Antibiotic tolerant B. subtilis cells
The antibiotic tolerance of persisters ensues from the fact that these cells are in a dormant state 5,6. It has been shown that Escherichia coli cells in the stationary growth phase display an antibiotic tolerance that is reminiscent of persister cells 22. To examine whether the same is true for B. subtilis, we used a sporulation deficient mutant ∆spoIIE, since B. subtilis spores are insensitive to antibiotics and, thus, would confound the measurements. The ∆spoIIE deletion mutant was grown overnight (18 h) in LB medium to stationary phase, and subsequently treated with 10 x minimal inhibitory concentrations of either vancomycin, kanamycin or ciprofloxacin for 10 h (see Table S4 for MIC). Samples for CFU measurements were taken every 2 h, and an exponentially growing culture was used as non-dormant comparison. As shown in Fig. 1, the latter culture was sensitive for all three antibiotics whereas the overnight culture was not, indicating that non-sporulating stationary phase B. subtilis cells can be used as a simple model system for antibiotic tolerant persister cells.
These stationary cells maintained membrane potential levels comparable to actively growing cells (Fig. S1). Many peripheral membrane proteins use an amphipathic alpha helix domain as a reversible membrane anchor and we have shown that the binding of such an anchor domain is strongly stimulated by the membrane potential 13. Depolarization of the membrane potential might therefore disturb the normal localization of different peripheral membrane proteins, which, in turn, could affect the viability of persister cells. To test whether dormant B. subtilis cells are sensitive for membrane depolarization, we exposed them to the potassium ionophore valinomycin, which specifically dissipates the transmembrane potential 23,24 (Fig. S1). To assure full dissipation over a 10 h incubation period, we used 100 μM valinomycin which is 10 x the MIC for exponentially growing cells (Table S4). The addition of valinomycin reduced the viable count of exponentially growing cells by 90 % after 2 h incubation (Fig. 1c). The dormant culture showed some degree of resilience compared to its actively growing counterparts, but after some time the viability decreased and after 6 h approximately 90 % of the cells were killed (Fig. 1d), suggesting that membrane depolarizing compounds might indeed be useful to combat persisters.
Cause of killing
The killing kinetic of valinomycin is quite different between exponentially growing and dormant cells, and in the latter case showed a clear acceleration over time (Fig. 1c & 1d). This made us wonder what the actual cause is why dormant cells eventually succumb when their membrane is depolarized. It has been shown that depolarization of the B. subtilis cell membrane results in an uncontrolled autolysin activity and cell lysis 25. To determine whether this could be responsible for the killing of dormant cells, we exposed a mutant lacking all major autolysins (∆lytA, ∆lytB, ∆lytC, ∆lytD, ∆lytE and ∆lytF) to valinomycin. However, this deletion mutant did not show an increase in viability (Fig. 2a). E. coli persisters can be killed by the induction of cryptic prophages 26. However, a B. subtilis strain cured of all prophages exhibited a similar sensitivity for valinomycin as the wild type strain (Fig. 2b), indicating that the reduction in viable count is also not related to prophage activation.
During the viable count measurements we noticed that after prolonged valinomycin treatment the colony size started to vary and smaller colonies emerged (Fig. 2c). Such variation in colony size is also observed when bacterial cells are treated with DNA damaging agents 27, suggesting that membrane depolarization might cause DNA damage. If this is the case, then a DNA repair mutant, such as a ∆recA mutant, should be more sensitive to valinomycin. This was indeed the case (Fig. 2d).
Generating ROS
A common cause of DNA damage is the accumulation of reactive oxygen species (ROS) in the cell. However, it seems unlikely that dormant B. subtilis cells would produce ROS upon membrane depolarization, since the accumulation of ROS is normally associated with hyper respiration, which can be prevented by membrane potential dissipating agents 15,28,29. Moreover, antibiotics that have been shown to generate ROS, including norfloxacin, vancomycin and kanamycin 1–3, are not active in dormant cells. ROS is primarily a by-product of aerobic respiration and under anaerobic conditions ROS levels are generally much lower 30. When dormant B. subtilis cells were placed into an anaerobic chamber for 10 h only a fraction of cells survived (Fig. 3a). Although B. subtilis can grow anaerobically, it requires certain medium conditions and sufficient time to adapt 31, which likely explains this reduction in viability. Nevertheless, the fraction of surviving cells was still smaller when valinomycin was added (Fig. 3a).
