Isolation of persisters enabled by ß-lactam-induced filamentation reveals their single-cell awakening characteristics

When exposed to lethal doses of antibiotics, bacterial populations are most often not completely eradicated. A small number of phenotypic variants, defined as ‘persisters’, are refractory to antibiotics and survive treatment. Despite their involvement in relapsing infections caused by major pathogens, processes determining phenotypic switches from and to the persister state largely remain elusive. This is mainly due to the low frequency of persisters in a population and the lack of reliable persistence markers, both hampering studies of persistence at the single-cell level. Problematically, existing methods to enrich for persisters result in samples with very low persister densities and/or a too high abundance of other cell types. Here we present a novel and highly effective persister isolation method involving cephalexin, an antibiotic that induces extensive filamentation of susceptible cells. We show that antibiotic-tolerant cells can easily be separated by size after a short cephalexin treatment, and that the isolated cells are genuine persisters. We used our isolation method to monitor persister outgrowth at the single-cell level in a microfluidic device, thereby conclusively demonstrating that awakening is a stochastic phenomenon. We anticipate that our novel approach can have far-reaching consequences in the persistence field, by allowing single-cell studies at a much higher throughput than previously reported.


37
Persisters are phenotypically distinct variants in a microbial population that survive a lethal antibiotic 38 dose and are able to regrow after treatment ceases [1,2]. Given this population heterogeneity, 39 interrogation of the persister physiology should rely on single-cell studies to properly capture their 40 defining traits. However, these studies require considerable and fast enrichment of persisters as they 41 are usually present at low frequencies and known to be in a metastable state. Problematically, apart 42 from their antibiotic tolerance, no reliable marker currently exists to distinguish persisters from 43 normal, susceptible cells. The state-of-the-art method to enrich for persisters involves lysis of 44 susceptible cells by ampicillin, followed by sedimentation of intact persister cells [3]. Due to the poor 45 separation efficiency during sedimentation, this method fails to efficiently remove cell debris and 46 results in a persister density that is most often too low for microscopic studies. Furthermore, prolonged 47 exposure of the culture to antibiotics or dead cell material could potentially affect persister formation 48 [4][5][6]. The latter problem was addressed by Cañas-Duarte et al., who optimized a method to rapidly 49 lyse susceptible cells using a chemo-enzymatic lysis solution [7]. Problematically, they did not validate 50 antibiotic tolerance of their isolated cells, nor did they report the purity and density of the resulting 51 sample. Other approaches using GFP expression, RpoS::mCherry expression, or the RNA-binding 52 Thioflavin T as fluorescent markers for persistence, make too strong assumptions on the physiological 53 state of persisters and therefore generate samples that are highly contaminated with normal, 54 susceptible cells [8][9][10]. Attempts to enrich persisters using chemical pretreatment [6] or strains that 55 are engineered to accumulate toxins [11] potentially generate artefacts that confound insights in 56 naturally occurring persistence. 57 In this study, we established a novel, highly efficient persister isolation method that largely addresses 58 the challenges posed by single-cell persistence studies. We show that persisters can be effectively 59 isolated by filtration after ß-lactam-induced filamentation. Cells isolated in this way are bona fide 60 persisters as they survive during antibiotic treatment, regrow after treatment, and exhibit tolerance 61 4 towards antibiotics with different targets. We then used our isolation method to resolve a key 62 outstanding question in the persistence field. Single-cell recovery of persisters after treatment was 63 monitored in a microfluidic 'mother machine' device. These data show that persister awakening occurs 64 at a constant rate, reflecting stochasticity. Our novel approach might prove useful for future single-cell 65 studies of persistence. 66 normal cells based on their antibiotic tolerance, the core feature that universally characterizes all 70 persisters and does not make any assumptions on their physiological state or underlying mechanisms. 71

