Killing by Bactericidal Antibiotics Does Not Depend on Reactive Oxygen Species

See allHide authors and affiliations

Science  08 Mar 2013:
Vol. 339, Issue 6124, pp. 1213-1216
DOI: 10.1126/science.1232688

This article has a correction. Please see:

Antibiotic Mechanisms Revisited

Several recent studies have suggested that bactericidal antibiotics kill cells by a common mechanism involving reactive oxygen species (ROS). Two groups tested this hypothesis using diverse experiments, with both finding that quinolone, lactam, and aminoglycoside antibiotics had similar efficacy for killing in the presence or absence of oxygen (or nitrate). Liu et al. (p. 1210) saw no increase in hydrogen peroxide production in antibiotic-exposed cells and found no association between antibiotic exposure and the expected symptoms of oxidative damage, such as the breakdown of iron-sulfur clusters in enzymes or of hydroxyl radical injuries to DNA. Similarly, Keren et al. (p. 1213) found no correlation between the production of ROS, inferred from hydroxyphenyl fluorescein dye measurements, and bacterial survival, nor was there any significant protective effect engendered by thiourea. The results do not support a common mode of action for bactericidal antibiotics mediated by ROS.


Bactericidal antibiotics kill by modulating their respective targets. This traditional view has been challenged by studies that propose an alternative, unified mechanism of killing, whereby toxic reactive oxygen species (ROS) are produced in the presence of antibiotics. We found no correlation between an individual cell's probability of survival in the presence of antibiotic and its level of ROS. An ROS quencher, thiourea, protected cells from antibiotics present at low concentrations, but the effect was observed under anaerobic conditions as well. There was essentially no difference in survival of bacteria treated with various antibiotics under aerobic or anaerobic conditions. This suggests that ROS do not play a role in killing of bacterial pathogens by antibiotics.

Bactericidal antibiotics are an essential component of our antimicrobial arsenal. However, even the best bactericidal antibiotics have limited efficacy against dormant persister cells, specialized survivors that are phenotypic variants of the wild type (1, 2). Bactericidal antibiotics from each of the three main classes have unique mechanisms of killing. Fluoroquinolones convert DNA gyrase/topoisomerase into an endonuclease (3); aminoglycosides cause mistranslation, leading to production of toxic peptides (4, 5); and β-lactams lead to autolysis (6, 7). In dormant persisters, targets of these antibiotics have limited activity, resulting in tolerance (2). Understanding the detailed mechanism of killing is essential for developing better therapeutics. Recently, an alternative unified mechanism of antibiotic killing was proposed (8, 9), according to which bactericidal compounds, irrespective of their mode of action, induce formation of reactive oxygen species (ROS) by activating the electron transport chain, which kills bacterial cells. The initial study reported that antibiotics induce ROS production and that their quenching by thiourea protects Escherichia coli cells from killing (8). Subsequent studies report ROS-dependent killing in different bacterial species and detail the mechanism leading to cell death (1015). The ROS hypothesis of antibiotic killing became widely accepted but does not account for many observations. For example, mutants lacking ROS production have not been reported among drug resistant clinical isolates, and Streptococcus pneumoniae, which is highly susceptible to killing by bactericidal antibiotics, lacks an electron transport chain (16, 17), the proposed source of ROS. Thus, we decided to reexamine the role of ROS in cell death and consequently found that killing by antibiotics is unrelated to ROS production.

Thiourea was reported to protect E. coli from killing by norfloxacin, a fluoroquinolone antibiotic (8). Norfloxacin was used at a fairly low concentration (0.25 μg/ml), only two to four times as high as the minimal inhibitory concentration (MIC). The peak plasma concentration of norfloxacin has been reported to range from 1.3 to 1.6 μg/ml, with a half-life of 3 to 7 hours (18). We therefore examined the effect of thiourea on killing at a range of antibiotic concentrations that included clinically achievable levels.

Thiourea was added to an exponentially growing culture of E. coli together with antibiotic, and a sample was taken for a colony count 3 hours later, as described in (8). In the control without thiourea, the number of live cells decreased sharply with increasing concentrations of antibiotic, reaching a characteristic plateau of surviving persisters at 0.25 μg/ml of norfloxacin (Fig. 1A). Addition of thiourea increased the norfloxacin MIC twofold, from 0.06 μg/ml to 0.125 μg/ml. At low concentrations of norfloxacin, thiourea protected cells from killing, which was probably due to its effect on the MIC. This protective effect disappeared at higher concentrations, starting with 0.5 μg/ml. Protection by thiourea was only observed at 0.25 μg/ml but not at higher concentrations of norfloxacin in a time-dependent killing experiment (Fig. 1B). Thiourea was also reported to protect cells from killing by β-lactams and aminoglycosides administered at low levels (8). Similar to experiments with norfloxacin, we did observe a protection by thiourea at low concentrations of ampicillin or the fluoroquinolone ofloxacin, but this effect disappeared at higher, clinically relevant levels, and there was no protection against kanamycin (fig. S1).

