The bacterial strains and plasmids used in this study are listed in Table 1. All B. subtilis strains are isogenic with BG214. Escherichia coli BL21(DE3) cells harboring pBT290-borne ε gene under the transcriptional control of the T7 RNA polymerase-dependent promoter (P_T7), or pCB920-borne wild type (wt) ζ gene under the control of P_T7 and ε gene under its native RNA polymerase σ^A-dependent promoter (P_ω) were used for protein purification as described (Camacho et al., 2002; Tabone et al., 2014b).

The BG214 derivatives were grown to mid-exponential phase (~5 × 10⁷ cells ml⁻¹) at 37°C in minimal medium S7 (MMS7) supplemented with the necessary amino acid (Lioy et al., 2006). Except for ΔrelA, strains were grown in MMS7 with methionine and tryptophan at 50 μg ml⁻¹ each (Lioy et al., 2006). The ΔrelA strain shows an “auxotrophy phenotype” for valine, leucine, isoleucine and threonine, and was also supplemented with these amino acids (25 μg ml⁻¹ each) (Roche, Germany; Lioy et al., 2006).

BG1125 bearing lacI-P_hsp wt ζ and pCB799-borne xylR-P_xylA wt ε (Table 1), in which ζ gene expression (transcribed by P_hsp) is regulated by IPTG (Calbiochem, Spain) addition and the ε gene (transcribed by P_xylA) is regulated by xylose (Xyl, Sigma, USA) addition (Lioy et al., 2012), was grown in MMS7 supplemented with Xyl (0.05%). In the absence of IPTG [Sigma, USA] there are ~40 ζ toxin monomers/colony-forming units (CFU), which lead to genetic rearrangement. To titrate basal ζ toxin levels, traces of Xyl (0.05%) were added to allow synthesis of low but marked ε₂ antitoxin levels by the pCB799-borne ε gene. After IPTG addition, toxin ζ concentration increased in a very short time (10 min) up to ~1,500 ζ monomers/CFU, and its steady-state level remained for at least 240 min; these toxin levels are considered the “physiological concentration” (Lioy et al., 2012). At indicated times, 0.5% Xyl was added to induce antitoxin ε₂ expression, and the culture was incubated 15 min before being plated without inductor or with 0.5% Xyl (Lioy et al., 2012).

In BG689 or BG1145 bearing the xylR-P_xylA ζY83C cassette (Table 1), expression of the toxin ζY83C variant was induced by addition of 0.5% Xyl. BG689 or BG1145 cells were grown in MMS7 to ~5 × 10⁷ cells ml⁻¹ at 37°C. Xylose addition increased ζY83C levels to a plateau within the first 10 min, and the steady-state level of the toxin remained for at least 240 min (Tabone et al., 2014b).

Where indicated, toxin and/or antitoxin expression was induced by adding IPTG and/or Xyl. Before plating, cells were centrifuged and resuspended in fresh LB medium to remove the inductor or the antibiotic, and dilutions were plated on LB agar plates containing glucose (which switches off xylR-P_xylA cassette expression) or Xyl to express the ε₂ antitoxin. The survival rate was derived from the number of CFU in a given condition relative to CFU of the non-induced/non-antibiotic-treated control. Except ΔrelA, cells grew in MMS7 with a doubling time of 50–60 min. The doubling time of ΔrelA cells increased 1.4-fold compared to the BG689 strain. Normal-sized and small colonies were observed in the ΔrelA and ΔdisA codY contexts. All plates were incubated for 20 h at 37°C.

The minimum inhibitory concentration (MIC) of Amp [Sigma, USA] was estimated by exposing 1–3 × 10⁶ cells ml⁻¹ (16 h, 37°C) in MMS7 with shaking (240 rpm). The Amp concentration used (3 μg ml⁻¹) was twice the MIC (2x MIC). In the absence of inducer, the presence of the ζY83C (BG689 strain) or the ζ gene (BG1125 bearing pCB799) does not affect the MIC (Tabone et al., 2014b).

