Definitions and guidelines for research on antibiotic persistence

Increasing concerns about the rising rates of antibiotic therapy failure and advances in single-cell analyses have inspired a surge of research into antibiotic persistence. Bacterial persister cells represent a subpopulation of cells that can survive intensive antibiotic treatment without being resistant. Several approaches have emerged to define and measure persistence, and it is now time to agree on the basic definition of persistence and its relation to the other mechanisms by which bacteria survive exposure to bactericidal antibiotic treatments, such as antibiotic resistance, heteroresistance or tolerance. In this Consensus Statement, we provide definitions of persistence phenomena, distinguish between triggered and spontaneous persistence and provide a guide to measuring persistence. Antibiotic persistence is not only an interesting example of non-genetic single-cell heterogeneity, it may also have a role in the failure of antibiotic treatments. Therefore, it is our hope that the guidelines outlined in this article will pave the way for better characterization of antibiotic persistence and for understanding its relevance to clinical outcomes.

More than 70 years ago, Hobby 1 and Bigger 2 observed that antibiotics that are considered bactericidal and kill bacteria in fact fail to sterilize cultures. Bigger realized that the small number of bacteria that manage to survive intensive antibiotic treatments are a distinct subpopulation of bacteria that he named 'persisters' . Fuelled in part by increasing concerns about antibiotic resistance but also by technological advances in single-cell analyses, the past 15 years have witnessed a great deal of research on antibiotic persistence by investigators with different backgrounds and perspectives. As the number of scientists that tackle the puzzles and challenges of antibiotic persistence from many different angles has profoundly increased, it is now time to agree on the basic definition of persistence and its distinction from the other mechanisms by which bacteria survive exposure to bactericidal antibiotic treatments 3 . Several approaches have independently emerged to define and measure persistence. Research groups following seemingly similar procedures may reach different results, and careful examination of the experimental procedures often reveals that results of different groups cannot be compared. During the European Molecular Biology Organization (EMBO) Workshop 'Bacterial Persistence and Antimicrobial Therapy' (10-14 June 2018) in Ascona, Switzerland, which brought together 121 investigators involved in antibiotic persistence research from 21 countries, a discussion panel laid the main themes for a Consensus Statement on the definition and detection procedure of antibiotic persistence detailed below. In light of the potential role that antibiotic persistence can have in antibiotic treatment regimens, it is our hope that clarification and standardization of experimental procedures will facilitate the translation of basic science research into practical guidelines.

