Article

Limited impact of Salmonella stress and

persisters on antibiotic clearance

[Link] Joseph Fanous1,2, Beatrice Claudi1,2, Vishwachi Tripathi1, Jiagui Li1, Frédéric Goormaghtigh1 &
Dirk Bumann1 ✉
Received: 18 January 2023

Accepted: 10 December 2024

Antimicrobial compounds are essential for controlling bacterial infections. Stress-
Published online: 5 February 2025
induced bacterial tolerance and persisters can undermine antimicrobial activities
Open access
under laboratory conditions, but their quantitative effects under physiological
Check for updates conditions remain unclear1,2. Here we determined constraints on clearance of
Salmonella by antimicrobials in infected mice and tissue-mimicking chemostats.
The antibiotics enrofloxacin and ceftriaxone exhibited poor anti-Salmonella activity
under both conditions, primarily owing to severe nutrient starvation, which restricted
Salmonella replication3–5. Other infection-associated conditions, such as acidic pH,
glucose, oxidative stress, nitrosative stress, antimicrobial peptides, osmolarity,
oxygen limitation, carbon dioxide and carbonate, as well as drug efflux, toxin–antitoxin
modules and cell size had limited effects. A subset of resilient Salmonella appeared as a
key obstacle for clearance by enrofloxacin, based on the biphasic decline of Salmonella
colony-forming units. However, these data were misleading, because colony formation
was confounded by extensive post-exposure killing. More accurate single-cell, real-
time assays showed uniformly slow damage, indicating high resilience across the entire
Salmonella population. The resulting extensive survival of bulk bacteria minimized
the effect of hyper-resilient persisters. Thus, starvation-induced general resilience of
Salmonella was the main cause of poor antibiotic clearance. These findings highlight
the importance of quantifying antibiotic activity with real-time, single-cell assays
under physiological conditions.

Effective antimicrobial therapy is crucial for control of bacterial infec- and prodrug of ciprofloxacin), which is distributed organism-wide11
tions. Antimicrobial resistance can impede treatment success, but within approximately 10 min. After 1 h, we euthanized the mice and
eradication failures occur even in the absence of resistance6. These recovered an approximately 20-fold diminished number of Salmonella
failures are often attributed to host stresses that trigger bacterial drug colony-forming units (CFUs) from the major target organ spleen12. This
tolerance1 and/or subsets of bacteria, such as persisters, that are refrac- CFU loss was around 300-fold slower compared with standard labora-
tory to killing2. Supporting evidence comes mostly from laboratory tory conditions for antimicrobial susceptibility testing (exponential
conditions, and the relevance under physiological conditions remains growth in Mueller–Hinton broth under normoxia at 37 °C) with an equiv-
unclear. Here we determined the quantitative effect of bacterial stresses alent enrofloxacin concentration12 (1.5 mg l−1) (Fig. 1a). Similarly, a dose
and heterogeneity on antimicrobial efficacy in a mouse model of inva- of 50 mg kg−1 ceftriaxone diminished CFUs in spleen approximately
sive salmonellosis, a difficult-to-treat infection that affects around 8-fold over 4 h, around 40-fold slower than under standard laboratory
15 million patients and causes around 200,000 deaths annually7. Our conditions with 25 mg l−1 ceftriaxone. Thus, the two recommended
mouse model replicates treatment failures observed in humans8 and bactericidal antibiotic classes for treatment of invasive salmonellosis
exposes Salmonella to diverse stresses, leading to the formation of (fluoroquinolones and cephalosporins)7 had poor anti-Salmonella
heterogeneous bacterial subsets with divergent properties and fates9,10. activity in infected mice, consistent with slow antimicrobial clearance in
Thus, the model is suitable for assessing the effects of host stresses humans13. This was not owing to the emergence of resistance, because
and bacterial heterogeneity on antimicrobial clearance in a clinically Salmonella from treated mice12 and humans retain full susceptibility.
relevant context. Slow clearance is the primary reason for eventual treatment failures
because some Salmonella still survive in host tissues when the critical
support of host inflammation vanishes12.
Inefficient killing of Salmonella
We infected genetically susceptible mice with gfp-expressing Salmo-
nella enterica subsp. enterica serovar Typhimurium. Upon appearance Limited effect of diverse stresses
of clinical symptoms, we administered the recommended dose of 5 mg To identify constraints on anti-Salmonella activities, we mimicked
per kg (body weight) enrofloxacin (the widely used veterinary version the extensively characterized Salmonella tissue microenvironments

Biozentrum, University of Basel, Basel, Switzerland. 2These authors contributed equally: Joseph Fanous, Beatrice Claudi. ✉e-mail: [Link]@[Link]
1

Nature | Vol 639 | 6 March 2025 | 181

Article
a ENR CRO c ENR CRO
100 100 100

10–1 10–1 10–1 WT

P = 4.9 × 10–7
Normalized CFU

Normalized CFU
lexA3

P = 0.98
10–2 10–2 –2 tisB
10

P = 0.89
P = 0.90
–11 Standard Δ3T
P = 3.4 × 10 In vitro
10–3 –3
10 10–3
Standard P = 3.7 × 10–5
10–4 In vitro
10–4
0 2 4 0 2 4 1h 4h 4h

Time (h) Time (h)

b ENR CRO d ENR CRO

100 10 0 100

10–1

P = 1.1 × 10–7
10–1 10–1

Normalized CFU
Normalized CFU

10–2 10–2 10–2

P = 0.81
Standard
P = 3.9 × 10–14 In vitro
10–3

P = 0.83
P = 0.83
10–3 10–3
–5
Standard P = 5.5 × 10
10–4 In vitro –4
10
0 2 4 0 2 4 1h 4h 4h

Time (h) Time (h)

Fig. 1 | Poor anti-Salmonella activities of enrofloxacin and ceftriaxone. c, Survival of wild-type (WT) and indicated mutant Salmonella in spleen. Each
a, Salmonella survival in mouse spleen after an antibiotic dose (grey symbols symbol represents an individual mouse (enrofloxacin: wild type and lexA3 1 h,
show re-analysed data from ref. 12; enrofloxacin (ENR): 1 h, n = 10; 2 h, n = 3; 4 h, n = 3; wild type 4 h, n = 3; tisB, n = 4; Δ3T, n = 3; ceftriaxone: wild type and Δ3T,
n = 6; ceftriaxone (CRO), n = 6) or in Mueller–Hinton broth under normal n = 6). Horizontal bars represent geometric means. Enrofloxacin, one-way
atmosphere at 37 °C (red symbols (this study); enrofloxacin, n = 5; ceftriaxone, ANOVA of log-transformed data with comparisons to wild-type data and
n = 3). Each symbol represents an individual mouse or an independent in vitro Holm–Šídák correction for multiple comparisons; ceftriaxone, two-tailed
culture. Lines connect the geometric means. Two-tailed t-test of log-transformed t-test of log-transformed data. d, Survival of wild-type and indicated mutant
data. b, Salmonella survival in tissue-mimicking chemostat cultures (grey symbols Salmonella in chemostat cultures. Each symbol represents an individual
show re-analysed data from ref. 12; enrofloxacin: 1 h and 2 h, n = 12; 4 h, n = 7; chemostat reactor (enrofloxacin: wild type and lexA3 1 h, n = 5; wild type 4 h,
ceftriaxone 1 h, 2 h and 4 h, n = 4) or laboratory conditions (red symbols; number n = 16; tisB and Δ3T, n = 4; ceftriaxone: wild type, n = 8; Δ3T, n = 7). Horizontal bars
of samples as in a). Each symbol represents an individual chemostat reactor. represent geometric means (statistical tests as in c).
Lines connect the geometric means. Two-tailed t-test of log-transformed data.

in chemostats. We used 10% oxygen, 5% carbon dioxide balanced with tension, increased carbon dioxide tension and bicarbonate concentra-
10 mM bicarbonate at pH 5.6, and 23 nutrients available to Salmonella tion, osmolarity, glucose, oxidative stress and membrane-damaging
in the mouse spleen5. We set the inflow of fresh medium to 0.116 ml h−1 antimicrobial peptides (Supplementary Note 1). We also inactivated
per ml of culture, mimicking the severe nutrient starvation that limits AcrB, the main efflux pump for fluoroquinolones19 and a proposed cause
Salmonella growth in mice3 (approximately 0.16 divisions per hour, of delayed antimicrobial clearance20, and compared Salmonella cells
generation time approximately 6 h). Under these conditions, physi- with different cell size or cell age21 (Extended Data Fig. 1a). Some of these
ological concentrations of enrofloxacin12 (1.5 mg l−1) or ceftriaxone14 parameters affected Salmonella killing by enrofloxacin, but the effects
(25 mg l−1) killed Salmonella at rates similar to those in mice (Fig. 1b). were generally small (up to threefold; glucose has an approximately six-
A lexA3 allele, which blocks repair of DNA double-strand breaks (DSBs)15, fold effect) compared with the approximately 300-fold difference from
accelerated enrofloxacin killing approximately 100-fold in mice and standard laboratory conditions (Fig. 2a), consistent with their moderate
chemostats (Fig. 1c,d), indicating that enrofloxacin triggered DSBs effects under other in vitro conditions (Supplementary Note 1). Oxida-
(which are lethal without repair) in almost all Salmonella, consistent tive and nitrosative stresses were not required for poor enrofloxacin
with pharmacological target attainment12 under both conditions. Tox- and ceftriaxone activities, consistent with their minor effects in vitro
ins encoded by tisB16 or ecnB3, shpAB and phD-doc17 (deleted in Δ3T), (Supplementary Note 1) and in mice (Supplementary Note 2). Combin-
had no detectable effects on enrofloxacin and ceftriaxone activities ing conditions that promote survival (pH 5.6; 5% CO2/10 mM bicarbo-
in mice and chemostats18 (Fig. 1c,d). In summary, tissue-mimicking nate; 0.4% glucose; 401 mOsm) resulted in approximately fivefold
chemostats recapitulated the poor in vivo activities of both antibiotics. greater survival (Fig. 2a, yellow circles). Unknown stresses or stress
To identify factors that constrained antibiotic activity in the chemo- combinations may exist in vivo, but our results indicate the presence
stat experiments, we varied tissue-associated stresses that are known of a non-stress factor that dominates the high in vivo-like Salmonella
to affect fluoroquinolone efficacy, including acidic pH, reduced oxygen survival in chemostats.

