Single-cell protein induction dynamics reveals a period of vulnerability to antibiotics in persister bacteria

Abstract

Phenotypic variability in populations of cells has been linked to evolutionary robustness to stressful conditions. A remarkable example of the importance of cell-to-cell variability is found in bacterial persistence, where subpopulations of dormant bacteria, termed persisters, were shown to be responsible for the persistence of the population to antibiotic treatments. Here, we use microfluidic devices to monitor the induction of fluorescent proteins under synthetic promoters and characterize the dormant state of single persister bacteria. Surprisingly, we observe that protein production does take place in supposedly dormant bacteria, over a narrow time window after the exit from stationary phase. Only thereafter does protein production stop, suggesting that differentiation into persisters fully develops over this time window and not during starvation, as previously believed. In effect, we observe that exposure of bacteria to antibiotics during this time window significantly reduces persistence. Our results point to new strategies to fight persistent bacterial infections. The quantitative measurement of single-cell induction presented in this study should shed light on the processes leading to the dormancy of subpopulations in different systems, such as in subpopulations of viable but nonculturable bacteria, or those of quiescent cancer cells.

Recent technological advances have made possible the measurements of cell-to-cell variability in populations of cells (1) and revealed a wealth of phenomena previously concealed in bulk experiments (2, 3). That cells can behave differently, despite uniform genetic and environmental conditions, has profound implications for our understanding of evolutionary processes (4, 5), and of design principles of genetic networks (6). Nongenetic individuality in bacteria has been observed long ago in the response to external signals (7, 8). Recently, the heterogeneous response of bacterial populations to antibiotics, termed bacterial persistence, has been linked to their inherent nongenetic variability. It has become increasingly obvious that persistence is the main problem in diseases such as tuberculosis, where the immune system is not effective, and a single bacterium can start an infection (9), and in recurrent urinary tract infections (10). Persistence has also been invoked to explain the recalcitrance of biofilms to antimicrobials (11). Bacterial persistence is typically observed when a population of genetically identical cells is subjected to various antibiotic treatments (12). At first, most of the population is rapidly killed. After a few hours of continuous antibiotic treatment, the death rate decreases substantially, revealing the existence of a persistent subpopulation. Persistence to antibiotics is ubiquitous in WT strains of different bacteria (13, 14). It has been shown that cells regrown from such persistent subpopulation are as sensitive to the antibiotic treatment as the original population, suggesting that persister bacteria do not differ genetically from the normally growing cells (12). In previous work, we demonstrated that persistence originates in subpopulations of slowly or nongrowing bacteria that form part of the bacterial population already before the exposure to antibiotics (15). Two types of persistent bacteria were identified in WT and in the high-persistence (hip) strains of Escherichia coli (16, 17): (i) Type I persisters, which enter at stationary phase (18), a dormant state that protects them from the lethal action of several antibiotics known to affect mainly actively growing cells (15, 19); and (ii) Type II persisters, which do not require a starvation signal to enter the persister state and are continuously generated from the normally growing cells. In the present work, we focus on Type I persistence to antibiotic treatments such as those found in hipA7 (20), which has been observed in several WT bacterial strains and is today believed to be at the root of persistent infections (13, 21). The current understanding of Type I persistence is depicted in the scheme of Fig. 1a: persisters are cells that remained in the growth-arrested state acquired at stationary phase, despite the transfer to fresh medium. After many hours in fresh medium, persister cells eventually switch back to normally growing cells (15, 22). However, direct proof showing that the differentiation into the persister state occurs at the entry of stationary phase state was lacking, and attempts to study this dormant state remained elusive. Microarray analysis suggested that Type I persisters differ both from normal and from stationary phase cells (23).

Fig. 1.