Defining the persistence phenomena
We adopt here a phenomenological definition of antibiotic persistence that is based on a small set of observations that can be made from experiments performed in vitro and that does not assume a specific mechanism. We focus on the differences and similarities between antibiotic persistence and other processes enabling bacteria to survive exposure to antibiotic treatments that could kill them, such as resistance, tolerance and heteroresistance. We identify different types of persistence that should be measured differently to obtain meaningful results; therefore, the definition of these types goes beyond semantics. For the more mathematically oriented readers, we provide a mathematical definition of the Definitions and guidelines for research on antibiotic persistence Nathalie Q. Balaban  Abstract | Increasing concerns about the rising rates of antibiotic therapy failure and advances in single-cellanalyseshaveinspiredasurgeofresearchintoantibioticpersistence.Bacterial persister cells represent a subpopulation of cells that can survive intensive antibiotic treatment withoutbeingresistant.Severalapproacheshaveemergedtodefineandmeasurepersistence,and it is now time to agree on the basic definition of persistence and its relation to the other mechanisms bywhichbacteriasurviveexposuretobactericidalantibiotictreatments,suchasantibiotic resistance,heteroresistanceortolerance.InthisConsensusStatement,weprovidedefinitionsof persistencephenomena,distinguishbetweentriggeredandspontaneouspersistenceandprovide aguidetomeasuringpersistence.Antibioticpersistenceisnotonlyaninterestingexampleof non-geneticsingle-cellheterogeneity,itmayalsohavearoleinthefailureofantibiotictreatments. Therefore,itisourhopethattheguidelinesoutlinedinthisarticlewillpavethewayforbetter characterizationofantibioticpersistenceandforunderstandingitsrelevancetoclinicaloutcomes.
Persistence versus resistance. 'Resistance' is the ability of bacteria to replicate and not just survive in the presence of a drug (Box 2). The most common measure of the level of resistance is the minimum inhibitory concentration (MIC), which is the lowest concentration of the antibiotic required to prevent the replication of the bacteria. A higher MIC corresponds with a higher level of resistance (Fig. 2a). Resistance is inherited and may be acquired by horizontal gene transfer of resistanceencoding genes (for example, encoding antibiotic inactivating enzymes 10 or efflux pumps 11 ) or mutations (for example, leading to modification of the antibiotic target) that confer the resistance phenotype to the bacterial population 12 .
'Persistence' is the ability of a subset of the population to survive exposure to a bactericidal drug concentration (Fig. 2). Therefore, persistence is defined only for bactericidal antibiotics. Several features distinguish persistence from resistance. First, the hallmark of antibiotic persistence is the biphasic killing curve (Fig. 2c); that is, the observation that not all bacteria in a clonal culture are killed at the same rate. Second, when persister cells regrow without antibiotics (see below), their progeny give rise to a population that is as susceptible to drugs as the parental population it was isolated from. Third, the level of persistence, namely, the size of the persister subpopulation, will only weakly depend on the concentration of the drug as long as it is far above the MIC. In addition, the survival advantage of persister bacteria is often observed for antibiotic treatments belonging to different classes of antibiotics, for example, β-lactams and fluoroquinolones 13 . Fourth, in contrast to resistant cells, persister bacteria cannot replicate in the presence of the drug any better than the non-persister cells but are killed at a lower rate than the susceptible population from which they arose. This property also distinguishes persistence from heteroresistance, a phenomenon in which a small subpopulation transiently displays a substantially (more than eightfold) higher MIC 14 (see Box 1).
Persistence and tolerance. 'Tolerance' and persistence are similar phenomena of increased survival in the presence of an antibiotic without an increase in the MIC. In studies that focus on only a qualitative understanding of the molecular mechanisms, the two terms are often interchangeable 15 . However, persistence has the added attribute of affecting only a subpopulation of cells, whereas tolerance is the general ability of a population to survive longer treatments, for example, by having a lower killing rate (see Fig. 2b), but without a change in the MIC 16 . Persister cells are simply a subpopulation of tolerant bacteria, and persistence could also be called 'heterotolerance' . Tolerant populations survive the period of antibiotic treatment better, with, typically, a weak dependence on the antibiotic concentration. Therefore, the MIC of tolerant cells is unchanged compared with non-tolerant strains. What characterizes their slower killing, even at high concentrations of the drug, is the time required to kill a large fraction of the population, for example, the MDK 99 , which is the minimum duration of treatment that kills 99% of the bacterial population. Persistence is a special case of tolerance in which a subpopulation of persister cells can survive the antibiotic treatment much better than the majority of the population, as reflected in the biphasic killing curve. Not surprisingly, mechanisms linked to tolerance, such as dormancy (see definitions in Box 3), reduced metabolism and ATP levels, have also been identified in persistence 9 . Therefore, when studying persistence, two mechanisms are of interest, and the first one overlaps with tolerance research whereas the second is specific to persistence: (1) the molecular mechanism of tolerance that enables the persister bacteria to survive, for example, a reduction in their metabolism, and (2) the mechanism that generates heterogeneity in the population 17 , for example, nonlinear mechanisms leading to bimodality by amplifying stochasticity 18,19 . Finally, several persister subpopulations may coexist; therefore, a multimodal killing curve may occur.

