The ISME Journal (2019) 13:1239–1251
https://doi.org/10.1038/s41396-019-0344-9
ARTICLE
Bacterial persistence promotes the evolution of antibiotic resistance
by increasing survival and mutation rates
Etthel Martha Windels1,2 ●
Joran Elie Michiels1,2 ●
Maarten Fauvart1,3 ●
Tom Wenseleers 4 ●
Bram Van den Bergh 1,2,5 ●
Jan Michiels1,2
Received: 23 March 2018 / Revised: 9 October 2018 / Accepted: 23 December 2018 / Published online: 15 January 2019
© International Society for Microbial Ecology 2019
Abstract
Persisters are transiently antibiotic-tolerant cells that complicate the treatment of bacterial infections. Both theory and
experiments have suggested that persisters facilitate genetic resistance by constituting an evolutionary reservoir of viable
cells. Here, we provide evidence for a strong positive correlation between persistence and the likelihood to become
genetically resistant in natural and lab strains of E. coli. This correlation can be partly attributed to the increased availability
of viable cells associated with persistence. However, our data additionally show that persistence is pleiotropically linked
with mutation rates. Our theoretical model further demonstrates that increased survival and mutation rates jointly affect the
likelihood of evolving clinical resistance. Overall, these results suggest that the battle against antibiotic resistance will benefit
from incorporating anti-persister therapies.
Introduction
Ever since the introduction of antibiotics in clinical practice,
clinicians have been faced with the emergence and spread
of genetic mutants resistant to the drugs’ therapeutic effects
[1]. Resistance mechanisms allow bacteria to grow in elevated
drug concentrations and are the main culprit for
antibiotic treatment failure [2]. However, the awareness
grows that other mechanisms besides genetic resistance can
enable bacteria to survive in the presence of antibiotics. In
addition to population-wide tolerance mechanisms [3] and
shielding effects of in vivo physical niches [4], the formation
of a small number of drug-tolerant cells or “persisters”
in antibiotic-sensitive populations is being recognized as a
frequent cause of treatment failure [5–10]. Theory and
experiments have shown that depending on the treatment
schedule, either resistance or tolerance/persistence can be
the optimal bacterial strategy to cope with antibiotic stress.
Resistance is likely to evolve when antibiotics are applied
continuously and/or at low doses, while intermittent treatment
with high drug concentrations tends to select for
increased tolerance and persistence [11–15].
Importantly, tolerance and persistence have also been
suggested to have the potential to act as a stepping stone
towards the evolution of genetic resistance, as persisters
provide a viable cell reservoir from which resistant mutants
can emerge by horizontal gene transfer or de novo chromosomal
mutations [16, 17]. This idea was first put forward
more than 30 years ago, based on the frequent cooccurrence
of penicillin tolerance and multiple drug resistance
in clinical isolates [15, 18]. Later, antibiotic tolerance
was indeed shown to be advantageous for the acquisition of
resistance genes by transformation in Streptococcus pneumoniae
[19]. Recent experiments have lent support to these
early findings. Sebastian et al. showed that de novo resistant
These authors contributed equally: Etthel Martha Windels, Joran Elie
Michiels
* Jan Michiels
jan.michiels@kuleuven.vib.be
1
Centre of Microbial and Plant Genetics, KU Leuven,
Leuven, Belgium
2
VIB Center for Microbiology, Flanders Institute for
Biotechnology, Leuven, Belgium
3
imec, Leuven, Belgium
4
Laboratory of Socioecology and Social Evolution, KU Leuven,
Leuven, Belgium
5
Douglas lab, Department of Entomology, Cornell University,
Ithaca, NY, USA
Supplementary information The online version of this article (https://
doi.org/10.1038/s41396-019-0344-9) contains supplementary
material, which is available to authorized users.
1234567890();,:
1234567890();,:
mutants arise with high frequency from the persister
population in Mycobacterium tuberculosis [20], and the
Balaban lab reported that tolerance, provoked by an
extended lag time, increases the opportunity for resistanceconferring
mutations to occur and spread in Escherichia coli
populations under intermittent ampicillin exposure [21].
Furthermore, levels of persistence and resistance were
shown to positively correlate across Pseudomonas strains
[22].
However, the following two lines of reasoning suggest
that the link between persistence and resistance is not
merely based on the increased number of cells available for
mutation, but that persistence could also be directly linked
with altered mutation rates and increased probabilities of
mutating into an antibiotic-resistant genotype. First, both
mutation and persistence have been suggested to be induced
by the activity of different stress responses [23–25]. Hence,
high persister fractions and high mutation rates could
coincide in stressful environments due to a partial overlap in
underlying mechanisms [17]. Second, both high-persistence
and mutator phenotypes are known to thrive in fluctuating
environments and upon sudden environmental changes [11,
26–32], pointing to a possible evolutionary link between
both traits. As yet, however, it is unknown whether persistence
is associated with increased mutation rates and how
this could affect the likelihood for genetic resistance to
evolve.
Our aim was to investigate this outstanding issue.
Therefore, we tested if persister fractions are correlated with
the likelihood to form resistant colonies and if the emergence
of resistant mutants from growing and non-growing
populations is pleiotropically linked with persistence. We
found that persistence is strongly linked with resistance
development, even when the number of surviving cells is
taken into account, revealing a pleiotropic link of persistence
with the probability to acquire resistance-conferring
mutations. Finally, we also developed a mathematical
model to investigate how persistence and altered mutation
rates would affect the emergence of antibiotic resistance in
the setting of a typical clinical infection, where the immune
system is also involved.
Materials and methods
Strains, culture conditions, and antibiotics
E. coli SX43, a derivative of BW25993, is the ancestor of
oppB*, a high-persistence mutant resulting from a recent
evolution experiment [11]. The de-evolved oppB* strain
results from experimental evolution of the oppB* mutant in
antibiotic-free conditions [11]. This strain has five additional
mutations, causing decreased persister fractions as
compared to the wild type strain. The hipA7 strain and its
isogenic wild type used in this work were constructed by
Pearl et al. [33] Strains were grown in Mueller-Hinton broth
(shaking, 200 rpm) and on Mueller-Hinton or Luria-Bertani
(LB) agar at 37 °C. For time-kill curves and persistence
assays, final concentrations of 5 μg/ml ciprofloxacin, 5 μg/
ml norfloxacin, and 400 μg/ml kanamycin were applied.
Mannitol was used at a final concentration of 10 mM.
Minimum inhibitory concentrations (MICs), time-kill
curves, and persistence assays
MICs were determined using the microdilution method. For
time-kill curves of stationary phase populations, an overnight
culture was diluted 1:100 in 100 ml fresh medium.
After 16 h, the initial cell number was determined by plating.
1 ml aliquots were incubated with 5 μg/ml ciprofloxacin
and the number of surviving cells was determined by plating
at specified time points. Persistence assays were performed
by determining survival after 5 h of antibiotic
exposure. Mannitol was added 1 h before initiating antibiotic
treatment.