182 | Nature | Vol 639 | 6 March 2025

 a 0.28 0.055 0.068 0.17 0.64 0.019 0.64 0.87 2.6 × 10–6 0.19 5.0 × 10–9 Padj
100
No killing

10–1
Normalized CFU

10–2

10–3

10–4 Standard in vitro

3

0
10
100
10
21

5.6
6.5

0.1
0.2
0.4

115
201
401

Std

Std
0.03

Small

Large
ΔacrB

Medium

CS

CS

High
Low
0.3/0
5/10

–
+
Oxygen (%) pH 16 μM
mOsm H2O2 (μM) Low High
CO2 (%)/BIC Glucose (%) LL-37 MHB
(mM) Nutrient supply
Size

b 1 h ENR 4 h CRO
c
100 100 1
P = 2.2 × 10–6 Chemostat
P < 1 × 10–15
Batch culture
10–1

Normalized CFU
Normalized CFU

10–1 Tissue mimic

MHB
10–2 0.1
10–2
10–3
Infected mice

10 –4 10–3 Slc11a1s/WT
0.01
0.1 1 0.1 1 Slc11a1s/hipAD88N 0 0.4 0.8 1.2
Division rate (h–1) Division rate (h–1) Slc11a1r r/WT Number of generations

Fig. 2 | Modulation of Salmonella killing by stresses and nutrient supply. ceftriaxone for division rates 0.083 h−1, 0.17 h−1 and 0.33 h−1, n = 5) or in individual
a, Salmonella survival after 1 h exposure to enrofloxacin in chemostat cultures batch cultures (enrofloxacin: tissue-mimicking medium, n = 3; Mueller–Hinton
with varying conditions. Each symbol represents an individual chemostat broth, n = 5; ceftriaxone: tissue-mimicking medium and Mueller–Hinton broth,
reactor. P values adjusted for multiple comparisons (Holm–Šídák) (two-tailed n = 3). One-way ANOVA with test for linear trend for log-transformed data for
t-test on log-transformed data for pH; ANOVA on log-transformed data tissue-mimicking medium or Mueller–Hinton broth cultures. c, Relationship
for glucose and osmolarity). BIC, bicarbonate; CS, combined stresses; between exposure time to ceftriaxone to number of generations and Salmonella
MHB, Mueller–Hinton broth; Std, standard tissue-mimicking conditions. survival for different proliferation rates and exposure times (geometric
b, Survival of Salmonella in mice (squares, geometric mean ± geometric s.d.; mean and geometric s.d.; upward triangles, division rate 0.17 h−1 and exposure
enrofloxacin: Slc11a1 s/WT, n = 10; Slc11a1 s /hipA D88N, n = 4; Slc11a1 r /WT, n = 8; for 1 h (n = 4), 2 h (n = 4) or 4 h (n = 9); downward triangles, 4 h exposure and
ceftriaxone, n = 6), or in chemostats in tissue-mimicking medium (geometric division rates 0.08 (n = 5) or 0.33 h−1 (n = 5)). The dashed line represents a
mean ± geometric s.d.; enrofloxacin for division rates 0.083 h−1, 0.17 h−1 and monoexponential fit and the shaded area shows the 95% confidence interval.
0.33 h−1, n = 10; ceftriaxone, n = 4), Mueller–Hinton broth (enrofloxacin or

To determine the role of nutrient access, we grew Salmonella in achieve ~0.16 divisions per hour, as seen in vivo3. These conditions
low-density batch cultures, keeping all other parameters the same as resulted in high Salmonella survival (approximately 3% after 1 h of
in tissue-mimicking chemostats. Facile nutrient access under these enrofloxacin and 30% after 4 h of ceftriaxone; Fig. 2a–c) as in mice,
conditions led to approximately tenfold faster replication (around indicating that starvation at levels similar to those experienced by
1.5 divisions per hour) and increased killing by enrofloxacin and cef- Salmonella in mice3,5 was sufficient to achieve in vivo-like antibiotic
triaxone by about 200-fold (Fig. 2a,b), highlighting the critical role of survival rates, even under non-physiological, stress-free conditions.
nutrient access in antibiotic killing. Multiple linear regression of the This was consistent with previous evidence linking starvation and slow
data shown in Fig. 2a, with survival as dependent variable and nutrient replication with increased resilience against antibiotic activity18,22.
access (which controlled the replication rate), oxygen, pH, CO2, glu- To determine the effect of replication rate in mice, we used two differ-
cose, LL-37, osmolarity and H2O2 as independent variables, explained ent approaches. First, we expressed the Escherichia coli toxin gene hipA
90.4% (R2) of the variance in survival (Extended Data Table 1). Nutrient in Salmonella. HipA inhibits translation and replication by phospho-
access and replication rate (P < 10−15), osmolarity (P = 8.6 × 10−7) and H2O2 rylating GltX, but this can be prevented by the antitoxin HipB23. When
(P = 0.020) had a significant effect; of these factors, nutrient access HipA and HipB levels are similar, fluctuations create a bimodal distribu-
and replication rate had by far the largest effect size (165-fold versus tion of growing ([HipA] < [HipB]) and non-growing ([HipA] > [HipB])
3-fold (osmolarity) and 1.5-fold (H2O2); Fig. 2a). Thus, nutrient access cells23. Without HipB, HipA impairs growth but does not create this
controlling replication rate dominated antibiotic survival. heterogeneity. We expressed the partially detoxified allele24 hipAD88N
To confirm this, we starved Salmonella under non-physiological in Salmonella (which naturally lacks hipAB). This did not affect the
stress-free conditions (Mueller–Hinton broth, pH 7.4, normoxia) in enrofloxacin mean inhibitory concentration (MIC) (0.06 mg l−1),
chemostats, limiting Salmonella nutrition by slow medium inflow to but reduced the replication rate in vitro (to around 47%) and in vivo

Nature | Vol 639 | 6 March 2025 | 183

Article
(to around 71%; Extended Data Fig. 1b,c; competitive index in mice at response and DNA repair28–31. Expressing recAR29A-mCherry from the
day 4 post-infection, 0.028 ± 0.009, n = 6, P = 4 × 10−5, one-sample t-test PrecA promoter provides basal RecA–mCherry for visualizing emerging
of log-transformed data). In mice, Salmonella hipAD88N survived enro- DSBs, and increasing RecA–mCherry levels during the subsequent SOS
floxacin exposure around threefold better than parental Salmonella response (Extended Data Fig. 1e).
(Fig. 2b and Extended Data Fig. 1d). In a second approach, we tracked We imaged Salmonella expressing recAR29A-mCherry in a microfluidic
Salmonella killing in Slc11a1r mice, which encode an active metal trans- device during slow growth in tissue-mimicking, nutrient-poor medium.
porter SLC11A1 (also known as NRAMP1)4. SLC11A1 deprives Salmo- RecA–mCherry occasionally formed short-lived (less than 5 min) foci
nella of magnesium, reducing its replication rate to around 50% of (Fig. 3a) indicating spontaneous, rapidly repairable damage during
that in standard laboratory mice (Slc11a1s) with dysfunctional SLC11A1 chromosomal replication29. Long-lasting RecA foci (lasting more than
(Extended Data Fig. 1c) without affecting stress levels or access to most 5 min; Fig. 3a), indicating more persistent DSBs28, occurred at a rate of
other nutrients4. In Slc11a1r mice infected with wild-type Salmonella, around 0.03 h−1 (Extended Data Fig. 1f), mostly without triggering a sub-
we observed around 15% CFU recovery after 1 h enrofloxacin treat- sequent SOS response. These data were consistent with slow baseline
ment (approximately fourfold more than in Slc11a1s mice; Fig. 2b and DNA damage in untreated bacteria29,31.
Extended Data Fig. 1d). Plating on low-nutrient minimal medium—which Exposure to 1.5 mg l−1 enrofloxacin accelerated formation of
improves CFU recovery (see below)—resulted in a 1 h survival rate of long-lasting RecA foci to around 0.21 h−1 (Extended Data Fig. 1f). Each
about 32% (n = 4) in Slc11a1r mice. This high survival is physiologically focus was followed by a SOS response, indicating severe DNA damage.
relevant because humans and wild mice carry Slc11a1r. Thus, slowing The fraction of ‘undamaged’ cells (those that had not yet experienced
Salmonella replication in vivo using two different methods significantly any DSB), declined monoexponentially to around 40% after 4 h of enro-
increased antibiotic survival consistent with the growth–killing rela- floxacin exposure. These real-time data were inconsistent with the CFU
tionship observed in vitro. data because: (1) only around 15% cells experienced the critical primary
These data indicate that the scarcity of carbon and energy sources DSB within 1 h of exposure, yet plating such cells showed a 20-fold
(standard laboratory Slc11a1s mice) and magnesium (in Slc11a1r mice), decrease in CFU (Fig. 1a,b); and (2) damage was monoexponential,
which limit Salmonella replication4,5, are the primary cause of Salmo- whereas the decrease in CFU was biexponential. The discrepancies were
nella resilience to antibiotics in vivo. This resilience could be owing not explained by insufficient enrofloxacin exposure in the microfluidic
to reduced DNA replication and cell wall synthesis (the targets of fluo- device (for example, due to drug absorption in the microfluidic device),
roquinolones and cephalosporins) at slow growth and/or metabolic because increasing enrofloxacin concentration to 5 mg l−1 had only a
effects of limited nutritients25. Further research is required to disen- minor effect on damage kinetics (Extended Data Fig. 1f), consistent
tangle these mechanisms. with nearly saturating enrofloxacin exposure12 (we used 5 mg l−1 for all
subsequent experiments). The discrepancies were not caused by fila-
mentation of enrofloxacin-exposed bacteria, which might make them
Biphasic loss of CFUs more prone to pipetting damage, because Salmonella in low-nutrient
Ceftriaxone killed Salmonella continuously by 0.90 ± 0.14 log per medium and in mice did not form filamented cells during enrofloxacin
generation (Fig. 2c), suggesting slow loss of viability across the entire exposure (Extended Data Fig. 1g and Supplementary Video 2). The
Salmonella population26. By contrast, enrofloxacin killed around 95% discrepancies were also not caused by insufficient sensitivity of DSB
of Salmonella within the first 1 h of exposure (killing rate approximately detection, because each SOS-responding cell had previously exhib-
3 h−1) (Fig. 1a,b), and the remaining 5.4 ± 1.8% (in vivo) or 7.4 ± 3.3% ited a long-lasting RecA focus, indicating comprehensive detection of
(chemostat) lost viability at a slower rate over the next 3 h (killing rate relevant DNA damage28,29,31. RecA forms foci around all newly formed
0.35 h−1). These biphasic kinetics suggested that a subset of 5–7% resil- DSBs within minutes28, arguing against relevant delays in detection of
ient Salmonella was a key obstacle for enrofloxacin-induced clearance the DNA damage. Thus, DNA damage during exposure was inconsist-
in mice and in chemostats (antibiotic persistence)2. ent with CFU data.
According to consensus guidelines2, antibiotic persistence is meas-
ured by exposing bacteria to antibiotics for various time intervals,
followed by washing and plating on agar plates. After overnight incuba- Massive post-exposure damage
tion, colonies are counted and the data are plotted against exposure The discrepancies might arise during post-exposure regrowth on lysog-
time. A biphasic decline of CFUs (as observed here; Fig. 1a,b) is consid- eny broth (LB) agar plates until visible colonies form. We first assessed
ered to be evidence that not all bacteria in the population are killed at the effect of switching from nutrient-poor medium to LB without prior
the same rate. However, this strategy uses a late and indirect readout antibiotic exposure (simulating regrowth of untreated Salmonella
(colony formation overnight) to infer killing kinetics during antibi- on LB plates). Switching to LB triggered a burst of long-lasting RecA
otic exposure, and relies on the assumption that each bacterium that foci in around 40% of cells, suggesting substantial DNA stress in the
is viable at the time of plating will form a colony. This assumption is initial 10 min after nutrient upshift (‘nutrient shock’32) (Extended Data
violated in diverse pathogen–antibiotic combinations (Supplemen- Fig. 1h,i and Supplementary Video 1). However, only few of these events
tary Note 3), indicating a need for more direct, real-time readouts of triggered SOS responses and around 98% of DSB+ cells restarted divi-
antimicrobial activity. sion after a lag of 1.5 h on LB in synchrony with undamaged DSB− cells
(Extended Data Fig. 1h–j and Supplementary Video 1), indicating effi-
cient repair and negligible fitness loss.
Slow monophasic DNA damage We then simulated the entire CFU assay by exposing Salmonella for
To monitor enrofloxacin activity in real-time, we detected DSBs, the 1, 2 or 4 h to enrofloxacin in nutrient-poor medium, followed by drug
crucial initial damage of fluoroquinolone action27, using Salmonella washout for 30 min with antibiotic-free nutrient-poor medium, and
expressing a recAR29A-mCherry fusion. RecA–mCherry forms visible regrowth in LB. Under these conditions, the shift to LB triggered vigor-
foci around all newly formed DSBs within <4 min (refs. 28,29) without ous RecA localization dynamics and SOS responses for several hours
interfering with subsequent DNA repair29,30. The R29A mutation pre- in all analysed cells, indicating severe DNA damage after the actual
vents DSB-independent RecA aggregation31. Formation of RecA foci antibiotic exposure (Fig. 3b,c and Supplementary Video 2). Almost
relies on the diffusion of constitutively present proteins with no need all cells increased in size, and many formed long filaments. Some cells
for de novo gene expression. RecA foci, therefore, serve as an early, reinitiated cell division about 3.5 h after the initial DSB (Extended Data
sensitive, and reliable readout for DSBs before the initiation of SOS Fig. 1j) and formed expanding microcolonies, indicating successful

