Observation of persister cell histories reveals diverse modes of survival in antibiotic persistence

To visualize the dynamics of low-frequency persister cells, we used a microfluidic device equipped with a membrane-covered microchamber array (MCMA; Figure 1). E. coli cells were enclosed in the 0.8-µm deep microchambers etched on a glass coverslip by covering microchambers with a cellulose semipermeable membrane via biotin-streptavidin bonding (Figure 1A–C; Inoue et al., 2001). E. coli cells grow in a monolayer and form two-dimensional microcolonies in the microchambers (Figure 1B–D). We can control medium conditions around cells flexibly by the medium flow above the membrane (Figure 1B–D). The medium exchange rate across the membrane has been evaluated in a similar device previously (Shimaya et al., 2021); it is reported that the medium in the microchamber is exchanged within 5 min, which is sufficiently shorter than the periods of antibiotic treatment and the periods of post-antibiotic treatment of observing regrowth in this study. Similar microfluidic devices with membrane-covered microchambers have been utilized for analyzing bacterial growth and starvation (Inoue et al., 2001; Umehara et al., 2003; Shimaya et al., 2021).

Figure 1 with 4 supplements see all

Download asset Open asset

We mainly used two E. coli strains in the single-cell analysis: MG1655, a strain derived from a wildtype K-12 isolate of E. coli (Blattner et al., 1997); and MF1, an MG1655-derived strain that expresses RpoS-mCherry from the native rpoS locus on the chromosome and green fluorescent protein (GFP) from a low copy plasmid (Zaslaver et al., 2006). RpoS is a specialized sigma factor controlling the general stress response (Battesti et al., 2011). We originally constructed the MF1 strain to study the correlation between the expression levels of RpoS and the persister trait at the single-cell level. However, a previous report clearly demonstrated that the fluorescent fusion of RpoS is functionally defective (Patange et al., 2018). We also confirmed this functional issue of RpoS of MF1; although the dependence on growth phase and culture media of RpoS-mCherry expression was similar to that characterized previously using lacZ fusion (Figure 1—figure supplement 1A and B; Lange and Hengge-Aronis, 1994), this strain was significantly more sensitive to H₂O₂ oxidative stress than MG1655 and as sensitive as the MG1655 ΔrpoS strain (Figure 1—figure supplement 1C). Therefore, we considered MF1 as a strain with impaired RpoS function. Despite this difference, the minimum inhibitory concentrations (MICs) against Amp and CPFX were almost identical between MG1655 and MF1 (Figure 1—figure supplement 2).

The population killing curves of the MG1655 and MF1 strains (Figure 1—figure supplement 3) exhibit biphasic or multiphasic decay when exponentially growing cell populations are treated with 200 µg/mL of Amp (12.5×MIC, Figure 1—figure supplement 2A; see Figure 1—figure supplement 4 for the growth curves of these strains) or with 1 µg/mL of CPFX (32×MIC, Figure 1—figure supplement 2B), confirming that these E. coli strains exhibit persistence against these antibiotics (Balaban et al., 2019). In the case of Amp treatment, the frequency of surviving cells of MG1655 was higher than that of MF1 up to 3 h of treatment, but became lower in the treatment longer than 3 h (Figure 1—figure supplement 3A) due to the delayed shift of the decay rate. Similarly, in the case of CPFX treatment, the frequency of surviving cells of MG1655 was slightly higher than that of MF1 up to 1 h, but then became lower at least up to 7 h of treatment (Figure 1—figure supplement 3B). These results confirm that intact RpoS function is not a prerequisite for the occurrence of persistence, but may affect the frequency of surviving cells and the timing of the decay rate shift.

Figure 2 shows examples of single-cell observation of persister cells using the MCMA device. In this observation, we monitored 110–170 microchambers with a 100× objective in each time-lapse experiment to observe individual cells of the MF1 strain sampled from exponential-phase batch cultures (‘post-exponential phase’; Figure 1C and D, Figure 1—figure supplement 4) with a high spatial resolution. After initial pre-culturing periods of 1.5 h without Amp in the device, we exposed cell populations to 200 µg/mL of Amp for 3.5 h. The Amp exposure lysed most cells (Figure 2, Videos 1 and 2). However, we detected two persister cells that survived and regrew after the Amp exposure in the ten repeated experiments conducted with the MCMA device. Interestingly, one of the persister cells of MF1 grew and divided actively before the Amp exposure (Figure 2A-D and Video 1). In response to the Amp exposure, the cell radically changed its cell morphology from a rod shape to a round shape with membrane blebbing and transitted to an L-form-like cell (Errington et al., 2016; Figure 2B and Video 1). The other persister cell of MF1 exhibited no growth and division before and during the Amp exposure and restarted growth after drug removal (Figure 2E–H and Video 2). This sequence of growth behavior is consistent with the expected survival dynamics for persistence caused by pre-existing non-growing cells (Balaban et al., 2004). The expression levels of RpoS-mCherry were low in all the observed cells, including the persister cells characterized above (Figure 2A and E, Videos 1 and 2). The suppression of RpoS expression in exponential phases in LB media is consistent with the literature (Battesti et al., 2011; Jishage and Ishihama, 1995). These results demonstrate that the MCMA device can unravel the single-cell history of low frequency persister cells over the course of antibiotic pre-exposure, exposure, and post-exposure periods.

