluor RNA System (Promega, WI, United States). The RNA quality was evaluated using the 5,200 Fragment Analyzer System and the Agilent HS RNA Kit (Agilent Technologies, Inc., CA, United States). After the ribosomal RNA depletion with the RiboPOOL kit (siTOOLs Biotech GmbH, Planegg, Germany), the cDNA libraries were prepared using the MGIEasy RNA Directional Library Prep Kit (MGI-Tech, Shenzhen, China). The concentration of libraries was determined as described above. A library quality check was performed using the dsDNA 915 reagent kit (Agilent Technologies) and the Fragment Analyzer. Circularized DNA and DNA nanoballs were prepared using the MGIEasy circularization kit and the DNBSEQ-G400RS high-throughput sequencing kit (MGI-Tech). The DNBs were sequenced using the DNBSEQ-G400 (MGI-Tech). The cutadapt (ver. 4.0) and sickle tools (ver. 1.33) were used to remove adapter sequences and low-quality sequence reads. The resulting reads were mapped using Bowtie 2 (ver. 2.5.0) to the E. coli BL21-AI genome (CP047231.1) and the plasmid ASKA-(tyrS). The mapped reads were quantified with FeatureCounts (ver.2.0.3). The row counts were normalized using the transcripts-per-million (TPM) method. The expression level was evaluated relative to gltA, a constitutively expressed gene (Denger et al., 2014).

The ASKA plasmid is a multi-copy plasmid containing the colE1 origin (Kitagawa et al., 2005). There are two types of ASKA plasmid: ASKA+ in which gfp is fused at the 3’ terminal end of the ORF, and ASKA-which excludes the gfp. We examined growth and GFP expression of E. coli AG1 harboring ASKA+ plasmids in overexpression induction with 1 mM IPTG (Supplementary Table S1). Total 4,269 ORFs were tested. Growth inhibition was observed in 3,301 ORFs (77%), of which 2,149 (50%) were severely inhibited with almost no growth, as previously reported (Kitagawa et al., 2005) (Figure 2). Of the ORFs that severely inhibited growth, 156 were essential genes. This represents 52% of all 300 essential genes. Some of these severely growth inhibitory essential genes did not interfere with GFP expression, suggesting that the host viability is well maintained, i.e., “bacteriostatic,” presumably. By contrast, 112 of the highly growth inhibitory essential genes expressed little or no GFP, indicating a bactericidal or severely damaging effect. Thus, there are more than 10² essential genes that are strongly harmful upon overexpression.

For the 112 essential genes that exhibited both strong growth inhibition and suppression of GFP expression, we evaluated the survivor frequency upon overexpression (Supplementary Table S2). To avoid the effects of gfp-tag, the ASKA-plasmids were used. The survivor frequency showed significant variability. For example, with dnaX and tilS, no colonies were detected in the presence of 1 mM IPTG after inoculating around 10⁷ CFU (Supplementary Table S2; Supplementary Figure S2). In contrast, some survivors were detected in other ORFs such as murG and ftsE. There were 28 (25%) ORFs that showed a survival frequency of <30 × 10⁻⁶. Of these, 16 were cytoplasmic proteins, 9 were integral membrane proteins, and 3 were membrane anchored proteins (Supplementary Table S2).

To realize the “demon and angel” kill-switch, we examined which essential gene would be a suitable model. The kill-switch must provide a small amount of functional product by leakage expression in the OFF state to maintain host viability. Membrane proteins are often difficult to express functionally and injure the host (Mathieu et al., 2019). Therefore, we decided to select a candidate essential gene among genes encoding cytoplasmic proteins. We focused on aminoacyl-tRNA synthetase (aaRS) genes. A total of 26 aaRS genes, including 22 essential genes, were found in the ASKA+ library. Of the 24 genes tested, 23 showed toxicity upon overexpression (Supplementary Table S3). Nine genes, pheT, leuS, tyrS, ileS, tilS, proS, argS, glyS and poxA, were highly growth inhibitory with no GFP expression. Among these toxic aaRS genes, we focused on the essential gene encoding the tyrosyl-tRNA synthetase tyrS (Mirande, 1991). The survivor frequency of AG1 [ASKA-(tyrS)] upon overexpression with 1 mM IPTG was only 22.5 × 10⁻⁶, indicating potent toxicity (Supplementary Tables S2, S3). After the exposure to 5 mM IPTG for 1 h or less, AG1 [ASKA-(tyrS)] rapidly and irreversibly lost the ability to form colonies, suggesting that the effect of tyrS overexpression is bactericidal and suitable for host-killing (Supplementary Figure S3). The availability of temperature-sensitive mutants and the presumed low impact of genomic gene deletion, as described below, are advantages for the application of tyrS to a proof-of-concept experiment of the kill-switch.