The S. pyogenes pSM19035-encoded ζ gene was overexpressed in E. coli BL21(DE3) cells from a rifampicin-resistant T7 RNAP-dependent promoter as reported (Tabone et al., 2014a). In short, IPTG was added to induce the expression of T7 RNAP that transcribed wt ζ toxin, and 30 min later rifampicin (Fluka, USA), was added to selectively block the expression of the ω and ε genes. After 120 min of incubation and full decay of the ε₂ antitoxin, the cells were harvested. The over-expressed long-living ζ toxin was purified in two steps as described (Tabone et al., 2014a). The fractions containing the ζ protein were dialyzed against buffer A (50 mM Tris-HCl pH 7.5, 80 mM NaCl) containing 50% glycerol and stored at −20°C. The ε gene was overexpressed in E. coli BL21(DE3) cells harboring pBT290 under the control of rifampicin-resistant P_T7 (Ceglowski et al., 1993), and antitoxin ε₂ was overexpressed, and purified as described (Camacho et al., 2002). The purified protein was stored in buffer A containing 50% glycerol at −20°C (Camacho et al., 2002).

The ATPase, dATPase or GTPase activities of ζ toxin were measured using a (d)NTP/NADH-linked assay (De La Cruz et al., 2000; Yadav et al., 2012). Reactions (50 μl) contained the indicated concentration of ζ toxin and the NADH enzyme mix (310 μM NADH [Roche, Germany], 100 U ml⁻¹ lactic dehydrogenase [Sigma, USA], 500 U ml⁻¹ pyruvate kinase [Roche, Germany], and 2.5 mM phosphoenolpyruvate [Roche, Germany]) in buffer B (50 mM Tris-HCl pH 7.5, 50 mM NaCl, 10 mM MgOAc, 1 mM DTT, 50 μg/ml BSA) with the indicated concentration of ATP, GTP or dATP, and 10 mM UNAG or uridine diphosphate-N-acetylgalactosamine (UNAGal) [Sigma, USA]. We determined the specific (d)NTPase activity (in μM) by measuring the (d)NDP production rate using a Shimadzu CPS-20A dual-beam spectrophotometer as described (Yadav et al., 2012). A standard curve with known amounts of NADH was obtained and used to convert the rate of ADP/GDP/dADP production from absorbance/time to concentration/rate (De La Cruz et al., 2000; Yadav et al., 2012).

Toxin ζ hydrolyses ATP, even in the presence of a 10- to 15-fold excess of cold GTP (Tabone et al., 2014a), suggesting that toxin ζ prefers ATP to GTP (Tabone et al., 2014a). To examine these reactions, we purified toxin ζ in the absence of its cognate antitoxin ε₂.

In the absence of UNAG, toxin ζ does not undergo autophosphorylation or hydrolyze NTP; with UNAG (2 mM) and 500 nM toxin ζ, only traces of the γ-phosphate of ATP (0.5 mM) were transferred to UNAG (Tabone et al., 2014a). We tested directly for nucleotide used preferentially by toxin ζ. Limiting ζ concentrations (60 nM) were used to analyze ζ-mediated ATP, GTP or dATP hydrolysis in physiological concentrations of UNAG and of nucleotides. The B. subtilis intracellular UNAG, ATP, GTP, and dATP pools approached ~10, ~10, ~5 and ~0.02 mM, respectively (Lopez et al., 1979; Lioy et al., 2012; Bittner et al., 2014).

Toxin ζ did not hydrolyze purine nucleotide when UNAG was omitted (Figure 2A). At physiological UNAG and ATP concentrations (10 mM each), toxin ζ (60 nM) hydrolyzed ATP in a reaction that rapidly reached saturation, which suggested that ζ is a UNAG-dependent NTPase. The final rate of ζ ATP hydrolysis approached the maximum rate (K_cat) of 1520 ± 120 min⁻¹ (Figure 2A).

The UNAG-dependent ζ ATPase activity was then compared with a bona fide ATPase enzyme. When the single-stranded DNA-dependent RecA ATPase was measured in parallel, B. subtilis RecA hydrolyzed ATP at near the previously observed K_cat of 9 ± 0.3 min⁻¹ (Yadav et al., 2014; Carrasco et al., 2015), which suggested that ζ is a very robust ATPase. UNAG-dependent ζ-mediated ATP hydrolysis was nonetheless sensitive to variations in ATP concentration, because when ATP was reduced to half (5 mM), the K_cat was reduced ~3-fold (510 ± 44 min⁻¹).