Induction of single cells in a microfluidic device. (a) Current model for Type I persisters showing the formation of persisters at stationary phase. Cells arrest their growth at stationary phase. Upon transfer to fresh medium, normal cells grow and become sensitive, whereas persister cells remain in the growth arrested state acquired at stationary phase. (b) Expected fluorescence response to induction of normal cells (black curve) and persister cells (red curve), based on the current model for Type I persisters shown in a. (c) Schematic layout of the induction network: YFP is constitutively expressed from the chromosome; the addition of aTc releases the repression of the tet promoter and mCherry proteins are produced. (d–k) Induction experiment in a microfluidic device: at t = 0, hipA7 bacteria from O/N culture were introduced in the microfluidic device and subjected directly to fresh growth medium + inducer (aTc). (d) All bacteria constitutively express the YFP (here shown in green). (e) Induction of the mCherry protein is seen in the red fluorescence increase. (f–h) Addition of 100 μg/ml ampicillin kills growing bacteria. (i–k) Antibiotics washed and a single persister identified (white arrow). (Scale bar: 5 μm.) (l) mCherry fluorescence increase of typical normal (black squares) and persister (red squares) cells. Initial fluorescence increases as fast in persisters and normal cells, in contrast to the expected behavior shown in b. The delay in detection is mainly because of the relatively high background fluorescence (a.u., arbitrary units).

Our goal in the present work was to follow the dynamics of the differentiation process in single cells and to determine whether persisters are indeed cells that remained at stationary phase and did not respond to the external cues provided by the transfer to fresh medium. For this purpose, we developed a methodology to follow the induction process of fluorescent proteins under inducible promoters (24) in microfluidic devices. According to current understanding, we expected a rapid increase in fluorescence in the normal cells and a weak or no response to the induction signal in the persister cells until they switch back to normal growth (Fig. 1b).

Results and Discussion

Measurement of the Induction Dynamics of Persisters in Microfluidic Devices.

We compared the induction curves of normal and persister cells in a population of the high-persistence strain hipA7 harboring a plasmid with an inducible promoter Ptet, which controls the expression of fluorescent proteins (MGYA7Z1/Ptet-mCherry) (16). The activity of Ptet fused to the mCherry gene (Ptet-mCherry) was measured by tracking the increase in the red fluorescence of single cells, after induction with anhydrous tetracycline (aTc) (Fig. 1c). All cells were also constitutively expressing the yellow fluorescence protein (YFP) to facilitate their identification with our automated tracking and detection program based on the yellow channel [supporting information (SI) Text and Movie S1].

In our initial assay, the MGYA7Z1/Ptet-mCherry population was taken from an overnight (O/N) culture at stationary phase, which can generate both normal and nongrowing cells. All cells had a high level of YFP that was expressed before entrance to stationary phase. This population was placed in fresh medium with inducer (aTc) in our microfluidic devices, where the induction process in single cells can be tracked, while the cells are intermittently subjected to antibiotic treatments (see Materials and Methods). The time course of the experiment is presented in Fig. 1 d–k. From the start (Fig. 1d), the bacteria expressed YFP and were detected in the yellow channel only (here presented in green). Cells then responded to the induction by aTc and mCherry protein expression ensued, as indicated by the increasing red fluorescence (Fig. 1e). Then, antibiotic was added, killing most of the cells (Fig. 1 f–h). After 5 h of continuous antibiotic treatment, the antibiotic was washed away and persister cells identified (Fig. 1 i–k). As observed in our previous work, these persister cells were not growing before the antibiotic treatment, and their ability to survive extensive antibiotic treatments was attributed to this preexisting dormancy (15). Typical induction curves of persisters (red) and normal cells (black) are shown in Fig. 1l. The induction dynamics of persisters over the whole experiment revealed an unexpected feature: those persisters had not been dormant during the first 1.5 h after their transfer to fresh medium, despite being in a nongrowing state. In fact, during that initial period, all cells exhibited a similar induction dynamics, suggesting that the initial production rates of both normal and nongrowing cells were similar (Fig. 1l) and in contrast to the expected behavior shown in Fig. 1b.

Quantitative Analysis of Single-Cell Induction Curves.