Resistance evolution on agar plates
Overnight cultures were resuspended in 10 mM MgSO4 and
100 μl was plated on Mueller-Hinton agar containing antibiotics
corresponding to 2x MIC (ciprofloxacin, norfloxacin,
spectinomycin), 3x MIC (chloramphenicol), or 4x
MIC (kanamycin). The number of initially plated cells was
determined by plating on non-selective LB agar. Stable
genetic resistance of the colonies was verified by streaking
on fresh antibiotic-containing plates and MIC-testing (3–5
colonies per strain). All colonies grew after transfer and had
increased MIC values, ranging from 2x to >64x the original
MIC. Stability of ciprofloxacin in agar plates was tracked
daily during long-term 37 °C incubation by streaking
resistant (E. aerogenes ATCC 13048; MIC = 0.064 μg/ml)
and non-resistant strains (SX43 and MG21) on incubated
plates containing an initial concentration of 2x MIC. Even
after 20 days of incubation, non-resistant strains were not
able to grow on the plates, indicating that the effective
concentration remained above the MIC. Time-kill kinetics
of the plated population were determined by excising pieces
of agar without resistant colonies, vigorously vortexing in
10 mM MgSO4, and plating on LB agar.
Resistance evolution in liquid medium
An overnight culture was diluted in fresh medium containing
ciprofloxacin (2x MIC) to an inoculum size of 106
bacteria
and incubated (37 °C, linear shaking) in an automated plate
reader. Optical density (OD600) was measured every 15 min
1240 E. M. Windels et al.
for 8 days. When a cell acquired a resistance-conferring
mutation, its outgrowth could be monitored and the time until
the mutational event was estimated from the resulting growth
curve. MICs of colonies isolated from growth-positive
populations at the end of the experiment were 2 to 16-fold
increased, confirming genetic resistance. For each out of two
independent experiments, 16 or 32 parallel populations were
started from an overnight culture in individual wells, totaling
32 or 64 replicates for each strain.
Sequencing
Resistance-conferring mutations in gyrA were identified by
PCR amplification and Sanger sequencing of the gyrA
QRDR using primers described elsewhere [34].
Rifampicin-based fluctuation assays
From an overnight culture in Mueller-Hinton broth, 10 or
20 parallel cultures were grown from a small inoculum
(5000 CFU/ml) in 96-well plates. After 24 h, the whole
population was plated on LB agar supplemented with
rifampicin (100 μg/ml). Additionally, 4 or 8 cultures per
experiment were grown under the same conditions and used
to estimate the final total cell number by plating on nonselective
LB agar. Colonies on non-selective plates and
rifampicin plates were scored after 24 and 48 hours of
incubation, respectively. Experiments were repeated 2–4
times, yielding 30 or 40 replicate populations per strain per
condition (SX43, n = 30; oppB*, n = 40; de-evolved
oppB*, n = 30; MG21, n = 30; hipA7, n = 30).
Experiments with ECOR strains
20 ECOR strains were randomly selected from the total
collection comprising 72 strains. For time-kill curves, strains
were inoculated in 150 μl Mueller-Hinton broth in a 96-well
plate and grown for 24 h. This culture was diluted 1:100 in
1.5 ml fresh medium in a 24-well plate and grown for 16 h.
Cultures were subsequently diluted 1:100 in 1 ml fresh
medium in 24-well plates and incubated for 2.5 h to obtain
exponential growth. An aliquot was plated to determine the
initial cell number, antibiotics were added, and the surviving
cell number was determined at specified time points by spot
plating. Exponential cultures were used because stationary
phase killing was very limited and showed little variation
between strains. Resistance evolution of ECOR isolates on
ciprofloxacin agar was performed as stated above, except
that 4x MIC instead of 2x MIC was used.
Mathematical model
To simulate bacterial population dynamics during antibiotic
treatment of an infection, we constructed a mathematical
model formulated as a system of differential equations
containing the bacterial subpopulations depicted in Fig. 4a
as well as a population of innate immune effector cells. This
system was solved numerically in Wolfram Mathematica
10.0 using NDSolve[]. For different combinations of the
switching rate and the mutation rate, the time until resistant
cells take over the population (>108
cells) was calculated
using NMinimize[] (default line search method). We chose
realistic parameter values within the range reported for
E. coli [11, 35] and in accordance with our own observations,
rather than specific values for any antibiotic-strain
combination. We verified that qualitatively, the model
predictions did not change when parameter values were
varied within a reasonable range (Fig. S12). A full
description of the model and justification of the chosen
parameter values is given in Supplementary Methods.
Data analysis and statistics
Time-kill curves
Biphasic killing parameters were determined by fitting a
non-linear mixed model to the Log10-transformed, normally
distributed number of surviving cells per milliliter (CFU/ml)
using the R package nlme (https://cran.r-project.org/web/pa
ckages/nlme/index.html). The model was based on the
equation Log10(CFU/ml) = Log10((N0−P0).e−kn.τ
+ P0.e−kp.τ
),
with τ the treatment time (in hours), N0 and P0 the number
of normal and persister cells per ml at τ = 0, and kn and kp
the rate at which normal and persister cells are killed (per
hour) [36]. Strain was coded as a fixed factor and replicate
was coded as a random factor. Best-fit parameters were
compared statistically with Wald F-tests. For all time-kill
curves, biphasic fits were superior over monophasic fits
(Akaike Information Criterion).
Phylogeny of ECOR strains
A maximum likelihood phylogenetic tree of the 20 ECOR
strains was inferred from three concatenated MLST markers
(tonB, dinG, and DPP) identified with Phylomark [37] and
aligned with ClustalW in Bio-Edit [38]. The tree was constructed
using MEGA version 4 [39] and was based on the
Kimura two-parameter model, using complete deletion of
sites with gaps. Escherichia fergusonii was used as an
outgroup to root the resulting tree.
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1241
Persistence and resistance in ECOR strains
Estimated persister fractions from biphasic fittings were
used as a quantitative parameter for persistence. The average
number of ciprofloxacin-resistant mutants per plated
cell per day, from day 3 onwards, was used as a parameter
reflecting resistance evolution. Both parameters were Log10transformed
in order to obtain normally distributed data.
Based on the obtained phylogenetic tree, a phylogenetic
generalized least squares (PGLS) regression assuming a
Brownian evolution model was performed using the R
packages ape (https://cran.r-project.org/web/packages/ape/
index.html) and nlme (https://cran.r-project.org/web/packa
ges/nlme/index.html). Outliers with a Cook’s distance
greater than 1 were omitted from the analysis.
Resistance evolution on agar plates
A linear model was fitted to the number of new antibioticresistant
colonies per plated cell (Log10-transformed to
obtain normally distributed data) as a function of time.
Significant differences between the model parameters were
assessed with unpaired, two-sided t-tests. S.e.m.’s were
calculated on the Log10-transformed data and transformed
back to the original scale for representation in Fig. 2b and
Fig. S4.
Resistance evolution in liquid medium
A cubic smoothing spline function was fitted to the OD600values.
The smoothing parameter was chosen by crossvalidation.