When ATP was replaced by physiological GTP concentrations (5 mM), ζ was able to hydrolyze GTP in a UNAG-dependent manner and the reaction reached saturation in ~7 min. The final steady state rate of GTP hydrolysis was reduced by ~5-fold (K_cat 280 ± 47 min⁻¹) compared with physiological ATP concentrations (Figure 2A). Increasing the GTP concentration to 10 mM did not improve the reaction.

We analyzed the potential role of dATP as a substrate (Figure 2A). In the presence of physiological UNAG and dATP concentrations (10 and 0.02 mM, respectively), we observed no ζ-mediated UNAG-dependent dATP hydrolysis (Figure 2A). To test whether ζ catalyzes dATP hydrolysis, we increased its concentration artificially. At a 10-fold excess of dATP (0.2 mM), ~3 min lag time was needed to reach the steady state rate of ζ-mediated dATP hydrolysis; however, saturation was not reached in 30 min reaction (Figure 2A). With a 10-fold excess of dATP, its hydrolysis was reduced by ~20-fold compared with ATP. ATP is probably the preferred ζ nucleotide cofactor.

Toxin ζ interacts specifically with UNAG rather than UDP-glucose (Mutschler et al., 2011); in addition, B. subtilis GalE is able to interconvert UNAG and UDP-N-acetylgalactosamine (UNAGal), and the cell wall contains N-acetylglucosamine and N-acetylgalactosamine (Soldo et al., 2003). To determine whether UNAGal, a C-4 epimer of UNAG, can activate toxin ζ ATPase activity, we carried out ATPase assays with increasing UNAGal concentrations.

In the absence of UNAG or UNAGal (minus UNAhexamines), ATP hydrolysis by toxin ζ was at background level (Figure 2B). Quantitative analysis of these reactions showed that at physiological UNAGal concentrations, the final ζ-mediated ATP hydrolysis rate was ~85-fold lower (K_cat 20 min⁻¹) than ζ in the presence of UNAG. In the presence of a UNAGal excess (10 mM), the K_cat was slightly increased (28 min⁻¹), but was still ~60-fold lower than that at physiological UNAG concentrations (Figure 2B); this result indicates that ζ ATPase activity is specifically stimulated by UNAG rather than by UNAGal. PetZ similarly accumulates UNAG-3P after 60 min, and UNAGal-3P after 720 min incubation (Mutschler et al., 2011).

In vitro, the ζε₂ζ complex is reported to hydrolyze ATP and phosphorylate UNAG to form inactive UNAG-3P (Mutschler et al., 2011). In contrast, in vivo experiments showed that the ε₂ antitoxin inhibits the effect of toxin ζ, perhaps by forming the inactive ζε₂ζ complex (Lioy et al., 2006, 2010). To test whether toxin ζ hydrolyzes ATP in the presence of the antitoxin ε₂, both proteins were purified separately (Camacho et al., 2002; Tabone et al., 2014a) and UNAG-dependent ATPase activity measured.

The antitoxin ε₂, alone or with UNAG, did not hydrolyze ATP (Figure 2C). In the presence of UNAG and ATP, the rate of UNAG-dependent ζ-mediated ATP hydrolysis was reduced by increasing antitoxin ε₂ concentrations (Figure 2C). At ζ:ε₂ ratios of 1:0.5 or 1:1, the kinetics of ζ-mediated ATP hydrolysis was initially unaltered, but ATP hydrolysis was inhibited after 5 min. At a slight ε₂ excess (1:2 ratio), the antitoxin inhibited ζ ATPase activity (Figure 2C). Results were similar when both proteins were preincubated (5 min) at a 1:1 ζ:ε₂ ratio (ζε₂ζ complex; not shown), which suggests that when it interacts with ζ, the antitoxin occupies the ATP binding pocket (Meinhart et al., 2003) and inhibits toxin ATPase activity. This is consistent with the crystal structure of the biologically inactive ζε₂ζ complex and with the interpretation that antitoxin ε₂ is necessary and sufficient to inactivate toxin ζ. It is likely that the long reaction incubation time (24 h) and/or low ε₂ stability could explain discrepancies with the previous report (Mutschler et al., 2011).

