PERSPECTIVES OPINION
Distinguishing between resistance, tolerance and persistence to antibiotic treatment Asher Brauner, Ofer Fridman, Orit Gefen and Nathalie Q. Balaban
Abstract | Antibiotic tolerance is associated with the failure of antibiotic treatment and the relapse of many bacterial infections. However, unlike resistance, which is commonly measured using the minimum inhibitory concentration (MIC) metric, tolerance is poorly characterized, owing to the lack of a similar quantitative indicator. This may lead to the misclassification of tolerant strains as resistant, or vice versa, and result in ineffective treatments. In this Opinion article, we describe recent studies of tolerance, resistance and persistence, outlining how a clear and distinct definition for each phenotype can be developed from these findings. We propose a framework for classifying the drug response of bacterial strains according to these definitions that is based on the measurement of the MIC together with a recently defined quantitative indicator of tolerance, the minimum duration for killing (MDK). Finally, we discuss genes that are associated with increased tolerance — the ‘tolerome’ — as targets for treating tolerant bacterial strains. The isolation and genetic characterization of antibiotic-resistant bacterial strains has uncovered many molecular mechanisms of resistance1, including mutations in the drug target, enzymatic activity that directly inactivates the antibiotic and the activation of efflux pumps that pump out the antibiotic2. The genes that are involved in these mechanisms are termed the ‘resistome’ (REF. 3). However, as long ago as 1944 it was observed that bacteria were able to survive extensive antibiotic treatments without acquiring resistance mutations4,5. The terms ‘tolerance’ (REF. 6) and ‘persistence’ (REF. 4) were coined to distinguish these modes of survival from ‘resistance’, but the definitions of these different terms, and their distinction from one another, have remained somewhat ambiguous7,8. ‘Resistance’ is used to describe the inherited ability of microorganisms to grow at high concentrations of an antibiotic2, irrespective of the duration of treatment, and is quantified by the minimum inhibitory concentration (MIC)
of the particular antibiotic (FIG. 1a), whereas ‘tolerance’ is more generally used to describe the ability, whether inherited or not, of microorganisms to survive transient exposure to high concentrations of an antibiotic without a change in the MIC, which is often achieved by slowing down an essential bacterial process (FIG. 1b). In this Opinion article, we follow the tolerance terminology defined by Kester and Fortune8, namely that tolerance enables bacterial cells to survive a transient exposure to antibiotics at concentrations that would otherwise be lethal9. For example, tolerance to β‑lactams may occur when bacteria grow slowly 10, which is associated with slower cell wall assembly. As β‑lactams require active cell wall assembly to kill bacteria, slower growth will result in a longer minimum treatment duration to achieve the same level of killing, regardless of the concentration of the antibiotic. Dormancy may be viewed as an extreme case of slow growth, and dormancy that leads to tolerance may also be termed ‘drug indifference’ (REF. 11).
320 | MAY 2016 | VOLUME 14
Tolerance may be acquired through a genetic mutation or conferred by environmental conditions11; for example, poor growth conditions have been shown to increase tolerance to several classes of antibiotic. This tolerance was exploited by Lederberg and Zinder to isolate auxotrophic mutants, as only non-growing auxotrophs are able to survive when a mutagenized bacterial population is exposed to penicillin in the absence of an amino acid12. A non-growing state that leads to tolerance can also be induced by the antibiotic itself. This drug-induced tolerance subsequently protects the bacteria from the lethal activity of the antibiotic9. In contrast to resistance and tolerance, which are attributes of whole bacterial populations, ‘persistence’ is the ability of a subpopulation of a clonal bacterial population to survive exposure to high concentrations of an antibiotic13. Persistence is typically observed when the majority of the bacterial population is rapidly killed while a subpopulation persists for a much longer period of time, despite the population being clonal. The resulting time–kill curve will be biphasic14, owing to the heterogeneous response of persistent and non-persistent subpopulations. The slower rate of killing of the persistent subpopulation is non-heritable: when persistent bacterial cells are isolated, regrown and re-exposed to the same antibiotic treatment, the same heterogeneous response to the drug will be observed as in the original population, with the division of the population into persistent and non-persistent subpopulations4 (FIG. 1c). The first direct observations of persistence at the single-cell level showed that slow growth, as well as dormancy, of a small subpopulation of bacterial cells can underlie the high rate of survival of a whole population14. Additional, generally dose-dependent, mechanisms of persistence that also display biphasic killing have been observed subsequent to these initial observations15. Experimental discrimination between the different strategies used by bacterial cells for survival during exposure to antibiotics is important for several reasons. First, these survival strategies, despite superficial similarities, differ in their www.nature.com/nrmicro
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES a
b
Susceptible versus resistant bacterial strains 0
2
4
8
16
32
c
Susceptible versus tolerant bacterial strains
62 128
0
2
4
8
16
32
4
8
16
32
4
8
16
32
62 128 μg ml–1
MIC 62 128
0
2
4
MIC 8
16
32
62 128
μg ml–1
Susceptible Resistant
2
μg ml–1
MIC 2
0
62 128
μg ml–1
0
Susceptible versus persistent bacterial strains
0
2
4
8
16
32
62 128
μg ml–1
MIC
MIC
μg ml–1 MIC
Susceptible Tolerant
Susceptible Persistent
100 0
10–1
Fraction of survivors
Fraction of survivors
10
Tolerant Susceptible
10–2
10–2
Persistent
10–4
Susceptible 10–3
0
2 MDK99
4
6 MDK99
8
Time (hours)
Figure 1 | Characteristic drug responses of resistance, tolerance and persistence. The survival strategies of resistance, tolerance and persistence to antibiotic treatment each manifest as a characteristic drug response. a | The minimum inhibitory concentration (MIC) for a strain of bacteria that is resistant to an antibiotic is substantially higher than the MIC for a susceptible strain. Coloured wells represent bacterial growth, whereas wells in which the antibiotic concentration is high enough to kill the bacteria are in light brown. b | The MIC for a tolerant strain of bacteria
basic mode of action, which means that a treatment will often be ineffective if it is applied irrespective of the survival strategy. Second, the underlying mechanisms, and the experiments that are required to investigate them, may be very different for each strategy. Third, the range of antibiotics that is affected by the drug response can differ according to the survival strategy. For example, tolerance by slow growth will often confer an advantage to several classes of antibiotic, whereas most resistance mechanisms are specific to one class of antibiotics. Finally, the quantitative measurement of resistance, tolerance or persistence requires different metrics and experimental procedures for each survival strategy. In this Opinion article, we discuss the current basis for and the strategies used to distinguish between resistance, tolerance and persistence to antibiotics in bacterial strains, without any a priori knowledge
0
2
4
6
MDK99 MDK99.99 MDK99 Time (hours)
8
10
MDK99.99
is similar to that of a susceptible strain; however, the minimum duration Nature Reviews | Microbiology for killing (MDK; for example for 99% of bacterial cells in the population (MDK99)) for a tolerant strain is substantially higher than the MDK99 for a susceptible strain. c | A persistent strain of bacteria has a similar MIC and a similar MDK99 to a susceptible strain; however, the MDK for 99.99% of bacterial cells in the population (MDK99.99) is substantially higher for a persistent strain than the MDK99.99 for a susceptible strain. Concentrations and timescales are chosen for illustration purposes only.
of the molecular mechanisms that are involved. These terms have often been used interchangeably in the literature, but we propose a clear and distinct definition for each term, and an experimental framework for distinguishing between these phenotypes that uses a standardized and measureable metric to detect tolerance to drug exposure — the minimum duration for killing (MDK). We hope that the combination of the MIC and the MDK may be used as standards for the in vitro characterization of sensitivity to antibiotics, which ultimately may lead to better treatments for recalcitrant infections.
Resistance or tolerance? Resistance. Resistance to antibiotics, which is typically caused by inherited mutations, is associated with numerous molecular mechanisms that have been comprehensively reviewed elsewhere16,17. It is important to
NATURE REVIEWS | MICROBIOLOGY
note that mechanisms of bacterial resistance decrease the effectiveness of the antibiotic; that is, a higher concentration of the antibiotic is required to produce the same effect in a resistant strain as is produced in a susceptible strain18 (FIG. 1a). Resistance is quantified by the MIC, which can be defined as the minimum concentration of an antibiotic that is required to prevent net growth of the culture. In practice, the MIC is measured by exposing a bacterial population to increasing concentrations of the antibiotic in a standardized growth medium. This enables the measurement of the minimum concentration at which growth is not detected, typically after 16–20 hours of exposure to the antibiotic19. The range of concentrations that is tested in a clinical microbiology laboratory is usually limited to the concentrations of the antibiotic used in the clinic. The MIC that is determined by these tests is viewed as a convenient metric VOLUME 14 | MAY 2016 | 321
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES for resistance, and a bacterial strain with a higher MIC than another strain will be regarded as more resistant2. Measurements of the MIC that indicate total insusceptibility to an antibiotic may be viewed as an extreme case of resistance. The MIC has two major limitations as a general metric for measuring the response of a bacterial strain to an antibiotic. First, it is not informative for bacterial strains that are tolerant, rather than resistant. Second, the MIC measured in vitro can vary according to the experimental conditions that are used, which may affect the usefulness of this metric as a predictor of the effectiveness of the antibiotic in vivo20. However, the ease of measuring the MIC means that it is currently the only metric that is routinely used to inform treatment decisions for bacterial strains isolated in the clinic21. Tolerance. Tolerance is generally understood to be the ability of a bacterial population to survive a transient exposure to antibiotics9, even at concentrations that far exceed the MIC. Unlike resistance, tolerance applies only to bactericidal antibiotics and not to bacteriostatic antibiotics, as all bacteria are expected to survive transient exposure to bacteriostatic antibiotics8,9, which are not lethal and instead merely arrest growth. Therefore, all discussion of drug exposure in this Opinion article should be assumed to refer to concentrations of bactericidal antibiotics. Importantly, a longer exposure to an antibiotic, rather than a higher concentration of an antibiotic, is required to produce the same level of killing in a tolerant strain as is produced in a susceptible strain (FIG. 1b). As tolerant bacteria can have the same MIC as non-tolerant bacteria, the MIC is not informative as a metric to evaluate tolerance22,23. One suggested approach for the evaluation of tolerance is the measurement of time–kill curves at different concentrations of an antibiotic9. However, without a standard method for interpreting these curves the results that are obtained in different laboratories are difficult to compare24. Another measure of tolerance that has been proposed is the MBC/MIC ratio25, where MBC represents the minimum bactericidal concentration, namely the concentration of an antibiotic that is required to kill ≥99.9% of cells in a bacterial culture, typically after 24 hours of incubation. For cases in which concentrations of antibiotic that are near the MIC cause only growth arrest but the MBC results in death, the MBC/MIC ratio will produce a large value.
