A portable epigenetic switch for bistable gene expression in bacteria

We describe a portable epigenetic switch based on opvAB, a Salmonella enterica operon that undergoes bistable expression under DNA methylation control. A DNA fragment containing the opvAB promoter and the opvAB upstream regulatory region confers bistability to heterologous genes, yielding OFF and ON subpopulations. Bistable expression under opvAB control is reproducible in Escherichia coli, showing that the opvAB switch can be functional in a heterologous host. Subpopulations of different sizes can be produced at will using engineered opvAB variants. Controlled formation of antibiotic-resistant and antibiotic-susceptible subpopulations may allow use of the opvAB switch in the study of bacterial heteroresistance to antibiotics.

We describe a portable epigenetic switch based on opvAB, a Salmonella enterica operon that undergoes bistable expression under DnA methylation control. A DnA fragment containing the opvAB promoter and the opvAB upstream regulatory region confers bistability to heterologous genes, yielding off and on subpopulations. Bistable expression under opvAB control is reproducible in Escherichia coli, showing that the opvAB switch can be functional in a heterologous host. Subpopulations of different sizes can be produced at will using engineered opvAB variants. controlled formation of antibiotic-resistant and antibiotic-susceptible subpopulations may allow use of the opvAB switch in the study of bacterial heteroresistance to antibiotics.
Biosensors able to detect environmental signals are made of a sensor that detects a given input and a reader that responds to the input generating a detectable signal in a quantitative or semi-quantitative fashion 1 . Classical sensors employ enzymes or whole cells. Enzyme-based biosensors present the advantage of high selectivity but the need for purification can be a drawback due to technical difficulties and high cost. In contrast, whole-cell sensors are often easy to use and inexpensive, especially if microbial strains are used 2 . A common type of microbial biosensor is an engineered strain that responds to physical or chemical inputs generating electrochemical or optical signals. Sensors of this type often employ a promoter sensitive to a specific input and a reporter gene that produces a detectable signal 1,3 . The literature contains multiple examples of sensors that detect electrochemical and optical signals, and use of fluorescent proteins has become widespread in the last decade 4 . An alternative to genetic circuits able to process information in living cells is the design of epigenetic switches. This approach has received special attention to develop diagnostic tests for human diseases [5][6][7] , while synthetic biology based on bacterial epigenetics remains largely unexplored. A relevant exception is the recent development of biosensors based on DNA adenine methylation using Escherichia coli as host 8 .
In this study, we describe the construction and application of an epigenetic switch that drives gene expression in a bistable fashion. Bistability generates bacterial subpopulations that differ in a specific phenotypic trait (e. g., antibiotic resistance) and have defined sizes. The switch is based on opvAB, a bacterial operon subjected to epigenetic control by DNA adenine methylation [9][10][11] . Transcription of opvAB is bistable, with concomitant formation of OpvAB OFF and OpvAB ON cells 9 . Bistability is controlled by binding of the OxyR transcription factor to a regulatory region upstream of the opvAB promoter (Fig. 1A) 10 . This region contains four sites for OxyR binding and four GATC motifs. OpvAB OFF and OpvAB ON cell lineages display alternative patterns of OxyR binding, which in turn cause alternative patterns of GATC methylation: in the OFF state, GATC 2 and GATC 4 are methylated; in the ON state, GATC 1 and GACT 3 are methylated 10 . Here, we show that a cassette of 689 nucleotides containing the opvAB promoter and the upstream regulatory region confers bistability to heterologous genes, and describe examples of opvAB-based constructs that produce bacterial subpopulations with distinct phenotypes. One of the examples involves formation of an antibiotic-resistant subpopulation upon cloning of an antibiotic resistance gene downstream of the opvAB promoter. This construct may provide an experimental system to study bacterial heteroresistance (HR) to antibiotics under highly controlled conditions 12 . HR is a phenotype where a bacterial isolate is characterized by the presence of a main susceptible population and a subpopulation with higher antibiotic resistance. Increasing evidence suggests that heteroresistance can lead to treatment failure [12][13][14][15][16][17] . Yet, little is known regarding the characteristics of the heteroresistance phenotypes (i.e. the size of the resistant subpopulation or its level of resistance) that are linked to treatment failure. Animal experiments, where infections are started with bacterial cultures that carry an antibiotic resistance gene under control of the opvAB switch, would allow www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ control of the frequency of the resistant subpopulation and determination of how different ratios of resistant:susceptible bacteria influence treatment outcome 17 . Other potential uses of the opvAB switch in synthetic biology are discussed.