Since the viability of B. subtilis is greatly affected by anaerobic conditions, even in the absence of valinomycin, this experiment did not reveal whether the DNA damage caused by valinomycin treatment was due to the accumulation of ROS. Therefore, we took another approach and tested the sensitivity of a spx deletion mutant. Spx is the key regulator of the oxidative stress response in B. subtilis and is required for survival in the presence of strong ROS-inducing compounds such as paraquat 32. Interestingly, a ∆spx deletion mutant was even more sensitive to valinomycin than the ∆recA mutant (Fig. 3b), supporting the idea that ROS is a key contributor to the killing of depolarized cells. Finally, to directly show that ROS was generated, we used the specific fluorescent ROS probe 2',7'-dichlorodihydrofluorescein diacetate (H2DCFDA) 3. Dormant B. subtilis cells were incubated with H2DCFDA and exposed to either valinomycin or the well-known ROS inducer paraquat for 2 and 4 h 33. The increase in fluorescence in cells was measured using fluorescence light microscopy. As shown in Fig. 3c, incubation with valinomycin for 2 h resulted in a strong increase in ROS that remained high over a 4 h period and even exceeded the effect of 1 mM paraquat. Thus, membrane depolarization of dormant B. subtilis cells does lead to the accumulation of lethal levels of ROS.
Determining ROS type
Hydroxyl radicals (•OH) and superoxide radicals (O2•-) are the main reactive oxygen species formed during aerobic growth 34. Incubation with antibiotics like norfloxacin, ampicillin and kanamycin results primarily in the formation of hydroxyl radicals made from hydrogen peroxide by the Fenton reaction 1–3. This conclusion was based on the fact that Fe2+-specific chelators like ferrozine or bipyridyl and the hydroxyl radical scavenger thiourea reduced the formation of radicals and diminished the killing efficiency of these antibiotics 1–3. Moreover, increased levels of catalase, which removes hydrogen peroxide 35 also suppressed the killing by either ampicillin, gentamicin or norfloxacin 3. Interestingly, the addition of 0.5 mM ferrozine had no effect on the viability of dormant cells when incubated with valinomycin, and 0.5 mM bipyridyl actually reduced the viability (Fig. 4a). Moreover, the presence of 150 mM thiourea also did not mitigate the effect of valinomycin and in fact enhanced its activity (Fig. 4b). In contrast to this, the presence of 10 mM tiron, a superoxide scavenger, almost completely inhibited the killing by valinomycin (Fig. 4b). Moreover, inactivation of KatA, the main catalase of B. subtilis 36, had no effect on the killing efficiency of valinomycin (Fig. 4c), whereas deleting sodA, encoding the main superoxide dismutase, strongly reduces the viable count upon valinomycin treatment (Fig. 4c). Indeed, ΔsodA cells show almost a doubling in fluorescence of the ROS probe H2DCFDA compared to wild type cells (Fig.4d). Apparently, depolarization of dormant B. subtilis cells triggers the production of lethal levels of superoxide radicals, without involvement of the Fenton reaction, which is different from the ROS generation caused by antibiotics like norfloxacin, vancomycin and kanamycin.
Possible source of superoxide production
The accumulation of lethal levels of ROS upon exposure to bactericidal antibiotics is believed to be triggered by a surge in NADH consumption that induces a burst in superoxide generation via the respiratory chain. These superoxide radicals then destabilize iron-sulfur clusters, leading to free ferrous iron which enables the Fenton reaction, thus creating deadly levels of hydroxyl radicals 1–3. The TCA cycle is the main source of NADH (Fig. 5a), and it was shown that inactivation of either isocitrate dehydrogenase or aconitase reduces the killing activity of norfloxacin, vancomycin and kanamycin 1. When we inactivated B. subtilis pyruvate dehydrogenase, which fuels the TCA cycle, cells became not less but more sensitive to valinomycin (Fig. 5b, ∆pdhB), suggesting that there is no surge in NADH levels upon membrane depolarization. The other well-known source of ROS is the electron transport chain 37–40. B. subtilis contains a classic electron transport chain composed of NADH dehydrogenase, succinate dehydrogenase, cytochrome bc complex, cytochrome c oxidase, also known as complex I, II, III and IV, respectively (Fig. 5a). Complex I and II feed electrons to the menaquinone pool. Inactivating one of them, by either deleting ndh, ndhF or sdhC, did not mitigate the killing by valinomycin but made it worse (Fig. 5b). Inactivation of glycerol-3-phosphate dehydrogenase, which reduces glycerol-3-phosphate to dihydroxyacetone phosphate using menaquinone 41, had no effect on the viability count (Fig. 5b, ∆glpD).