Therefore, this metric may accurately evaluate the level of drug-induced tolerance but was shown to correlate poorly with other forms of tolerance22,23,26,27. Recently, the MDK was described as a quantitative measure of tolerance that can be extracted from time–kill curves, based on the notion that a tolerant bacterial strain requires more exposure time to be effectively killed than a susceptible strain. The MDK is defined as the typical duration of antibiotic treatment that is required to kill a given proportion of the bacterial population28 at concentrations that far exceed the MIC (that is, when the rate of cell death is independent of the concentration of antibiotic). For example, the minimum duration of treatment that is required to kill 99% of a bacterial population (MDK99), which can be extracted from a time–kill curve (FIG. 1b). The assumption that underlies the MDK as a measure of tolerance is that the killing effect reaches saturation at high concentrations of an antibiotic so that it is almost insensitive to concentration and dependent only on the duration of exposure29. Similarly to the MIC, which can be used to compare the level of resistance between bacterial strains, the MDK can be used to compare the level of tolerance between strains. In contrast to the killing rate (that is, the rate at which survival decreases exponentially), which can only be extracted from exponential killing curves, the MDK simply integrates all of the different factors that together underlie a faster or slower overall killing efficacy, such as delays in killing or killing curves that are not exponential. Therefore, this quantification of tolerance is not dependent on any particular molecular mechanism. We argue that the MDK should be the preferred metric for the measurement of tolerance, as it enables a quantitative comparison between different bacterial strains and conditions. Furthermore, an evaluation framework that measures both the MDK and the MIC would enable a clear distinction to be made between resistance (an increase in the MIC) and tolerance (an increase in the MDK). Reports of tolerance in the literature are generally associated with slow growth and reduced metabolism14,30–33. As in the β‑lactam example, the slowing or complete cessation of growth results in a reduced or diminished susceptibility to many antibiotics. This is a direct result of the evolution of these antibiotics in microorganisms competing for resources, in which the production of antibiotics that target fast growing bacterial cells, which are the most competitive for
322 | MAY 2016 | VOLUME 14
resources, is selected for 34. Different classes of antibiotic have evolved to target different processes that are required for growth and it is sometimes possible to artificially decouple the efficacy of the antibiotic from the growth rate (that is, decouple target production and growth), once the process that is targeted by the antibiotic is known. For example, in Escherichia coli cells that are growth-arrested by the stringent response, treatment with chloramphenicol enables cell wall assembly to resume without the full resumption of cell growth. As a result of the resumption of cell wall assembly, the bacterial cells are sensitive to β-lactams, even though they remain essentially growth-arrested35. However, reports that E. coli cells can be killed by β‑lactams during growth arrest are often based on experiments that measure the growth arrest of the batch culture, which means that a dynamic balance of growth and death — in which β‑lactams only target growing cells but the overall growth of the culture is stationary — cannot be ruled out. Although some studies have assayed growth arrest in single cells, these studies assayed the absence of growth at the beginning of the treatment36 and cannot rule out that growth occurred during treatment.
Which form of tolerance? We identify two main forms of tolerance, which we term ‘tolerance by slow growth’ and ‘tolerance by lag’. Although these two forms of tolerance share an increased MDK compared with susceptible bacterial cells, the mathematical description and measurement protocol differ between them. The distinction arises because tolerance by slow growth occurs at steady state, whereas tolerance by lag is a transient state that is induced by starvation or stress. Tolerance by slow growth. Conditions that decrease the rate of growth have long been known to increase tolerance to numerous antibiotics10,37–40, as the mechanisms of action of these drugs share a requirement for growth. For example, the mechanism of action of β‑lactams relies on the disruption of bonds in the peptidoglycan layer that occurs during bacterial growth. β‑lactams exploit this process by preventing the reassembly of the peptidoglycan bonds, which eventually leads to cell lysis6. Therefore, the number of defects in the peptidoglycan layer increases in proportion with the growth rate. Indeed, the killing rate of bacteria that are exposed to β‑lactams has even been shown to be proportional to the growth rate, which demonstrates the strength of correlation www.nature.com/nrmicro
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES between slow growth and tolerance for these antibiotics10. Similarly, the exposure of bacterial cells to DNA gyrase inhibitors, such as fluoroquinolones, results in DNA damage and killing at a rate that is proportional to the growth rate41. Indeed, plotting the MDK99 versus the growth rate, using values extracted from time–kill curves published in the literature10,29,37–40, confirms the positive correlation of killing activity and growth rate for several species of bacteria and different classes of antibiotic (FIG. 2a). Tolerance by slow growth can either be inherited, when a bacterial species or strain has an inherently slow growth rate, or non-inherited, when slow growth occurs because the conditions for growth are poor. Species with inherently slow growth rates include Mycobacterium tuberculosis, which has a doubling time in nutrient-rich medium of approximately 24 hours42. This doubling time is approximately 40 times longer than that of E. coli and, not surprisingly, the MDK99 of M. tuberculosis strains is also approximately 40 times longer than the MDK99 of E. coli. Auxotrophs and other bacterial strains with mutations that reduce their intrinsic growth rate also show inherited tolerance12. Non-inherited tolerance by slow growth occurs when bacterial growth is impaired, such as by poorer growth conditions43,
the location of a cell within a biofilm or exposure to inhibitors44. When the antibiotic is added in the presence of these growth-reducing conditions, killing will be reduced. Note that dormancy can be viewed as the extreme case of slow growth, in which the growth rate is zero. Importantly, a decrease in growth rate has been observed for intracellular bacteria when within a host cell. For example, Salmonella enterica subsp. enterica serovar Typhimurium cells with arrested growth have been detected in infected macrophages31. Accordingly, infections by intracellular pathogens are notoriously difficult to eradicate, even when treated with antibiotics that readily penetrate host cells. The notion that tolerance rather than resistance underlies the resilience of these infections is supported by in vitro assays that showed that intracellular Staphylococcus aureus45 treated with dicloxacillin had a fivefold increase in the MDK, but no change in the MIC, compared with extracellular S. aureus, which suggests that tolerance enables this intracellular pathogen to survive treatment with dicloxacillin. A special case of tolerance by slow growth is drug-induced tolerance, which occurs when bacterial cells respond to antibiotic exposure by reducing or arresting their growth. This effect has been observed in
a
various Gram-negative and Gram-positive bacteria when exposed to antibiotics, including an antibiotic that inhibits cell wall synthesis9, whereby the growth arrest was mediated by the defective induction of autolysins9,46, and the fluoroquinolone ciprofloxacin, whereby the growth arrest was mediated by the induction of a stress response47. In summary, non-inherited tolerance can be triggered by external stress factors that include starvation48, host factors42 and even the antibiotic itself 47,49. As might be expected, tolerance by slow growth also occurs when antibiotics are added at the stationary phase of growth, in which the net growth rate of the bacterial population is zero (but conditions are permissive for a balance between the growth and death of individual cells). In addition, in what may be viewed as an extreme case of tolerance by slow growth, tolerance at the stationary phase can occur when the growth rate of individual bacterial cells is zero50, which can produce an extremely long MDK51. The protective effect of growth arrest as a passive survival strategy can be enhanced by the activation of stress response mechanisms that provide further protection from antibiotic stress52. Some of these additional protective mechanisms, such as the production of efflux pumps, may also reduce
b Growing Non-growing
108
CFU ml–1 Tolerance by lag Tolerance by slow growth
10–1 100
101
1 in 100
106
100
1 in 100
1 in 100,000
MDK99 (hours)
101
104
0
2
Doubling time (hours)
4
6
8
Time (hours)
Figure 2 | Tolerance arises from slow growth or lag phase. a | The minimum duration for killing (MDK) for 99% of bacterial cells in a population (MDK99) is plotted against doubling time for several combinations of bacterial strain or species and antibiotic, as extracted from time–kill curves in the literature10,26,28,29,37,38,40,56. The dashed line shows the best fit for the relationship between the MDK99 and the doubling time for strains of bacteria that are tolerant by slow growth, which demonstrates the correlation between these two variables. The shaded area highlights the distribution of bacterial strains that are tolerant by lag; these strains were detected by exposure to the drug directly on dilution from the stationary phase.
b | A schematic growth curve that shows the importance of subculturing Nature Reviews | Microbiology to reach strictly exponential growth. An initial 1 in 100,000 dilution of a bacterial population from a culture in the stationary phase of the growth cycle is followed by serial 1 in100 dilutions; in each instance, the colony is grown until the population density reaches 107 colony forming units (CFU) ml –1 before dilution. Each dilution reduces the number of residual non-growing bacterial cells — that is, cells in the lag phase — in the population and several dilution steps may be required until the population is composed only of cells in the exponential growth phase, with no cells remaining in the lag phase.