Fig. 1

Killing and HPF fluorescence. E. coli wild-type strain BW25113 was cultured to mid-exponential phase in LB medium and treated with norfloxacin for 3 hours. After treatment, one sample was washed, diluted, and plated for colony count, and the other was analyzed for single-cell HPF fluorescence. (A) Colony count before (0) and after antibiotic treatment (blue bars, no thiourea; red bars, with thiourea). Thiourea was added at 150 mM. (B) Time-dependent killing with norfloxacin. Thiourea was added together with antibiotic where indicated at 150 mM. (C) HPF fluorescence after treatment with norfloxacin. HPF was present at 5 μM, and thiourea was added at 150 mM together with antibiotic (right panel). (D) Relationship between HPF fluorescence and the concentration of norfloxacin. The graph is based on data from (C). (E) Lack of a relationship between HPF fluorescence and killing (blue circles, norfloxacin; blue triangles, norfloxacin with thiourea; green circles, ampicillin; green triangles, ampicillin with thiourea, red circles, kanamycin). The graph is based on data from (A) to (C).

Antibiotics have been reported to increase production of ROS, as measured by increased fluorescence of a hydroxyphenyl fluorescein (HPF) dye (8). We similarly examined samples from cultures treated with norfloxacin for ROS production using HPF as a detector. Single-cell analysis by a flow cytometer showed a shift in HPF fluorescence in response to antibiotic treatment (Fig. 1C and fig. S2). The greatest shift was observed at 0.25 μg/ml of norfloxacin, the level at which it was used in the previous studies (8, 10). As the concentration of norfloxacin was increased above 0.25 μg/ml the shift in fluorescence decreased. At the highest concentration of antibiotic tested, 2.5 μg/ml, there was no difference in HPF fluorescence between treated and untreated cells (Fig. 1D). Addition of thiourea quenched the HPF signal (Fig. 1C). However, there was no correlation between the level of HPF fluorescence and the extent of killing by norfloxacin (Fig. 1, D and E) or ampicillin (Fig. 1E); HPF fluorescence did not change in cells treated with kanamycin (Fig. 1E). Similar results were observed with a different strain of E. coli (fig. S3).

To examine the relationship between ROS and killing more directly, cells from a population treated with norfloxacin were sorted onto nutrient medium to determine the level of survival in relation to HPF fluorescence. In a control, 384 cells from a culture containing HPF but no antibiotic were sorted onto a rectangular LB agar plate and produced 384 colonies (red arrow, Fig. 2, A and B), showing the high fidelity of the process. After treatment with norfloxacin at 0.25 μg/ml, 0.9% of the population representing highly fluorescent cells, and 0.9% of low-fluorescent cells, were sorted. At this concentration, norfloxacin killed 99.9% of the population, which required us to sort 50 cells per spot (19,200 total cells per plate) to observe survivors with a reasonable probability. The plates were incubated overnight, and the surviving cells formed colonies (Fig. 2A). There was no difference in recovery; 34 spots (out of 384 seeded) yielded colonies from the low-fluorescence fraction (0.18% survival), and 37 spots from the high-fluorescence fraction produced growth (0.19% survival). The 10-fold difference in HPF fluorescence among these groups of cells, however, was substantial, showing no relationship between survival and ROS. Similar results were obtained with ampicillin (fig. S4).

Fig. 2

HPF fluorescence is not an indicator of cell death. Mid-exponential phase cultures were treated with 0.25 μg/ml norfloxacin for 3 hours. (A) Single cells were sorted with a fluorescence-activated cell sorter based on fluorescence. The HPF fluorescence histogram before addition of antibiotic is shown in red, and the one after treatment with norfloxacin is in blue. The red rectangle indicates the gate corresponding to the part of the population from which single cells were spotted on an agar plate. The blue rectangles indicate the two gates corresponding to the part of the population from which cells were spotted on an agar plate. Each gate corresponds to 0.9% of the population. Fifty cells per spot were sorted from each gate onto a 384-point grid of an LB agar plate. (B) The average level of survival of each population (n = 3 ± SD), calculated based on data from Fig. 3A.