Further analysis of higher-magnification observations on agarose without antibiotics (Fig. 2 a–c) allowed us to follow the induction of normal cells for longer times and compare quantitatively the initial induction dynamics of normally growing bacteria to those of persister bacteria, characterized by their late growth. At each time point, a phase-contrast image was acquired as well as fluorescence images in the YFP and mCherry channel. Fig. 2d shows the automatic detection and tracking of the fluorescent bacteria in pseudocolors (see Movie S1). The output curves of the mCherry fluorescence increase for one experiment are shown in Fig. 2e.

Fig. 2.

Automated quantitative analysis of single-cell induction curves. (a–e) High-magnification microscopy on agarose+aTc of MGYA7Z1/Ptet-mCherry. (a–c) Simultaneous pictures of the same field of view, where the growth of microcolonies initiated by two normally growing cells are monitored. (a) Phase contrast. (b) YFP (here shown in green). (c) mCherry-induced fluorescence. (d) Automatic detection and tracking of the fluorescent bacteria in pseudocolors. Each new cell is assigned an arbitrary color by the detection algorithm. Similar shades indicate bacteria that are closely related in their pedigree (see also Movie S1). (Scale bar: 5 μm.) (e) Output curves of the mCherry fluorescence increase from the automated procedure. Each curve represents one bacterium and its progeny (a.u., arbitrary units).

Late-growing cells were identified by following the growth in the area of each bacterium progeny. Typical growth curves are shown in Fig. 3a, where late-growing cells can be identified (red curves). The initial fluorescence increases of both subpopulations overlap (Fig. 3b). Automated analysis of such experiments shows that the histograms of the times to cross a fluorescence threshold five times above background are similar, for normally growing and late-growing cells (Fig. 3c), with median times of induction of 51 ± 5 and 56 ± 5 min, respectively. After the initial response, the fluorescence increase in the nongrowing cells stopped, indicating that fluorescent protein production has shut down, and dormancy has set in. The shutdown in protein production can be also seen in the absence of YFP fluorescence increase in the nongrowing bacteria (Fig. S1). To verify that this behavior was general and not linked to the chosen inducer (25) or fluorophore, we repeated those experiments with the lac promoter, induced by isopropyl β-d-thiogalactoside and controlling YFP instead of mCherry expression. The results are shown in Fig. 3 a and b Insets. Again, nongrowing cells follow a dynamics similar to that of normal cells, in the initial induction phase.

Fig. 3.

Comparison of fluorescence induction between normally growing and nongrowing cells. (a) Growth of microcolonies starting from single bacteria. Late-growing cells are marked in red. (Inset) Similar growth data for the induction of the Plac promoter with IPTG. (b) mCherry fluorescence increase for normal (black) and persister bacteria (red). (Inset) Similar fluorescence data for the induction of the Plac promoter with IPTG. (c) Histogram of the times to cross a fluorescence threshold of five times the background level for normally growing (black) and slow-growing cells (red). The two peaks overlap and have similar median times. Histograms normalized to the total number of cells in a single experiment (136). Five independent experiments were conducted, all leading to similar results. (a.u., arbitrary units).

Measurement of the Induction Dynamics of Persisters After the Onset of Dormancy.

The arrest in protein production, 1.5 h after transfer to fresh medium, suggested that persister cells might not respond to an induction signal, provided the inducer is introduced after dormancy has set in. We then repeated our microfluidic assay, subjecting the bacteria from the O/N culture to fresh medium, this time subjecting the population first to fresh medium without inducer and adding the inducer only after the typical time for the setting in of dormancy. The results obtained are presented in Fig. 4. Now, we detected no increase in the red fluorescence of persister cells upon exposure to aTc (Fig. 4h, red curve), whereas normally growing cells were induced and died (Fig. 4h, black curve). The fluorescence increased only when persister cells reverted to normal growth. These results suggested that the dormancy that protects persister bacteria from the lethal action of antibiotics also prevents induction and is not yet present at stationary phase but rather develops after the exit from stationary phase. Furthermore, our observations revealed the existence of a restricted time window of ≈1.5 h, starting at the exit from stationary phase, when the dormancy that characterizes persistence is not yet fully developed. They also suggest that subjecting the bacterial populations to antibiotics within this time window, before the onset of full dormancy, might prevent persistence.