Results
Bistable expression of lacZY under opvAB transcriptional control. The ability of the opvAB epigenetic switch to confer bistable expression to a heterologous locus was tested by engineering a strain that harbored the E. coli lacZY operon downstream of the opvAB promoter and its upstream regulatory region (P opvAB - Fig. 1A,B). To avoid cell-to-cell heterogeneity associated with variations in plasmid copy number, the construct was engineered in the S. enterica chromosome. Construction involved replacement of the opvAB coding region with a promoterless lacZY operon, leaving the opvAB promoter and upstream regulatory region intact. The construct harbored the opvA ribosome binding site (RBS). Plating of the engineered strain on LB containing X-gal yielded Lac + (blue) and Lac − (white) colonies, thus revealing bistable expression (Lac OFF or Lac ON ) of the heterologous lacZY operon (Fig. 1C). Streaking of either Lac + or Lac − colonies on X-gal agar yielded a mixture of Lac + and Lac − colonies, thus indicating the occurrence of reversible bistability ("phase variation") as previously described for the native opvAB locus 9 .
Calculation of phase variation frequencies indicated a frequency of 1.1 × 10 −4 ± 0.3 per cell and generation for the OFF→ON transition, and 3.4 ± 0.1 × 10 −2 per cell and generation for the ON→OFF transition. The 300-fold difference between switching rates was two-fold lower than in the native opvAB locus (OFF → ON, 6.1 ± 1.7 × 10 −5 ; ON → OFF 3.7 ± 0.1 × 10 −2 ; 600-fold difference between switching rates) 9 . The increased size of the Lac ON subpopulation may result from multiple factors including potential differences in mRNA stability and codon usage constraints.
Variants of the P opvAB ::lacZY construct were engineered to further explore the ability of opvAB-driven transcription to confer bistable expression to a heterologous locus. One such variant involved the use of a mutant opvAB regulatory region lacking GATC sites 1 and 2 (GATC 1,2 ), previously shown to increase the size of the OpvAB ON subpopulation 10 . As expected, a higher proportion of Lac + colonies was detected (Fig. 1C). Another variant, used as control, lacked all opvAB GATC sites (GATC-less) and locked lacZY transcription in the ON state ( Fig. 1C) as previously described for the native opvAB operon 9 .
Variants carrying a green fluorescent protein gene (gfp) dowstream of the lacZY operon were also engineered, and assessment of subpopulation sizes by flow cytometry confirmed that the Lac ON subpopulation formed by the wild type opvAB switch was smaller than that formed by the GATC 1,2 variant (Fig. 1D). Furthermore, only cells in the Lac ON state were detected in the strain that harbored the GATC-less construct, and subpopulation formation was abolished as above (Fig. 1D).
The ability of the opvAB switch to permit selection of one of the subpopulations was examined by testing the ability of strains carrying P opvAB ::lacZY::gfp and P opvAB GATC 1,2 ::lacZY::gfp constructs to grow in minimal medium with lactose as sole carbon source. As above, a strain carrying the GATC-less P opvAB ::lacZY::gfp construct was included as a control. Assessment of the growth patterns of these strains revealed that the time required for culture saturation was dependent on the size of the Lac ON subpopulation present at the start of the culture (Fig. 1E). Reversibility of the Lac ON state was confirmed by growth on NCE-glucose (Fig. 1F).
Bistable expression of the chimaeric opvAB::lacZY operon in a heterologous host, E. coli. The functionality of the opvAB switch in a heterologous host was tested in E. coli. For this purpose, the P opvAB ::lacZY::gfp construct and its GATC 1,2 and GATC-less variants were introduced into the chromosome of E. coli DR3 (ΔlacZY). Strains carrying the P opvAB ::lacZY-gfp and P opvAB GATC 1,2 ::lacZY::gfp constructs (DR22 and DR23, respectively) formed Lac + and Laccolonies on X-gal agar, and the number of Lac + colonies was higher in the strain carrying the P opvAB GATC 1,2 ::lacZY::gfp construct. The strain carrying the GATC-less construct (DR24) formed Lac + colonies only ( Fig. 2A). Flow cytometry assessment of GFP expression upon growth in LB confirmed the occurrence of subpopulations of Lac OFF and Lac ON cells in the strains carrying the P opvAB ::lacZY::gfp and P opvAB GATC 1,2 ::lacZY::gfp constructs but not in the strain carrying the GATC-less construct (Fig. 2B). As above, growth pattern assessment revealed that the time required for culture saturation was dependent on the initial size of the Lac ON subpopulation (Fig. 2C). Altogether, these observations indicated that the opvAB switch is functional in E. coli.