NATURE REVIEWS | MICROBIOLOGY
VOLUME 14 | MAY 2016 | 323 . d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES the effective concentration of the antibiotic, which increases the MIC and results in a mixed phenotype of resistance and tolerance. Tolerance by lag. In addition to the stationary phase, another phase of the bacterial growth cycle during which bacteria do not grow, and may therefore be transiently protected from killing by antibiotics, is the lag phase. The lag phase is defined as the time it takes for growth-arrested bacterial cells (for example, under starvation conditions) to resume exponential growth when adjusting to an environment that is permissive for growth (for example, when starved bacterial cells are diluted into fresh nutrient conditions). The typical mean lag time of E. coli K-12 populations when diluted from an overnight culture is 30 minutes, but this time can be substantially longer when a culture has been in the stationary phase for several days53,54 before dilution into fresh nutrients. Although these two growth phases are often thought to be similar to each other, the lag phase has now been shown to be a distinct metabolic state from the stationary phase30,55, in which bacterial cells must first adapt to the increase in nutrient concentration before resuming exponential growth55. Therefore, tolerance by lag differs from the extreme case of tolerance by slow growth that occurs at the stationary phase. Similarly to the lag phase that occurs after stationary growth, transient growth arrest can also occur during a lag phase that follows transitions between growth conditions — for example, when bacterial cells enter the host environment or switch between different niches32. Importantly, tolerance by lag is a transient phenotype that is not sustained when the culture has sufficient time to fully adjust to the new conditions. Therefore, tolerance by lag is modelled as a decay process28, whereas tolerance by slow growth is a steady-state phenotype that is characterized by a reduction in killing rate. Thus, the mathematical descriptions of these two forms of tolerance are inherently different28. Tolerance by lag occurs when the antibiotic treatment is shorter than the duration of the growth arrest14,56. The protective effect of the lag phase on the survival of the bacterial population is very broad, as it can enable tolerance to different antibiotics23,28, in addition to other stresses, including exposure to the host immune system57 and the induction of prophages58. Despite the transience of growth arrest at the lag phase, tolerance by lag can be very effective, reaching an MDK of many hours or days. For example, it has been shown that the intermittent
exposure of E. coli to a β‑lactam antibiotic can select for a lag phase that is 10 times longer than the lag phase of the ancestral population, reaching an MDK99 of more than a day 28. Remarkably, the duration of the lag phase evolved to match the duration of antibiotic treatment in as few as eight exposures to the drug. Owing to the tolerance that was conferred by the extended lag time, antibiotic treatment eventually became ineffective, even when changing the class of antibiotic, as long as the duration of the treatment was the same28. Several genes were repeatedly mutated in these populations (BOX 1), which led to an inherited tolerance by lag. By contrast, no change in the MIC was detected, which suggests that the phenotype of tolerance by lag may evolve more rapidly than the emergence of resistance. The rapid evolution to tolerance by lag that was observed in this in vitro assay, in which bacterial populations adapted to the duration of the treatment rather than to its chemical composition, calls for an evaluation of the importance of the evolution of tolerance in the host environment. Measuring tolerance. It is important to realize that the experimental protocol for the in vitro measurement of tolerance differs according to whether the measurement is for tolerance by slow growth or tolerance by lag (BOX 2). To measure tolerance at the lag phase, exposure to the antibiotic must occur directly on transition to the lag phase (generally, when diluting from a stationary-phase culture). If the culture is instead first diluted into fresh medium for an undetermined period of time before exposure to antibiotics, a mixed population of exponentially growing and non-dividing bacterial cells arises (FIG. 2b). The proportion of the bacterial population that survives exposure to the antibiotic will then depend on the time between dilution into fresh medium and exposure to the drug, owing to the complex population dynamics of the exit from the lag phase, which is heterogeneous at the single-cell level59. The dependence on experimental parameters that can be challenging to control may result in non-reproducible results and ambiguous measurements of tolerance or persistence (see below). By contrast, tolerance by slow growth should be measured in a steady-state culture during exponential growth, ensuring that no lagging bacterial cells are carried over from the stationary phase. This steady-state culture can be achieved in chemostats or in cultures that are sub-cultured several times during the exponential phase. Note that true
324 | MAY 2016 | VOLUME 14
exponential growth ends long before the transition to the stationary phase that occurs at high cell density, typically already at an OD600 (optical density at 600 nm) of 0.1 in rich medium60.
Persistence and heterogeneity For those antibiotic treatments that effectively kill the majority of the bacterial population, subpopulations that are not killed by the antibiotic can nevertheless emerge4,61, even in clonal cultures. When these surviving subpopulations are grown in the presence of the same antibiotic, the heterogeneous response is repeated51,62. This phenomenon is termed ‘bacterial persistence’ and the surviving bacterial cells are referred to as persisters. We note that ‘persistence’ is also used more generally to describe infections that are not cured effectively and persist in the host63, including those infections that may be unrelated to the definition of persistence used in this Opinion article to denote the presence of a subpopulation of persisters in a clonal population of bacteria. As opposed to tolerance and resistance, persistence only occurs in a subpopulation of bacterial cells. Persistence can be detected by the presence of a bimodal (or multimodal) time–kill curve that deviates from the simple decay expected from a uniform bacterial population13. In the simple case of two coexisting subpopulations, persistence is characterized by a switching between two phenotypes — susceptible and persistent. Persisters constitute the less numerous subpopulation (typically less than 1%) and are killed at a slower rate than the susceptible cells13. We propose that the first step towards characterizing the heterogeneity of bacterial populations under antibiotic treatment is to determine whether persisters survive the exposure to the antibiotic because they are transiently more resistant or because they are transiently more tolerant than the majority of the population (BOX 2). Time-dependent persistence. Time-dependent persistence is characterized by the presence of a subpopulation of tolerant bacteria, which typically has either a longer lag time (tolerance by lag) or slower growth rate (tolerance by slow growth) than the majority of the population. These two types of persistence have very different dynamics and were previously defined as Type I persistence and Type II persistence, respectively 14. All of the characteristics of tolerance that are described above for a whole bacterial population can also be applied to a subpopulation with time-dependent www.nature.com/nrmicro
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES Box 1 | The tolerome: genetic factors that increase the MDK An increase in the minimum duration for killing (MDK) occurs when the killing rate of bacterial cells that are exposed to antibiotics is slowed down by one or more of numerous mechanisms. Non-inherited tolerance can be triggered by external stress factors, such as starvation48, low temperature11, host factors42 and even the antibiotic itself47,49. However, we use the term ‘tolerome’ to describe the genetic factors that have been repeatedly shown to increase tolerance or time-dependent persistence. For the β-lactam and fluoroquinolone classes of antibiotic, the stringent response, which inhibits bacterial growth, has been shown to have a central role in tolerance. During nutritional stress, the decrease in the availability of amino acids leads to an accumulation of uncharged tRNAs that triggers the production of guanosine tetraphosphate (ppGpp), an alarmone stress signal that mediates the stringent response89. The first high-persistence mutants to be isolated in the laboratory66 were Escherichia coli mutants found to have mutations in the hipBA toxin–antitoxin module, which encodes HipA, a toxin, and its cognate antitoxin, HipB. HipA was later shown to inactivate an essential amino-acyl tRNA synthetase, glutamate–tRNA ligase (GltX)90,91, thus producing high levels of ppGpp owing to the accumulation of uncharged tRNAs. When hipA is expressed above a threshold set by the abundance of HipB, a stringent response is induced. Importantly, the stringent response involves the induction of a lag phase (that is, a transient growth arrest), as has been shown in both Gram-negative90 and Gram-positive bacteria92. As the expression of hipA increases, the lag phase becomes longer, which results in a longer MDK and a phenotype of tolerance by lag64. Aside from hipA, the overexpression of other toxin genes in toxin–antitoxin modules can also produce similar tolerance phenotypes93. The tolerome has been studied using mutational screens for tolerance, which have identified numerous mutations that are related to the activation of the stringent response, including mutations in hipA66, hipB94 and methionine–tRNA ligase (metG)94, as well as many global regulators95 and metabolic genes, such as glycerol‑3‑phosphate dehydrogenase (glpD)54,94,96. A study that used experimental evolution also identified metG as a gene that is associated with tolerance, as well as prsA and the toxin–antitoxin module vapBC28. High expression of other toxins97 and virulence factors82 may also transiently arrest growth to trigger a phenotype of tolerance by lag97. A recent study found that genes
persistence. Indeed, the only difference between time-dependent persistence and tolerance is the fact that only part of the bacterial population is responsible for the slower killing that is observed in persistence (FIG. 1c). Therefore, the molecular mechanisms that lead to tolerance are expected to be relevant for time-dependent persistence (BOX 1). For example, inducing the expression of toxins in toxin–antitoxin modules results in either the growth arrest of a subpopulation of induced bacterial cells (that is, time-dependent persistence), when expressed at low levels, or in dormancy of all induced bacterial cells, when expressed at high levels64,65. A well-characterized example of time-dependent persistence is the hipA7 allele of the hipA gene, the presence of which produces a high-persistence mutant66 that generates two subpopulations with very different lag time distributions to one another 64. The subpopulation with the longer
with functions that are related to amino acid synthesis and genes that encode toxin–antitoxin modules were among hundreds of genes implicated in tolerance to a drug that belongs to the aminoglycoside class of antibiotics98. Interestingly, the number of genes identified for the tolerome is substantially larger94 than the number of genes identified for the resistome3, which suggests that the evolution of increased tolerance may occur faster than the evolution of increased resistance, as observed in an experimental evolution study based on intermittent in vitro exposures to a β-lactam antibiotic28. The molecular mechanisms of tolerance that slow down killing by antibiotics are also associated with time-dependent persistence, which applies to a heterogeneous clonal bacterial population in which tolerance is present in a subpopulation but not in the majority of bacterial cells. However, persistence poses an additional intriguing question: how can a clonal bacterial population spontaneously differentiate into subpopulations with different tolerance levels? The role of molecular noise in generating variability that leads to persistence has been reviewed elsewhere99,100 but can be briefly summarized as stochastic fluctuations in the concentration of cellular factors that affect growth. These fluctuations may be the outcome of changes in production and degradation rates or uneven partitioning following cell division101, and may then be further amplified by regulatory feedback circuits102. For example, toxin–antitoxin modules can contribute to persistence through a threshold mechanism that amplifies noise64,103 to result in stochastic activation of the stringent response33,90. Accordingly, the deletion of toxin–antitoxin modules104,105 or stringent response genes65 leads to a decrease in persistence. Examples of persistence that arise from fluctuations that are produced by asymmetric cell division have been reported in mycobacteria106–108. Note that the tolerome does not include genes that are associated with dose-dependent persistence, such as those encoding efflux pumps69,109 or catalase–peroxidase (katG)15, as dose-dependent persistence is associated with genes that are implicated in resistance rather than tolerance. prsA110 has been reported to belong to both the resistome and the tolerome, but future work will be required to carefully determine whether it is a bona fide genetic determinant of both the minimum inhibitory concentration (MIC) and the MDK.