Single-cell versus population phenotype.
Because the definition of antibiotic persistence is anchored in the heterogeneity of the response to antibiotics in the population, it is a population-level phenotype. However, tolerance can be the attribute of a whole population that is killed at a slow rate as well as of a single cell that manages to survive an extensive treatment (see definitions in Box 3).
Genetic mutations can increase the tolerance of a strain if they result in slower killing. Similarly, genetic mutations can increase the persistence of a strain either by reducing the killing rate of the persistent subpopulation even more or by increasing the fraction of that subpopulation, as, for example, in the hipA7 high persistence mutant 20 . The population level of high persistence is then genetically inherited.

Types of persister bacteria
Whether a single general or multiple specific molecular mechanisms underlie persistence is still under debate 21-23 and therefore will not be discussed in this article. However, distinct ways for generating persister bacteria in a culture have been identified. Distinguishing between the types of persistence identified thus far is crucial because each type requires a different procedure to measure the persistence level.
Triggered persistence. In most observations of persistence described to date, the fraction of persister bacteria is generated upon a stress signal, the most common one being starvation (Fig. 3). This type of persistence, here termed triggered persistence, was previously called type I persistence 24 . Even when the signal is removed, for example, by diluting a starved overnight culture in fresh volume 17 | JulY 2019 | 443 NATuRe RevIeWS | Microbiology C o n S e n S u S S tat e m e n t

Box1| Mathematical distinctions between antibiotic resistance, tolerance, persistence and heteroresistance
Predictive models of the survival of microorganisms exposed to cidal drugs show that measuring the minimum inhibitory concentration (mIC) is not enough to characterize the behaviour, although it is widely used 53,54 . Common phenomenological models for the relationship of the survival, S, with the concentration of the drug, c, or duration of treatment, t, are the Zhi function 55 , emax or Hill model 56 . In these frameworks, the killing rate, ψ, is described by three main parameters that represent distinct underlying physicochemical mechanisms: the mIC; the minimum duration to kill 99% of the population, MDK 99 ; and the Hill coefficient for the steepness of the concentration dependence, k.
This general function predicts how the concentration of the antibiotic and its duration will affect the growth or death of a strain with growth rate without antibiotic, ψ max . Note that the common notation of the model uses the following: min 99 In the framework of this model, resistance is defined as an increase in the mIC, whereas tolerance is defined as an increase in the mDK 99 . Thus far, the parameters describe a uniform population. When the population is heterogeneous, it means that at least one of the parameters is heterogeneous.
Heteroresistance entails that subpopulations of cells have a higher MIC than the majority of the population. In typical reports of heteroresistance, it is also assumed that the heritability of the increased mIC is long enough to create detectable colonies 57 .
Antibiotic persistence (which in this context could have been called heterotolerance) entails that a subpopulation of cells have a higher mDK 99 than the majority of the population. If we assume that the fraction of persisters is α, the survival can be written as the sum of the survival of two subpopulations with different killing rates: C o n S e n S u S S tat e m e n t medium, persister cells may still linger for extensive periods and be the ones found in the surviving fraction. Even when the culture is allowed to resume growth for a few hours and to reach what seems to be 'exponential growth' , a fraction of the persisters triggered by the previous starvation may remain in a lag phase. Therefore, the lag time distribution of single cells after starvation or exposure to a stress is an important factor to take into account as it may determine the persistence level 25,26 . Many stress conditions have been shown to generate triggered persistence, including limitation of different nutrients 27 , high cell number 28 , acid stress, immune factors 29 and exposure to immune cells 30 .
Confounding results can occur when the antibiotic itself serves as a trigger for growth arrest, causing druginduced persistence 31 and sometimes paradoxical lower killing at high drug concentration 32 . In this case, instead of killing the cells, a bactericidal antibiotic becomes bacteriostatic for a subpopulation of cells that respond to the antibiotic signal itself, for example, by activating a stress response that enables them to survive 31,33 . This type of persistence does not depend on the history of the culture before exposure to the drug 34 and therefore may be attributed to spontaneous persistence. However, because in this case persistence relies on the response of the cells to the antibiotic, it may be more specific to the class of antibiotic used and its concentration than other forms of persistence.
Spontaneous persistence. Persistence may be observed without any trigger and when the culture is in steadystate exponential growth and all parameters are kept constant, that is, during balanced growth (Fig. 3). In this case, persisters may occur spontaneously, and their fraction remains constant as long as the steady-state growth is maintained. Spontaneous persistence was previously called type II persistence 24 . This form of persistence seems to be much less common than triggered persistence.