Bistable expression of antibiotic resistance genes under opvAB control.
An additional test of the ability of the opvAB bistable switch to generate bacterial subpopulations was performed by cloning antibiotic resistance genes downstream of the opvAB promoter in the S. enterica chromosome. The antibiotic resistance genes chosen for these experiments were aac3-Ib (henceforth, aac3) and aac(6′)-Ib-cr (henceforth, aac6), which encode aminoglycoside acetyl transferases 18 , and bla CTX-M-15 (henceforth, ctxM), which encodes an extended-spectrum β-lactamase 19 . In these constructs, the native ribosome binding sites were replaced with a stronger RBS, named BI 20 to adjust the sensitivity of the switch to a level that could permit unambiguous detection of the antibiotic resistance phenotype under study, thus facilitating discrimination between OFF and ON cells. Experiments with strains carrying P opvAB ::aac6::gfp and P opvAB ::ctxM::gfp fusions (strains SV9703 and SV9706, respectively) yielded bacterial subpopulations resistant to kanamycin and to cefotaxime, respectively (Fig. 3A). Controls using strains that constitutively expressed aac6 and ctxM (SV9705 and SV9707, respectively) showed that the concentrations of antibiotics used permitted growth (Fig. 3A). The wild type strain ATCC 14028 failed to grow under such conditions, confirming that the concentrations of antibiotics used were bactericidal.
Flow cytometry analysis confirmed that growth in the presence of kanamycin and cefotaxime was a consequence of subpopulation selection (Fig. 3B), excluding the idea that growth might result from selection of mutants present in the inoculum. This conclusion was further strengthened by the observation that growth in LB restored the initial sizes of ON and OFF subpopulations (Fig. 3B).
www.nature.com/scientificreports www.nature.com/scientificreports/ Use of the opvAB synthetic switch in generating antibiotic heteroresistance. As a proof of concept, we examined the utility of the OpvAB switch to address antibiotic heteroresistance and the question of what proportions of resistant subpopulations might lead to clinical treatment failure. Specifically, we tested whether the OpvAB switch could generate, in a susceptible main population, defined subpopulations of cells with increased antibiotic resistance. For this purpose, we used a S. enterica strain harboring a P opvAB ::BI-aac3::gfp construct (SV9776). Expression of aac3 (Aac3 ON ) leads to kanamycin resistance (Km r ). The frequency of Km r cells formed by a pure culture of SV9776 was 1 × 10 −2 (Fig. 4A), similar to the frequency of ON cells detected when gfp was cloned behind the opvAB promoter (1.1%: Fig. 1D). To obtain smaller subpopulation sizes without altering other phenotypic traits of the strain, SV9776 was mixed with an isogenic strain that expressed P opvAB ::gfp (SV9777) and did not produce any Km r resistant subpopulation. Mixtures of cells were prepared from overnight cultures in Mueller-Hinton (MH) broth at proportions 1:10, 1:100, 1:1,000, 1:10,000 and 0:1. Population analysis profile (PAP) tests were then performed by plating on MH agar containing increasing concentrations of kanamycin. After overnight incubation, the number of resistant cells and total number of cells were determined to allow calculation of the fraction of resistant cells. The numbers of Km r colonies detected in the PAP tests were proportional to the amounts of the Aac3 ON subpopulations present in each mixture, and ranged from 1 × 10 −2 to 1 × 10 −6 (Fig. 4A). Epsilometer tests (Etests) further confirmed that the size of the Km r subpopulation decreased in a manner proportional to dilution (Fig. 4B).

Discussion
In its native host, the opvAB operon undergoes bistable transcription, which generates OpvAB ON and OpvAB OFF subpopulations 9 . Bistability is reversible ("phase-variable") and the switching rate is skewed to OFF in the wild type 9,11 . In this study, we show that a 689 bp DNA fragment containing the opvAB promoter and the opvAB upstream activating sequence (UAS) confers bistability to genes cloned downstream. For instance, an engineered P opvAB ::lacZY operon produces Lac OFF and Lac ON subpopulations (Fig. 1C), and addition of a gfp reporter gene www.nature.com/scientificreports www.nature.com/scientificreports/ permits discrimination of Lac OFF and Lac ON cells by flow cytometry (Fig. 1D). Utilization of L-lactose sustains growth of Lac ON cells (Fig. 1E), thereby producing increased fluorescence. However, because the opvAB switch is reversible, in the absence of L-lactose the system slowly returns to its initial state, with a strong predominance of Lac OFF cells (Fig. 1F).