184 | Nature | Vol 639 | 6 March 2025

 32 min ENR 41 min LB 218 min LB 368 min LB
a b
00:32 02:11 05:08 07:38

Short 1 min

Division
Time
Long
DSB

c 1 e
00:35 03:57 13:00 Division 26:26

0.1
DSB
ENR

0.01
LB DSB
Pre-activated
0 2 4 6
1
Undamaged fraction
Regrowing fraction

f 5 1 0.2 0.04 mg l–1 ENR g

P = 0.0085
0 0
0.1
Undamaged fraction
Regrowing fraction

0.01

Regrowing fraction
ENR LB
0.1 0.1
0 2 4 6 8
1

0.1
0.01 0.01

0.01
ENR LB
0 4 8 12 16 20 24 28 32

IVM
4 h/LB
0 2 4 6 8 10
Time (h)
Time (h)

d h 4,000 P = 1.6 × 10–7 i j

Exposure P = 1.5 × 10–6
MFI GFP+

1 1 SOS+
SOS–
Non-responder fraction
Undamaged fraction

1,000
Post- Contribution to CFU
Normalized CFU

0.8
0.1 exposure 0 1 2 3 4 P = 6.3 × 10–12
Frequency

Time (h) 4h 0.1

3h
0.4
Regrowing WT: 0 h, 1 h, 31h
h, 4 h
0.01 lexA3
lexA3: 1h
LB Ctrl
0.01
0
0 1 2 3 4 1h 4h
0 1 2 3 4 102 103 104 105
Exposure time
Exposure time (h) Green fluorescence Exposure time (h)

Fig. 3 | Salmonella DNA damage during and after exposure to enrofloxacin. same data as in d) or under in vivo-mimicking conditions (IVM) with declining
a, Time-lapse gallery of RecA foci in enrofloxacin-exposed Salmonella. Scale enrofloxacin concentrations and regrowth in nutrient-poor medium (276, 271
bar, 1 μm. b, Snapshots of Salmonella exposed for 1 h to enrofloxacin, followed and 378 cells). Circles represent independent experiments. Two-tailed t-test on
by drug washout for 30 min and incubation in LB. Scale bar, 5 μm. c, Fraction of log-transformed data. h, Fluorescence of Salmonella/pPcad-gfp or Salmonella
undamaged and regrowing cells during 1 h (top), 2 h (middle) or 4 h (bottom) lexA3/pPcad-gfp in untreated (0 h) or enrofloxacin-treated mice. The vertical
enrofloxacin exposure, washout and LB incubation (250 (top), 255 (middle) and line separates GFP+ responders from GFP− non-responders. The inset shows
250 (bottom) cells; dotted lines show monoexponential fits for damage during the median fluorescence (MFI) of GFP+ cells, the line connects the geometric
exposure and washout; summary data and independent replicates in d,g). means (test for non-zero slope in a linear regression of log-transformed
d, Fractions of undamaged cells at the end of enrofloxacin exposure, in LB, values). The histograms represent pooled data, circles in the inset graph
and regrowing survivors (two independent experiments, 755 and 1,233 cells). represent individual mice (1 h, n = 5; 3 h and 4 h, n = 4). i, Blue, fraction of GFP−
e, Snapshots of Salmonella exposed for 7 h to decreasing concentrations of non-responders in enrofloxacin-treated mice (0 h, n = 2; 1 h, n = 6; 2 h and 4 h,
enrofloxacin, followed by drug-free nutrient-poor medium (Supplementary n = 4). Each symbol represents an individual mouse. Brown, CFUs recovered
Video 3). Scale bar, 5 μm. f, Fraction of undamaged and regrowing cells during from similar samples (same data as in Fig. 1a; geometric mean ± geometric
and after 7 h exposure to decreasing concentrations of enrofloxacin (ENR) in s.d.; 1 h, n = 10; 2 h, n = 3; 4 h, n = 6). Two-way ANOVA for difference between
nutrient-poor medium (276 cells; dotted lines show monoexponential fits for SOS response and CFU recovery. j, Contribution of responders (SOS+) and
damage during exposure, post-antibiotic damage or baseline damage; summary non-responders (SOS−) to CFU counts (Supplementary Note 5; data are
data for this and two independent replicates in g). g, Fractions of regrowing mean ± s.d. for data from independently infected mice; 1 h, n = 6; 4 h, n = 4).
cells after 4 h enrofloxacin exposure and a switch to LB (4 h/LB; 250 and 444 cells; Two-way ANOVA for difference between SOS+ and SOS−.