The time needed for a resistant mutant to emerge
was estimated empirically as the time at which the first
derivative of the fitted growth curve exceeded 0.1. A nonparametric
Cox proportional hazards model was fitted to the
right-censored distribution of lag times for each strain (n =
64 for SX43, oppB*, and de-evolved oppB*; n = 32 for
MG21 and hipA7). Mutant and wild type lag times were
compared by Wald tests on coefficients of the fitted model.
Determination of mutation rates
Data from rifampicin-based fluctuation assays and day 2
counts of ciprofloxacin plate assays were analyzed by
fluctuation analysis with the R package ‘flan’ version 0.4,
using a maximum likelihood function that takes into
account final cell counts [40–42]. For experiments with
ciprofloxacin, only 100 μl of a 5 ml culture was plated. To
avoid underestimating the number of mutations in the total
population, we used the correction factor proposed by
Stewart [43]. Mutations in persister cells surviving exposure
to lethal antibiotic stress were calculated as the number of
resistant colonies divided by the viable cell count two days
prior (mutations per cell per day), and the mutation rate was
averaged over day 3-10. Viable cell counts were estimated
from the fitted time-kill curves (Fig. S8). Statistical comparison
was done with unpaired, one-sided t-tests with
Welch’s correction in the case of unequal variances (F-test).
Confidence intervals for relative differences between
mutants and wild type were determined according to Fieller’s
theorem.
Results
Persistence and the evolution of resistance are
positively correlated in natural isolates and lab
strains of E. coli
As a first probe for the link between persistence and the
likelihood to evolve genetic resistance, we focused on a set of
20 natural isolates randomly picked from the ECOR collection,
the standard panel of environmental diversity in E. coli
[44]. The ECOR collection is ideal for this, as it displays
substantial differences in persister fractions, whereas MIC
variation is limited [45]. First, MICs were measured to confirm
that all isolates are sensitive towards ciprofloxacin (all
MICs ≤ 0.016 μg/ml, Table S1). Next, we determined two
quantitative parameters: one reflecting persistence and one
reflecting the likelihood for resistance to evolve. Persister
fractions were derived from a biphasic exponential fit to timekill
data of cultures challenged with a high dose of ciprofloxacin
(Fig. S1a, Table S2). Persistence is highly variable
among strains (p < 0.0001, likelihood ratio test) and shows no
correlation with MICs (Pearson’s r = 0.148, p = 0.535), corroborating
previous findings [45]. To determine the rate of
resistance evolution, we monitored the emergence of resistant
colonies on ciprofloxacin-supplemented Mueller-Hinton agar
(4x MIC) over a course of 20 days (Fig. S1b). Resistant
mutants that were formed in the preculture (pre-exposure)
were excluded by ignoring colonies that appeared during the
first two days of incubation, as described previously [46–48].
That way, we focused only on mutants that arose de novo
after antibiotic exposure, and excluded variation caused by the
presence of mutants that originated during growth in the
preculture. The average number of mutants per plated cell
per day was subsequently used as a quantitative parameter for
the evolution of resistance. In order to investigate the relationship
between persistence and resistance evolution while
accounting for phylogenetic non-independence among strains,
we performed a phylogenetic generalized least squares
(PGLS) regression (Fig. 1). Calculation of Cook’s distances
indicated that four observations deviate strongly from the
other data and severely influence the regression analysis
(Fig. S2). The coefficient of determination (R²
) for the 16
remaining isolates is 0.498, meaning that the variation in
1242 E. M. Windels et al.
persister fractions explains 49.8 % of the variation in resistance
evolution (p = 0.002). This result indicates that the
persister fraction of a strain is a factor that strongly determines
its tendency to evolve resistance.
Next, we investigated if the same trend could be
observed in well-characterized persistence mutants in E.
coli lab strains. We selected three previously characterized
persistence mutants: the hipA7 and oppB* high-persistence
mutants, and the de-evolved oppB* mutant which shows a
decreased persister fraction due to evolution in prolonged
drug-free conditions. oppB* carries a single amino acid
substitution in OppB, one of five subunits of the Opp oligopeptide
permease system, which confers high persistence
through a currently unknown mechanism [11]. In addition
to the mutation in OppB, the de-evolved oppB* strain has
five mutations, but it is currently unknown how these are
responsible for the decreased persister fraction. hipA7 was
the first recognized high-persistence allele and encodes the
HipA toxin harboring two amino acid substitutions, causing
persistence by inactivation of glutamyl-tRNA synthetase
and subsequent induction of the stringent response [49, 50].
All strains are equally sensitive towards ciprofloxacin (MIC
= 0.008 μg/ml, Table S4) and stationary phase time-kill
curves of these mutants and their isogenic wild types show
the expected persistence phenotypes (Fig. 2a, Table S3). As
a first step, we wanted to thoroughly characterize resistance
evolution in these strains.
As for the natural E. coli isolates, we quantified resistance
evolution by following the emergence of resistant
colonies on ciprofloxacin-containing Mueller-Hinton agar
(2x MIC) over time. In agreement with our hypothesis, we
found that the number of resistant colonies per plated cell is
increased in hipA7 and oppB*, while it is decreased in deevolved
oppB* (Fig. 2b). We then investigated whether
resistance evolution in high-persistence mutants is still
increased when the number of persisters is decreased
through the addition of mannitol, which is known to sensitize
persisters to antibiotics [51]. Mannitol indeed suppresses
the high-persistence phenotype of oppB*, while
hipA7 still forms more persisters than its wild type under
this condition (Fig. S3). Correspondingly, resistance evolution
on mannitol-supplemented agar reverts to wild type
levels in the case of oppB* but not hipA7 (Fig. S4).
In Gram-negative bacteria, the DNA gyrase subunit GyrA
constitutes the main target for ciprofloxacin and other quinolones
[52]. Mutations in a particular N-terminal region, the
“quinolone resistance-determining region” (QRDR), are the
most common mechanism of resistance, both in clinical isolates
and in vitro evolution experiments [52–54]. To determine
if mutations in GyrA are also involved in the evolution
of antibiotic resistance in our experiments, we sequenced the
gyrA QRDR in 10–11 ciprofloxacin-resistant colonies per
strain and identified mutations causing amino acid changes at
residues 81, 82, 83, and 87 in 35 out of 64 colonies (Fig. 2d).
The remaining colonies bear no mutations in the gyrA QRDR,
but presumably have mutations in another gyrA region or in
parC, encoding the topoisomerase IV subunit A [52]. The
amino acid substitutions that we found are commonly
observed in clinical isolates of quinolone-resistant E. coli as
well as other Enterobacteriaceae and Gram-positive organisms
[52, 55, 56].
To further confirm our finding of persistence being
strongly linked to the likelihood of evolving antibiotic
resistance, we also quantified antibiotic resistance development
using an independent approach. We inoculated a
population of 106
bacteria in ciprofloxacin-containing
Mueller-Hinton broth (2x MIC) and measured the time
until the emergence of resistance by tracking growth. As
expected, the number of populations that had evolved
resistance after eight days was strongly linked with the
persistence phenotypes: 29/32 for hipA7 (14/32 for the wild
type), 48/64 for oppB* and 15/64 for de-evolved oppB*
(37/64 for the wild type) (Fig. 2c). Hence, the rate of
resistance evolution is increased in hipA7 and oppB*, and
decreased in de-evolved oppB*.