The fact that the opvAB cassette is functional in E. coli (Fig. 2) suggests that the switch can be used to generate bistability in other heterologous hosts. However, the need of both Dam methylation and OxyR may be an obvious limitation. Aside from this caveat, the versatility of the switch is reinforced by an additional example of subpopulation formation presented in Fig. 3: P opvAB -driven bistable expression of kanamycin and cefotaxime resistance genes permitted selection of antibiotic-resistant subpopulations in a reversible fashion.
Introduction of mutations in the upstream regulatory region of the native opvAB operon alters the switching rate, yielding OpvAB ON and OpvAB OFF subpopulation sizes that are different from those of the wild type 10,11 . Hence, variants of the opvAB switch can be engineered to modulate subpopulation sizes at will. For instance, a variant (GATC 1,2 ) that lacks two of the four GATC sites present in the wild type increases the initial size of the ON subpopulation (Figs 1 and 2). Additional UAS variants that yield subpopulations of different sizes have been described 10 , and their use may allow choice of other switching frequencies. Modification of the ribosome-binding site of genes under P opvAB control can also contribute to adjust the sensitivity of the switch, facilitating detection of the phenotype under study. For instance, use of the BI ribosome binding site 20 permitted unambiguous detection of aac3-mediated kanamycin resistance, thereby facilitating discrimination of Km r cells (Fig. 4).
As a proof of concept, we have used the opvAB switch to produce antibiotic-resistant and antibiotic-susceptible bacterial subpopulations of predetermined sizes. The aim of these experiments was to mimic under laboratory conditions bacterial heteroresistance to antibiotics, a phenomenon where small subpopulations of cells show higher antibiotic resistance than the main population 12 . Heteroresistance is difficult to detect and study in clinical samples 12 , and accurate assessment of the frequencies of subpopulation formation and of their antibiotic resistance levels may improve our understanding of heteroresistance as a cause of clinical treatment failure 15 . Experiments shown in Fig. 4 provide evidence that subpopulation formation under opvAB control allows accurate modulation of the number of resistant cells present in a population. In principle, the method should be applicable to any antibiotic resistance gene. Because we were able to specifically vary the frequency of resistant bacteria in www.nature.com/scientificreports www.nature.com/scientificreports/ the population, this approach provides a proof of concept to study how different frequencies of resistant subpopulations may affect the outcome of antimicrobial treatment in vivo (e. g., in a murine model). In theory, mixing constitutively resistant and susceptible strains that are otherwise isogenic would also lead to bacterial cultures with pre-defined amounts of resistant bacteria. However, to reach specific frequencies of resistant bacteria our OpvAB-based approach requires mixing bacteria at frequencies 100-fold lower (e. g., to reach frequencies of 1 × 10 −6 Km r resistant bacteria, the P opvAB ::BI-aac3 strain was mixed at a frequency of 1 × 10 −4 ). Thus, an advantage of our opvAB switch-based approach is that it can be expected to be less affected by infection bottlenecks that could otherwise eliminate very small subpopulations of bacteria present in the inoculum 21 . For example, one such bottleneck is observed during cecum colonization by Salmonella in mice 2-4 days after oral infection, and is dependent on the inflammatory response induced by S. enterica invading epithelial cells 22,23 .
Additional applications of the opvAB switch can be envisaged, including the design of bistable biosensors. For instance, a strain harboring an P opvAB ::gfp fusion might be useful to detect bacteriophages in environmental samples using flow cytometry 24,25 , and to identify DNA methylation inhibitors in screens for novel antimicrobial drugs 26,27 . Sensors of this kind can be expected to be selective as growth will occur under specific circumstances only. Furthermore, use of fluorescence to monitor growth of ON cells can be expected to be sensitive and rapid, and constitutive expression may contribute to robustness, avoiding the problem of instability of transcription-based gene circuits 28 . Besides biosensor design, formation of phenotypic subpopulations under epigenetic control might have additional applications in synthetic biology: for instance, division of labour between subpopulations performing distinct segments of a catabolic pathway might optimize biodegradation processes 29 .