Figure 2

Download asset Open asset

Video 1

Download asset

Video 2

Download asset

Hereafter, we refer to the surviving cells that showed growth and division before drug exposure as growing persisters (Figure 2A–D) and those that exhibited no growth and division as non-growing persisters (Figure 2E–H), focusing on growth traits during the pre-treatment period in the MCMA device without antibiotic.

To further scale up single-cell observation, we next used a 40× objective and acquired only bright-field images in the time-lapse measurements (Video 3). These changes in the microscopy configuration reduced the spatial resolution of acquired images and did not involve acquisition of fluorescence images. However, most single-cell lineages could still be tracked, and the drug response of more than 10⁵ individual cells was observable with a time-lapse interval of 3 min.

Video 3

Download asset

We first observed the MG1655 cells sampled from a batch culture in the exponential phase in LB (post-exponential phase condition; Figure 1—figure supplement 4A). The cells were initially allowed to grow in the microchambers for about 1.5 h by flowing a fresh LB medium over the membrane (Figure 1C and D). Then the cells were exposed to 200 µg/mL Amp for 6 h by flowing an Amp-containing LB medium. After exposure, we switched the flowing medium back to LB and monitored cell regrowth typically for 12 h. We note that when some cells rapidly resumed proliferation after treatment, their progeny filled chambers by growth, eventually pushing up the semipermeable membrane covering the chambers, growing into neighboring chambers and preventing further observation. In this case, we stopped observing the invaded chambers earlier than 12 h. We set the above duration of antibiotic exposure to be longer than the time of the decay rate shift of the population killing curve (Figure 1—figure supplement 3A) to ensure that only persister cells are allowed to re-proliferate after the drug removal.

In this post-exponential phase condition, we detected six persister cells (defined as drug-exposed cells that survived Amp exposure and re-proliferated after drug removal) among 3.8×10⁵ drug-exposed cells (Figures 3A and 4A–D, Figure 4—figure supplement 1 and Video 4). All persister cells were actively growing and dividing before the Amp exposure. Since it was reported that peristers of E. coli hipQ mutant against Amp grew at suppressed rates in LB medium (0.2∼0.3 doublings/h; Balaban et al., 2004), we quantified the pre-exposure division rates of the observed cells, finding no significant difference between persister cells and non-surviving cells: 2.36±0.26 doublings/h for growing persisters and 2.26±0.04 doublings/h for non-surviving cells (Figure 4E; ${\displaystyle {\displaystyle p}}$ = 0.74, Wilcoxon rank sum test). Therefore, these persister cells grew as fast as the non-surviving cells prior to the Amp treatment.

Figure 3 with 1 supplement see all

Download asset Open asset

Figure 4 with 2 supplements see all

Download asset Open asset

Video 4

Download asset

We also analyzed the post-exponential phase cells of MF1 exposed to 200 µg/mL of Amp for 3.5 h after 1.5 h of pre-exposure cultivation in the device and found 12 persister cells among 3.0×10⁵ drug-exposed cells (Figure 3—figure supplement 1, Videos 5 and 6). The duration of the Amp treatment was again determined to be longer than the time of the decay rate shift of the population killing curve of this strain (Figure 1—figure supplement 3A). Analysis of the pre-exposure growth dynamics revealed that 10 of the 12 persister cells were growing persisters and the other two persister cells were non-growing persisters (Figure 3—figure supplement 1, Figure 4—figure supplement 2A, Videos 5 and 6). In contrast to MG1655, the pre-exposure division rate of the growing persisters of MF1 was slightly lower than that of non-surviving cells (2.00±0.22 doublings/h for growing persisters; and 2.48±0.06 doublings/h for non-surviving cells; Figure 4—figure supplement 2B; ${\displaystyle {\displaystyle p}}$ = 0.011, Wilcoxon rank sum test). However, this division rate of the growing persisters of MF1 was significantly larger than the growth rate of the hipQ mutant reported in Balaban et al., 2004. Therefore, the growing persisters are dominant among the persisters in the post-exponential phase cell populations of MF1 and grow slightly slower than non-surviving cells before Amp treatment.