To assess the generality of our observations, we further
tracked the emergence of resistant colonies on MuellerHinton
agar supplemented with other antibiotics: norfloxacin,
kanamycin, chloramphenicol, and spectinomycin.
MICs of these antibiotics are given in Table S4. Norfloxacin
and kanamycin are bactericidal drugs, and we confirmed the
high-persistence phenotypes of hipA7 and oppB* upon
Fig. 1 Across natural E. coli isolates, persister fractions are strongly
predictive of the likelihood to evolve genetic antibiotic resistance.
Persister fractions (n = 3 per strain) and the average number of
resistant mutants per plated cell per day (n = 5 per strain) on
ciprofloxacin-containing Mueller-Hinton agar (4x MIC) are plotted for
16 ECOR isolates on a log-log scale. To account for the phylogenetic
non-independence among strains, PGLS regression was used to fit a
linear model onto the log-log transformed data (R²
= 0.498, p = 0.002)
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1243
treatment with these antibiotics (Fig. S5). The persister
fraction of de-evolved oppB* is only very slightly decreased
as compared to the wild type. Chloramphenicol and spectinomycin
are bacteriostatic antibiotics, which inhibit E. coli
growth but do not induce killing. The positive correlation
between persister fractions and resistance development
holds true for each antibiotic, except for spectinomycin.
Only very few resistant colonies appeared on
spectinomycin-supplemented plates, resulting in large
variations and no significant differences for oppB* and deevolved
oppB* as compared to their wild type (Fig. S6).
A link between persistence and mutation rates
The observed correlation between persistence and resistance
development could be due to an increased viable cell
reservoir, a pleiotropic effect of persistence on mutation
rates, or a combination of both. To investigate the
Fig. 2 Ciprofloxacin resistance development is increased in highpersistence
mutants and decreased in a low-persistence mutant. a
Ciprofloxacin time-kill curves of stationary phase cultures show
increased persister fractions in hipA7 and oppB*, and decreased persistence
in de-evolved oppB*. Individual data points from three
independent replicates are shown. Lines represent a biphasic exponential
fit to the data. Parameter estimates for persister fractions were
compared with Wald F-tests. b Cumulative number of resistant
mutants per plated cell on Mueller-Hinton agar supplemented with
ciprofloxacin (2x MIC). Data points represent averages ± s.e.m. (n =
10). Statistical comparison was done by two-sided t-tests on coefficients
of a log-linear model. c Cumulative proportion of wells with
ciprofloxacin-containing Mueller-Hinton broth (2x MIC) in which
resistant mutants emerged (n = 32-64). Statistical comparison was
done by Wald tests on coefficients of a Cox proportional hazards
model, to take into account censoring. d Amino acid changes caused
by gyrA mutations detected in ciprofloxacin-resistant colonies. *p ≤
0.05; **p ≤ 0.01; ***p ≤ 0.001, and ns non-significant
1244 E. M. Windels et al.
contribution of mutagenesis, we compared mutation rates in
the well-defined high and low-persistence mutants. Resistant
mutants can be generated from growing bacteria as well
as from non-growing but surviving bacteria exposed to
growth-restricted conditions [47]. Based on earlier studies,
resistant colonies that emerge on antibiotic-containing agar
during the first two days of incubation can be traced back to
resistant mutants that emerged during growth in the preculture
[46, 47]. Thus, from the number of colonies on day
2, the mutation rate during growth (mutations per cell per
division) was estimated by fluctuation analysis. The results
from these experiments demonstrate that the mutation rate is
increased in hipA7 (2.8-fold) and oppB* (5.0-fold), while it
is decreased in de-evolved oppB* (2.0-fold) (Fig. 3a,
Table S5). We further confirmed these results using a
rifampicin-based fluctuation assay to determine the mutation
rate at another, independent genomic locus (Fig. S7,
Table S5).
Apart from the resistant colonies that appear within two
days of incubation and acquired a mutation during growth
in the preculture, mutants are also formed de novo from the
surviving subpopulation on the ciprofloxacin-containing
agar plate. To quantify the mutation rate of cells from the
surviving reservoir of non-growing persisters, the number
of resistant colonies has to be normalized for the remaining
number of viable cells on the plate. To estimate the size of
the viable cell reservoir, we measured time-kill kinetics of
the plated populations by determining the number of surviving
cells at different time points after plating (Fig. S8). In
agreement with earlier reports [46–48], a biphasic killing
pattern was observed for all strains except hipA7. The bulk
of the population (>99.9%) is killed within the first day after
plating, while a small subpopulation of cells is killed at a
much slower rate. When we correct the number of resistant
colonies for the number of surviving cells in the persister
reservoir, we still observe a correlation between persistence
and resistance. The average mutation rate per viable cell
per day is significantly increased in hipA7 and oppB*, and
significantly lower in de-evolved oppB*, compared to their
respective wild types (Fig. 3b). Together, these experiments
reveal that persistence correlates with mutation rates under
various growth conditions and in this way has a strong
effect on the likelihood to evolve genetic resistance.
Increased survival and mutation rates jointly drive
evolution of clinical resistance
To assess the implications of our findings under clinical
conditions, we developed a mathematical model to simulate
antibiotic treatment of an in vivo bacterial infection. In this
model, bacteria continuously switch between the normal
and persister states following switching rates a and b.
Growth and killing of non-persisters is related to the
antibiotic concentration according to a pharmacodynamic
Hill function [35]. Persisters are assumed to be non-dividing
and recalcitrant to killing by antibiotics. Both normal and
persister cells can be killed by the action of an innate
immune response, although the killing rate is taken to be
lower for persister cells [57, 58]. As the immune system
alone is not sufficient to control this infection within a
reasonable amount of time, bacterial clearance is highly
dependent on the action of the antibiotic [59]. The antibiotic
is added every 12 h at a peak concentration of 25x MIC
(Amax) and the effective concentration declines exponentially
with a half-life of 2.5 h (Fig. S9). As such, the model
incorporates realistic clinical pharmacokinetics and pharmacodynamics
for a broad range of antibiotics (see Supplementary
Methods and Table S6 for details).
Our model focuses on resistance evolution via chromosomal
mutations. In this case, clinical resistance is mostly
achieved by multiple mutations accumulated in a stepwise
fashion [60]. Therefore, we implemented resistance evolution
as a gradual process, where subsequent mutations
generate mutants showing 4, 20, and 50-fold increased MIC
values. Replicating cells develop resistance according to the
mutation rate µgrowing, which can be increased up to twofold
by the presence of sublethal concentrations of antibiotics.
Non-dividing persisters acquire resistance-conferring
mutations at a constant rate µpersisters. A schematic diagram
showing the possible interconversions between all different
phenotypes and genotypes is shown in Fig. 4a.