Nature | Vol 639 | 6 March 2025 | 185

Article
DNA repair. The frequency of these survivors (Fig. 3c,d, 1 h enrofloxacin tissues. Instead, we monitored the Salmonella SOS response to DNA
exposure, about 10% survival; 4 h, about 3% survival) was similar to damage using a reporter strain carrying a sensititve Pcad-gfp fusion36
CFU data from enrofloxacin-treated chemostat cultures (Fig. 1b, 1 h, (Extended Data Fig. 2b,c). In the spleen of mice, the reporter showed
about 7% survival; 4 h, about 2% survival). Thus, enrofloxacin induced negligible SOS activity before enrofloxacin administration (Fig. 3h),
DSBs and killed Salmonella primarily post-exposure during regrowth indicating limited host-induced DSBs (in contrast to infections of mac-
on LB, explaining the discrepancy between limited damage during rophages in cell culture; Supplementary Note 4).
exposure and low CFU recovery. Enrofloxacin administration activated around 25% Salmonella SOS
Nutrient-rich LB is a poor mimic of relevant regrowth conditions reporter cells within 1 h (Fig. 3h,i). This was a specific SOS response,
in tissues. To approximate the in vivo situation, we grew Salmonella because SOS-defective Salmonella lexA3 carrying the same reporter
in nutrient-poor medium, exposed them to diminishing enrofloxa- construct showed no response. Prolonged in vivo exposure to enro-
cin concentrations approximating enrofloxacin pharmacokinetics floxacin increased the brightness and proportion of SOS+ Salmonella
in mice11, and maintained them in nutrient-poor medium after expo- cells, resulting in a monoexponential decay of non-responders (SOS−)
sure. Under these conditions, Salmonella exhibited long-lasting RecA at a rate of 0.24 h−1. Non-responders retained a functional reporter con-
foci with monoexponential kinetics at 0.18 h−1 throughout the 7 h of struct and experienced enrofloxacin poisoning in vivo, because most
enrofloxacin exposure and about 1 h after washout, followed by slower reporter cells exhibited a SOS response post-exposure upon ex vivo
monoexponential kinetics at 0.07 h−1 for around 14 h in drug-free incubation in LB15 (Extended Data Fig. 2d). Ex vivo sorting and plating
medium. Thereafter, focus formation returned to the basal rate of of SOS+ and SOS− Salmonella revealed that most colonies originated
0.02 h−1 (Fig. 3e,f and Supplementary Video 3). All cells with long RecA from SOS− Salmonella cells that had not yet responded in vivo (Fig. 3j,
foci showed vigorous SOS responses and RecA localization dynamics Extended Data Fig. 2e and Supplementary Note 5). These observations
indicating severe DNA damage. However, growth and filamentation were inconsistent with the CFU data because: (1) after 1 h of exposure
were less extensive than in LB. Re-initiation of division and formation only around 25% Salmonella responded to DNA damage, whereas the
of microcolonies started 4 to 30 h after the initial DSB (Extended Data CFU dropped around 20-fold; (2) the response to DNA damage was
Fig. 1j), and around fivefold more cells survived than in 4 h exposure monoexponential, whereas the CFU decline was biexponential; (3) most
followed by LB regrowth (Fig. 3g), despite more damage at the end of colonies originated from in vivo non-responders (SOS−), although an
exposure. Thus, scarce nutrition improved repair and survival, reach- active SOS response was essential for formation of more than 99% of all
ing levels (14 ± 4%; regrowing survivors) consistent with Salmonella colonies (based on the phenotype of SOS-deficient Salmonella lexA3;
loads in the spleen 24 h after enrofloxacin doses (34 ± 13% of pre-dose Fig. 1c). These data suggested that most DNA damage, SOS response and
levels; this includes regrowing survivors and their daughter cells)12. viability loss occurred after enrofloxacin exposure during regrowth on
Together, our results demonstrate that enrofloxacin continued to LB (Extended Data Fig. 2a), consistent with our observations in micro-
damage Salmonella after the end of exposure. This was consistent with fluidic devices (Fig. 3d).
the half-life of ternary enrofloxacin–gyrase–DNA complexes of 2.5 h To test the idea that regrowth on LB compromised the colony-forming
causing extended post-exposure growth inhibition (post-antibiotic ability of initially viable but enrofloxacin-poisoned Salmonella, we
effect)33. A shift to nutrient-rich LB caused extensive DSBs, explaining explored alternative plating media to reduce post-exposure damage.
the need for DNA repair after, but not during, fluoroquinolone exposure The addition of thiourea and dipyridiyl to mitigate oxidative stress had
in CFU assays15,34. Vigorous growth of enrofloxacin-poisoned cells on no effect on colony recovery, consistent with oxygen-independent kill-
LB might cause biosynthetic imbalances that lead to viability loss35. ing by second-generation fluoroquinolones at concentrations achieved
Physiological starvation slowed regrowth3,5, resulting in less damage in vivo12,37. Lowering the incubation temperature diminished colony
and allowing more time for repair. counts. Plating on the nutrient-poor medium used in the microfluid-
These findings indicated that plotting CFU against exposure time ics devices yielded no visible colonies even after 7 days of incubation.
was misleading, because most viability loss occurred not during expo- However, after increasing carbon and energy sources tenfold, this
sure (as implicitly assumed in CFU-based assays), but after exposure. medium supported formation of tiny colonies after 3 days at 37 °C.
Thus, the approximately 20-fold drop in CFU on LB plates after 1 h of Plating spleen homogenates from enrofloxacin-treated mice on this
exposure did not represent the killing of 95% of the Salmonella pop- medium yielded 2.2 ± 0.4-fold higher CFU than on LB (11 independent
ulation within that 1 h, and the gradual CFU decline with extended infections, P < 0.0001, two-tailed t-test on log-transformed values).
exposure did not indicate a refractory subpopulation. Instead, enro- This confirmed that a majority of enrofloxacin-poisoned but initially
floxacin exposure itself inflicted slow damage, but around 90% of the alive Salmonella died on LB plates, whereas some of them could repair
yet undamaged, enrofloxacin-poisoned Salmonella died post-exposure the DNA damage and survive during slow regrowth on nutrient-poor
during regrowth, irrespective of prior exposure times (Fig. 3d and medium34. The nutrient-poor medium and standard LB yielded sim-
Extended Data Fig. 2a). Thus, the biphasic CFU kinetics did not reflect ilar CFU for Salmonella from untreated control mice (3 infections;
heterogeneous antibiotic survival but rather a confounded viability 1.1 ± 0.1; P = 0.19), indicating that slow growth enhanced specifically
readout. the survival of enrofloxacin-poisoned Salmonella (as in microfluidic
Real-time imaging revealed slow, monoexponential damage of Sal- devices; Fig. 3g). Further optimization of regrowth conditions might
monella by enrofloxacin, suggesting low but uniform and constant yield even higher CFUs from enrofloxacin-treated mice, but they would
damage probabilities for individual cells (akin to the radioactive decay still underestimate the number of initially alive Salmonella owing to
of unstable atomic nuclei). Thus, the entire Salmonella population, and inevitable post-exposure damage by still-bound enrofloxacin (Fig. 3f).
not just a small subset, was rather refractory to enrofloxacin-inflicted Thus, CFU data would remain a confounded readout.
damage. Some damaged Salmonella were able to repair the DSBs and The slow monoexponential SOS kinetics (Fig. 3i) suggested that
survived, further delaying clearance. We cannot exclude that some enrofloxacin damaged Salmonella in vivo with low but rather homog-
cells had lower damage probabilities or superior repair capabilities, enous and constant single-cell probabilities. Thus, the Salmonella
but even the bulk population died at rates below 0.2 h−1. bulk population, rather than only a minor persister subset, was highly
refractory to DNA damage. Against this background of high resilience
and massive survival, minor subsets with even superior survival would
Slow monophasic SOS responses in mice have limited impact. This was consistent with the only about threefold
We next investigated Salmonella killing in mice. Detecting in vivo clearance per day during a 4-day treatment with daily enrofloxacin
DNA damage by tracking RecA foci over hours is challenging in mouse administration12. A Salmonella subset with even slower clearance exists

186 | Nature | Vol 639 | 6 March 2025

in the splenic white pulp. This subset experiences sufficient enrofloxa- pre-existing heterogeneity. In mice, non-replicating Salmonella are
cin exposure, but insufficient local inflammation provides inefficient observed in the first two days following infection with 2 × 1010 Salmo-
support for antibiotic clearance. Eventually, inflammation resolves nella17, a dose >107 times higher than typically encountered in natural
across the entire spleen, resulting in overall eradication failure. By human infections48 (Supplementary Note 8). Under more clinically rel-
contrast, sustaining inflammation throughout antimicrobial treat- evant infection and treatment conditions, non-replicating Salmonella
ment enables eradication from all spleen compartments, confirming are rare and have no detectable role in eradication failures3,12. Instead,
the central role of the host immune system and the limited effect of slow bacterial clearance results from slowly replicating Salmonella,
nonreplicating Salmonella persisters12. which respond poorly to antibiotic treatment3,12,49.

No effect of auxotrophy and persisters Broad post-exposure killing

To explore further the effect of Salmonella physiology on antibiotic Our findings with enrofloxacin-treated Salmonella raise concerns
clearance in vivo, we tested two gene alleles reportedly affecting per- about interpreting CFU-based killing assays. To test an unrelated
sister formation. First, we examined a hisGP69L allele that repairs the pathogen–antibiotic combination, we imaged gfp-expressing Staph-
histidine auxotrophy of Salmonella strain SL134438 used throughout ylococcus aureus cells in microfluidic devices during and after expo-
this study. Histidine auxotrophy has been suggested to favour antibi- sure to flucloxacillin, a first-line β-lactam antibiotic. S. aureus cells
otic tolerance over persistence in vivo, where histidine-limiting condi- started to release their cytosolic contents—indicating death50—after
tions are assumed to exist39. However, auxotrophic Salmonella SL1344 approximately 1 h exposure to flucloxacillin (Fig. 4a–c and Supple-
and prototrophic Salmonella SL1344 hisGP69L have undistinguishable mentary Video 4). After 4 h of exposure, 15% remained alive, but kill-
replication dynamics and net growth in mice3, indicating that host ing continued post-exposure for several hours, consistent with the
tissues supply sufficient histidine to support the biomass needs of long-lived transpeptidase inhibition of the covalent penicillin-binding
SL1344. Auxotrophic Salmonella SL1344 and prototrophic Salmo- protein–β-lactam complexes51. Eventually, only around 0.5% of cells
nella SL1344 hisGP69L also exhibited superimposable killing kinetics in survived and formed microcolonies. The standard CFU plot versus
enrofloxacin-treated mice (Fig. 1a and Extended Data Fig. 2f; SL1344, exposure time would suggest monoexponential killing with 30-fold
survival at 1 h, 5.4 ± 1.8%; 4 h, 2.2 ± 1.2%; SL1344 hisGP69L: 1 h, 4.3 ± 1.1%; inflated rates compared to actual killing during exposure, which began
4 h, 1.5 ± 0.4%;). Thus, the histidine auxotrophy of SL1344 does not only after a delay50 (Fig. 4c). Thus, CFU again provided a confounded
affect its antibiotic survival in vivo (Supplementary Note 6). and misleading readout for antibiotic activity.
In addition, we tested a variant of the antitoxin ShpB Q97* (ShpB1) Previous reports indicate that more than 90% of bacteria that survive
that lacks the last four amino acids, which increases persister frequency initial exposure to fluoroquinolones, β-lactams, aminoglycosides, cidal
in Salmonella40. Salmonella hisGP69L shpBQ97* exhibited an unaltered macrolides, nalidixic acid or trimethoprim may die during regrowth on
growth rate and enrofloxacin MIC in vitro (Extended Data Fig. 1b, standard agar plates (Supplementary Note 3), consistent with exten-
MIC = 0.06 mg l−1) but survived enrofloxacin exposure in vitro around sively documented post-antibiotic effects of these and other antibiot-
30-fold better than wild type (Extended Data Fig. 2g). In mice, Salmo- ics52. Specialized regrowth conditions can improve survival by more
nella hisGP69L shpBQ97* showed normal fitness (competitive index versus than 10-fold in most of these cases, but substantial post-exposure death
wild type at day 4 post-infection, 0.79 ± 0.3; n = 6; P = 0.12, one-sample might be unavoidable owing to ongoing damaging effects of remaining
t-test of log-transformed data) and had killing kinetics during enrofloxa- target-bound antibiotic (Fig. 3f). As we show here, this post-exposure
cin exposure that were undistinguishable from wild type (Extended killing during regrowth can generate artificial biphasic CFU kinet-
Data Fig. 2f; Supplementary Note 7). Thus, a mutation increasing the ics (Fig. 3d,i), undermining the reliability of the standard persister
frequency of persisters had no detectable impact on antimicrobial assay. Additionally, CFU assays may suggest exaggerated killing rates
survival in vivo, supporting our model that extensive survival of bulk (Figs. 3d,g,i and 4c) and may conflate tolerance (slower killing dur-
Salmonella minimizes the impact of minor hyper-resilient persister ing exposure) with reduced post-exposure killing (for example, due
subsets. to slower regrowth; Fig. 3g). Thus, commonly used yet potentially
In contrast to our findings, it has been reported that infected mac- misleading CFU assays should be replaced with real-time, single-cell
rophages trigger Salmonella to express toxin–antitoxin modules that monitoring of antibiotic action to quantify bacterial killing, tolerance
arrest growth in a subset of Salmonella. These non-replicating per- and persistence. For several antibiotic classes, this will require novel
sisters maintain an active type 3 secretion system 2 (T3SS-2) which reporters for antibiotic-induced damage and viability.
is essential for their survival41 and are proposed to cause eradication
failures in infected mice receiving daily 300 mg kg−1 enrofloxacin (30- to
60-fold more than recommended42) for 5 days, starting one day after Discussion
infection17—before disease symptoms appear, which does not reflect Invasive salmonellosis is a life-threatening disease that requires anti-
typical clinical scenarios (Supplementary Note 8). Some of these data microbial chemotherapy. The recommended bactericidal antibiotics,
have been difficult to replicate3,18,43, the role of non-mutated toxin– fluoroquinolones and cephalosporins, achieve only slow Salmonella
antitoxin modules for persister formation is debatable18,44,45 (Fig. 1c,d), clearance and are prone to eradication failures, even when the causa-
Salmonella does not require T3SS-2 for antibiotic survival in mice12, tive Salmonella strain tests as susceptible in the laboratory. This poor
and the evidence for any Salmonella surviving such excessive enro- efficacy is usually attributed to bacterial stress-induced tolerance and
floxacin doses is questionable3,12,43 (Supplementary Note 8). Notably, persisters. However, our quantitative comparison under physiological
macrophages do not induce a distinct subset of non-replicating Salmo- conditions suggests that the main reason is nutrient starvation, which
nella when infected with homogeneously growing Salmonella in vivo restricts Salmonella replication in infected tissues3–5. Salmonella also
or in vitro (using Salmonella cultured under T3SS-2-inducing condi- experiences various stresses in vivo. Although some stresses affect
tions to simulate ongoing intracellular replication while preventing antibiotic survival, their overall effects are limited compared with
rapid macrophage pyroptosis)3. A distinct subset of non-replicating the dominant role of starvation. Thus, future antibiotic development
intracellular bacteria, largely incapable of resuming growth after being should mimic nutrient-scarce conditions.
released from host cells, only emerges when macrophages are infected Antimicrobial treatment failures are often attributed to a small sub-
with stationary-phase Salmonella cultures3,17,41. Stationary cultures are set of non-replicating, hyper-resilient bacteria known as persisters. Per-
highly heterogeneous46,47, and macrophages may solely amplify this sisters are implicated when antibiotic exposure causes an initial rapid