E. coli W3110 ts 29-13 strain is a temperature-sensitive mutant of tyrS (https://shigen.nig.ac.jp/ecoli/strain/locale/change?lang=en). This strain shows normal growth at 30°C but fails to form colonies at 42°C (Supplementary Figure S4A). We transformed the plasmid ASKA-(tyrS) into this strain. The resulting strain, W3110 ts 29-13 [ASKA-(tyrS)], recovered colony formation at 42°C in the absence of IPTG, suggesting that ASKA-(tyrS) can supply TyrS at levels required for host survival and proliferation by leakage expression (Supplementary Figure S4A). No colonies were observed in the presence of 3 mM IPTG at both 30°C and 42°C, suggesting that tyrS overexpression kills the host (Supplementary Figure S4B). These results suggest that the “demon and angel” kill-switch using tyrS is feasible to construct, as we confirmed the rescue of host survival and growth by leakage expression in the OFF state and the host-killing by overexpression in the ON state.

Next, we constructed the kill-switch in a tyrS deletion mutant strain. tyrS is part of the pdxH-tyrS-pdxY operon (Lam and Winkler, 1992; Yang et al., 1998). pdxH and pdxY are non-essential genes, suggesting that disruption of the pdxH-tyrS-pdxY operon by tyrS deletion is unlikely to impair the function of other essential genes (Tramonti et al., 2021). First, we transformed ASKA-(tyrS) into the unmodified BL21-AI to obtain BL21-AI [tyrS ⁺ ; ASKA-(tyrS)] (Figure 3A) (Bhawsinghka et al., 2020). The ORF of tyrS in the genome was subsequently deleted by substituting the neomycin resistance gene. The deletion was confirmed by genome sequencing. The resulting BL21-AI [tyrS ^- ; ASKA-(tyrS)] strain survived and grew in the absence of IPTG, presumably due to leakage expression of the plasmid tyrS (Figures 3B, C). Meanwhile, BL21-AI [tyrS ^- ; ASKA-(tyrS)] was killed in the presence of IPTG, as well as BL21-AI [tyrS ⁺ ; ASKA-(tyrS)].

Passage tests showed that BL21-AI [tyrS ⁺ ; ASKA-(tyrS)] strain, which retains tyrS in the genome, lost the ASKA plasmid after six to seven passages (60–70 generations) in the absence of plasmid selection marker chloramphenicol (Figure 3D). In contrast, the ASKA plasmid was stably maintained for at least 120 generations in BL21-AI [tyrS ^- ; ASKA-(tyrS)] in which the genomic tyrS was deleted, suggesting that the plasmid tyrS is essential for survival in this strain.

BL21-AI [tyrS ^- ; ASKA-(tyrS)] was killed in the presence of IPTG. In a passage test, the IPTG sensitivity was preserved well through more than 30 passages, equivalent to about 300 generations (Figures 3C, E, left panel). Interestingly, the survivor frequency showed complex changes over the generations, with a range of less than 30 × 10⁻⁶. Survivors isolated from passages 14 and 30 (P14R and P30R, respectively) were IPTG-resistant, indicating the emergence of loss-of-function mutants (Figure 4A). Notably, most bacteria (>99.997% at all time points of the 30 passages) maintained the kill-switch function and no rapid replacement with the loss-of-function mutants was observed (Figure 3E, left panel).

To identify the IPTG-resistant mechanisms, the transcriptomes in the IPTG-sensitive strains isolated from passage 0 and 31, P0S and P31S, and the IPTG-resistant strains, P14R and P30R, were analyzed using RNA-seq. We found that the induction of the IPTG-regulated tyrS on the ASKA plasmid remarkably decreased in the IPTG-resistant strains although the IPTG-inducibility still remained partial (Figure 4B). The tyrS induction ratio was 836.3, 499.1, 18.1 and 42.8, for P0S, P31S, P14R and P30R, respectively. Meanwhile, the basal expression of the plasmid tyrS was similar in both the IPTG-sensitive and resistant strains. Interestingly, the IPTG-induced expression level of the native IPTG-inducible genes in the genome, lacA, lacY and lacZ, was comparable between IPTG-sensitive and resistant strains. The basal expression of those genomic genes remarkably decreased in P14R and P30R although the expression of lacI, that negatively regulates the expression of the IPTG-inducible genes, was not clearly affected (Supplementary Figure S5).

Next, we sequenced the total DNAs containing both genomic and plasmid DNA isolated from P0S, P14R and P30R, using Illumina sequencing. Severe mutations were found in the regulatory region of tyrS in the ASKA plasmid in P14R and P30R whose sequences were completely identical (Figure 4C). The coliphage T5 promoter region was largely mutated, but the −10 and −35 regions were relatively conserved. Although lacO ₁ sequence embedded in the promoter was almost destroyed, lacO ₂ located between the promoter and the translation initiation site was completely conserved. The plasmid lacI was intact. Some additional genomic mutations were also detected in P14R and P30R (Supplementary Table S4). Finally, the ASKA plasmid was extracted from P14R and transformed into unmodified BL21-AI. The resulting strain was IPTG-resistant, suggesting that the mutations on the plasmid are sufficient to cause loss of the kill-switch function (Figure 4A).

Notably, the bacterial density clearly started to increase from around the passage 18 and grew to the passage 29-30 (Figure 3E, right panel). The growth rate of P31S (doubling time = 30.3 min) was greater than those of P0S (41.5 min) and P30R (33.3 min).