The absence of either the cytochrome b subunit (∆qcrB) or the cytochrome b/c subunit (∆qcrC) from complex III also strongly reduced the survival chance of valinomycin treated cells (Fig. 5c). Interestingly, when we deleted qcrA, encoding the Rieske-type iron-sulfur subunit of complex III, cells became more resilient to valinomycin, and after 10 h the viable count was more than 20-fold higher compared to wild type cells (Fig. 5c).
Impairing the expression of one of the cytochrome-c oxidase subunits of complex IV, by deleting either ctaC, ctaD, ctaE or ctaF, made dormant cells slightly more sensitive for membrane depolarization (Fig. 5d). B. subtilis contains two cytochrome-c electron carriers, cytochrome c550, which contains a transmembrane anchor, and cytochrome c551 that is anchored to the cell membrane via a diacyl glycerol tail (Fig. 5a) 42. Deleting one of them did not mitigate the killing by valinomycin (Fig. 5e, ∆cccA or ∆cccB). Inactivating both, by removing enzymes involved in their biogenesis (∆ccdA, ∆resA), strongly increased the sensitivity to valinomycin (Fig. 5e). B. subtilis possesses two alternative cytochrome oxidases that use quinol for their oxidation reaction, cytochrome bd oxidase and cytochrome aa3 oxidase (Fig. 5a). Inactivation one of them, by deleting either qoxB or cydA, considerably reduced the viability of dormant cells when incubated with valinomycin (Fig. 5f).
These surprising results suggest that not so much an active but rather an intact electron transport chain is important for the survival of dormant B. subtilis cells upon membrane depolarization, and that QcrA is a source of superoxide radicals. This Rieske protein contains a unique 2Fe-2S cluster in which one of the two iron atoms is held in place by two histidines rather than two cysteines. This cluster accepts an electron from the quinol anion and transfers it to the cytochrome heme iron 16,43,44, and in fact this step is a well-known source of superoxide radicals in mitochondria 45. To provide further support that QcrA is a likely source of the observed ROS when the membrane potential is dissipated, we deleted qcrA in the ΔrecA background strain, which has an impaired DNA repair system and has been shown to be especially sensitive for valinomycin (Fig. 2d). As shown in Fig. 5g, the absence of QcrA clearly attenuated the valinomycin susceptibility of dormant ΔrecA cells. Finally, we directly measured ROS production in the ∆qcrA mutant using the fluorescent probe H2DCFDA (Fig, 5h). Indeed, in the absence of the Rieske protein the average fluorescence signal after 4 h valinomycin treatment was considerably lower.
Cellular distribution of QcrA
It is difficult to understand how QcrA can produce lethal levels of superoxide radicals upon membrane depolarization. As mentioned in the introduction, depolarization of the cell membrane leads to the delocalization of different membrane proteins, including some transmembrane proteins 13,46. Possibly, the localization of electron transport chain components is also affected. To examine this, we constructed GFP fusions with the transmembrane proteins QcrA, QcrB and QcrC of complex III, CtaC and CtaD of complex IV, and the main cytochrome C (CccA), and expressed these fusions from an ectopic locus in the genome. Of note, we have not checked whether the GFP fusions influenced their enzymatic activities, since we were only interested in their localization. As shown in Fig. 6a, all dormant cells showed a more or less uniform fluorescent membrane stain. Interestingly, incubation with valinomycin caused a strong clustering of GFP-QcrA over time, whereas the localization of the other fusions were unaffected. When we repeated the experiment with exponentially growing cells, the clustering of GFP-QcrA was already visible within 10 min after the addition of valinomycin, whereas the other fusions showed no delocalization, also not after 30 min (Fig. 6b). Possibly, the distinct clustering of QcrA upon membrane depolarization is somehow responsible for the production of superoxide radicals.