Nature | Vol 639 | 6 March 2025 | 187

Article
a
04:00 06:00 07:20 10:00

240 min 2% BHI 120 min FLX 80 min BHI 240 min BHI

b c
1 1

Survival at end of exposure

Colony-forming fraction
Survivng fraction

0.1
0.1
Post-
exposure

0.01
0.01 P = 7 × 10–8
FLX

BHI FLX BHI FLX BHI

0 2 4 6 8 0 2 4 6 8 0 2 4 6 8 10 12 14
Time (h) Time (h) Time (h) 0 1 2 3 4
Exposure time (h)

Fig. 4 | Killing of S. aureus by flucloxacillin. a, Snapshots of gfp-expressing counted as one survivor. Summary data for individual replicates are shown in c.
S. aureus (inverted fluorescence) before, during and after a 2 h exposure to c, Surviving fractions of S. aureus after 1 h, 2 h or 4 h exposure to flucloxacillin
flucloxacillin (FLX) and switching to antibiotic-free brain–heart infusion (BHI) and after switching to antibiotic-free BHI medium. Data from three independent
medium (Supplementary Video 4). Scale bar, 5 μm. b, Surviving fractions of experiments (619, 745 and 917 cells). Lines connect the geometric means.
S. aureus after 1 h (left), 2 h (middle) or 4 h (right) exposure to flucloxacillin Two-way ANOVA for difference between survival at the end of exposure and
followed by switching to BHI medium (937, 745 and 599 cells pooled from three colony-forming fractions. The arrow depicts the post-exposure loss of viability.
independent experiments). A growing colony originating from a single cell is

decline in CFUs, followed by a slower decrease. Salmonella exhibited tuberculosis, deep-seated S. aureus infections and brucellosis. Inef-
such a biphasic CFU decline during enrofloxacin treatment in mice. ficient drug delivery to infected tissues, biofilms and noncompliant
However, we found that standard CFU assays were misleading because patients may contribute to this problem. These infections also have long
enrofloxacin remains bound to its bacterial target gyrase even after incubation periods54–57, suggesting slow pathogen growth in human
washing, continuing to kill the bacteria during regrowth on the plates, tissues. Quantifying the effects of slow growth, stresses and persisters
which results in low CFU counts even for short treatment intervals. will require real-time, single-cell assays under physiologically relevant
Similar issues may occur with a wide range of other antibiotics. Thus, conditions.
CFU-based killing assays can provide confounded and misleading
results, suggesting non-existing distinct subsets of bacteria with dif-
ferent susceptibilities, underestimating the number of survivors, Online content
and misrepresenting their relationship to the bulk population. Using Any methods, additional references, Nature Portfolio reporting summa-
more appropriate single-cell, real-time assays, we found that damage ries, source data, extended data, supplementary information, acknowl-
kinetics were rather uniform across the Salmonella population and edgements, peer review information; details of author contributions
much slower than CFU data suggested. More than 35% of Salmonella and competing interests; and statements of data and code availability
did not experience any serious damage during 4 h of enrofloxacin are available at [Link]
exposure, minimizing the influence of rare hyper-resilient persisters.
Ceftriaxone also exhibited homogeneously slow killing activity. Thus, 1. Bjarnsholt, T. et al. The importance of understanding the infectious microenvironment.
the bulk of Salmonella, not just a small subset, were difficult to kill. Lancet Infect. Dis. 22, e88–e92 (2022).
This was consistent with the slow, steady Salmonella clearance over 2. Balaban, N. Q. et al. Definitions and guidelines for research on antibiotic persistence.
Nat. Rev. Microbiol. 17, 441–448 (2019).
several days of treatment with no detectable effect of non-replicating 3. Claudi, B. et al. Phenotypic variation of Salmonella in host tissues delays eradication by
persisters3,12. antimicrobial chemotherapy. Cell 158, 722–733 (2014).
Our results with two recommended classes of bactericidal antibiot- 4. Cunrath, O. & Bumann, D. Host resistance factor SLC11A1 restricts Salmonella growth
through magnesium deprivation. Science 366, 995–999 (2019).
ics for treating invasive salmonellosis challenge common assump- 5. Steeb, B. et al. Parallel exploitation of diverse host nutrients enhances Salmonella virulence.
tions about bacterial resilience and highlight the dominant role of PLoS Pathog. 9, e1003301 (2013).
starvation of the entire Salmonella population. The limited effects of 6. Doern, G. V. & Brecher, S. M. The clinical predictive value (or lack thereof) of the results of
in vitro. J. Clin. Microbiol. 49, S11–S14 (2011).
stresses and persisters observed in our study are consistent with the 7. WHO Bacterial Priority Pathogens List, 2024: Bacterial Pathogens of Public Health
broad clinical efficacy of short antibiotic treatments for most infections Importance, to Guide Research, Development and Strategies to Prevent and Control
caused by non-resistant bacteria53, suggesting limited interference from Antimicrobial Resistance (World Health Organization, 2024).
8. Newson, J. P., Gaissmaier, M. S., McHugh, S. C. & Hardt, W. D. Studying antibiotic
inflammation-associated stresses and persisters. Longer treatment is persistence in vivo using the model organism Salmonella Typhimurium. Curr. Opin.
required for difficult-to-treat infections such as invasive salmonellosis, Microbiol. 70, 102224 (2022).