lag time will not be detected in standard measurements of the culture lag time, as the exit of the culture from the lag phase will be dominated by the subpopulation with a short lag time59. However, the heterogeneity of the lag times between the two subpopulations translates into a bimodal killing curve (FIG. 1c), which in our proposed framework is termed ‘persistence by lag’. The key features of the time–kill curve for time-dependent persistence, whether by lag or by slow growth, are bimodality and insensitivity to the concentration of antibiotic (assuming that the concentration is substantially higher than the MIC). Time-dependent persistence can be measured by extracting the MDK from the time–kill curve obtained from a bacterial population that is exposed to a concentration of antibiotic that is high enough to reach saturation. Importantly, for the heterogeneous bacterial populations that are relevant to persistence, the measurement will also rely on the percentile of cells killed that is chosen to
NATURE REVIEWS | MICROBIOLOGY
define tolerance. For example, the MDK99 is only sensitive when it is applied to bacterial populations in which more than 1% of cells are persisters; therefore, the detection of smaller subpopulations may require a different choice of percentile for the MDK measurement, such as the MDK99.99, which is the duration of treatment that is required to kill 99.99% of a bacterial population (FIG. 1c). Similarly to the measurement of tolerance by lag and tolerance by slow growth, it is important to note that a different protocol is required for the in vitro measurement of persistence depending on whether the measurement is for persistence by lag or persistence by slow growth67 (BOX 2). Dose-dependent persistence. Although most studies of persistence relate to time-dependent persistence (BOX 1), subpopulations of persisters have also been reported that instead have a transient decrease in antibiotic sensitivity. For example, VOLUME 14 | MAY 2016 | 325
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES Box 2 | Framework for the measurement of resistance, tolerance and persistence We have developed an experimental framework for the use of batch culture measurements to distinguish between the various possible strategies for survival under antibiotic stress. The framework is based on a flowchart (see the figure) that classifies each bacterial strain as more resistant, tolerant or persistent than a wild-type reference strain, and further classifies tolerant and persistent strains into the subtypes of tolerance by lag, tolerance by slow growth, time-dependent persistence by lag, time-dependent persistence by slow growth and dose-dependent persistence. The individual steps of the flowchart are, for the most part, not very different from existing protocols, but are organized together into a single framework. The framework is designed to distinguish between the survival strategies without accounting for the very different molecular mechanisms that may be involved, and we hope that this will enable the comparison of results between different laboratories, or even stimulate the development of improved definitions to those proposed in this Opinion article. The framework involves up to five experimental tests (see the figure, in orange). First, the minimum inhibitory concentration (MIC) is measured for both a susceptible reference strain of bacteria and for a strain of interest (for example, a wild-type strain and a mutant strain, respectively). In common with the standard approach for identifying resistant strains of bacteria in the clinic, this step classifies a mutant strain as resistant if the MIC is substantially higher in the mutant strain than in the wild-type strain. Strains of bacteria that have the same MIC as the reference strain are further characterized in a second step that measures the minimum duration for killing (MDK) for 99% of the population (MDK99), which is a value extracted from a time–kill curve at a concentration at which the killing efficacy of the antibiotic reaches saturation. The choice of 99% for the percentile that is measured in this step is designed to evaluate the tolerance level of the bulk population, as the percentile is low enough to be relatively insensitive to persister Strain to subpopulations characterize (unless they are highly enriched in the population) but MIC Resistant high enough to High capture the MIC Low MIC MDK99 Low MDK99 MDK99.99 Low MDK99.99 Susceptible
dynamics of effective killing by the antibiotic. A strain with an MDK99 that is substantially higher than the reference strain, but with an equal MIC, is characterized as tolerant. For these tolerant strains, a third step is then used to distinguish between tolerance by lag and tolerance by slow growth. In this step, survival under treatment with an antibiotic is compared between a bacterial culture that is inoculated from the stationary phase and a bacterial culture that is inoculated from a strictly exponential phase (FIG. 2b). For strains in which the MDK99 is high only for the culture inoculated from the stationary phase, and thus the duration of killing is dependent on the duration of the lag phase but not on the rate of growth, the form of tolerance is classified as tolerance by lag. By contrast, when the MDK99 is high for both cultures, tolerance by lag can be ruled out and the form of tolerance is thus classified as tolerance by slow growth. For strains of bacteria with both an equal MIC and an equal MDK99 to the reference strain, an alternative third step is used to establish whether persistence is present in a subpopulation of bacteria too small (less than 1%) to be detected by the MDK99 measurement. In this step, higher percentiles are used to measure the MDK. For example, for a bacterial population in which 0.2% of cells are persistent, an increased exposure time to the antibiotic is required to kill 99.99% (MDK99.99) of the population than 99.99% of the reference strain population; therefore, a strain with a substantially longer MDK99.99 than the reference strain, but with both an equal MIC and an equal MDK99 to the reference strain, will be classified as persistent (FIG. 1c). For strains of bacteria that have been identified as persistent, a fourth step is required to distinguish between dose-dependent persistence (owing to a resistance mechanism transiently present in a subpopulation of bacteria) and time-dependent persistence (owing to a tolerance mechanism transiently present in a subpopulation of bacteria). For dose-dependent persistence, the higher MDK values in the previous two steps are not caused by the presence of slow growth or lag phase persisters, but by the presence of a subpopulation of bacterial cells that transiently express a resistance factor that better enables them to survive the concentration of the antibiotic used to measure the MDK. Therefore, the MDK99.99 measurements are repeated at a concentration of antibiotic that is increased twofold compared with the previous step, to determine whether the high MDK values are due to dose-dependent persistence, which is indicated by a strong dependence on the concentration of antibiotic, or time-dependent persistence, which is indicated by a weak dependence on the concentration of antibiotic. Tolerant High Finally, as with tolerant strains of bacteria, for strains that are MDK99 shown to have time-dependent persistence, a further step is required to determine whether time-dependent persistence is due Concentration Time-dependent to a subpopulation with a long lag phase (persistence by lag; also dependency persisters High known as ‘Type I’ persistence) or due to a Low MDK99.99 dependency subpopulation that has slow growth Stationary (persistence by slow growth; also known as versus exponential High High MDK ‘Type II’ persistence). In this fifth step, the same High MDK inoculum dependency only in in both test is used to distinguish between the two stationary phenotypes as is used to distinguish between Dose-dependent Tolerance (or persistence) Tolerance (or tolerance by lag and tolerance by slow growth persisters persistence) by lag by slow growth (see above). Nature Reviews | Microbiology
this can occur when a resistance factor, such as an efflux pump, is transiently overexpressed in a subpopulation of bacterial cells68,69. The overexpression of a resistance factor in these subpopulations of persisters causes a reduction in the effective intracellular concentration of the antibiotic and thus results in a lower antibiotic sensitivity, as a higher
concentration is required to achieve the same rate of killing. When the bacterial population is exposed to a concentration of antibiotic that is high enough to reach saturation for the majority of cells in the population but not for the subpopulation of persisters, a bimodal time–kill curve will be observed. We propose that these persisters are classified as dose-dependent persisters.