The release of toxins from their cognate antitoxins [or induction of toxin expression (+Drug in blue in Figure 1)], should lead to a bimodal time-inactivation curve if persisters appeared (dotdashed line). This deviates from the simple decay, anticipated for a population of only susceptible cells (dashed line) or for a uniformly tolerant bacterial population (dotted line, in Figure 1; Brauner et al., 2016; Harms et al., 2016). Inactivation of the toxin by expression of the antitoxin (+Drug in red) should lead to recovery of the plating efficiency (red solid line) if the toxin is bacteriostatic (Figure 1). To test whether expression of physiological levels of free toxin ζ induce persistence (dotdashed line) or tolerance (dotted line) and to study the mechanism used for such a phenotype (bacteriostasis or bacteriolysis) we performed long term survival assays. Toxin ζ was induced for a long period, and then antitoxin ε₂ expression was induced (Figure 3A).

Bacillus subtilis BG1125 bearing the ζ gene under the control of IPTG induction is prone to rearrangement in the absence of IPTG (Lioy et al., 2012). To overcome this effect, the pCB799-borne ε gene under the control of Xyl was transferred into the background (Table 1; see Section Materials and Methods). BG1125 cells bearing pCB799 were grown in MMS7 supplemented with 0.05% Xyl, to ~5 × 10⁷ cells ml⁻¹ (OD₅₆₀ = 0.2), and expression of the ζ gene was induced by IPTG addition (time zero). Cells, which formed colonies after plating on LB agar without IPTG, showed a bimodal time-inactivation curve suggesting the presence of persisters (Figure 3A), rather than showing a uniform simple decay, expected for tolerant cells (Figure 1, dotted line).

To test whether the persisters are due to noise that causes instability in a bacterial population (a reduced cell fraction transiently insensitive to toxin action) or noisy gene expression (a reduced fraction with no toxin expression), we maintained IPTG induction up to 900 min, after which cells were plated in the absence of the inducer. In the former case, only a fraction of the non-replicating dormant cells would exit the arrest state and resume growth after plating without IPTG, whereas in the latter case, cell proliferation of persisters is predicted to increase 8- to 16-fold. After IPTG addition, the small persister subpopulation increased slightly (~3-fold) during the first 240 min, to later remain apparently constant (Figure 3A); this suggested negligible biological noise during the first 240 min, and persisters were transiently insensitive to toxin action.

Massive expression of toxin PezT or ζ triggers an irreversible bactericidal effect in E. coli grown in rich medium or B. subtilis grown in minimal medium, respectively (Zielenkiewicz and Ceglowski, 2005; Mutschler et al., 2011), but physiological concentrations of free toxin ζ induce a reversible bacteriostatic state (Lioy et al., 2012). To identify the source of these discrepancies, we tested whether IPTG-induced growth arrest in B. subtilis cells is fully reversible after antitoxin ε₂ expression triggered by 0.5% Xyl (15 min), followed by plating in LB agar with 0.5% Xyl but lacking IPTG. Antitoxin ε₂ expression was sufficient to reverse growth arrest, and most cells recovered proliferation, even after 900 min of toxin ζ action (Figure 3A). Although, a reduced fraction (10 to 15% of total cells) were stained with propidium iodide, suggesting a membrane compromise in these cells. It is likely that toxin ζ induces a reversible inhibition of cell growth, and that antitoxin ε₂ expression is necessary and sufficient to switch off toxin-induced responses, with cells awakening and forming colonies even after 15 h of toxin incubation (see Figure 1, solid red line), but 10 to 15% of total cells might loss cell viability.

Bacillus subtilis encodes one bifunctional RelA synthase-hydrolase and two mono-functional SasA (also termed YwaC/RelP/Sas1) and SasB (YjbM/RelQ/Sas2) synthases [see Nanamiya et al., 2008; Srivatsan and Wang, 2008]. In the absence of RelA, an excess of GTP over GDP as well as baseline levels of (p)ppGpp “dysregulated” by the SasA and SasB synthases, increase toxin persistence or tolerance by >150-fold (Tabone et al., 2014b). This effect correlates with dysregulated (p)ppGpp levels. Lowering the GTP levels without affecting (p)ppGpp, by treating cells with decoyinine (a GMP synthetase inhibitor), the persistent rate was indistinguishable between treated or untreated ΔrelA cells (Lioy et al., 2012).