188 | Nature | Vol 639 | 6 March 2025

9. Bumann, D. Heterogeneous host-pathogen encounters: act locally, think globally. 38. Hoiseth, S. K. & Stocker, B. A. Aromatic-dependent Salmonella typhimurium are non-
Cell Host Microbe 17, 13–19 (2015). virulent and effective as live vaccines. Nature 291, 238–239 (1981).
10. Fang, F. C., Frawley, E. R., Tapscott, T. & Vazquez-Torres, A. Bacterial stress responses 39. Michaux, C., Ronneau, S., Giorgio, R. T. & Helaine, S. Antibiotic tolerance and persistence
during host infection. Cell Host Microbe 20, 133–143 (2016). have distinct fitness trade-offs. PLoS Pathog. 18, e1010963 (2022).
11. Liu, X. et al. Pharmacokinetics and pharmacodynamics of enrofloxacin treatment of 40. Slattery, A., Victorsen, A. H., Brown, A., Hillman, K. & Phillips, G. J. Isolation of highly
Escherichia coli in a murine thigh infection modeling. BMC Vet. Res. 17, 212 (2021). persistent mutants of Salmonella enterica serovar typhimurium reveals a new toxin–
12. Li, J. et al. Tissue compartmentalization enables Salmonella persistence during antitoxin module. J. Bacteriol. 195, 647–657 (2013).
chemotherapy. Proc. Natl Acad. Sci. USA 118, e2113951118 (2021). 41. Stapels, D. A. C. et al. Salmonella persisters undermine host immune defenses during
13. Dobinson, H. C. et al. Evaluation of the clinical and microbiological response to Salmonella antibiotic treatment. Science 362, 1156–1160 (2018).
Paratyphi A infection in the first paratyphoid human challenge model. Clin. Infect. Dis. 64, 42. Slate, A. R. et al. Efficacy of enrofloxacin in a mouse model of sepsis. J. Am. Assoc. Lab.
1066–1073 (2017). Anim. Sci. 53, 381–386 (2014).
14. Wang, E., Bergeron, Y. & Bergeron, M. G. Ceftriaxone pharmacokinetics in interleukin-10- 43. Griffin, A. J., Li, L. X., Voedisch, S., Pabst, O. & McSorley, S. J. Dissemination of persistent
treated murine pneumococcal pneumonia. J. Antimicrob. Chemother. 55, 721–726 (2005). intestinal bacteria via the mesenteric lymph nodes causes typhoid relapse. Infect. Immun.
15. Dorr, T., Lewis, K. & Vulic, M. SOS response induces persistence to fluoroquinolones in 79, 1479–1488 (2011).
Escherichia coli. PLoS Genet. 5, e1000760 (2009). 44. Jurėnas, D., Fraikin, N., Goormaghtigh, F. & Van Melderen, L. Biology and evolution of
16. Dorr, T., Vulic, M. & Lewis, K. Ciprofloxacin causes persister formation by inducing the TisB bacterial toxin–antitoxin systems. Nat. Rev. Microbiol. 20, 335–350 (2022).
toxin in Escherichia coli. PLoS Biol. 8, e1000317 (2010). 45. LeRoux, M. & Laub, M. T. Toxin–antitoxin systems as phage defense elements. Annu. Rev.
17. Helaine, S. et al. Internalization of Salmonella by macrophages induces formation of Microbiol. 76, 21–43 (2022).
nonreplicating persisters. Science 343, 204–208 (2014). 46. Ryall, B., Eydallin, G. & Ferenci, T. Culture history and population heterogeneity as
18. Pontes, M. H. & Groisman, E. A. Slow growth determines nonheritable antibiotic resistance determinants of bacterial adaptation: the adaptomics of a single environmental transition.
in Salmonella enterica. Sci. Signal. 12, eaax3938 (2019). Microbiol. Mol. Biol. Rev. 76, 597–625 (2012).
19. Chen, S. et al. Contribution of target gene mutations and efflux to decreased susceptibility 47. Ciolli Mattioli, C. et al. Physiological stress drives the emergence of a Salmonella
of Salmonella enterica serovar typhimurium to fluoroquinolones and other antimicrobials. subpopulation through ribosomal RNA regulation. Curr. Biol. 33, 4880–4892.e14
Antimicrob. Agents Chemother. 51, 535–542 (2007). (2023).
20. Bergmiller, T. et al. Biased partitioning of the multidrug efflux pump AcrAB–TolC underlies 48. Woodward, W. E. Volunteer studies of typhoid fever and vaccines. Trans. R. Soc. Trop.
long-lived phenotypic heterogeneity. Science 356, 311–315 (2017). Med. Hyg. 74, 553–556 (1980).
21. Aldridge, B. B. et al. Asymmetry and aging of mycobacterial cells lead to variable growth 49. Kaiser, P. et al. Cecum lymph node dendritic cells harbor slow-growing bacteria
and antibiotic susceptibility. Science 335, 100–104 (2012). phenotypically tolerant to antibiotic treatment. PLoS Biol. 12, e1001793 (2014).
22. Kaldalu, N. & Tenson, T. Slow growth causes bacterial persistence. Sci. Signal. 12, eaay1167 50. Giesbrecht, P., Kersten, T., Maidhof, H. & Wecke, J. Staphylococcal cell wall: morphogenesis
(2019). and fatal variations in the presence of penicillin. Microbiol. Mol. Biol. Rev. 62, 1371–1414
23. Rotem, E. et al. Regulation of phenotypic variability by a threshold-based mechanism (1998).
underlies bacterial persistence. Proc. Natl Acad. Sci. USA 107, 12541–12546 (2010). 51. Davoodi, S., Daryaee, F., Chang, A., Walker, S. G. & Tonge, P. J. Correlating drug-target
24. Korch, S. B., Henderson, T. A. & Hill, T. M. Characterization of the hipA7 allele of residence time and post-antibiotic effect: insight into target vulnerability. ACS Infect. Dis.
Escherichia coli and evidence that high persistence is governed by (p)ppGpp synthesis. 6, 629–636 (2020).
Mol. Microbiol. 50, 1199–1213 (2003). 52. MacKenzie, F. M. & Gould, I. M. The post-antibiotic effect. J. Antimicrob. Chemother. 32,
25. Lopatkin, A. J. et al. Bacterial metabolic state more accurately predicts antibiotic lethality 519–537 (1993).
than growth rate. Nat. Microbiol. 4, 2109–2117 (2019). 53. Grant, J. & Saux, N. L. Duration of antibiotic therapy for common infections. J. Assoc. Med.
26. Tuomanen, E., Cozens, R., Tosch, W., Zak, O. & Tomasz, A. The rate of killing of Escherichia Microbiol. Infect. Dis. Can. 6, 181–197 (2021).
coli by beta-lactam antibiotics is strictly proportional to the rate of bacterial growth. J. Gen. 54. Awofisayo-Okuyelu, A., McCarthy, N., Mgbakor, I. & Hall, I. Incubation period of typhoidal
Microbiol. 132, 1297–1304 (1986). salmonellosis: a systematic review and meta-analysis of outbreaks and experimental
27. Aldred, K. J., Kerns, R. J. & Osheroff, N. Mechanism of quinolone action and resistance. studies occurring over the last century. BMC Infect. Dis. 18, 483 (2018).
Biochemistry 53, 1565–1574 (2014). 55. Behr, M. A., Edelstein, P. H. & Ramakrishnan, L. Revisiting the timetable of tuberculosis.
28. Wiktor, J. et al. RecA finds homologous DNA by reduced dimensionality search. Nature BMJ 362, k2738 (2018).
597, 426–429 (2021). 56. Smit, J. et al. Onset of symptoms, diagnostic confirmation, and occurrence of multiple
29. Amarh, V., White, M. A. & Leach, D. R. F. Dynamics of RecA-mediated repair of replication- infective foci in patients with Staphylococcus aureus bloodstream infection: a look into
dependent DNA breaks. J. Cell Biol. 217, 2299–2307 (2018). the order of events and potential clinical implications. Infection 46, 651–658 (2018).
30. Lesterlin, C., Ball, G., Schermelleh, L. & Sherratt, D. J. RecA bundles mediate homology 57. Hasanjani Roushan, M. R. & Ebrahimpour, S. Human brucellosis: an overview. Caspian
pairing between distant sisters during DNA break repair. Nature 506, 249–253 (2014). J. Intern. Med. 6, 46–47 (2015).
31. Renzette, N. et al. Localization of RecA in Escherichia coli K-12 using RecA-GFP.
Mol. Microbiol. 57, 1074–1085 (2005). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
32. Azevedo, N. F. et al. Proposal for a method to estimate nutrient shock effects in bacteria. published maps and institutional affiliations.
BMC Res. Notes 5, 422 (2012).
33. Wiedemann, B. & Kratz, B. in Ciprofloxacin iv. Defining its Role in Serious Infection Open Access This article is licensed under a Creative Commons Attribution
(ed. Garrard, C.) 13–19 (Springer, 1993). 4.0 International License, which permits use, sharing, adaptation, distribution
34. Mok, W. W. K. & Brynildsen, M. P. Timing of DNA damage responses impacts persistence and reproduction in any medium or format, as long as you give appropriate
to fluoroquinolones. Proc. Natl Acad. Sci. USA 115, E6301–e6309 (2018). credit to the original author(s) and the source, provide a link to the Creative Commons licence,
35. Pontes, M. H. & Groisman, E. A. A physiological basis for nonheritable antibiotic resistance. and indicate if changes were made. The images or other third party material in this article are
mBio 11, e00817–e00820 (2020). included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
36. Norman, A., Hestbjerg Hansen, L. & Sorensen, S. J. Construction of a ColD cda promoter- to the material. If material is not included in the article’s Creative Commons licence and your
based SOS–green fluorescent protein whole-cell biosensor with higher sensitivity toward intended use is not permitted by statutory regulation or exceeds the permitted use, you will
genotoxic compounds than constructs based on recA, umuDC, or sulA promoters. Appl. need to obtain permission directly from the copyright holder. To view a copy of this licence,
Environ. Microbiol. 71, 2338–2346 (2005). visit [Link]
37. Malik, M., Hussain, S. & Drlica, K. Effect of anaerobic growth on quinolone lethality with
Escherichia coli. Antimicrob. Agents Chemother. 51, 28–34 (2007). © The Author(s) 2025