Fig. 4b depicts bacterial population dynamics for
pathogens exhibiting different persister fractions, with and
without an associated increase in mutation rates (for separate
dynamics of all different subpopulations, see Fig. S10).
In the baseline situation, the infection is cleared within
5 days of antibiotic treatment. A small subpopulation of
resistant cells emerges, but it cannot acquire the multiple
subsequent mutations necessary to obtain high-level resistance,
and therefore it does not impair treatment success.
When we increase switching to the persister state 10-fold,
the infection is still cured, although bacterial clearance is
delayed. Similarly, increasing the mutation rates by a factor
in line with our experimental findings, i.e., twofold for
µgrowing and fourfold for µpersisters, has no effect on the
infection dynamics. However, when we concomitantly
increase persistence and mutation rates, antibiotic therapy
fails as highly resistant mutants can now take over the
population. Only upon a higher, 50-fold increase in persister
switching, high-level resistance can already emerge with
baseline mutation rates, and even then, an associated
increase in mutation rate causes high-level resistant mutants
to take over the population sooner. These examples clearly
illustrate that increased persister fractions and mutation
rates jointly drive the evolution of genetic antibiotic resistance.
To investigate this effect over a larger range of
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1245
parameter values, we calculated the time it takes for
the resistant population to reach a density of 108
cells,
given different combinations of persister fractions and
mutation rates. The resulting heat map confirms that both
parameters simultaneously drive the evolution of resistance
(Fig. 4c).
To verify that a mutant with a pleiotropically linked
persister fraction and mutation rate is actually favored over
Fig. 4 A mathematical simulation model shows that increased persistence
and mutation rates jointly drive the evolution of resistance
under clinical conditions. a Schematic diagram of the mathematical
model showing all possible transitions between the different genotypes
and phenotypic states. b Population dynamics for wild type and
resistant cells during antibiotic treatment for increasing persister
fractions with or without associated increased mutation rates. The
separate dynamics of all different subpopulations under the same sets
of parameters are given in Fig. S10. c Heat map of the time at which
the resistant cells take over the population (>108
resistant cells), as a
function of the mutation rate and the persister fraction (switching rate).
The mutation rate is shown for normal cells, the mutation rate for
persisters is 104
-fold higher. See text and Supplementary Methods for
details on the mathematical model
Fig. 3 Persistence is correlated
with elevated mutation rates. a
Mutation rate (mutations per cell
per division) of growing cells
determined from the number of
ciprofloxacin-resistant mutants
after two days using fluctuation
analysis, relative to the wild
type. Error bars show 95% c.i.
(n = 10), p-values are from onesided
t-tests. b Mutations in nongrowing
cells surviving lethal
antibiotic exposure (mutations
per cell per day), relative to the
wild type. Error bars show 95%
c.i.’s (n = 10), p-values are from
one-sided t-tests. *p ≤ 0.1; **p ≤
0.05; ***p ≤ 0.01
1246 E. M. Windels et al.
the wild type, we calculated the invasion potential of rare
mutants with different combinations of switching rates and
mutation rates in a wild type population. The results of this
simulation suggest that the selective advantage of a mutant
is proportional to its persister fraction as well as its mutation
rate. Moreover, mutants with high values for both parameters
exhibit the strongest invasion potential, pointing
towards a possible evolutionary link between persistence
and mutation rates (Fig. S11).
Discussion
To date, many lines of evidence strongly suggest that
persisters contribute to treatment failure of bacterial infections
[6]. In this paper, we show that the clinical burden of
persisters probably extends beyond their direct role in
chronic and recurrent infections because they strongly
contribute to the evolution of genetic resistance. By tracking
the appearance of resistant colonies on ciprofloxacincontaining
agar, we found that persistence is a crucial factor
driving the acquisition of genetic resistance by chromosomal
mutations. Targeted sequencing revealed that the
responsible mutations are similar to those encountered in
fluoroquinolone-resistant clinical isolates in a number of
bacterial species [52, 55, 56]. The effect of persistence was
also not only observed with ciprofloxacin, but is valid for
various antibiotics with different modes of action. Moreover,
the quantitative relationship between persister fractions
and ciprofloxacin resistance evolution in 80% of the
tested natural isolates of E. coli suggests that the link is
intrinsic and widespread.
The obvious mechanism by which persistence can
facilitate resistance development is by providing more
viable cells eligible for mutation. The number of surviving
cells will then influence the likelihood that resistance
emerges. Indeed, Levin and Rozen illustrated mathematically
that increasing the number of persisters augments the
likelihood of evolving resistance [16]. Recent experimental
work has confirmed that resistant mutants can be generated
de novo from the persister population during antibiotic
treatment of M. tuberculosis [20]. Levin-Reisman et al.
showed that mutants exhibiting increased levels of ampicillin
tolerance or persistence evolved resistance earlier than
the wild type [21].
In addition to this straightforward mechanism, we
hypothesized that persistence and resistance could also be
linked in a less obvious way. Indeed, two different lines of
reasoning suggest that persistence correlates with mutation
rates. First, the involvement of stress responses in persistence
points to a link with mutation rates of cells growing in
stressful environments [17]. According to a recent study,
this effect could be especially important when the stress
response is activated by DNA damage, because erroneous
repair of DNA damage is a major source of mutations [61].
A second argument stems from evolutionary theory.
Mathematical models indicate that persistence may have
evolved as a bet-hedging strategy in environments characterized
by fluctuating episodes of stress [26, 62–64].
Recent experimental evidence lends strong support to these
theoretic considerations, showing that persistence is disadvantageous
in constant environments but favored during
periodic stress [11, 13, 14]. Mathematical models have also
been used to show that high mutation rates are favored
when the environment shows periodic changes [27, 28].
However, models describing mutagenesis are mathematically
equivalent to models of phenotypic switching, as they
consider a single fitness-determining selective locus with
only two possible allelic states that can mutate back and
forth. Nevertheless, both theory and experiments have
shown the adaptive benefit of increased mutation rates to
cope with stress or a sudden environmental change. Maladapted
genotypes are predicted to experience selection for
increased mutation rates in both constant and fluctuating
environments [32]. Evolution experiments within laboratory
[29] or host environments [31] confirmed these theoretical
predictions by showing selection for increased mutation
rates in conditions of near-lethal stress and colonization of a
new host. Moreover, it was shown that antibiotic treatment
can select mutator alleles, especially during chronic infections
that are repeatedly challenged with antibiotics [30].
We tested this putative association between persistence
and mutation by directly measuring mutation rates in highand
low-persistence mutants. Using fluctuation assays with
ciprofloxacin and rifampicin selection, we found that persistence
is pleiotropically linked with mutation rates in
growing cells. Moreover, the number of resistant colonies
emerging from the viable cell reservoir shows the same
correlation, even when a correction for the number of surviving
cells is applied. Thus, persister cells formed by highpersistence
strains have an increased likelihood to acquire
resistance-conferring mutations, while persisters from lowpersistence
strains are less likely to mutate and hence to
acquire resistance. The positive association between persistence
and mutation rates is important, since even modest
increases in the mutational supply can dramatically improve
evolvability [65–67]. Interestingly, these findings also
imply that the clinical relevance of persistence is not limited
to infections treated with bactericidal antibiotics. Highpersistence
strains, which have repeatedly been identified in
chronic infections [68, 69], can even complicate treatment
in the absence of bactericidal drugs because they are more
likely to evolve resistance due to their increased mutation
rates.