Fig. 4.

No response to induction in persister cells after the onset of dormancy. At t = 0, MGYA7Z1/Ptet-mCherry bacteria from O/N culture were introduced in the microfluidic device and subjected first to fresh growth medium without inducer. The inducer (aTc) was added only after 2.5 h. (a) All bacteria constitutively express YFP (here shown in green). (b and c) Induction and production of the mCherry protein are seen in the red fluorescence increase and in normal cells only. (d) Growing bacteria die because of addition of 100 μg/ml ampicillin. (e–g) Removal of ampicillin: a persister (white arrow) reverts to normal growth and divides. (Scale bar: 5 μm.) (h) Analysis of mCherry fluorescence increase of typical normal (black squares) and persister cells (red squares). Persister bacteria do not respond to the induction signal when applied, but only when they switch to normally growing cells. (a.u.: arbitrary units).

A Period of Vulnerability to Antibiotics in Persister Bacteria.

To test this prediction, we monitored the number of persisters to antibiotic exposure, during the first hours after transfer to fresh medium. Our goal was to subject the population to antibiotics before the onset of full dormancy, without first growing the cells as typically done in persistence assays (15, 26–28). We diluted the O/N culture into fresh medium, while maintaining constant temperature. At each time point after the dilution, an aliquot of the culture was directly exposed to ampicillin for 5 h, and the number of persisters was determined by the most probable number-counting method (MPN). The results are plotted in Fig. 5a. The number of persisters obtained in the aliquots subjected to ampicillin after >2 h in fresh medium amounted to a few percentage of the initial population, as expected for the hipA7 mutant. However, the number of persisters was 1 order of magnitude lower in the aliquots exposed to ampicillin within a time window similar to the one observed in the induction of fluorescence experiment, i.e., before the halt in fluorescent proteins production. These nontrivial results revealed that nongrowing cells that generate Type I persistence are still vulnerable during the first 1.5 h after transfer to fresh medium. The results also indicate that the total number of persisters measured in the population can be very different, depending on the timing of antibiotic exposure. These findings might explain the large variability in the number of persisters reported in experiments that did not take this effect in account (27).

Fig. 5.

Decreased persistence level at the exit from stationary phase. (a) An O/N culture of MGYA7Z1/Ptet-mCherry bacteria was diluted in fresh LB medium and its growth monitored by MPN (black). At each time point, an aliquot of the culture was subjected to ampicillin and the number of persisters counted by MPN (red). During the lag period, an increase in the number of persisters is observed. Error bars are calculated from eight MPN replicates. Similar results were obtained for the hipA7 strain without plasmid (data not shown). (b) New model for Type I persisters showing the different stages in the formation of persisters: Cells arrest their growth at stationary phase. Upon transfer to fresh medium, all cells become sensitive and respond to the induction signal. After a typical time of ≈ 1 h, persisters fully differentiate into the dormant state that protects them from ampicillin, whereas normal cells continue to grow. (c) Expected response to induction applied directly at the exit from stationary phase, for normal cells (black curve), and persister cells (red curve), based on the new model for Type I persisters shown in b. (a.u., arbitrary units).

Our results challenge the current view of bacterial persistence and suggest an alternative model for the differentiation of normal cells into persisters (Fig. 5 b and c): similarly to normal cells, persister cells become sensitive to antibiotics as soon as they are transferred to fresh medium; only after 1.5 h attempting to exit stationary phase do persisters enter the dormant state that protects them from the lethal action of antibiotics, such as ampicillin, that kills actively growing cells. The vulnerability to antibiotics achieved here is a predictable event that occurs faster, by 1 order of magnitude, than the stochastic switching from persistent to normal cells.