Video 5

Download asset

Video 6

Download asset

The non-growing persisters detected in the post-exponential phase cell populations of MF1 (Figure 3—figure supplement 1 and Video 6) may be long-lagging cells that remained non-growing throughout the period from the previous stationary phase in the first pre-culture to the Amp treatment (Bakshi et al., 2021; Levin-Reisman et al., 2010), or they may have resumed growth during the second pre-culture but stopped growing spontaneously before the start of time-lapse measurements (see Methods and Figure 1C for sample preparation procedures for time-lapse measurements). To test which scenario is more plausible, we extended the duration of the exponential phase by preparing exponential phase cell populations in the second pre-culture from a 1000-fold diluted inoculum of stationary phase cells and allowed them to grow to the same cell density before loading the cells into the MCMA device. With this change in the cell sample preparation procedure, the expected frequency of long-lagging cells at the time of Amp treatment becomes approximately 10⁻⁸, which is well below the detection limit of our single-cell measurements (see Methods). Despite this change, we still found non-growing persisters at a frequency comparable to the undiluted inoculum experiment: We detected 23 persister cells among 9.9×10⁵ observed cells, and 4 out of 23 persister cells were non-growing persisters (Figure 3—figure supplement 1A and Video 7). This result shows that non-growing persisters are observed even without the carry-over of lagging cells and are therefore likely to be generated spontaneously in the exponentially growing cell populations (Balaban et al., 2019).

Video 7

Download asset

To characterize the dynamics of individual persister cells, we quantified transitions in cell size and cell shape (cell body circularity, ${\displaystyle 4\pi \times [\mathrm{C}\mathrm{e}\mathrm{l}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{a}]/[\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{i}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{r}{]}^2}$) during pre-exposure, exposure, and post-exposure periods (Figures 1D and 4A–D, Figure 4—figure supplement 1, and Figure 4—figure supplement 2A). The analysis reveals heterogeneous responses of growing persisters of MG1655 to Amp exposure. For example, the growing persister shown in Figure 4A and B suppressed growth and stopped division after the first division under the exposure to Amp. This cell resumed division with a long lag approximately 6 h after drug removal (Figure 4B). On the other hand, the growing persister shown in Figure 4C and D continued to grow throughout the period of Amp treatment with deformation of the cell shape from rod to L-form-like shape. Cell body fission also continued during Amp exposure and gradually restored normal rod shape through multiple fissions after drug removal (Figure 4C and D). The other persister cells exhibited intermediate behavior between these two, with growth and cell body deformation in some time periods and growth suppression in other time periods (Figure 4—figure supplement 1). Surprisingly, some of the L-form-like cells crawled around like amoeba cells within the microchambers, detaching and leaving some parts of their cell bodies behind (Video 4). Furthermore, rod-shaped regrowing cell populations emerged even from the small and left-behind cell bodies (Video 4). These observations reveal the abilities of E. coli cells to sustain viability with abnormal morphologies and to restore rod shapes in the absence of agents that induced morphological changes.

The growth and cell shape dynamics of growing persister cells of MF1 were similar to those of MG1655. We detected persister cells that stopped growing and dividing after the first divisions in response to Amp exposure, as well as those that continued growth and cell body fission with deformed cell shape throughout the period of Amp exposure, and those that exhibit intermediate behavior (Figure 4—figure supplement 2A). These results reveal heterogeneous responses of growing persisters to Amp treatment with respect to continued growth and division and the occurrence of extensive cell shape changes.

To test whether medium conditions influence persistence dynamics, as previously suggested (Harms et al., 2017), we repeated the experiments with M9 minimal medium. The killing curves of both MG1655 and MF1 cell populations sampled from exponential phase cultures in M9 minimal medium (post-exponential phase condition in M9) show biphasic curves when exposed to the same concentration of Amp as used in LB medium (200 µg/mL), confirming the occurrence of persistence in this condition (Figure 1—figure supplement 3C–E; see Figure 1—figure supplement 4B for the growth curve). The frequency of surviving cells of MG1655 in M9 was consistently higher than that of MF1 (Figure 1—figure supplement 3C). The frequency of persister cells in the post-exponential phase cell populations in M9 was higher than the frequency in the LB medium, approximately 100-fold and 10-fold for MG1655 and MF1, respectively (Figure 1—figure supplement 3D and E).