The continuous emergence of resistant colonies on
antibiotic-containing agar plates over the course of several
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1247
days or weeks is usually explained by de novo mutation in
non-growing, surviving cells [46–48]. Indeed, spontaneous
or antibiotic-induced DNA damage in non-growing cells
may lead to mutations because, even in the absence of
chromosomal replication, these cells still display DNA
turnover [70, 71]. On the other hand, others have proposed
that these and similar observations do not involve mutations
in the absence of growth, but can instead be explained by
pre-existing small-effect mutants that grow slowly and give
rise to faster-growing mutants that subsequently form
colonies [72]. These two contrasting views are subject of
considerable debate [71–73]. Cirz et al. have shown that
deletion or mutation of several genes can dramatically
decrease late colony formation on ciprofloxacin-containing
agar plates, without affecting the number of early colonies,
fitness under normal or growth-restricted conditions, or the
selective advantage of resistance-conferring mutations [46].
These results indicate that de novo mutation in non-growing
cells is the most likely explanation for the gradual accumulation
of colonies on ciprofloxacin plates.
Taken together, our results lead us to propose a model in
which two factors constitute the link between persistence
and resistance. When a population is treated with bactericidal
antibiotics (Fig. 5, red frame, non-growing conditions),
only persisters will survive. Increased persistence
leads to more cells in the viable cell reservoir and an
increased likelihood for the occurrence of resistanceconferring
mutations. Moreover, the mutation rate of persisters
is increased in high-persistence strains. When antibiotics
are absent or present at low concentrations (Fig. 5,
green frame, growing conditions), the mutation rate of
growing cells is correlated with persistence. Elevated
mutation rates in turn increase the likelihood for the emergence
of resistance.
To explore the implications of our findings in a clinical
context, we developed a mathematical model simulating
bacterial infection dynamics during antibiotic treatment.
Our model significantly extends on previous modeling
efforts by incorporating step-wise resistance evolution and a
flexible mutation rate under antibiotic stress, leading to a
model that is closer to a realistic clinical situation. Under
conditions that mimic those encountered in the clinic and
using parameter values in range for E. coli and in line with
our findings, the model predicts that the impact of increased
persistence on the evolution of resistance is considerably
greater when increased persister fractions are correlated
with increased mutation rates. As our simulations also
demonstrate, these mutants with pleiotropically linked high
persistence and mutation rates are likely to emerge in
fluctuating conditions.
Current efforts to tackle the problem of antibiotic resistance
are insufficient, as antibiotic-resistant pathogens pose
a growing public health problem [2, 74]. As the pipeline of
new antibiotics active against these pathogens is very thin,
inhibiting the evolution of resistance could be an important
part of the solution. Indeed, directly intervening with evolution
is regarded as a promising therapeutic strategy in
bacterial infections and beyond [75–77]. Antibiotic resistance
evolution is a complex process, influenced by many
factors including stress responses, mismatch repair, iron
homeostasis, and translation fidelity [25, 65]. Proposed
strategies to inhibit antibiotic resistance evolution include
combination therapy [78], drug cycling protocols [79],
interference with the SOS response [46], and smallFig.
5 Proposed model explaining how persistence accelerates the
evolution of genetic resistance. Persistence refers to the fraction of
cells in a sensitive population that survives otherwise lethal antibiotic
exposure. High persistence accelerates the evolution of antibiotic
resistance by increasing the number of viable cells available for
mutation, thus increasing the likelihood for a resistance-conferring
mutation to arise. Additionally, persistence is linked with higher
mutation rates in growing cells and cells that survive lethal antibiotic
exposure but cannot grow (persisters). Together, these two factors
have the potential to accelerate the evolution of antibiotic resistance in
high-persistence strains in different environments with or without
antibiotics. Red depicts environments where the antibiotic concentration
is above the MIC, while green depicts environments with subinhibitory
or zero antibiotic concentrations
1248 E. M. Windels et al.
molecule RecA inhibitors [48]. Our results indicate that
bacterial persistence also plays a pivotal role in the development
of genetic antibiotic resistance and could represent a
novel and innovative way to combat it. Interestingly, it was
recently shown that the molecular mechanisms underlying
drug-tolerant persisters in cancer could be a novel and
promising target to prevent acquired therapy resistance [80,
81]. In conclusion, we propose that anti-persister strategies
may be an important part of the solution to the ongoing
antibiotic crisis, not only by shortening treatment durations
but also by slowing down resistance development.
Acknowledgements We thank Chris Michiels (KU Leuven) for providing
ECOR strains, Abram Aertsen (KU Leuven) for sharing the
hipA7 strain, and Bernard Ycart and Adrien Mazoyer (Grenoble Alpes
University) for their assistance with fluctuation analysis. EMW and
BVdB are Research Foundation—Flanders (FWO)-fellows and JEM is
a fellow of the Agency for Innovation by Science and Technology
(IWT). This research was funded by the KU Leuven Research Council
(PF/10/010; C16/17/006), the Belgian Science Policy Office (IAP P7/
28), FWO (G047112N; G0B2515N; G055517N), and the Flemish
Institute for Biotechnology (VIB).
Compliance with ethical standards
Conflict of interest The authors declare that they have no conflict of
interest.
Publisher’s note: Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
References
1. Davies J, Davies D. Origins and evolution of antibiotic resistance.
Microbiol Mol Biol Rev. 2010;74:417–33.
2. World Health Organization. Antimicrobial resistance: global
report on surveillance. WHO press, Geneva, Switzerland; 2014.
3. Brauner A, Fridman O, Gefen O, Balaban NQ. Distinguishing
between resistance, tolerance and persistence to antibiotic treatment.
Nat Rev Microbiol. 2016;14:320–30.
4. Bumann D. Heterogeneous host-pathogen encounters: act locally,
think globally. Cell Host Microbe. 2015;17:13–19.
5. Conlon BP. Staphylococcus aureus chronic and relapsing infections:
evidence of a role for persister cells. Bioessays.
2014;36:991–6.
6. Michiels JE, Van den Bergh B, Verstraeten N, Michiels J.
Molecular mechanisms and clinical implications of bacterial persistence.
Drug Resist Updat. 2016;29:76–89.
7. Van den Bergh B, Michiels JE, Fauvart M, Michiels J. Should we
develop screens for multidrug tolerance? Expert Rev Anti Infect
Ther. 2016;14:613–6.
8. Van den Bergh B, Fauvart M, Michiels J. Formation, physiology,
ecology, evolution and clinical importance of bacterial persisters.
FEMS Microbiol Rev. 2017;41:219–51.
9. Harms A, Maisonneuve E, Gerdes K. Mechanisms of bacterial
persistence during stress and antibiotic exposure. Science.
2016;354:aaf4268.