Effect of mutations.
Importantly, apart from the environmental triggers mentioned above, all types of persistence may be increased (or reduced) by mutations. Although within a clonal population, persister and non-persister bacteria are typically isogenic, mutations have been identified that are able to increase the level of persistence or tolerance. For example, one of the first identified high persistence mutations, the hipA7 mutation 20 , increases the level of triggered persistence in Escherichia coli by orders of magnitude, reaching persistence levels of about 20% of the population, whereas it is typically below 0.1% for wild-type strains. A high persistence mutation can be viewed as a tolerance mutation with partial penetrance in the whole population; only the subpopulation of persister bacteria will exhibit a phenotype owing to the mutation and will die slower. A high tolerance mutation reduces the killing rate of the whole population. In other words, 100% persistence is equivalent to tolerance.
We stress the fact that the definition of persistence presented here is not directly linked to a specific mechanism or to a physiological state of the bacteria. Rather, it is defined by the time-kill assay; therefore, we outline below some important considerations for increasing the reproducibility and reliability of persistence detection. Typical microbiology procedures developed for measuring uniform bulk phenomena need to be carefully re-evaluated when measuring survival that is dominated by a small subpopulation of cells.
Although we focus here on antibiotic persistence of bacteria, we believe that the definitions and guidelines should be relevant for heterogeneous responses to other drugs such as antifungals 35 and anticancer treatments 36 .
A guide to measuring persistence The starting point for identifying persistence is the timekill assay, which measures survival of bacteria at different time points during exposure to the antibiotic. Survival is defined as the ability to regrow when the antibiotic

Box2| The mechanistic distinction between resistance and tolerance
There are numerous mechanisms of antibiotic resistance. The main types of resistance are a reduction in intracellular drug levels (due to reduced uptake or increased efflux), inactivation of the antibiotic or target modification to reduce drug binding. Although mechanisms of resistance are diverse, they typically achieve the same result -reduced antibiotic binding to the target 12 , which allows bacteria to grow. In order to understand tolerance, we need to consider that bactericidal antibiotics kill not by inhibiting the targets, but by corrupting them 58 , leading to toxic products 42,59 . By slowing down these processes in persister bacteria, the activity of targets is diminished, leading to higher survival. is removed. The hallmark of persistence is the bimodal (or multimodal) killing curve (Fig. 1). However, observing bimodal killing alone is not enough, and several additional steps are required to evaluate whether the bimodality results from persistence or from resistance and to differentiate between the different types of persistence mentioned above. A standardization of the assays and a clear description of the conditions used are required to enable comparing different strains or conditions and results from different laboratories.