To test whether the CFU increase correlates with a simple decay curve predicted from a uniform tolerant bacterial population or with a biphasic time-inactivation curve due to persistence (Figure 1), we analyzed toxin expression in the relA⁺ (BG689) or ΔrelA (BG1145) cells bearing the xylR-P_xylA ζY83C cassette (Table 1). The relA⁺ and ΔrelA cells were grown in MMS7 to ~5 × 10⁷ ml⁻¹, expression of the ζY83C gene was induced with 0.5% Xyl, and the time-inactivation curve was analyzed. In the absence of RelA, a typical biphasic curve was observed upon expression of physiological concentrations of the toxin, with an ~160-fold (~5 × 10⁻³) increase in the rate of persisters after plating on LB agar without Xyl (Figure 3B).

The second messenger c-di-AMP, which is at the heart of cell wall homeostasis, is produced mainly by Gram-positive bacteria of the phyla Firmicutes and Actinobacteria, and by some species of the δ-Proteobacteria class (Corrigan and Grundling, 2013). In Firmicutes, high or low c-di-AMP levels indirectly increase (p)ppGpp (Rao et al., 2010; Corrigan et al., 2015), whereas in Staphylococcus aureus, they, respectively, increase or decrease β-lactam tolerance/resistance (Corrigan et al., 2011, 2015). It is not known whether c-di-AMP has a role in toxin ζ responses to stress.

In Firmicutes, intracellular c-di-AMP levels are precisely regulated by two sets of enzymes with opposite effects and by two purine nucleotides. The diadenylate cyclases (DAC) synthesize c-di-AMP from two ATP molecules, the phosphodiesterases (PDE) degrade c-di-AMP into pApA; (p)ppGpp and pApA inhibit PDE enzyme activity (Rao et al., 2010; Corrigan and Grundling, 2013; Huynh and Woodward, 2016), which predicts that c-di-AMP levels increase during starvation. Exponentially growing B. subtilis cells express two DAC (DisA and CdaA) and two PDE enzymes (GdpP and PgpH; Rao et al., 2010; Corrigan et al., 2011; Corrigan and Grundling, 2013; Commichau et al., 2015; Huynh and Woodward, 2016). The absence of both DAC or of both PDE causes aberrant physiology and synthetic lethality when the medium contained high K⁺ (5 mM KCl), but one representative of each family can be deleted with no apparent effect (Corrigan et al., 2011; Corrigan and Grundling, 2013; Commichau et al., 2015; Huynh and Woodward, 2016; Gundlach et al., 2017). C-di-AMP levels vary marginally (2- to 3-fold) in cells lacking DisA, CdaA or GdpP compared to the wt strain (Oppenheimer-Shaanan et al., 2011; Gándara and Alonso, 2015).

To determine how B. subtilis cells respond to toxin-mediated stress, we induced transient toxin ζY83C expression and studied the effect of disturbing the c-di-AMP metabolic balance by deleting one DAC (DisA) or one PDE (GdpP) enzyme on an isogenic background (Table 1). In parallel, Amp was used as a second stressor at twice the MIC, in anticipation that toxin expression and Amp would respond to different physiological cues. The MIC of Amp was similar in all strains tested. After Amp exposure, a subpopulation of clonal cells yielded a biphasic time-kill curve (twodotted dashed line), which indicated that they persisted rather than becoming Amp-tolerant (dotted line; Figure 1). Similar biphasic curves are reported for other bacterial species after Amp exposure (Lewis, 2010; Amato et al., 2013; Brauner et al., 2016; Harms et al., 2016).