Apart from reports of a correlation between HPF measurements and killing and protective effects of thiourea, several observations have linked ROS production more directly with the action of bactericidal antibiotics. It has been proposed that antibiotics induce the tricarboxylic acid (TCA) cycle and respiratory chain–dependent ROS production, and TCA mutants show improved survival (8). Antibiotic-dependent ROS production has been reported to oxidize the guanine pool, leading to double-strand DNA breaks and death (10). It was also claimed that antibiotic tolerance in Pseudomonas aeruginosa biofilms was due to increased oxidative stress tolerance (12) and that killing of E. coli, Mycobacterium tuberculosis, and P. aeruginosa is potentiated in the presence of oxygen and reduced in an oxygen-limited environment (11). We reasoned that examining the effectiveness of antibiotics under aerobic compared to anaerobic conditions would be the most direct way to test these claims.

E. coli grows well under anaerobic conditions using fermentation to produce adenosine triphosphate and can also use alternative electron acceptors such as nitrate for anaerobic respiration (19). To investigate the action of antibiotics under anaerobic conditions, we used ofloxacin, which is employed in treating both aerobic and anaerobic pathogens. Killing by ofloxacin was more effective under anaerobic conditions, in contradiction to the ROS hypothesis (Fig. 3A). Inclusion of nitrate had no effect; why ofloxacin killed better under anaerobic conditions is unclear. Norfloxacin MIC was 0.125 μg/ml under aerobic conditions and 0.25 μg/ml under anaerobic conditions. As expected, there was no killing by norfloxacin below MIC at anaerobic conditions, but killing was comparable at aerobic or anaerobic conditions at ≥0.5 μg/ml (Fig. 3B). There was more killing by ampicillin under anaerobic as compared to aerobic conditions (Fig. 3C). Killing by kanamycin was slightly higher under aerobic conditions at a low concentration of antibiotic (Fig. 3D). This is expected because aminoglycosides require proton motive force for transport into the cell (20). At higher levels of kanamycin, there was no difference in killing between aerobic and anaerobic conditions. We also tested killing of P. aeruginosa under anaerobic conditions. Killing by ofloxacin was the same under aerobic versus anaerobic conditions (Fig. 3E).

Fig. 3

Antibiotic killing under aerobic and anaerobic conditions. A culture of E. coli BW25113 was split, diluted, and grown overnight in LB medium either aerobically or anaerobically in an anaerobic chamber. The respective overnight cultures were diluted 1:500; grown to mid-exponential phase, either aerobically or anaerobically; and treated with antibiotics for 3 hours. A sample was then removed, washed, diluted, and plated for colony count. (A) Ofloxacin (solid blue bars, aerobic LB; hollow blue bars, aerobic LB+KNO3; solid red bars, anaerobic LB; hollow red bars, anaerobic LB+KNO3). (B) Norfloxacin (blue bars, aerobic; red bars, anaerobic). (C) Ampicillin (blue bars, aerobic; red bars, anaerobic). (D) Kanamycin (blue bars, aerobic; red bars, anaerobic). (E) P. aeruginosa PAO1 was cultured similarly to E. coli in LB medium, aerobically or anaerobically (in the presence of KNO3), and treated with ofloxacin (yellow bars, aerobic; green bars, anaerobic).

Our findings raise the question, what is the source of discrepancy between this study and the reports linking cell death to ROS production by antibiotics? There seem to be several factors responsible for this disagreement. The use of very low concentrations of antibiotics, at or slightly above the MIC, is prone to producing errors, because the accepted error in determining the MIC is 100% (21). In some experiments performed with the same concentration of antibiotic, no killing would result because of variation in the MIC. The variation in the level of surviving persisters can be high among biological replicates (22). The use of thiourea to quench ROS is also problematic. According to our data, thiourea mildly inhibits cell growth (Fig. 1B), indicating that this compound may have nonspecific effects, such as slowing down cell metabolism, which would lead to increased tolerance to killing. Indeed, we observed protective effect of thiourea from antibiotic killing even under anaerobic conditions (fig. S5). Reports of mutants in the TCA cycle being more resistant to killing could similarly result from these cells having a slower metabolism. Finally, the use of HPF as a detector of ROS is only valid if this dye is a specific detector. However, we find that antibiotics cause a shift in HPF fluorescence under anaerobic conditions (fig. S6). Dying cells apparently produce some products to which HPF responds.

Taken together, our results show that killing by antibiotics is unrelated to ROS production. This finding will refocus efforts on unanswered questions on the mechanism of killing by antibiotics. For example, we do not know how β-lactams induce autolysis, nor how exactly mistranslation caused by aminoglycosides leads to cell death. Better understanding of cell death will guide the development of advanced cures to treat recalcitrant infectious diseases (23).

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Reference (24)

References and Notes

  1. Acknowledgments: The authors thank V. Isabella for helpful discussion of the manuscript. This work was supported by NIH grant T-R01AI085585-01 and by Army Research Office grants W9911NF-09-1-0265 and 55631-LS-RIP.
View Abstract

Stay Connected to Science

Navigate This Article