326 | MAY 2016 | VOLUME 14
The difference between resistance and dose-dependent persistence resides in the transient heritability of the overexpression of the resistance factor. When dose-dependent persisters are regrown to a full population, the resistance factor will only be overexpressed in a subpopulation of bacterial cells in the new population, so that the new population is also heterogeneous www.nature.com/nrmicro
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES with regard to antibiotic sensitivity. A decrease in the effectiveness of antibiotics that we attribute to dose-dependent persistence has been associated with the transient overexpression of efflux pumps69 and the multiple antibiotic resistance (marRAB) operon68, as well as the transient decrease in expression of an enzyme that activates the antibiotic isoniazid15. In this latter work, persisters in a population of Mycobacterium smegmatis had decreased levels of pulsed expression of catalase–peroxidase (katG)15, and were therefore able to grow in the presence of isoniazid. Consistent with dose-dependent — rather than time-dependent — persistence, growth rate and survival were not correlated in this population, which suggests that a decrease in the growth rate was not responsible for the increase in persistence. Similarly to time-dependent persistence, dose-dependent persistence can be detected by the presence of a bimodal time–kill curve (FIG. 1c), which is the hallmark of persistence, as well as by the transiency of the survival effect (as regrowth will produce another heterogeneous population of bacterial cells rather than a population that is uniformly resistant to antibiotics). It should be noted that dose-dependent persistence, when inherited for a sufficient number of generations for colonies to become visible on plates that are treated with antibiotics, has sometimes been referred to as heteroresistance61,70. Heteroresistance has mostly been described in S. aureus but requires further characterization, as recently reviewed elsewhere71. The defining feature of dose-dependent persistence is that higher doses of the antibiotic decrease survival more effectively than a longer duration of exposure, in contrast to the longer treatment duration that is required to decrease survival in populations with time-dependent persistence, such as those with a subpopulation of dormant persisters (BOX 2). In addition, as with resistance, dose-dependent persistence typically increases survival to a specific class of antibiotic and, furthermore, is often independent of the rate of cell growth. By contrast, time-dependent persistence provides a more general protection against several classes of antibiotic that target mechanisms associated with cell growth, such as β‑lactams and quinolones23,51. A systematic classification of which antibiotics are more prone to dose-dependent or time-dependent persistence awaits further characterization of drug responses.
Conclusion and future prospects In this Opinion article, we propose that bacterial survival under antibiotic stress is characterized by two major factors — resistance and tolerance. We suggest that these factors can be quantitatively estimated through the measurement of two parameters: the MIC for resistance and the MDK for tolerance. Finally, we propose a classification framework that we argue will enable not only the identification of resistant and tolerant bacterial strains but also the clarification of complex cases that include at least one tolerant or transiently resistant subpopulation of bacterial cells — for example, persisters in heterogeneous clonal populations. We predict that this classification will provide a useful approach to identify and distinguish between the different survival strategies. Furthermore, it may help to define a ‘tolerome’ that is composed of gene targets that have been shown to affect the MDK (BOX 1). In the clinic, these insights may be useful for establishing more effective treatment regimens that are tailored to the specific survival strategies used by the infecting pathogen. For example, dose-dependent persistence might be targeted by known inhibitors72 of resistance, such as efflux pump inhibitors, whereas time-dependent persistence might be countered by an extension of the treatment duration. For strains with tolerance or time-dependent persistence, the MDK can provide clear predictions of the duration of treatment that is required to treat an infection, and could thus be combined with current pharmacokinetic and/or pharmacodynamic models to guide treatment regimens. Indeed, current practice has empirically extended the duration of treatment for bacterial strains that are notoriously slow growing 2. However, in the case of tolerant strains of bacteria in which the MDK is very high, the toxicity of the antibiotic to the host may limit the duration of treatment2. In addition, ambulatory treatments are rarely capable of maintaining a constant level of antibiotic concentration in the body, as this would require constant administration for the majority of antibiotics that are typically removed from the serum within a few hours after administration73. Therefore, alternative strategies against tolerant bacterial pathogens are required74. One avenue to be explored is the use of existing antibiotics for which the drug response has been found to be less prone to tolerance, such as daptomycin, which is
NATURE REVIEWS | MICROBIOLOGY
effective even when applied to stationary- phase cultures39. Systematic screens have been carried out to search for compounds that are more effective against tolerant strains. For example, a recent screen for US Food and Drug Administration (FDA)-approved compounds that remain effective at the stationary phase has led to the identification of promising candidates for treating tolerance by slow growth in Borrelia burgdorferi 75. Other systematic screens have searched for new compounds that can be used in combination with conventional antibiotics to decrease tolerance. Several compounds have been identified in these screens that are effective against time-dependent persisters76 or against tolerance in biofilms77; however, the effectiveness of these compounds has not yet been assessed in the clinic. As an alternative to systematic screening, targeted treatment design can make use of recent insights into the major pathways that lead to tolerance and into the metabolism of tolerant cells (BOX 1). For example, understanding the role of the protein degradation pathway in persistence has already led to the targeting of this pathway, showing promising results both in vitro and in vivo78. Although we expect our proposed framework, which simplifies the characterization of time–kill curves under antibiotics to two main parameters, to be powerful for distinguishing between resistance and tolerance, more subtle effects may not be fully captured by the MIC and the MDK metrics. For example, heteroresistance, which we briefly mentioned above, can be considered to be dose-dependent persistence that is heritable for sufficient generations to enable colony growth. The characteri zation of heteroresistance requires additional measurements to those in our framework; in particular, the switching rate, namely the number of generations over which the resistance is heritable, is an additional key parameter that needs to be evaluated to predict the outcome of treatment for heteroresistant bacterial strains. Finally, we note that drug-induced tolerance (or drug-induced persistence), namely the ability of some microorganisms to arrest growth in response to antibiotic stress47,49, can result in an MDK that is very long and therefore difficult to measure. The challenge of drug-induced tolerance is that the non-growing state can be induced for the duration of the exposure to the antibiotic. For antibiotics that do not kill non-growing bacteria at all, the MDK may become too long to measure for practical reasons. VOLUME 14 | MAY 2016 | 327
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES An important question has been raised as to whether the in vitro study of tolerance or persistence is relevant to the failure of antibiotic treatments in vivo8. The question can be applied to our framework by asking whether an extended MDK in vitro is relevant to the likelihood of treatment failure for the same bacterial strain in vivo. Methods that have been developed for the study of single cells (BOX 3), specifically for the detection of single persisters with a high MDK in vitro30,31,79, have recently been applied to the detection of persistence
in vivo in mouse models of infection, which demonstrated that dormant persisters are present in infections with S. Typhimurium31 or M. tuberculosis 42. Interestingly, both persistence by slow growth and persistence by lag can be detected in the same bacterial infection42,80. Dormancy and the resulting phenotype of persistence by slow growth were attributed to stresses that are induced by host factors and nutrient deprivation. Infections with S. Typhimurium showed a clear correlation between single-cell growth rate and survival under treatment with
Box 3 | Single-cell measurements of persistence In contrast to the measurement of resistance and tolerance across whole populations of bacterial cells, which has been possible since the 1940s, the measurement of the heterogeneous response to antibiotics observed in populations with persisters was only made possible with the development, in the past two decades, of single-cell techniques111. The first direct identification of persistence at the single-cell level used a microfluidic device to study the hipQ and hipA7 mutants of Escherichia coli that had been identified in genetic screens for persistence to antibiotic treatment14. The bacterial cells were grown in the device, which is able to keep single cells within the observation field of the microscope, and exposed to a transient antibiotic treatment. By tracking individual bacterial cells, microscopy images before exposure to the antibiotic could be matched to the small number of bacterial cells that survived treatment with the antibiotic (that is, persisters), which revealed that persisters were either slow growing14 or had a long lag phase before treatment with the antibiotic. Therefore, these persisters were a subpopulation of tolerant bacterial cells (that is, time-dependent persisters); specifically, persisters in the hipQ mutant population were tolerant by slow growth, whereas persisters in the hipA7 population were tolerant by lag. In subsequent work, microfluidics was used in combination with dynamic fluorescence microscopy to identify the window of time in which the differentiation into persisters fully develops, by using the abundance of an induced fluorescent protein that was expressed from a synthetic promoter as a proxy for metabolic activity30. In another example of time-dependent persistence, an imaging study that used a microfluidic device known as the mother machine112 showed that the expression of virulence genes was correlated with a decrease in growth rate, and a higher minimum duration for killing (MDK), in a subpopulation of Salmonella enterica subsp. enterica serovar Typhimurium cells expressing a fluorescent marker for virulence82. Microfluidics was also used to observe time-dependent and dose-dependent persistence in Mycobacterium smegmatis15,106. Using a reporter for the activator of the antibiotic, it was shown that persistence to isoniazid was associated with variations in the concentration of the activator between bacterial cells, possibly leading to variations in the effective concentration of the antibiotic. A non-microscopy method that has been developed for the detection of dose-dependent persistence is a high-throughput assay that uses femtolitre droplets formed on a hydrophilic-in‑ hydrophobic micropatterned surface to enclose single bacterial cells pre-incubated with fluorescein-di‑β‑d‑galactopyranoside (FDG)113. FDG, which is a precursor of the fluorescent dye fluorescein, is hydrolysed inside the cell by β‑galactosidase; however, efflux pumps can efficiently export FDG before hydrolysis can occur. By measuring the fluorescence signal, it was possible to quantify the activity of efflux pumps in individual bacterial cells, and thereby infer variability in the expression of efflux pump genes, which can lead to dose-dependent persistence. Time-dependent persistence can also be measured without microscopy, as the duration of the lag phase in single cells can be measured using the ScanLag technique28,53. High throughput can be obtained by fluorescence-activated cell sorting (FACS), which has been used to enrich for time-dependent persisters that are persistent by lag79. To enrich for these cells, the expression of a fluorescent protein is induced in all cells. When the cells are moved into an inducer-free medium, in which the expression of the fluorescent protein is repressed, non-growing cells will maintain a high level of fluorescence, whereas the fluorescent proteins will be diluted in growing cells. This method has been used to show that non-replicating (that is, time-dependent) persisters arise in populations of S. Typhimurium upon internalization by macrophages31,105. Persistent subpopulations enriched by FACS can be further analysed by microarray114 or phenotypic assays, which may shed light on the underlying metabolism of each form of persistence115. FACS can also be used to identify dose-dependent persistence at the single-cell level; for example, fluorescent antibiotics were used to detect dose-dependent persistence in methicillin-resistant Staphylococcus aureus (MRSA) cells116.