Results and discussion
Our approach is specifically aimed at limiting the amount of cell debris in the resulting sample, as well 72 as shortening the antibiotic exposure time. To this end, we benefit from the killing characteristics of 73 cephalexin, a ß-lactam that does not immediately induce lysis, but first induces severe filamentation 74 of susceptible cells before lysis is initiated (Figure 1a; Suppl. Movie 1). Cephalexin targets penicillin-75 binding protein (PBP) 3, also known as FtsI, a transpeptidase that is essential for peptidoglycan 76 synthesis during cell division [12]. Drug-tolerant persisters are not affected by cephalexin and 77 therefore do not filament in its presence, enabling their isolation from a culture by filtration ( Figure  78 1b). 79 ß-lactams only exhibit effective activity at low cell densities [13], implying that cultures should be in 80 early exponential phase when cephalexin treatment starts. The biphasic killing pattern resulting from 81 a long-term treatment with cephalexin confirms the presence of persisters in this low-density culture 82 (Suppl. Figure S1a). By comparing the number of persisters isolated with our filtration method to the 83 total number of persisters at the plateau of the time-kill curve (Suppl. Figure S1a), we estimated that 84 isolation occurs with an average efficiency of 28 %. The remaining persisters are presumably lost during 85 filtration, as filamented cells cause clogging of the filter. 86 Notably, the fact that filamentation occurs at a much shorter timescale than lysis considerably reduces 87 the antibiotic exposure time as compared to the ampicillin lysis method. This was confirmed by 88 performing our filtration protocol at different time points during a longer-term cephalexin treatment 89 (Suppl. Figure S2a). These data show that the number of isolated cells does not change significantly for 90 cephalexin treatments longer than one hour (p=0.17), implying that a one-hour treatment is sufficient 91 to obtain the persisters of the culture by filtration. Any treatment shorter than one hour results in 92 contamination with susceptible cells, while longer treatments successively generate more debris of 93 dead cells in the sample (Suppl. Figure S2b). Indeed, an optimal treatment time of one hour results in 94 a final sample that contains short, antibiotic-tolerant persisters and very little cell debris (Figure 1c). 95 The purity of the resulting samples was also confirmed by the side scatter distributions of samples 96 before and after filtration (Figure 1d). 97

119
Importantly, these microscopic observations additionally demonstrate that the isolated cells cannot 120 grow in the presence of cephalexin, implying that they are not genetically resistant. We then also 121 validated that cells isolated by filtration are able to reinitiate growth, by seeding them onto an agarose 122 pad supplemented with rich medium (Figure 2c be partially attributed to killing of persisters as they wake up during treatment. However, these data 140 presumably also imply that not all persisters are tolerant to all antibiotics, but rather represent a 141 heterogeneous pool of cells with partially overlapping tolerance to various antibiotics. Together, our 142 data show that cells isolated by cephalexin treatment and filtration are tolerant towards a longer-term 143 cephalexin treatment, that they are able to reinitiate growth when treatment ceases, and that they 144 show a high degree of multidrug tolerance. All these characteristics are key to the persister phenotype 145 and make us confident that our isolated cells are bona fide persister cells. 146 Single-cell analysis of isolated persisters in the mother machine reveals that persister awakening is 147 a stochastic process 148 To microscopically examine single persister cells and their regrowth, a major drawback of using 149 agarose pads is that early-dividing cells quickly overgrow the whole pad.

166
We isolated persisters from a culture using the filtration method described above, inserted these cells 167 into the channels of the mother machine, and provided them with fresh nutrients (Figure 3a). Most of 168 the channels contained either no or only one cell, allowing to track single cells. Based on a few hundred 169 individual observations, we derived a distribution of single-cell persister awakening times for the wild-170 type E. coli strain K-12 MG1655 and the well-known high-persistence strain hipA7 (Figure 3b). A similar 171 distribution was obtained for both strains. This distribution shows a surprisingly high cell-to-cell 172 variability in awakening times, ranging from a few minutes to up to 13 hours. In both cases, an 9 exponential curve fits well to the data, indicative of a high degree of stochasticity involved in persister 174 awakening. The persister awakening rate in fresh medium without antibiotics (b = 0.31-0.35; Figure  175 3b) is higher than in the presence of cephalexin (kp = 0.04-0.18; Figure 2a; Suppl. Figure S1), although 176 both rates were measured in different setups and therefore not perfectly comparable. While these 177 findings corroborate existing assumptions and hypotheses about persister awakening [17,18], this is, 178 to our knowledge, the first study that provides conclusive experimental evidence of stochastic 179 awakening at single-cell level with such a high throughput. In addition to the awakening times, we also 180 derived individual growth rates of freshly-awakened persisters (Suppl. Figure S3). Strikingly, these data 181 reveal that persisters instantaneously divide at a rate that does not differ from the growth rate of 182 normal cells. Furthermore, individual growth rates do not correlate with awakening times, indicating 183 that cells with a long lag time do not necessarily grow slower than cells with a short lag time (Figure  184 3c). It should be noted that the majority of the cells did not start dividing within the course of the 185 experiment (20 hours). As these cells are too numerous to be completely covered by the tail of the 186 exponential distribution, they can presumably be classified as viable but non-culturable cells (VBNCs). 187 The high abundance of VBNCs in E. coli cultures has already been reported before [19,20] and 188 represents a prominent source of contamination in most persistence enrichment protocols, including 189 ours. As VBNCs cannot be distinguished from persisters based on their antibiotic tolerance, our method 190 is only able to discriminate between both by visualizing regrowth in fresh medium.