Nature | Vol 639 | 6 March 2025 | 189

Article
Methods the antibiotic. Plating of the resuspended pellets on LB plates without
diaminopimelic acid enabled growth of Salmonella but not the mixed-in
Microbiology and molecular biology diaminopimelic acid-dependent E. coli strain.
Salmonella strains used in this study were based on S. enterica serovar We used S. aureus strain PROSA28, a strain of clonal complex 45 iso-
Typhimurium SL1344 hisGL69P xyl38,58. A histidine-prototrophic deriva- lated from a patient with an implant-associated joint infection at Univer-
tive SL1344 hisGP69L and mutants ΔtisB and ΔecnB ΔshpAB ΔphD-doc sity Hospital Basel. We transformed this strain with a pRN11-derivative67
have been previously described3. Salmonella lexA3 were generated carrying a transcriptional fusion of the PpdhABCD promoter and gfp-mut3.1
by generalized transduction using P22 phage JS841 ΔlexA33::[Cm obtained from plasmid pC183-S368. S. aureus was grown in BHI medium
lexA3(Ind−)](sw)59,60. gfp-expressing strains carried gfp.mut2 in the chro- containing 10 mg l−1 chloramphenicol, or in BHI medium diluted 1:50
mosomal sifB locus that has homogenous high activity in vivo38,58. The with saline (for microfluidics experiments).
recAR29A-mCherry DNA damage sensor strain carried a pSC101-derivative
with the native PrecA promoter (SL1344 chromosome positions 2,998,612 Mouse infections and tissue collection
to 2,998,642) with a modified ribosomal binding site (AGGAA instead of All animal experiments were approved (license 2239, Kantonales
AGGAG) to reduce translation to levels just sufficient for imaging; the Veterinäramt Basel) and performed according to local guidelines
entire SL1344 recA gene without stop codon and a change of codon 29 (Tierschutz-Verordnung, Basel) and the Swiss animal protection law
from CGT (encoding arginine) to GCG (encoding alanine)31; a GGGAGC (Tierschutz-Gesetz). We estimated sample size by a sequential statis-
ATC linker encoding Gyl-Ser-Ile31; and mCherry without the start ATG. tical design. We first infected two to three mice each based on effect
The SOS reporter strain carried mCherry in the chromosomal virK sizes and variation observed in our previous studies5, and used the
locus61 for detecting all bacteria regardless of their SOS response and results to estimate group sizes for obtaining statistical significance
a pSC101-derivative with a transcriptional fusion of the SOS-inducible with sufficient power.
Pcad promoter36 of colicin D62 to gfp-ova coding for a degradable green Female 10- to 16-week-old BALB/c mice (Charles River Laborato-
fluorescent protein variant61 (half-life in the range of 30 min). The Pcad ries) were infected by tail-vein injection of ~1,000 CFU Salmonella
promoter was obtained as synthetic DNA corresponding to bases 6389 grown to late-log phase in Lennox LB containing 10 mM MgCl2 (which
to 6675 of pColD-157 (Genbank Y10412.1)63 with base 6621 ‘C’ instead increased consistency across experiments). The inoculum size was
of ‘T’ and base 6628 ‘T’ instead of ‘C’36 followed by TTAAAAGTCAAAG determined by plating. Intravenously infected mice show similar Sal-
AGGTGTTTTTGC, containing the ribosomal binding site upstream of monella growth rates3 and Salmonella localization in spleen69 compared
cda in pColD-CA2336,62. hipA from E. coli BW25113 was mutagenized to orally infected mice but exhibit less variation in Salmonella tissue
to obtain hipA D88N and expressed in Salmonella hisGP69L from the loads between individual mice, and thus require fewer experimen-
constitutive PybaJ promoter64 with a suboptimal ribosomal binding tal animals for detecting differences with the same statistical power.
site AAGAG together with PybaJ-timerbac (ref. 3) on a pSC101-derivative. Some mice received at day 5 post-infection an intraperitoneal injection
Salmonella hisGP69L shpB1 was constructed by two consecutive single of 5 mg kg−1 enrofloxacin or 50 mg kg−1 ceftriaxone. For plating, mice
cross-overs65 to change codon 97 from CAA to TAA, resulting in a Q97* were euthanized with carbon dioxide and spleen was homogenized
mutation, which truncates ShpB by 4 C-terminal amino acids40. in PBS containing 0.2% Triton X-100. Salmonella load was determined
Salmonella was grown in Lennox LB containing 90 mg l−1 strepto- by plating. Salmonella survival after enrofloxacin administration was
mycin and 50 mg l−1 kanamycin, or in cation-adjusted Mueller–Hinton determined by comparing flow cytometry counts and corresponding
broth. To mimic in vivo conditions, we cultured Salmonella in chemo- CFU counts of sorted Salmonella populations as described12. Salmonella
stat medium (100 mM MES, 5 mM KCl, 15 mM NH4Cl, 0.5 mM K2SO4, survival after ceftriaxone administration was determined by plating.
1 mM KH2PO4, 50 μM MgSO4, 0.02% casamino acids, 0.02% glycerol,
0.0042% N-acetyl-glucosamine, 0.003% glucose, 0.0018% glucosamine, Randomization. Control and experimental animals were co-housed.
0.005% histidine, 25 mM NaHCO3, pH 5.5; sterile filtered) with a constant We ensured that litter mate or age-matched and healthy animals with
stream of 10% O2/5 % CO2 in mini-chemostats66 at various dilution rates identical sex were used in all experiments. In order to reduce the impact
(modifications to these conditions are explained in Fig. 2a). Hydrogen of covariates such as housing and litter size, animals were recruited in a
peroxide was added 30 min before antibiotic treatment. Growth con- partially randomized manner while taking these factors into account.
ditions were maintained for 72 h before addition of antibiotics. For Control and treated animals were infected with the same inoculum
growth in microfluidics devices, we used the same chemostat medium to control for bacterial variation. Data derived from animals were
with 1,000-fold decreased levels of all carbon and energy sources. pooled by genotype and/or condition after analyses were completed.
To determine killing of diluted batch cultures, we diluted overnight Comparisons of isogenic Salmonella strains in mice were performed
cultures 1:1,000 in fresh pre-warmed medium followed by growth for using competitive (mixed) infections to control for variation between
3 h before antibiotic treatment. For comparing Salmonella hisGP69L and experimental animals.
Salmonella hisGP69L shpBQ97*, we grew day cultures for only 1 h before
antibiotic treatment as described40. To minimize experimental variation Blinding. Data acquisition for drug treatments could not be blinded
in these experiments due to undefined stationary cultures, we stand- because the comparisons were made as pre- versus post-drug treat-
ardized the preparation of stationary cultures by inoculating primary ment, the treatment sequence was essential, and often only one drug
overnight cultures from 1-day plate cultures, diluting the cultures after was applied. For animal welfare reasons, researchers were not blinded to
overnight growth 1:1,000 in fresh pre-warmed medium, growing these mouse genotype during study and data collection. Specifically, Slc11a1s
primary day cultures until exponential phase to an OD600 of ~0.2, adjust- mice have to be euthanized at day 4 post-infection to prevent high sever-
ing them to OD600 = 0.1 followed by dilution 1:1,000 in fresh pre-warmed ity grades, whereas Slc11a1r mice reach similar bacterial loads only at
medium, growth for 14 h (secondary defined overnight culture), dilut- day 6 post-infection. All other data were collected and analysed without
ing them again 1:1,000 in fresh pre-warmed medium and treated them blinding but objectively, using instruments without bias and analysis
with enrofloxacin 1 h later. To remove antibiotics from dilute cultures definitions that were uniformly applied to all data sets.
with minimal loss of Salmonella during washing, we mixed the bacterial
culture with ~109 CFU diaminopimelic acid-dependent E. coli JKe20165. Flow cytometry
These additional bacteria provided sufficient cell mass for a visible and Spleen was homogenized in ice-cold PBS containing 0.2% Triton X-100.
physically stable cell pellet after centrifugation, permitting recovery All samples were kept on ice until analysis. Large host cell fragments
of >90% of the surviving Salmonella after two washing steps to remove were removed by centrifugation at 500g for 5 min. Relevant spectral
parameters were recorded in a FACS Fortessa II operated using BD
FACSDIVA V8.0.1 software and equipped with 405 nm, 488 nm and Reporting summary
561 nm lasers (Becton Dickinson), using a threshold on side scatter Further information on research design is available in the Nature Port-
to exclude electronic noise. We used the following channels: GFP and folio Reporting Summary linked to this article.
green TIMER component, excitation 488 nm, emission 502–525 nm;
mCherry and orange TIMER component, excitation 561 nm, emission
595–654 nm with red laser (633 nm) switched off; red autofluorescence Data availability
channel, excitation 488 nm, emission 663–677 nm (the gating strategy Data points generated for this study are included in the figures when-
for mCherry-Salmonella is shown in Extended Data Fig. 2c). Salmonella ever possible. Tabulated data for all figures, videos of the microfluidics
cells were purified from infected spleen homogenates using an Aria experiments and flow cytometry data are available at [Link]
IIIu cell sorter (BD Biosciences) using excitation 488 nm and emission [Link]/biostudies/studies/S-BSST1727. Source data are provided
channels 499–529 nm (predominantly GFP) and 573–613 (predomi- with this paper.
nantly host autofluorescence) as well as excitation 561 nm and emis-
sion 595–617 nm. Defined sample volumes of the sorted Salmonella 58. Kroger, C. et al. The transcriptional landscape and small RNAs of Salmonella enterica
serovar Typhimurium. Proc. Natl Acad. Sci. USA 109, E1277–E1286 (2012).
were re-analysed by flow cytometry and plating to determine survival 59. Bunny, K., Liu, J. & Roth, J. Phenotypes of lexA mutations in Salmonella enterica: evidence
rates12. Data were processed with FlowJo 10.6.1 and further analysed for a lethal lexA null phenotype due to the Fels-2 prophage. J. Bacteriol. 184, 6235–6249
with OriginPro 2019 (64-bit) [Link]. (2002).
60. Craig, M. & Slauch, J. M. Phagocytic superoxide specifically damages an extracytoplasmic
target to inhibit or kill Salmonella. PLoS ONE 4, e4975 (2009).
Microfluidics experiments 61. Rollenhagen, C., Sorensen, M., Rizos, K., Hurvitz, R. & Bumann, D. Antigen selection based
on expression levels during infection facilitates vaccine development for an intracellular
Salmonella expressing recAR29A-mCherry were grown overnight in che­
pathogen. Proc. Natl Acad. Sci. USA 101, 8739–8744 (2004).
mostat medium, diluted 1:50 in chemostat medium with 100-fold 62. Roos, U., Harkness, R. E. & Braun, V. Assembly of colicin genes from a few DNA fragments.
decreased concentrations of carbon and energy sources, grown for 6 h, Nucleotide sequence of colicin D. Mol. Microbiol. 3, 891–902 (1989).
63. Hofinger, C., Karch, H. & Schmidt, H. Structure and function of plasmid pColD157 of
inoculated in CellASIC ONIX B04A-03 microfluidic plates (Millipore)
enterohemorrhagic Escherichia coli O157 and its distribution among strains from patients
operated in a CellASIC ONIX Microfluidic System, and grown for 6 h in with diarrhea and hemolytic–uremic syndrome. J. Clin. Microbiol. 36, 24–29 (1998).
chemostat medium with 1,000-decreased concentrations of carbon/ 64. Rollenhagen, C. & Bumann, D. Salmonella enterica highly expressed genes are disease
specific. Infect. Immun. 74, 1649–1660 (2006).
energy sources (generation time ~3 h), before exposing them to enro-
65. Cianfanelli, F. R., Cunrath, O. & Bumann, D. Efficient dual-negative selection for bacterial
floxacin in the same medium. After various time intervals bacteria were genome editing. BMC Microbiol. 20, 129 (2020).
switched back to chemostat medium with 1,000-fold decreased con- 66. Nanchen, A., Schicker, A. & Sauer, U. Nonlinear dependency of intracellular fluxes on
growth rate in miniaturized continuous cultures of Escherichia coli. Appl. Environ. Microbiol.
centrations of carbon/energy sources, and to Lennox lysogeny broth.
72, 1164–1172 (2006).
gfp-expressing S. aureus were grown overnight in BHI medium diluted 67. de Jong, N. W., van der Horst, T., van Strijp, J. A. & Nijland, R. Fluorescent reporters
1:50 with saline, diluted 1:50 in fresh medium, grown for 4 h, and inocu- for markerless genomic integration in Staphylococcus aureus. Sci. Rep. 7, 43889
(2017).
lated in the microfluidic plates. After growth for 4 h in 1:50 BHI (generation
68. Burian, M. et al. Temporal expression of adhesion factors and activity of global regulators
time ~1.5 h), bacteria were exposed to various time intervals to 12 mg l−1 during establishment of Staphylococcus aureus nasal colonization. J. Infect. Dis. 201,
flucloxacillin in the same medium, and then switched to normal BHI. 1414–1421 (2010).
69. Burton, N. A. et al. Disparate impact of oxidative host defenses determines the fate of
Bacteria were imaged using a Nikon Eclipse Ti2 inverted micro- Salmonella during systemic infection in mice. Cell Host Microbe 15, 72–83 (2014).
scope equipped with Perfect Focus System and Plan Apo 100× Oil 70. Schneider, C. A., Rasband, W. S. & Eliceiri, K. W. NIH Image to ImageJ: 25 years of image
Ph3 DM (NA 1.45) objective lens (MRD31905), SPECTRA X light engine analysis. Nat. Methods 9, 671–675 (2012).
71. Thévenaz, P., Ruttimann, U. E. & Unser, M. A pyramid approach to subpixel registration
(Lumencor), phase contrast (8% light intensity, 200 ms exposure) and based on intensity. IEEE Trans. Image Process. 7, 27–41 (1998).
a Dualband EGFP/mCherry filter set (Chroma 59022 ET; for S. aureus, 72. Becker, D. et al. Robust Salmonella metabolism limits possibilities for new antimicrobials.
470 nm excitation, 4% light intensity, 150 ms exposure; for Salmonella, Nature 440, 303–307 (2006).
73. Volkmer, B. & Heinemann, M. Condition-dependent cell volume and concentration of
575 nm excitation, 5% light intensity, 250 ms exposure), a Hamamatsu Escherichia coli to facilitate data conversion for systems biology modeling. PLoS ONE 6,
ORCA-Flash4.0 V3 CMOS camera (C13440-20CU) with pixel size of e23126 (2011).
65 nm, and NIS-Elements (Nikon). Images were acquired at 1 min (Salmo-
nella) or 5 min (S. aureus) intervals at 37 °C with 95% humidity controlled
Acknowledgements The authors thank J. Slauch for the lexA3 phage lysate; R. Kühl and
by Okolab T-unit (Okolab). Images were analysed with ImageJ 1.53q70 N. Khanna for strain S. aureus PROSA28; R. Nijland for plasmid pRN11; A. Peschel for plasmid
using plugins MultiStackReg71 and FeatureJ-Laplacian (Smoothing pC183-S3; and M. Basler, U. Jenal, R. Kühl, C. Dehio, M. Zampieri, T. Tenson, N. Kaldalu and
scale 2.0) (developed by E. Meijering). M. Putrinš for helpful discussions. D.B. received support from Schweizerischer Nationalfonds
(51NF40_180541 NCCR AntiResist, 310030_156818 and NRP 72−177449).
The decay of undamaged cells in Fig. 3i appeared to occur in three
stages based on discontinuities in the first derivative of log-transformed Author contributions Experiments were conceived and designed by J.F., V.T., F.G. and D.B.
data. We fitted each stage separately with monoexponential decays Chemostat and batch culture experiments were performed and analysed by J.F. and B.C.
Mouse infections were carried out and analysed by B.C. and J.L. Microfluidics experiments
using GraphPad Prism 9.3.1. Data were further analysed with OriginPro
were done by V.T. and F.G. and analysed by V.T. and D.B. Figure design, manuscript writing and
2019 (64-bit) [Link]. editing were carried out by D.B. with input from all authors. Project supervision and funding
were undertaken by D.B.
Statistics
Funding Open access funding provided by University of Basel.
Statistical tests were performed with GraphPad Prism 9.3.1 as indi-
cated in the figure legends. We always used two-tailed tests because Competing interests The authors declare no competing interests.
we were interested to also test for effects that go into the opposite
Additional information
direction to what we might have predicted. CFU counts approximate Supplementary information The online version contains supplementary material available at
normal distributions after log transformation72, thus permitting para- [Link]
metric test statistics. Multiple linear regression was carried out after Correspondence and requests for materials should be addressed to Dirk Bumann.
Peer review information Nature thanks the anonymous reviewer(s) for their contribution to the
log-transforming values for dependent and independent variables peer review of this work.
(except for pH). Reprints and permissions information is available at [Link]
 Article
a Re-analysis
Large
d 1
PsifB-gfp
b