Does the bimodal killing curve really reflect persistence?
First, a bimodal killing curve may be due to resistant mutants. To rule out this effect, surviving bacteria that are clearly in the tail of the survival curve should be regrown in the same conditions and exposed again to the same antibiotic treatment. Persistence requires that the killing curve remains the same as in the initial inoculation 37 . If resistant mutants were responsible for the slower killing rate in the first killing curve, the second assay will show reduced killing of a much higher proportion of the population than in the first assay.
Second, in order to distinguish persistence from transient modes of resistance such as heteroresistance 14 , the killing curve should be performed at high antibiotic concentration, at least several times the MIC, to efficiently kill bacteria that have a higher MIC than the rest of the population. The killing rate should only weakly depend on the antibiotic concentration. Strong dependence on the antibiotic concentration (scaled with MIC) reflects phenomena linked to resistance.
Third, experimental pitfalls that may result in bimodal killing should be ruled out. One of the most common reasons for a decrease in the killing rate is degradation of the drug with time. Therefore, it is important to test that the killing efficacy of the drug itself does not decrease with time because of natural degradation of the drug, uptake by bacteria or changes in the medium. Another reason for survival of some bacteria may be their adhesion to the walls of the culture vessel 38 , where the antibiotic may not efficiently kill them.
Finally, in addition to clearly stating in which conditions the time-kill assay is performed, care should be given to the recovery conditions, when bacteria are allowed to grow after removal of the antibiotic. The precise conditions for the evaluation of survival after treatment should be described, such as the washing out of the antibiotics, the medium in which the bacteria are recovered and the time that has passed from the exposure to the antibiotics until the exposure to the recovery conditions. For example, it has been shown that keeping the bacteria in non-growing conditions after treatment may increase their survival 39 . In addition, bacteria recovering from an antibiotic treatment may have a delayed regrowth either because they are in the tail of the lag time distribution 26,40 or because of the post-antibiotic effect 41,42 , which results in the delayed growth of bacteria after treatment. Therefore, evaluating the survival by counting colonies should be done not only after the typical appearance time of colonies but also several days later.

Measuring triggered persistence.
As the trigger is an integral part of triggered persistence, the trigger duration, intensity and exact conditions should be clearly mentioned and kept the same between experiments. For example, one of the most common triggers of persistence is starvation. Many reports used an 'overnight culture' as inoculum. This overnight culture has been exposed to several stress signals during starvation, such R e s i s t a n t S u s c e p t i b l e T o l e r a n t P e r s i s t e n t C o n S e n S u S S tat e m e n t as high cell density, stringent response and altered pH, that may trigger persistence, and therefore inevitably still contains bacteria that experienced a trigger. In this example of triggered persistence, the persistence level will depend strongly on several parameters, including the size of the inoculum, the time that has elapsed since the inoculum was regrown and the duration of starvation during the previous overnight culture 26,40 . Typically, to obtain reproducible results, the time between the trigger for persistence and the exposure to antibiotics should be minimized to avoid the uncontrolled loss of persister bacteria that switch back to normal cells.
Measuring drug-induced persistence. The conditions for measuring drug-induced persistence are the same as for measuring spontaneous persistence, namely, steadystate growth, as the trigger is the drug itself and should be applied in steady-state conditions to avoid stationaryphase-induced persistence. Without further characterizations, the spontaneous and drug-induced persistence are difficult to distinguish. In this case, direct observation of single cells as they respond to the antibiotics is needed 43,44 , or a dissection of the molecular mechanism that allows the bacteria to respond to the drug by activating a stress response 31 . Earlier attempts to characterize drug-induced tolerance or drug-induced persistence made use of the minimum bactericidal concentration (MBC) 38 . The MBC is the concentration required to kill bacteria. Some drugs may arrest the growth at the MIC but require a higher concentration to kill. If the drug itself induces persistence at the MIC, a higher concentration may be required to reach killing. Drug-induced persister bacteria have a higher MBC than the rest of the population, but their MIC is unchanged.
Measuring spontaneous persistence. In contrast to triggered persistence, which is determined by the history of the culture, the rare spontaneous persistence should be measured in conditions of steady-state (also called balanced) growth so as to avoid the effect of the past growth conditions. This measurement can be achieved in a chemostat or by subdiluting the culture several times 37 before performing the time-kill assay, making sure to dilute the inoculum to below the persistence level 24 . As spontaneous persistence is a steady-state phenomenon, care should be taken to evaluate whether the culture remains in steady-state growth, also after the inoculum influence has been ruled out. For example, a common pitfall is to perform the time-kill assay without diluting and subculturing the bacteria for enough time to eliminate the persister bacteria triggered by past stationary phase growth. Another common pitfall is to perform the time-kill assay when the culture is too close to the next stationary phase, which again may trigger the formation of persister cells. Even if the culture seems to be growing exponentially, it may no longer be in balanced growth and persister formation may be already triggered at a cell density that is ten times lower than the maximal density 45 . The spontaneous persistence fraction should remain constant with time in steady-state growth conditions. A simple way to test that the results do not depend on the cell density is to perform the same experiment at a twofold lower density and verify that the persistence fraction remains the same.
Regrowth of persister bacteria. An inherent part of the persistence phenomenon is the ability of persisters to eventually resume growth. As evidenced by the low killing rate displayed in the second phase of biphasic killing curves (Fig. 2c), persisters may resume growth at a low and constant rate, independently of the presence of the drug. Only persisters resuming growth after cessation of the antibiotic treatment will give rise to a new population of susceptible bacteria.
Single-cell observation often shows non-growing cells that remain intact during exposure to the drug. However, regrowth must be documented 30 to illustrate that bacteria have survived exposure to the drug before those can be dubbed persisters.