Isolation of persisters 216
A 20-hour overnight culture was diluted 1:10,000 in 100 ml Mueller-Hinton broth (MHB) and incubated 217 for 20 hours. This culture was diluted 1:5,000 in 100 ml MHB and grown for 90 minutes, to a density 218 of 1-2 x 10 6 CFU/ml. Next, the culture was treated with cephalexin (50 µg/ml) for 60 minutes, after 219 which it was poured twice over a polyvinylidene fluoride membrane filter (Merck Millipore) with a pore 220 size of 5 µm. The filtrate was collected in falcons and spun down (4,000 rpm -5 min). After pouring off 221 the supernatant, the remaining volume was transferred to a microcentrifuge tube and centrifuged 222 twice (6,000 rpm -5 min) to wash away the remaining antibiotic. The pellet was resuspended in MgSO4 223 (10 mM). 224

Flow cytometry 239
A 20-hour overnight culture was diluted 1:10,000 in 100 ml MHB and incubated for 20 hours. This 240 culture was diluted 1:5,000 in 100 ml MHB and grown for 90 minutes, to a density of 1-2 x 10 6 CFU/ml. 241 Next, the culture was treated with cephalexin (50 µg/ml) for 60 minutes.

Microscopy of agarose pads 249
To visualize killing by cephalexin, a 20-hour overnight culture was diluted 1:5,000 and incubated for 90 250 minutes. The resulting exponential phase culture was washed with MgSO4 and 2 µl of cells was seeded 251 onto an MHB+agarose pad (2% w/v) containing cephalexin (50 µg/ml). Cells were incubated at 37 °C 252 and killing was monitored for 6 hours. Images were obtained using a Nikon Ti-E inverted microscope 253 with a 60x objective. 254 13 To visualize persisters, cells were isolated as described above. The resulting sample was resuspended 255 in 10 µl MgSO4 and 2 µl of cells was seeded onto an MHB+agarose pad (2% w/v) with or without 256 cephalexin (50 µg/ml). Cells were incubated at 37 °C and growth was monitored for 12 hours. Images 257 were obtained using a Nikon Ti-E inverted microscope with a 60x objective. 258 project.org/web/packages/lme4/index.html) was used to fit the equation Log10(CFU)=Log10((N0+P0).e -286 k.τ ). AIC (Akaike Information Criterion) was used to assess both models. 287

Multidrug tolerance 288
Surviving fractions were compared between conditions using one-way ANOVA and post-hoc 289 comparisons with Sidak's correction for multiple testing. 290

Distribution of awakening times 291
The variable 'persister awakening time' was split into bins of 60 minutes. The number of observations 292 in each bin was normalized, to obtain relative frequencies (freq) of awakening events. The nls function 293 in R was used to fit an exponential distribution with equation Log10(freq) = Log10(b.e -b.τ ) onto the data, 294 with τ the time in fresh medium and b the rate of awakening. After checking normality with a Shapiro-295 Wilk test, Log10-transformed awakening times were compared statistically among different strains or 296 cell types using an unpaired, two-sided t-test, with Welch correction in the case of unequal variances 297 (checked with an F-test). 298

Growth rates 299
A piecewise linear function was fitted to the cumulative number of divisions over time, for each 300 individual cell. The number of knots was chosen by cross-validation and most often corresponds to 301 one, dividing the growth curve into a lag phase and exponential growth phase. The slope of the second 302 15 curve was then used to derive the average growth rate of individual cells. After checking normality 303 with a Shapiro-Wilk test, growth rates were compared statistically among different strains using an 304 unpaired, two-sided t-test, with Welch correction in the case of unequal variances (checked with an F-305 test). 306

307
The data that support the findings of this study are available from the corresponding author upon 308 request. 309