Methods
Strains and strain construction. Strains of Salmonella enterica serovar Typhimurium and Escherichia coli used in this study are listed in Table 1. Strain construction by targeted gene disruption was achieved using plasmids pKD3, pKD4 or pKD13 as templates to generate PCR products for homologous recombination 30 . Antibiotic resistance cassettes introduced during strain construction were excised by recombination with plasmid pCP20 30 . Primers used in strain construction are shown in Table 2. For the construction of translational lac fusions on the S. enterica chromosome, FRT sites generated by excision of Km r cassettes were used to integrate plasmid pCE40 31 . For construction of fluorescent fusions, a DNA fragment containing a promoterless green fluorescent protein (gfp) gene and a chloramphenicol resistance cassette was PCR-amplified from plasmid pZEP07 32 , and the resulting PCR product was integrated into the chromosome of each strain. For construction of strains that carry antibiotic resistance genes under P opvAB control, a counterselectable cassette containing sacB and Ap R genes was amplified from strain DA52596 using the oligos opvAB-ampsacB-F and ampsacB-gfp-R. The PCR product was integrated into the chromosome of SV6727 and SV6729 respectively, generating the intermediate strains MN441 www.nature.com/scientificreports www.nature.com/scientificreports/ and MN442, respectively. Antibiotic resistance genes were introduced into these strains by targeted gene disruption 30 , and transformants in which the ampicillin-sacB cassette had been excised were selected on minimal plates containing sucrose.
Transductional crosses using phage P22 HT 105/1 int201 33 were used for transfer of chromosomal markers between S. enterica strains 34 . To obtain phage-free isolates, transductants were purified by streaking on green plates 35 . Phage sensitivity was tested by cross-streaking with P22 H5.
Directed construction of point mutations was achieved using the QuikChange ® Site-Directed Mutagenesis Kit (Stratagene) using the suicide plasmid pDMS197 36 and propagated in E. coli CC118 λ pir. Plasmids derived from pMDS197 (pIZ2224 and pIZ2234) were transformed into E. coli S17-1 λ pir. The resulting strains were used as donors in matings with S. enterica SV9700, selecting tetracycline-resistant transconjugants on minimal plates. One transductant from each mating was propagated as strains SV9701 and SV9702. culture media and growth conditions. Bertani's lysogeny broth (LB) 37  Growth curves. Plates were incubated at 37 °C with shaking on an automated microplate reader (Synergy HTX Multi-Mode Reader, Biotek), and the absorbance at 600 nm for each well was measured every 30 min. Each sample was assayed by triplicate. Growth of SV9700, SV9701, SV9703, DR22, DR23 and DR24 strains was monitored in NCE-lactose and NCE-glucose. Growth of SV9704, SV9705, SV9706, SV9707 was monitored in LB broth with and without antibiotics. www.nature.com/scientificreports www.nature.com/scientificreports/ calculation of phase variation rates. Phase variation rates were estimated as described by Eisenstein 40 .
Briefly, a strain harboring a lacZY fusion was plated on LB + X-gal and colonies displaying an ON or OFF phenotype after 16 h growth at 37 °C were selected, resuspended in PBS and re-spread on fresh LB + X-gal plates. Phase variation frequencies were calculated using the formula (M/N)/g where M is the number of cells that underwent phase variation, N the total number of cells, and g the total number of generations that gave rise to the colony. epsilometer (e) tests of antibiotic resistance. Etest strips were purchased from bioMérieux. Mixtures of overnight cultures of bacteria grown in MH broth were diluted 1:25 in phosphate buffered saline (PBS) to reach cell densities of 0.5 MacFarland or about 1.5 × 10 8 CFU/mL. Bacteria were plated onto MH agar plates using sterile cotton swabs dipped in the cell suspensions, and a Etest strip was applied on top. Plates were incubated 18 h at 37 °C before reading the results and taking pictures.
Population analysis profile (PAP) tests. PAP tests were performed on MH agar plates supplemented with increasing amounts of kanamycin (Sigma Aldrich) as described previously 15 . Five µl of overnight cultures in MH broth (containing approx. 3 × 10 9 cells/ml) and serial dilutions (down to 10 −6 ) were spread on MH plates containing no antibiotics (for total CFU determination) or different concentrations of kanamycin. The plates were incubated overnight and the colonies were counted. Colony numbers were plotted in a graph to determine if the PAP fulfilled the criteria for heteroresistance (at least 8-fold difference in antibiotic concentration between the highest non-inhibitory concentration and the highest inhibitory concentration).
To prepare mixtures of resistant and susceptible cells, three isolated colonies of SV9776 (P opvAB ::BI-aac3::gfp, kanamycin resistant in the ON state) and SV9777 (P opvAB ::gfp, always kanamycin susceptible) were grown overnight in 2 mL MH broth at 37 °C under shaking. Pure cultures of each overnight or three independent sets of SV9776:SV9777 mixtures at proportions ranging from 1:10 to 1:10,000 were used for PAP tests.