Our results show the power of single-cell induction dynamics in microfluidic devices to follow in real time the differentiation process resulting in dormancy and to reveal new ways to reduce the persistence to antibiotics. This work presents a quantitative phenotypic characterization of nongrowing cells (29), which constitute the most common lifestyle in the wild and should find relevance in different ecological environments where growth arrest plays a major role, for example in biofilms (30, 31). An approach similar to the one presented in this study might shed light on the processes leading to the dormancy of subpopulations in different systems, such as in subpopulations of viable but nonculturable bacteria (32) or those of quiescent cancer cells (33).

Materials and Methods

Bacterial Strains and Media.

E. coli strain DH5αZ1 was obtained from H. Bujard (34). A P1_vir lysate of DH5αZ1 was used to transduce MGYA7 into MGYA7Z1, under a spectinomycin resistance screen, to move the lacI and tetR genes into the MGYA7 background under constitutive promoters. LB (Difco LB Broth), M9 (Difco M9 Minimal Salts, 5×).

Plasmids.

The plasmid encoding fluorescent protein mCherry was kindly given by the Tsien laboratory (35). Using this template, the mCherry was amplified by PCR, using primers, which introduced the flanking restriction sites KpnI and HindIII (GGGGTACCATGGTGAGCAAGGG and CCCAAGCTTTTACTTGTACAGC, respectively). It was inserted into plasmid of pZ serie (34), pZA21mCherry, and pZE21mCherry, at the multiple cloning site, using the original KpnI and HindIII sites of the plasmids.

Ptet and Plac correspond to PLtetO1 and PLlacO1 promoters (34, 36), respectively.

Microscopy.

Microscopy was performed by using a Leica DMIRE2 inverted microscope system with automated stage (Ludl) and shutters (Uniblitz). The microscope was placed in a large incubator box at 37°C (Life Imaging Systems) that controls the temperature to an accuracy of 0.1°C. All microfluidic supply lines and vessels were kept inside the incubator during the experiments. Autofocus and image acquisition were done by using custom macros in Scope-Pro (Media Cybernetics) to control the microscope, stage, shutters, and camera. Several different locations were monitored in parallel on the same device. Images were acquired by using either a ×63 long-range air or a ×100 oil objective and a cooled CCD camera (−75°C) (Orca II, Back-illuminated, Hamamatsu) and processed with Image-Pro (Media Cybernetics).

Fluorescence images were acquired with minimal excitation to minimize bleaching. No difference was noticed in the growth of cells exposed or unexposed to fluorescence illumination.

Automated Image Analysis.

Similarly to ref. 24, we performed segmentation and tracking of each bacterium across frames and created a lineage tree for each microcolony. In our method, we used the YFP images, as in ref. 37, for finding the segmentation, instead of phase-contrast images. This approach was substantially superior in our case, because the YFP images have a higher dynamic range and less noise. We used image reconstruction (38) for separating individual bacterium bodies from the darker background. Furthermore, our tracking algorithm performs multiresolution center-weighted motion detection based on ref. 39 for each bacterium, so we can track bacteria over large time periods and large accumulated movements.

Microfluidic Devices.

The microfluidic devices were fabricated as described (15). Briefly, the microfluidic devices consist of several layers clamped together: a thin patterned polydimethylsiloxane layer (Sylgard 184, Dow Corning) with microscopic grooves or rectangular boxes made by soft lithography techniques (40) by using a mold of AZ4110 (Clariant); a cellulose membrane; a thicker PDMS layer with flow channels patterned by using a mold of SU-8 2100 (MicroChem).

These devices enabled switching conditions under the microscope and the controlled growth of the bacteria in the patterned PDMS.

Growth in the Microfluidic Devices.

Bacteria were introduced in the microfluidic devices by putting a 4-μl drop of culture on the PDMS layer and closing the device by clamping. Experiments were started from aliquots.

Aerated LB with or without antibiotics and inducer was continuously flown in the flow channels. The channels were continuously washed once per second. Application and removal of antibiotics occurred within 1 min.