328 | MAY 2016 | VOLUME 14
fluoroquinolones80, which is indicative of persistence by slow growth or tolerance by slow growth. In addition, small subpopulations of cells with very long lag phases were detected42, which is indicative of persistence by lag. Finally, evidence for dose-dependent persistence was also observed in M. tuberculosis under treatment with isoniazid. The MDK may, in principle, be a useful indicator of which subpopulation of persisters is the dominant factor in treatment failure or relapse; however, measuring the MDK in vivo is technically extremely challenging, owing to the difficulty of controlling parameters, such as the level of antibiotic over time and spatial homogeneity, and of accurately determining the size of the bacterial population. An alternative to directly measuring the MDK in vivo is to determine the correlation between in vivo pharmacokinetic and pharmacodynamic measurements and in vitro measurements of the MDK, as has been done for the MIC81. For example, a study in S. aureus showed that strains that were identified as tolerant in vitro, with an MDK99 extended to 24 hours, were most effectively killed in vivo by a longer treatment duration rather than a higher antibiotic concentration45, which indicated that these strains were also tolerant in vivo. Therefore, the routine determination of the MDK of pathogens isolated in the clinic may help to direct more effective therapies, even when the measurement is made in vitro. However, the evaluation of the MDK currently requires the labour- intensive measurement of time–kill curves; thus, more practical methods to evaluate this metric would make our framework more amenable to clinical use (and would also be beneficial to the study of bacterial survival strategies in the research laboratory). A final caveat is that the in vitro evaluation of the MDK of pathogens in the host is limited to inherited tolerance, which arises from mutations that increase tolerance. Non-inherited tolerance in the host may be due to environmental factors, such as the complex interactions between bacterial pathogens and host cells80 and the immune system82, or the presence of biofilms83 and interactions with other bacterial species84. A major challenge would then be to develop in vitro assays that recreate the conditions that induce tolerance in vivo. Alternatively, direct measurement of the MDK in vivo may become possible, owing to the development of sequencing technologies that may enable the inference of time–kill curves from in vivo sequencing data. www.nature.com/nrmicro
. d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES The definitions used in the framework we introduce in this Opinion article are formulated to describe the response of bacterial populations to antibiotic stress; however, we propose that they are also applicable to a wide range of stresses and biological systems. For example, it has been shown that cancer cells can exhibit drug responses that, in our framework, would be categorized as tolerance by slow growth85 or dose-dependent persistence86. Finally, the survival of a bacterial population under conditions that are designed to kill may have far-reaching consequences for the subsequent emergence of resistance. For example, treatment with numerous antibiotics has been shown to increase the mutation rates of bacterial genomes; the survival of bacterial populations by tolerance may therefore constitute fertile ground for the subsequent development of resistance to the antibiotic. Understanding the bacterial survival strategies operating in different experimental systems should lead to a better understanding of how pathogens evolve resilience to treatment with antibiotics87,88. Asher Brauner, Ofer Fridman, Orit Gefen and Nathalie Q. Balaban are at the Racah Institute of Physics and the Harvey M. Kruger Family Center for Nanoscience and Nanotechnology, Edmond J. Safra Campus, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. Correspondence to N.Q.B.
[email protected] doi:10.1038/nrmicro.2016.34 Published online 15 Apr 2016 McKeegan, K. S., Borges-Walmsley, M. I. & Walmsley, A. R. Microbial and viral drug resistance mechanisms. Trends Microbiol. 10, S8–S14 (2002). 2. Scholar, E. M. & Pratt, W. B. (eds) The Antimicrobial Drugs (Oxford Univ. Press, 2000). 3. D’Costa, V. M., McGrann, K. M., Hughes, D. W. & Wright, G. D. Sampling the antibiotic resistome. Science 311, 374–377 (2006). 4. Bigger, J. W. Treatment of staphylococcal infections with penicillin by intermittent sterilisation. Lancet 244, 497–500 (1944). 5. Hobby, G. L., Meyer, K. & Chaffee, E. Observations on the mechanism of action of penicillin. Proc. Soc. Exp. Biol. Med. 50, 281–285 (1942). 6. Horne, D. & Tomasz, A. Tolerant response of Streptococcus sanguis to β-lactams and other cell-wall inhibitors. Antimicrob. Agents Chemother. 11, 888–896 (1977). 7. Balaban, N. Q., Gerdes, K., Lewis, K. & McKinney, J. D. A problem of persistence: still more questions than answers? Nat. Rev. Microbiol. 11, 587–591 (2013). 8. Kester, J. C. & Fortune, S. M. Persisters and beyond: mechanisms of phenotypic drug resistance and drug tolerance in bacteria. Crit. Rev. Biochem. Mol. Biol. 49, 91–101 (2014). 9. Handwerger, S. & Tomasz, A. Antibiotic tolerance among clinical isolates of bacteria. Annu. Rev. Pharmacol. Toxicol. 25, 349–380 (1985). 10. Tuomanen, E., Cozens, R., Tosch, W., Zak, O. & Tomasz, A. The rate of killing of Escherichia coli by β-lactam antibiotics is strictly proportional to the rate of bacterial growth. J. Gen. Microbiol. 132, 1297–1304 (1986). 11. McDermott, W. Microbial persistence. Yale J. Biol. Med. 30, 257–291 (1958). 1.
12. Lederberg, J. & Zinder, N. Concentration of biochemical mutants of bacteria with penicillin. J. Am. Chem. Soc. 70, 4267–4268 (1948). 13. Gefen, O. & Balaban, N. Q. The importance of being persistent: heterogeneity of bacterial populations under antibiotic stress. FEMS Microbiol. Rev. 33, 704–717 (2009). 14. Balaban, N. Q., Merrin, J., Chait, R., Kowalik, L. & Leibler, S. Bacterial persistence as a phenotypic switch. Science 305, 1622–1625 (2004). 15. Wakamoto, Y. et al. Dynamic persistence of antibiotic-stressed mycobacteria. Science 339, 91–95 (2013). 16. Depardieu, F., Podglajen, I., Leclercq, R., Collatz, E. & Courvalin, P. Modes and modulations of antibiotic resistance gene expression. Clin. Microbiol. Rev. 20, 79–114 (2007). 17. Blair, J. M., Webber, M. A., Baylay, A. J., Ogbolu, D. O. & Piddock, L. J. Molecular mechanisms of antibiotic resistance. Nat. Rev. Microbiol. 13, 42–51 (2015). 18. Chait, R., Craney, A. & Kishony, R. Antibiotic interactions that select against resistance. Nature 446, 668–671 (2007). 19. Wiegand, I., Hilpert, K. & Hancock, R. E. Agar and broth dilution methods to determine the minimal inhibitory concentration (MIC) of antimicrobial substances. Nat. Protoc. 3, 163–175 (2008). 20. Mattie, H. Antibiotic efficacy in vivo predicted by in vitro activity. Int. J. Antimicrob. Agents 14, 91–98 (2000). 21. Paterson, D. L. et al. Outcome of cephalosporin treatment for serious infections due to apparently susceptible organisms producing extended-spectrum β-lactamases: implications for the clinical microbiology laboratory. J. Clin. Microbiol. 39, 2206–2212 (2001). 22. Ishida, K., Guze, P. A., Kalmanson, G. M., Albrandt, K. & Guze, L. B. Variables in demonstrating methicillin tolerance in Staphylococcus aureus strains. Antimicrob. Agents Chemother. 21, 688–690 (1982). 23. Wolfson, J., Hooper, D., McHugh, G., Bozza, M. & Swartz, M. Mutants of Escherichia coli K-12 exhibiting reduced killing by both quinolone and β-lactam antimicrobial agents. Antimicrob. Agents Chemother. 34, 1938–1943 (1990). 24. Mueller, M., de la Pena, A. & Derendorf, H. Issues in pharmacokinetics and pharmacodynamics of antiinfective agents: kill curves versus MIC. Antimicrob. Agents Chemother. 48, 369–377 (2004). 25. Barry, L. A. et al. Methods for determining bactericidal activity of antimicrobial agents; approved guideline. (National Committee for Clinical Laboratory Standards, 1999). 26. Keren, I., Kaldalu, N., Spoering, A., Wang, Y. P. & Lewis, K. Persister cells and tolerance to antimicrobials. Fems Microbiol. Lett. 230, 13–18 (2004). 27. Pasticci, M. B. et al. Bactericidal activity of oxacillin and glycopeptides against Staphylococcus aureus in patients with endocarditis: looking for a relationship between tolerance and outcome. Ann. Clin. Microbiol. Antimicrob. 10, 26 (2011). 28. Fridman, O., Goldberg, A., Ronin, I., Shoresh, N. & Balaban, N. Q. Optimization of lag time underlies antibiotic tolerance in evolved bacterial populations. Nature 513, 418–421 (2014). 29. Regoes, R. R. et al. Pharmacodynamic functions: a multiparameter approach to the design of antibiotic treatment regimens. Antimicrob. Agents Chemother. 48, 3670–3676 (2004). 30. Gefen, O., Gabay, C., Mumcuoglu, M., Engel, G. & Balaban, N. Q. Single-cell protein induction dynamics reveals a period of vulnerability to antibiotics in persister bacteria. Proc. Natl Acad. Sci. USA 105, 6145–6149 (2008). 31. Helaine, S. et al. Dynamics of intracellular bacterial replication at the single cell level. Proc. Natl Acad. Sci. USA 107, 3746–3751 (2010). 32. Amato, S. M., Orman, M. A. & Brynildsen, M. P. Metabolic control of persister formation in Escherichia coli. Mol. Cell 50, 475–487 (2013). 33. Maisonneuve, E., Castro-Camargo, M. & Gerdes, K. (p)ppGpp controls bacterial persistence by stochastic induction of toxin–antitoxin activity. Cell 154, 1140–1150 (2013). 34. Chao, L. & Levin, B. R. Structured habitats and the evolution of anticompetitor toxins in bacteria. Proc. Natl Acad. Sci. USA 78, 6324–6328 (1981). 35. Rodionov, D. G. & Ishiguro, E. E. Effects of inhibitors of protein synthesis on lysis of Escherichia coli induced by β-lactam antibiotics. Antimicrob. Agents Chemother. 40, 899–903 (1996).