P= 3.4e-5

P= 1.5e-7
1

Normalized CFU
104
Green fluorescence

0.1
Sort Medium

OD600
0.1
0.01 WT
104 104
hisGP69L shpBQ97*
hipAD88N
0.001

Slc11a1s / WT
103 Small 0 5 10 15 20
Time in h Slc11a1s / hipAD88N

e
Sideward scatter
104 Phase contrast Slc11a1r / WT

c Slc11a1s / WT
103

Slc11a1s / hipAD88N 2 μm

P= 1.3e-6
Frequency

Divisions per h

Slc11a1r / WT
Fluorescence Laplacian
Frequency

0.10
P= 1.5e-4

0.05

0.00 False color False color

0
0.0 0.1 0.2 0.3 0.4

Divisions per h

f h
01:55 04:39 DSB 05:31 06:18
Control 5 μm
Undamaged fract.

1 DSB
1.5 mg/L DSB Division
5 mg/L DSB
DSB
0.1 DSB Division
DSB
0 1 2 3 4
Time in h
MM DSB 9 min LB 61 min LB 108 min LB
g
i j
P= 1.5e-8
LB
Chemostat 1
Sideward scatter

Mouse spleen 16
Dividing fraction

10,000
Undamaged fract.

P= 0.003
Repair time in h

0.1
4
P= 3.1e-9
P= 0.72
P= 0.65 0.01
LoD LB 1
1,000
- + - + - + 0 2 4 6
Enrofloxacin treatment Time in h DSB- DSB+ DSB+ DSB+

No ENR/LB 1h/LB IVM

Extended Data Fig. 1 | See next page for caption.

Extended Data Fig. 1 | Analysis of Salmonella in vitro. (a) Sorting of Salmonella exposure from t = 0 h to 1.5 or 5 mg/L enrofloxacin (130, 250 cells). The dotted
cells with different sizes. Salmonella expressing gfp under control of the lines represent monoexponential fits. The data for 5 mg/L are also shown in
chromosomal PsifB promoter61 showed correlated sideward-scatter (a proxy for Fig. 3c. Summary data for the 5 mg/L exposure and an independent replicate are
size73) and green fluorescence signals. The combination of the two detection shown in Fig. 3d. (g) Flow-cytometry analysis of sideward scatter as a read-out
channels increased resolution and enabled enrichment of subsets by flow for cell size73 of Salmonella before and after 1 h exposure to enrofloxacin during
cytometric sorting using the gate boundaries shown as dashed red lines. exponential growth in lysogeny broth (LB), slow growth in tissue-mimicking
(b) Growth curves in lysogeny broth in 96-well plates. Means and SDs for 4 chemostats, or in mouse spleen. Each symbol represents an independent
(wild-type, WT), 5 (hipAD88N), or 6 (shpBQ97*) biological replicates (each measured culture or an individual mouse (LB +/− n = 3; Chemostat +/− n = 5; Mouse spleen
in technical triplicates) are shown (OD600, optical density at 600 nm; Det. Lim., +/− n = 3; two-tailed t-test of log-transformed data with Holm-Šídák correction
detection limit). (c) Division rate distributions of Salmonella strains in different for multiple comparisons). (h) Snap-shots of Salmonella growing in nutrient-
mouse genotypes (data for wild-type Salmonella in SLC11A1r mice from4, poor minimal medium (MM) without enrofloxacin and then in lysogeny broth
n = 61,137 cells from seven independently infected mice; SLC11A1s, WT n = 34,851 (LB; video S1). (i) Fraction of undamaged cells before and after switching to LB,
from three independently infected mice; SLC11A1s, hipA D88N n = 832 from and dividing Salmonella after the LB switch (146 cells; LoD, limit of detection).
three independently infected mice). The inset shows medians for the individual RecA foci during the first 4 h are also shown in panel f (‘Control’). The dotted
mice (SLC11A1s WT/hipAD88N n = 3; SLC11A1r, WT n = 7; ANOVA with comparison line is a monoexponential fit for damage before the LB switch. (j) Time difference
against SLC11A1s, WT; P-values adjusted for multiple comparisons according between detection of a DNA double-strand bread (DSB) based on a long-lasting
to Holm-Šídák). (d) Survival of Salmonella strains in mouse spleen 1 h after RecA focus, and re-initiation of cell division. Control cells without enrofloxacin
administration of 0.1 mg enrofloxacin. Each circle represents an individual exposure that were switched to lysogeny broth (No ENR/LB: DSB- n = 40; DSB+
mouse (SLC11A1s: WT n = 10; hipAD88N n = 4; SLC11A1r, WT n = 7) two-tailed t-test, n = 69), cells exposed for 1 h to enrofloxacin followed by a switch to LB (1 h/LB
P-values adjusted for multiple comparisons according to Holm-Šídák). n = 35), and cells exposed to declining concentrations of enrofloxacin over 7 h
(e) Analysis of Salmonella/pRecA-mCherry fluorescence images. Detection followed by continuing incubation in nutrient-poor medium (‘in vivo-mimicking’,
of RecA-mCherry foci was enhanced by 2D Laplacian of Gaussian filtering IVM n = 115) are shown. Each circle represents an individual cell. The statistical
(approximating a second derivative) of red fluorescence and false-color difference between groups was test using the Kruskal-Wallis-test with Dunn’s
visualization. (f) Fraction of cells with no long-lasting RecA focus (“undamaged correction for multiple testing.
fraction”) during growth without enrofloxacin (Control, 146 cells), or
Article

Extended Data Fig. 2 | See next page for caption.

Extended Data Fig. 2 | Salmonella analysis in vivo. (a) Traditional model and green fluorescence as shown in Fig. 3h. (d) Green fluorescence of Salmonella
actual mechanism underlying rapid loss of colony-forming units. In the retrieved from mouse spleen 4 h after enrofloxacin administration and
traditional models, a small subset of persisters, some of which are triggered incubated for 3 h in lysogeny broth (LB). Control Salmonella without gfp are
by stresses, survive antibiotic exposure and give rise to colonies on plates. also shown. The data represent histograms pooled from 2 mice. (e) Colony
However, antibiotic exposure actually kills only few cells, but survivors are recovery of GFP− (SOS−) and GFP+ (SOS+) Salmonella sorted from spleen
poisoned by drug, which remains bound to its target (depicted as yellow homogenates from untreated mice (0 h), or from treated mice 1 h or 4 h after
symbols). Most of these poisoned cells die after plating. Thus, colony counts enrofloxacin administration. Survival was determined by plating on lysogeny-
reflect mostly post-exposure killing rather than viability loss during exposure. broth plates. Each circle represents an individual mouse (untreated n = 14; 1 h
(b) Plasmid map of the SOS-reporter construct. The Pcad promoter is repressed GFP +/− n = 6; 4 h GFP +/− n = 4). The statistical difference of GFP- and GFP+
by LexA 36. LexA self-cleavage (SOS-response) is associated with unbinding contributions from treated mice was tested by two-way ANOVA (column factor,
of LexA to DNA, de-repressing Pcad . gfp-ova encodes a destabilized GFP variant two-tailed, single comparison). (f) Survival of Salmonella strains in enrofloxacin-
to report current promoter activities. aphA encodes aminoglycoside treated mice (hisGP69L: 1 h n = 10; 4 h n = 6; hisGP69L shpBQ97* 1 h/4 h n = 3; geometric
phosphotransferase conferring resistance to kanamycin. repA encodes RepA means and geometric SDs; two-way ANOVA of log-transformed data).
which replicates plasmids with a oriSC101 origin (T, terminator). (c) Gating (g) Survival of Salmonella strains in lysogeny broth with enrofloxacin (geometric
strategy for identifying Salmonella reporter cells in spleen homogenates based means and geometric SDs; six independent cultures from two different
on the red fluorescence of chromosomally encoded mCherry (Ex, excitation experiments; two-way ANOVA of log-transformed data).
wavelength; Em, emission range). mCherry-Salmonella were then analyzed for
Article
Extended Data Table 1 | Impact of various stresses on fluoroquinolone activity in vitro

, maximum fold difference in survival within tested range; Std. Error, standard error.
1