10. Fisher RA, Gollan B, Helaine S. Persistent bacterial infections and
persister cells. Nat Rev Microbiol. 2017;15:453–64.
11. Van den Bergh B, Michiels JE, Wenseleers T, Windels EM,
Vanden Boer P, Kestemont D, et al. Frequency of antibiotic
application drives rapid evolutionary adaptation of Escherichia
coli persistence. Nat Microbiol. 2016;1:16020.
12. Michiels JE, Van den Bergh B, Verstraeten N, Fauvart M,
Michiels J. In vitro emergence of high persistence upon periodic
aminoglycoside challenge in the ESKAPE pathogens. Antimicrob
Agents Chemother. 2016;60:4630–7.
13. Fridman O, Goldberg A, Ronin I, Shoresh N, Balaban NQ.
Optimization of lag time underlies antibiotic tolerance in evolved
bacterial populations. Nature. 2014;513:418–21.
14. Mechler L, Herbig A, Paprotka K, Fraunholz M, Nieselt K, Bertram
R. A novel point mutation promotes growth phase-dependent
daptomycin tolerance in Staphylococcus aureus. Antimicrob
Agents Chemother. 2015;59:5366–76.
15. Moreillon P, Tomasz A. Penicillin resistance and defective lysis in
clinical isolates of pneumococci: evidence for two kinds of antibiotic
pressure operating in the clinical environment. J Infect Dis.
1988;157:1150–7.
16. Levin BR, Rozen DE. Non-inherited antibiotic resistance. Nat Rev
Microbiol. 2006;4:556–62.
17. Cohen NR, Lobritz MA, Collins JJ. Microbial persistence and the
road to drug resistance. Cell Host Microbe. 2013;13:632–42.
18. Liu HH, Tomasz A. Penicillin tolerance in multiply drug-resistant
natural isolates of Streptococcus pneumoniae. J Infect Dis.
1985;152:365–72.
19. Novak R, Henriques B, Charpentier E, Normark S, Tuomanen E.
Emergence of vancomycin tolerance in Streptococcus pneumoniae.
Nature. 1999;399:590–3.
20. Sebastian J, Swaminath S, Nair RR, Jakkala K, Pradhan A,
Ajitkumar P. De novo emergence of genetically resistant mutants
of Mycobacterium tuberculosis from the persistence phase cells
formed against antituberculosis drugs in vitro. Antimicrob Agents
Chemother. 2016;61:AAC.01343-16.
21. Levin-Reisman I, Ronin I, Gefen O, Braniss I, Shoresh N, Balaban
NQ. Antibiotic tolerance facilitates the evolution of resistance.
Science. 2017;355:826–30.
22. Vogwill T, Comfort AC, Furió V, MacLean RC. Persistence and
resistance as complementary bacterial adaptations to antibiotics. J
Evol Biol. 2016;29:1223–33.
23. Bjedov I, Tenaillon O, Gérard B, Souza V, Denamur E, Radman
M, et al. Stress-induced mutagenesis in bacteria. Science.
2003;300:1404–9.
24. Al Mamun, AAM Lombardo M, Shee C, Lisewski AM, Gonzalez
C, et al. Identity and function of a large gene network underlying
mutagenic repair of DNA breaks. Science. 2012;338:1344–8.
25. Poole K. Stress responses as determinants of antimicrobial resistance
in Gram-negative bacteria. Trends Microbiol. 2012;20:227–34.
26. Kussell E, Kishony R, Balaban NQ, Leibler S. Bacterial persistence:
a model of survival in changing environments. Genetics.
2005;169:1807–14.
27. Carja O, Liberman U, Feldman MW. Evolution in changing
environments: Modifiers of mutation, recombination, and migration.
Proc Natl Acad Sci. 2014;111:17935–40.
28. Ishii K, Matsuda H, Iwasa Y, Sasaki A. Evolutionarily stable
mutation rate in a periodically changing environment. Genetics.
1989;121:163–74.
29. Swings T, Van den Bergh B, Wuyts S, Oeyen E, Voordeckers K,
Verstrepen KJ, et al. Adaptive tuning of mutation rates allows
fast response to lethal stress in Escherichia coli. eLife. 2017;
6:1–24.
30. Giraud A, Matic I, Radman M, Fons M, Taddei F. Mutator bacteria
as a risk factor in treatment of infectious diseases. Antimicrob
Agents Chemother. 2002;46:863–5.
31. Giraud A, Matic I, Tenaillon O, Clara A, Radman M, Fons M,
et al. Costs and benefits of high mutation rates: adaptive evolution
of bacteria in the mouse gut. Science. 2001;291:2606–8.
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1249
32. Ram Y, Hadany L. The evolution of stress-induced hypermutation
in asexual populations. Evolution. 2012;66:2315–28.
33. Pearl S, Gabay C, Kishony R, Oppenheim A, Balaban NQ.
Nongenetic individuality in the host-phage interaction. PLoS Biol
2008;6:0957–64.
34. Wang H, Dzink-Fox J, Chen M, Levy SB. Genetic characterization
of highly fluoroquinolone-resistant clinical Escherichia coli
strains from China: role of acrR mutations. Antimicrob Agents
Chemother. 2001;45:1515–21.
35. Regoes RR, Wiuff C, Zappala RM, Garner KN, Baquero F, Levin
BR. Pharmacodynamic functions: a multiparameter approach to
the design of antibiotic treatment regimens. Antimicrob Agents
Chemother. 2004;48:3670–6.
36. Stepanyan K, Wenseleers T, Duéñez-Guzmán EA, Muratori F,
Van den Bergh B, Verstraeten N, et al. Fitness trade-offs explain
low levels of persister cells in the opportunistic pathogen Pseudomonas
aeruginosa. Mol Ecol. 2015;24:1572–83.
37. Sahl JW, Matalka MN, Rasko DA. Phylomark, a tool to identify
conserved phylogenetic markers from whole-genome alignments.
Appl Environ Microbiol. 2012;78:4884–92.
38. Hall TA. BioEdit: a user-friendly biological sequence alignment
editor and analysis program for Windows 95/98/NT. Nucleic
Acids Symp Ser. 1999;41:95–98.
39. Tamura K, Dudley J, Nei M, Kumar S. MEGA4: Molecular
Evolutionary Genetics Analysis (MEGA) software version 4.0.
Mol Biol Evol. 2007;24:1596–9.
40. Hamon A, Ycart B. Statistics for the Luria-Delbrück distribution.
Electron J Stat. 2012;6:1251–72.
41. Ycart B, Veziris N. Unbiased estimation of mutation rates under
fluctuating final counts. PLoS ONE. 2014;9:e101434.
42. Mazoyer A, Drouilhet R, Despréaux S, Ycart B. flan: An R
package for inference on mutation models. R J. 2017;9:334–51.
43. Stewart FM. Fluctuation analysis: the effect of plating efficiency.
Genetica. 1991;84:51–55.
44. Ochman H, Selander RK. Standard reference strains of Escherichia
coli from natural populations. J Bacteriol. 1984;157:690–3.