Bacillus subtilis cells that lack GdpP show intracellular c-di-AMP levels ~2-fold higher than those of the wt strain (Gándara and Alonso, 2015). Absence of GdpP indirectly increases (p)ppGpp pools (Gundlach et al., 2015; Zhu et al., 2016), which suggests that a small number of specific signaling nucleotides integrate and coordinate key metabolic intersections in response to variation of the intracellular c-di-AMP pool. We constructed and analyzed a strain bearing the xylR-P_xylA ζY83C cassette in the context of ΔgdpP (Table 1). Regulated ζY83C expression in the wt or ΔgdpP contexts produced a typical biphasic survival curve, with an initial rapid decrease in CFU and a persistent subpopulation with a stable number of CFU (between 10 and 300 min; not shown); for direct comparison of the various strains, data at 120 min are shown (Figure 4). In the wt strain, ζY83C expression used (p)ppGpp to mediate rapid inhibition of cell proliferation, and a cell subpopulation entered a toxin- (~7.2 × 10⁻⁵) or Amp-persistent state (~2.1 × 10⁻³; Figure 4), as reported (Lioy et al., 2012). In the ΔgdpP strain, after transient toxin expression, the persistence rate decreased by ~10-fold (7 × 10⁻⁶), but did not significantly affect the persistence rate after Amp addition (~1.7 × 10⁻³). In the absence of GdpP, exposure to Amp and Xyl decreased the rate of persisters by ~16-fold (~3 × 10⁻⁶ survivals) compared to the wt strain (Figure 4), which suggests that a subset of toxin or Amp persisters randomly switched to the susceptible state and were targeted by Amp or the toxin.

Cells lacking DisA have ~2-fold lower levels of intracellular c-di-AMP than wt cells (Oppenheimer-Shaanan et al., 2011; Gándara and Alonso, 2015). We constructed the xylR-P_xylA ζY83C ΔdisA strain (Table 1), and found that the persister cell rate was slightly affected by Amp addition (<2-fold, ~1.5 × 10⁻³ survivals). Transient toxin ζY83C expression increased persisters ~4-fold (~3 × 10⁻⁴) in the ΔdisA compared to the wt strain (Figure 4). Transient toxin ζY83C expression and Amp addition did not notably affect colony formation compared to addition of Xyl alone (Figure 4).

Transient toxin ζ expression decreased the GTP pool in exponentially growing B. subtilis cells (Lioy et al., 2012). Intracellular GTP levels have a central role in modulating the stringent response and in reprogramming gene regulation to allow appropriate adaptation to stress. CodY, a GTP-binding protein, is a pleiotropic regulator that senses intracellular branched chain amino acids and GTP levels (Sonenshein, 2007). Low GTP levels, as found during acute stress, release CodY from DNA, leading to deregulation of genes involved in adaptation to nutrient limitation (Ratnayake-Lecamwasam et al., 2001; Belitsky and Sonenshein, 2013; Bittner et al., 2014). The role of CodY in toxin ζ stress responses is unknown. To test whether CodY modulates toxin and/or antibiotic persistence, we constructed the xylR-P_xylA ζY83C codY strain (Table 1). Lack of CodY did not markedly alter the Amp persister rate (~2.5 × 10⁻³ survivals; Figure 4). After Xyl exposure, however, we observed a slight decrease in the toxin persister rate (~2-fold) compared to wt cells (Figure 4). Transient Xyl and Amp addition decreased CFU (~8 × 10⁻⁶ survivals) and the level of persisters decreased by ~5-fold compared to the wt strain (Figure 4), which suggest that cells lacking CodY adapt poorly to toxin and Amp stress.

The absence of DisA increased, and of CodY or GdpP decreased the rate of toxin persistence (Figure 4). Since lack of both CodY and GdpP strongly affected cell recovery, but the combined absence of CodY and DisA showed a less stringent phenotype, we constructed the xylR-P_xylA ζY83C ΔdisA codY strain (BG1527; Table 1). The BG1527 strain yielded colonies with diffuse borders, a 3:1 normal:small size ratio, and viability reduced by ~1.4-fold compared to parental BG689 strains, but lack of CodY and DisA did not notably alter the rate of Amp persisters (~2.5 × 10⁻³ survivals). Following toxin ζY83C expression, we observed a moderate decrease (~2-fold) in the toxin persister rate (~3.5 × 10⁻⁵ survivals) compared to wt cells, similar to the codY strain (Figure 4). Addition of Xyl and Amp greatly decreased the persistence rate (~3 × 10⁻⁶ survivals; Figure 4). Different clonal subpopulations of persisting cells thus probably evolved differently in response to the toxin and Amp.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer RDO and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.