Conclusion
There has been a sharp increase in the interest for antibiotic persistence in the past years in the background of growing concerns about antimicrobial resistance. The observation that triggered persistence evolves fast in vitro 46,47 and can be followed by the evolution of resistance 48 suggests that persistence may be evolving quickly in the host as well. It has been suggested that the presence of antibiotic persister cells is responsible, Box3| Definitions Antibiotic resistant cell An antibiotic resistant cell is a cell that survives antibiotic treatment by carrying a resistance factor (for example, an efflux pump). Resistance factors enable resistant bacteria to grow at antibiotic concentrations that would prevent the growth of more susceptible bacteria.

Antibiotic tolerant cell
An antibiotic tolerant cell is a cell that survives treatment with an antibiotic, without carrying a resistance factor, and that can regrow after removal of the antibiotic. Often, tolerant cells are non-growing before antibiotic exposure, but not necessarily. Tolerance factors enable bacteria to survive the duration of treatment that would kill more susceptible bacteria. These tolerance factors can be environmental or genetic.

Antibiotic persistence
Antibiotic persistence is a population-level phenomenon that historically has been derived from the observation of biphasic killing curves, indicating the presence of two subpopulations, consisting of cells that are killed fast by the antibiotic and tolerant cells that may survive. By definition, the term antibiotic persistence is always connected with a heterogeneous population, in which only a part of the population consists of tolerant cells.

Tolerance
Tolerance is a population-level phenomenon that enables the population to survive the duration of a transient antibiotic treatment several times above the minimum inhibitory concentration (mIC) without a resistance mechanism.

Persister cell
A persister cell is a tolerant cell originating from a population that displays antibiotic persistence.

Dormancy
Dormancy reflects the state of a bacterium that does not grow and has decreased activity when compared with growing cells or even typical stationary phase cells. This term is often also used for single cells that are viable but do not grow despite environmental conditions that support growth. Dormant bacteria are often tolerant to many antibiotics because of their growth arrest or their decreased metabolism 60 . However, tolerance and persistence may arise without dormancy.
at least partly, for lack of clearance of pathogenic bacteria by antibiotic treatment. Indeed, even in the absence of any antibiotic resistance, many bacterial infections are hard to treat and tend to relapse (such as tuberculosis, lung infections in people with cystic fibrosis, systemic infections with Salmonella, tonsillitis and urinary tract infections). The underlying reasons are most likely multifactorial, with suboptimal pharmacodynamics in the host probably playing a major role in some instances. However, it is also clear that non-growing bacteria 29 and high-persister-forming mutants are selected over time in patients exposed to repeated doses of antibiotics 49,50 . Further work is needed to evaluate the possible impact of persister cells on the treatment outcome of bacterial infections and to find ways to fight them 15 . As seen above, many pitfalls exist even for in vitro analysis of persistence, and controlling the experimental conditions is crucial. The understanding of persistence in the host, in which our knowledge of the conditions is scarce, is orders of magnitude more challenging 51,52 . It is our hope that the guidelines outlined in this article will enable a consensus on in vitro measurements and pave the way for designing protocols adapted to the clinical evaluation of antibiotic persistence.