45. Stewart B, Rozen DE. Genetic variation for antibiotic persistence
in Escherichia coli. Evolution. 2011;66:933–9.
46. Cirz RT, Chin JK, Andes DR, de Crécy-Lagard V, Craig WA,
Romesberg FE. Inhibition of mutation and combating the evolution
of antibiotic resistance. PLoS Biol. 2005;3:1024–33.
47. Riesenfeld C, Everett M, Piddock LJ, Hall BG. Adaptive mutations
produce resistance to ciprofloxacin. Antimicrob Agents
Chemother. 1997;41:2059–60.
48. Alam MK, Alhhazmi A, DeCoteau JF, Luo Y, Geyer CR. RecA
inhibitors potentiate antibiotic activity and block evolution of
antibiotic resistance. Cell Chem Biol. 2016;23:381–91.
49. Kaspy I, Rotem E, Weiss N, Ronin I, Balaban NQ, Glaser G.
HipA-mediated antibiotic persistence via phosphorylation of the
glutamyl-tRNA-synthetase. Nat Commun. 2013;4:3001.
50. Germain E, Castro-Roa D, Zenkin N, Gerdes K. Molecular
mechanism of bacterial persistence by HipA. Mol Cell.
2013;52:248–54.
51. Allison KR, Brynildsen MP, Collins JJ. Metabolite-enabled eradication
of bacterial persisters by aminoglycosides. Nature.
2011;473:216–20.
52. Piddock LJ. Mechanisms of fluoroquinolone resistance: an update
1994-1998. Drugs. 1999;58:11–18.
53. Oz T, Guvenek A, Yildiz S, Karaboga E, Tamer YT, Mumcuyan
N, et al. Strength of selection pressure is an important parameter
contributing to the complexity of antibiotic resistance evolution.
Mol Biol Evol. 2014;31:2387–401.
54. Lázár V, Nagy I, Spohn R, Csörgő B, Györkei Á, Nyerges Á,
et al. Genome-wide analysis captures the determinants of the
antibiotic cross-resistance interaction network. Nat Commun.
2014;5:4352.
55. Conrad S, Oethinger M, Kaifel K, Klotz G, Marre R, Kern WV.
gyrA mutations in high-level fluoroquinolone-resistant clinical isolates
of Escherichia coli. J Antimicrob Chemother. 1996;38:443–55.
56. Weigel LM, Steward CD, Tenover FC. gyrA mutations associated
with fluoroquinolone resistance in eight species of Enterobacteriaceae.
Antimicrob Agents Chemother. 1998;42:2661–7.
57. Ankomah P, Levin BR. Exploring the collaboration between
antibiotics and the immune response in the treatment of acute,
self-limiting infections. Proc Natl Acad Sci USA. 2014;
111:8331–8.
58. Putrinš M, Kogermann K, Lukk E, Lippus M, Varik V, Tenson T.
Phenotypic heterogeneity enables uropathogenic Escherichia coli
to evade killing by antibiotics and serum complement. Infect
Immun. 2015;83:1056–67.
59. Levin BR, Baquero F, Ankomah P, McCall IC. Phagocytes,
antibiotics, and self-limiting bacterial infections. Trends Microbiol.
2017;25:878–92.
60. Normark BH, Normark S. Evolution and spread of antibiotic
resistance. J Intern Med. 2002;252:91–106.
61. Yaakov G, Lerner D, Bentele K, Steinberger J, Barkai N. Coupling
phenotypic persistence to DNA damage increases genetic
diversity in severe stress. Nat Ecol Evol. 2017;1:16.
62. Kussell E, Leibler S. Phenotypic diversity, population growth, and
information in fluctuating environments. Science. 2005;309:2075–8.
63. Patra P, Klumpp S. Population dynamics of bacterial persistence.
PLoS ONE. 2013;8:e62814.
64. Gardner A, West SA, Griffin AS. Is bacterial persistence a social
trait? PLoS ONE. 2007;2:e752.
65. Méhi O, Bogos B, Csörgo B, Pál C. Genomewide screen for
modulators of evolvability under toxic antibiotic exposure. Antimicrob
Agents Chemother. 2013;57:3453–6.
66. Kohanski MA, DePristo MA, Collins JJ. Sublethal antibiotic
treatment leads to multidrug resistance via radical-induced mutagenesis.
Mol Cell. 2010;37:311–20.
67. Long H, Miller SF, Strauss C, Zhao C, Cheng L, Ye Z, et al.
Antibiotic treatment enhances the genome-wide mutation rate of
target cells. Proc Natl Acad Sci USA. 2016;113:E2498–E2505.
68. LaFleur MD, Qi Q, Lewis K. Patients with long-term oral carriage
harbor high-persister mutants of Candida albicans. Antimicrob
Agents Chemother. 2010;54:39–44.
69. Mulcahy LR, Burns JL, Lory S, Lewis K. Emergence of Pseudomonas
aeruginosa strains producing high levels of
persister cells in patients with cystic fibrosis. J Bacteriol. 2010;
192:6191–9.
70. Bridges BA. DNA turnover and mutation in resting cells. Bioessays.
1997;19:347–52.
71. Williams AB, Foster PL. Stress-induced mutagenesis. EcoSal
Plus. 2012;5:1–29.
72. Roth JR, Kugelberg E, Reams AB, Kofoid E, Andersson DI.
Origin of mutations under selection: the adaptive mutation controversy.
Annu Rev Microbiol. 2006;60:477–501.
73. Rosenberg SM, Shee C, Frisch RL, Hastings PJ. Stress-induced
mutation via DNA breaks in Escherichia coli: A molecular
mechanism with implications for evolution and medicine. Bioessays.
2012;34:885–92.
74. O’Neill J. Antimicrobial resistance: tackling a crisis for the health
and wealth of nations. Review on Antimicrobial Resistance. 2014.
75. Rosenberg SM, Queitsch C. Combating evolution to fight disease.
Science. 2014;343:1088–9.
76. Cirz RT, Romesberg FE. Controlling mutation: intervening in
evolution as a therapeutic strategy. Crit Rev Biochem Mol Biol.
2007;42:341–54.
77. Goldberg DE, Siliciano RF, Jacobs WR. Outwitting evolution: fighting
drug-resistant TB, malaria, and HIV. Cell. 2012;148:1271–83.
78. Fischbach MA. Combination therapies for combating antimicrobial
resistance. Curr Opin Microbiol. 2011;14:519–23.
1250 E. M. Windels et al.
79. Imamovic L, Sommer MOA. Use of collateral sensitivity networks
to design drug cycling protocols that avoid resistance
development. Sci Transl Med. 2013;5:204ra132.
80. Oxnard GR. The cellular origins of drug resistance in cancer. Nat
Med. 2016;22:232–4.
81. Hangauer MJ, Viswanathan VS, Ryan MJ, Bole D, Eaton JK,
Matov A, et al. Drug-tolerant persister cancer cells are vulnerable
to GPX4 inhibition. Nature. 2017;551:247–50.
Bacterial persistence promotes the evolution of antibiotic resistance by increasing survival and. . . 1251