Single-cell time-lapse measurements detected 18 persister cells of MG1655 that regrew and divided after the exposure to 200 µg/mL of Amp for 6 h among 3.5×10⁵ drug-exposed cells. Among them, we were able to determine the presence or absence of pre-exposure growth and division of 17 persister cells, of which 16 grew and divided before the Amp exposure, and the other one showed no growth and division (Figure 3A, Figure 5A-D, Figure 5—figure supplement 1, and Video 8). The presence or absence of growth and division of 1 persister cell was indeterminate due to translocation of cells within a microchamber. Similarly, we detected 24 persister cells of MF1 that survived and re-proliferated after the exposure to 200 µg/mL of Amp for 6 h among 1.1×10⁶ drug-exposed cells (Figure 3—figure supplement 1). The presence and absence of pre-exposure growth and division were identified for 19 persister cells, and we found that 17 of them were growing persisters (Figure 3—figure supplement 1, and Video 9). No growth or division was observed for the other two persister cells. These results in the M9 minimal medium are qualitatively consistent with those in LB; growing persisters are dominant in the post-exponential phase cell populations for both MG1655 and MF1, and non-growing persisters exist as small fractions.

Figure 5 with 1 supplement see all

Download asset Open asset

Video 8

Download asset

Video 9

Download asset

Despite the consistency of the dominant survival mode of persister cells, the pre-exposure division rate of growing persisters (0.46±0.07 doublings/h) of MG1655 in M9 was significantly lower than that of non-surviving cells (0.96±0.03 doublings/h, ${\displaystyle p<{10}^{-6}}$, Wilcoxon rank sum test; Figure 5E). Therefore, growing persisters in the M9 medium tend to grow more slowly than non-surviving cells.

The heterogeneous response of growing persisters to Amp exposure was also observed in the M9 medium; some persister cells suppressed growth and division to varying degrees in response to drug exposure, while others continued to grow and undergo cell body division with a change in cell shape from rod to L-form-like (Figure 5A–D, Figure 5—figure supplement 1, and Video 8). The transition to an L-form-like cell shape was prominent during the return to normal growth and rod shape in the post-exposure period. In contrast to the growing persisters in LB, we observed filamentation of some persister cells in M9 after exposure; normal-sized progeny cells produced from these filamentous cells by unequal cell division regained normal growth (Figure 5—figure supplement 1 and Video 8). Radical morphological changes of cell bodies were observed in many growing cells in the M9 medium, including non-surviving cells (Video 8). As observed in the post-exponential phase conditions in LB, some of the L-form-like cells crawled around like amoeba cells within the microchambers, detaching and leaving some parts of their cell bodies behind (Figure 5C, Videos 8 and 10). Rod-shaped and normally growing progeny cells emerged from these detached cell bodies.

Video 10

Download asset

In the single-cell analysis of the persister cells of MG1655 in the M9 medium, we found a single case where a pair of siblings produced during the pre-exposure period both survived the Amp treatment and re-proliferated after the drug removal (growing persister cell 1 in Figure 5A and B and growing persister cell 4 in Figure 5—figure supplement 1). If the survival of drug-exposed individual cells was determined independently, the probability of finding such a pair of siblings that both survived would be extremely low, and in our measurement virtually zero. Therefore, this result implies a correlation between cell lineage and the persistence trait, as well as some heritable factors contributing to the survival of individual cells, as previously suggested by a fluctuation test-based batch culture analysis (Hossain et al., 2023). However, our observation is limited to this single pair; higher throughput analysis is needed to clarify the cell lineage correlation of survival.

Several batch-culture assays have shown that stationary phase environments increase the frequencies of persister cells in bacterial cell populations (Spoering and Lewis, 2001; Keren et al., 2004; Balaban et al., 2019). To understand how the past cultivation history affects the survival dynamics of individual persister cells, we sampled cells from early and late stationary phases in LB medium (Figure 1—figure supplement 4A), allowed them to grow in the fresh medium for 2 h, and exposed them to 200 µg/mL of Amp in both batch and single-cell cultures (Figure 1C and D, Figure 6A).