NATURE REVIEWS | MICROBIOLOGY
36. Orman, M. A. & Brynildsen, M. P. Dormancy is not necessary or sufficient for bacterial persistence. Antimicrob. Agents Chemother. 57, 3230–3239 (2013). 37. Johansen, H. K., Jensen, T. G., Dessau, R. B., Lundgren, B. & Frimodt-Moller, N. Antagonism between penicillin and erythromycin against Streptococcus pneumoniae in vitro and in vivo. J. Antimicrob. Chemother. 46, 973–980 (2000). 38. Thonus, I. P., Fontijne, P. & Michel, M. F. Ampicillin susceptibility and ampicillin-induced killing rate of Escherichia coli. Antimicrob. Agents Chemother. 22, 386–390 (1982). 39. Mascio, C. T., Alder, J. D. & Silverman, J. A. Bactericidal action of daptomycin against stationaryphase and nondividing Staphylococcus aureus cells. Antimicrob. Agents Chemother. 51, 4255–4260 (2007). 40. de Steenwinkel, J. E. et al. Time–kill kinetics of antituberculosis drugs, and emergence of resistance, in relation to metabolic activity of Mycobacterium tuberculosis. J. Antimicrob. Chemother. 65, 2582–2589 (2010). 41. Evans, D. J., Allison, D. G., Brown, M. R. & Gilbert, P. Susceptibility of Pseudomonas aeruginosa and Escherichia coli biofilms towards ciproflaxin: effect of specific growth rate. J. Antimicrob. Chemother. 27, 177–184 (1991). 42. Manina, G., Dhar, N. & McKinney, J. D. Stress and host immunity amplify Mycobacterium tuberculosis phenotypic heterogeneity and induce nongrowing metabolically active forms. Cell Host Microbe 17, 32–46 (2015). 43. Kitano, K. & Tomasz, A. Escherichia coli mutants tolerant to β-lactam antibiotics. J. Bacteriol. 140, 955–963 (1979). 44. Bernier, S. P. et al. Starvation, together with the SOS response, mediates high biofilm-specific tolerance to the fluoroquinolone ofloxacin. PloS Genet. 9, e1003144 (2013). 45. Sandberg, A. et al. Intra- and extracellular activities of dicloxacillin against Staphylococcus aureus in vivo and in vitro. Antimicrob. Agents Chemother. 54, 2391–2400 (2010). 46. Dorr, T., Davis, B. M. & Waldor, M. K. Endopeptidasemediated β-lactam tolerance. PloS Pathog. 11, e1004850 (2015). 47. Dorr, T., Vulic, M. & Lewis, K. Ciprofloxacin causes persister formation by inducing the TisB toxin in Escherichia coli. PloS Biol. 8, e1000317 (2010). 48. Wiuff, C. & Andersson, D. I. Antibiotic treatment in vitro of phenotypically tolerant bacterial populations. J. Antimicrob. Chemother. 59, 254–263 (2007). 49. Johnson, P. J. T. & Levin, B. R. Pharmacodynamics, population dynamics, and the evolution of persistence in Staphylococcus aureus. PloS Genet. 9, e1003123 (2013). 50. Gefen, O., Fridman, O., Ronin, I. & Balaban, N. Q. Direct observation of single stationary-phase bacteria reveals a surprisingly long period of constant protein production activity. Proc. Natl Acad. Sci. USA 111, 556–561 (2014). 51. Lewis, K. Persister cells, dormancy and infectious disease. Nat. Rev. Microbiol. 5, 48–56 (2007). 52. Nguyen, D. et al. Active starvation responses mediate antibiotic tolerance in biofilms and nutrient-limited bacteria. Science 334, 982–986 (2011). 53. Levin-Reisman, I. et al. Automated imaging with ScanLag reveals previously undetectable bacterial growth phenotypes. Nat. Methods 7, 737–739 (2010). 54. Luidalepp, H., Joers, A., Kaldalu, N. & Tenson, T. Age of inoculum strongly influences persister frequency and can mask effects of mutations implicated in altered persistence. J. Bacteriol. 193, 3598–3605 (2011). 55. Madar, D. et al. Promoter activity dynamics in the lag phase of Escherichia coli. BMC Syst. Biol. 7, 136 (2013). 56. Joers, A., Kaldalu, N. & Tenson, T. The frequency of persisters in Escherichia coli reflects the kinetics of awakening from dormancy. J. Bacteriol. 192, 3379–3384 (2010). 57. Putrinš, M., Kogermann, K., Lukk, E. & Lippus, M. Phenotypic heterogeneity enables uropathogenic Escherichia coli to evade killing by antibiotics and serum complement. Infect. Immun. 83, 1056–1067 (2015). 58. Pearl, S., Gabay, C., Kishony, R., Oppenheim, A. & Balaban, N. Q. Nongenetic individuality in the host–phage interaction. PloS Biol. 6, 957–964 (2008).
VOLUME 14 | MAY 2016 | 329 . d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©
PERSPECTIVES 59. Baranyi, J. Stochastic modelling of bacterial lag phase. Int. J. Food Microbiol. 73, 203–206 (2002). 60. Akerlund, T., Nordstrom, K. & Bernander, R. Analysis of cell size and DNA content in exponentially growing and stationary-phase batch cultures of Escherichia coli. J. Bacteriol. 177, 6791–6797 (1995). 61. Hartman, B. J. & Tomasz, A. Expression of methicillin resistance in heterogeneous strains of Staphylococcus aureus. Antimicrob. Agents Chemother. 29, 85–92 (1986). 62. Levin, B. R. & Rozen, D. E. Non-inherited antibiotic resistance. Nat. Rev. Microbiol. 4, 556–562 (2006). 63. Nataro, J. P., Blaser, M. J. & Cunningham-Rundles, S. (eds) in Persistent Bacterial Infections. 3–10 (ASM Press, 2000). 64. Rotem, E. et al. Regulation of phenotypic variability by a threshold-based mechanism underlies bacterial persistence. Proc. Natl Acad. Sci. USA 107, 12541–12546 (2010). 65. Korch, S. B. & Hill, T. M. Ectopic overexpression of wildtype and mutant hipA genes in Escherichia coli: effects on macromolecular synthesis and persister formation. J. Bacteriol. 188, 3826–3836 (2006). 66. Moyed, H. S. & Bertrand, K. P. hipA, a newly recognized gene of Escherichia coli K-12 that affects frequency of persistence after inhibition of murein synthesis. J. Bacteriol. 155, 768–775 (1983). 67. Levin-Reisman, I. & Balaban, N. Q. in Bacterial Persistence: Methods and Protocols (eds Michiels, J. & Fauvart, M.) 75–81 (Humana Press, 2015). 68. El Meouche, I., Siu, Y. & Dunlop, M. J. Stochastic expression of a multiple antibiotic resistance activator confers transient resistance in single cells. Sci. Rep. 6, 19538 (2016). 69. Adams, K. N. et al. Drug tolerance in replicating mycobacteria mediated by a macrophage-induced efflux mechanism. Cell 145, 39–53 (2011). 70. Kayser, F. H., Benner, E. J. & Hoeprich, P. D. Acquired and native resistance of Staphylococcus aureus to cephalexin and other β‑lactam antibiotics. Appl. Microbiol. 20, 1–5 (1970). 71. El‑Halfawy, O. M. & Valvano, M. A. Antimicrobial heteroresistance: an emerging field in need of clarity. Clin. Microbiol. Rev. 28, 191–207 (2015). 72. Adams, K. N., Szumowski, J. D. & Ramakrishnan, L. Verapamil, and its metabolite norverapamil, inhibit macrophage-induced, bacterial efflux pump-mediated tolerance to multiple anti-tubercular drugs. J. Infect. Dis. 210, 456–466 (2014). 73. Mattie, H., Sekh, B. A., van Ogtrop, M. L. & van Strijen, E. Comparison of the antibacterial effects of cefepime and ceftazidime against Escherichia coli in vitro and in vivo. Antimicrob. Agents Chemother. 36, 2439–2443 (1992). 74. Coates, A. R. & Hu, Y. Targeting non-multiplying organisms as a way to develop novel antimicrobials. Trends Pharmacol. Sci. 29, 143–150 (2008). 75. Feng, J. et al. Identification of novel activity against Borrelia burgdorferi persisters using an FDA approved drug library. Emerg. Microbes Infect. 3, e49 (2014). 76. Kim, J. S. et al. Selective killing of bacterial persisters by a single chemical compound without affecting normal antibiotic-sensitive cells. Antimicrob. Agents Chemother. 55, 5380–5383 (2011). 77. Fleck, L. E. et al. A screen for and validation of prodrug antimicrobials. Antimicrob. Agents Chemother. 58, 1410–1419 (2014). 78. Conlon, B. P. et al. Activated ClpP kills persisters and eradicates a chronic biofilm infection. Nature 503, 365–370 (2013). 79. Roostalu, J., Jõers, A., Luidalepp, H., Kaldalu, N. & Tenson, T. Cell division in Escherichia coli cultures
monitored at single cell resolution. BMC Microbiol. 8, 68 (2008). 80. Claudi, B. et al. Phenotypic variation of Salmonella in host tissues delays eradication by antimicrobial chemotherapy. Cell 158, 722–733 (2014). 81. Mattie, H., Zhang, L. C., van Strijen, E., Sekh, B. R. & Douwes-Idema, A. E. Pharmacokinetic and pharmacodynamic models of the antistaphylococcal effects of meropenem and cloxacillin in vitro and in experimental infection. Antimicrob. Agents Chemother. 41, 2083–2088 (1997). 82. Arnoldini, M. et al. Bistable expression of virulence genes in Salmonella leads to the formation of an antibiotic-tolerant subpopulation. PLoS Biol. 12, e1001928 (2014). 83. Nickel, J. C., Ruseska, I., Wright, J. B. & Costerton, J. W. Tobramycin resistance of Pseudomonas aeruginosa cells growing as a biofilm on urinary catheter material. Antimicrob. Agents Chemother. 27, 619–624 (1985). 84. Hayes, C. S. & Low, D. A. Signals of growth regulation in bacteria. Curr. Opin. Microbiol. 12, 667–673 (2009). 85. Bhuyan, B. K., Fraser, T. J. & Day, K. J. Cell proliferation kinetics and drug sensitivity of exponential and stationary populations of cultured L1210 cells. Cancer Res. 37, 1057–1063 (1977). 86. Sharma, S. V. et al. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell 141, 69–80 (2010). 87. Jayaraman, R. Bacterial persistence: some new insights into an old phenomenon. J. Biosci. 33, 795–805 (2008). 88. Cohen, N. R., Lobritz, M. A. & Collins, J. J. Microbial persistence and the road to drug resistance. Cell Host Microbe 13, 632–642 (2013). 89. Potrykus, K. & Cashel, M. (p)ppGpp: still magical? Annu. Rev. Microbiol. 62, 35–51 (2008). 90. Kaspy, I. et al. HipA-mediated antibiotic persistence via phosphorylation of the glutamyl-tRNA-synthetase. Nat. Commun. 4, 3001 (2013). 91. Germain, E., Castro-Roa, D., Zenkin, N. & Gerdes, K. Molecular mechanism of bacterial persistence by HipA. Mol. Cell 52, 248–254 (2013). 92. Hahn, J., Tanner, A. W., Carabetta, V. J., Cristea, I. M. & Dubnau, D. ComGA–RelA interaction and persistence in the Bacillus subtilis K-state. Mol. Microbiol. 97, 454–471 (2015). 93. Gerdes, K. & Maisonneuve, E. Bacterial persistence and toxin–antitoxin loci. Annu. Rev. Microbiol. 66, 103–123 (2012). 94. Girgis, H. S., Harris, K. & Tavazoie, S. Large mutational target size for rapid emergence of bacterial persistence. Proc. Natl Acad. Sci. USA 109, 12740–12745 (2012). 95. Hansen, S., Lewis, K. & Vulic, M. Role of global regulators and nucleotide metabolism in antibiotic tolerance in Escherichia coli. Antimicrob. Agents Chemother. 52, 2718–2726 (2008). 96. Spoering, A. L., Vulic, M. & Lewis, K. GlpD and PlsB participate in persister cell formation in Escherichia coli. J. Bacteriol. 188, 5136–5144 (2006). 97. Vazquez-Laslop, N., Lee, H. & Neyfakh, A. A. Increased persistence in Escherichia coli caused by controlled expression of toxins or other unrelated proteins. J. Bacteriol. 188, 3494–3497 (2006). 98. Shan, Y., Lazinski, D., Rowe, S. E., Camili, A. & Lewis, K. Genetic basis of persister tolerance to aminoglycosides in Escherichia coli. mBio 6, e00078‑15 (2015). 99. Balaban, N. Q. Persistence: mechanisms for triggering and enhancing phenotypic variability. Curr. Opin. Genet. Dev. 21, 768–775 (2011).
330 | MAY 2016 | VOLUME 14
100. Casadesus, J. & Low, D. A. Programmed heterogeneity: epigenetic mechanisms in bacteria. J. Biol. Chem. 288, 13929–13935 (2013). 101. Huh, D. & Paulsson, J. Non-genetic heterogeneity from stochastic partitioning at cell division. Nat. Genet. 43, 95–100 (2011). 102. Tsimring, L. S. Noise in biology. Rep. Progress Phys. 77, 026601 (2014). 103. Gelens, L., Hill, L., Vandervelde, A., Danckaert, J. & Loris, R. A general model for toxin–antitoxin module dynamics can explain persister cell formation in E. coli. PLoS Comput. Biol. 9, e1003190 (2013). 104. Maisonneuve, E., Shakespeare, L. J., Jorgensen, M. G. & Gerdes, K. Bacterial persistence by RNA endonucleases. Proc. Natl Acad. Sci. USA 108, 13206–13211 (2011). 105. Helaine, S. et al. Internalization of Salmonella by macrophages induces formation of nonreplicating persisters. Science 343, 204–208 (2014). 106. Aldridge, B. B. et al. Asymmetry and aging of mycobacterial cells lead to variable growth and antibiotic susceptibility. Science 335, 100–104 (2012). 107. Kieser, K. J. & Rubin, E. J. How sisters grow apart: mycobacterial growth and division. Nat. Rev. Microbiol. 12, 550–562 (2014). 108. Vaubourgeix, J. et al. Stressed mycobacteria use the chaperone ClpB to sequester irreversibly oxidized proteins asymmetrically within and between cells. Cell Host Microbe 17, 178–190 (2015). 109. Wu, Y. X., Vulic, M., Keren, I. & Lewis, K. Role of oxidative stress in persister tolerance. Antimicrob. Agents Chemother. 56, 4922–4926 (2012). 110. Song, Y., Rubio, A., Jayaswal, R. K., Silverman, J. A. & Wilkinson, B. J. Additional routes to Staphylococcus aureus daptomycin resistance as revealed by comparative genome sequencing, transcriptional profiling, and phenotypic studies. PLoS ONE 8, e58469 (2013). 111. Brehm-Stecher, B. F. & Johnson, E. A. Single-cell microbiology: tools, technologies, and applications. Microbiol. Mol. Biol. Rev. 68, 538–559 (2004). 112. Ping, W. et al. Robust growth of Escherichia coli. Curr. Biol. 20, 1099–1103 (2010). 113. Iino, R., Matsumoto, Y., Nishino, K., Yamaguchi, A. & Noji, H. Design of a large-scale femtoliter droplet array for single-cell analysis of drug-tolerant and drug-resistant bacteria. Front. Microbiol. 4, 300 (2013). 114. Shah, D. et al. Persisters: a distinct physiological state of E. coli. BMC Microbiol. 6, 53 (2006). 115. Orman, M. A. & Brynildsen, M. P. Establishment of a method to rapidly assay bacterial persister metabolism. Antimicrob. Agents Chemother. 57, 4398–4409 (2013). 116. Jarzembowski, T., Wisniewska, K., Jozwik, A. & Witkowski, J. Heterogeneity of methicillin-resistant Staphylococcus aureus strains (MRSA) characterized by flow cytometry. Curr. Microbiol. 59, 78–80 (2009).
Acknowledgements
The authors thank N. Shoresh for illuminating discussions regarding this manuscript, and the members of the Balaban laboratory, I. Kaspy and G. Glaser for comments and suggestions. This work is supported by the Minerva Center for Stochastic Decision Making in Microorganisms, a European Research Council (ERC) Starting Grant (260871) and the Israel Science Foundation (492/15).
Competing interests statement
The authors declare no competing interests.
www.nature.com/nrmicro . d e v r e s e r s t h g i r l l A . d e t i m i L s r e h s i l b u P n a l l i m c a M 6 1 0 2 ©