Figure 6 with 2 supplements see all

Download asset Open asset

The population killing curves of MG1655 under these conditions (hereafter referred to as ‘post-early stationary phase’ and ‘post-late stationary phase’ conditions; Figure 1—figure supplement 4A) follow biphasic or multiphasic kinetics under the exposure to 200 µg/mL of Amp (Figure 6B). While the killing curve of the post-early stationary phase condition was similar to that of the post-exponential phase condition, that of the post-late stationary phase condition differed significantly from those of the other two coniditions with consistently higher frequencies of surviving cells (Figure 6B). The frequencies of surviving cells evaluated at the 6 h time point were 3.7×10⁻⁷ (1.7×10⁻⁷–8.1×10⁻⁷) for the post-early stationary phase cell populations and 5.1×10⁻² (3.9×10⁻²–6.6×10⁻²) for the post-late stationary phase cell populations (Figure 6B).

Comparison of the killing curves between MG1655 and MF1 shows that for the post-early stationary phase cell populations, the frequencies of surviving cells of MG1655 were higher than those of MF1 up to 2.5 h of Amp exposure, but became lower thereafter, resulting in fewer surviving cells at longer exposures (Figure 6—figure supplement 1A). This time-dependent reversal of surviving cell frequencies between MG1655 and MF1 was also observed for the post-exponential phase cell populations, as noted above (Figure 1—figure supplement 3A). Under post-late stationary phase conditions, the frequency of surviving cells was consistently higher in MG1655 than in MF1 (Figure 6—figure supplement 1B).

To characterize the dynamics of individual persister cells in the cell populations with these different pre-exposure cultivation histories, we next performed single-cell measurements with the MCMA device. We loaded cells in the corresponding stationary phase conditions (Figure 1—figure supplement 4A) into the MCMA device and allowed the cells to regrow by changing the media around the cells from the corresponding stationary phase conditioned medium to fresh LB medium at the beginning of the time-lapse measurement (Figure 1C and D). After a 2-h pre-exposure period, we exposed the cells to 200 µg/mL of Amp for 6 h and monitored the re-proliferation of surviving cells by flowing drug-free fresh medium (Figure 1C and D). Since all cells were maintained in a non-growing state by the corresponding stationary phase conditioned medium until the start of the time-lapse measurement, non-growing cells found in these experiments are by definition lagging cells that failed to rapidly resume growth.

Measurements of MG1655 with the 40× objective detected seven persister cells that survived and re-proliferated after Amp exposure among 7.7×10⁵ drug-exposed cells in the post-early stationary phase cell populations (Figure 3A and Video 11). We were able to determine the pre-exposure growth states of six persister cells. Among them, one persister cell was a growing persister and the other five were non-growing persisters (Figure 3A). Therefore, the proportions of non-growing persisters increased significantly in the post-early stationary phase cell populations despite the similarity of their killing curve to that of post-exponential phase cell populations.

Video 11

Download asset

For the post-late stationary phase cell populations, we detected 187 persister cells among 1.0×10⁵ drug-exposed cells, finding that all but 20 indeterminate cells were non-growing persisters (Figure 3A and Video 12). These results show that entry into stationary phase conditions rapidly changes the predominant pre-exposure trait of persister cells from growing to non-growing. We confirmed that the same conclusion holds true for MF1 persister cells in the post-early and post-late stationary phase cell populations (Figure 3—figure supplement 1, Videos 13 and 14).

Video 12

Download asset

Video 13

Download asset

Video 14

Download asset

We found that many cells sampled from late stationary phase remained non-growing in fresh medium during the pre-exposure period. Many, but not all, of these long lagging cells also appeared to be intact under Amp exposure, but did not regrow even after the drug removal for at least 12 h. Most of these non-regrowing cells may be viable, as they were negative in propidium iodide (PI) staining (Figure 6—figure supplement 2 and Video 15), although PI is known to be an imperfect marker of cell viability and the possibility that they were actually dead cannot be excluded (Kim et al., 2018; Wu et al., 2024). These cells may eventually regrow as persisters with long lags after drug removal or may not be able to regrow due to deepened dormancy (Pu et al., 2019; Bollen et al., 2021; Bollen et al., 2025). In fact, it has been shown by the single-cell measurements with a mother machine microfluidic device that the time to regrowth after drug removal is heterogeneous among non-growing cells (Bakshi et al., 2021). These late regrowers are detectable in the killing curve assay but not in the single-cell measurements with the MCMA microfluidic device due to the overgrowth of the early regrowers. Consequently, our single-cell measurements may underestimate the frequency of persister cells in the population sampled from late stationary phase. However, this potentially overlooked fraction of persister cells are also non-growing persisters; the predominance of non-growing persisters in the post-late stationary phase cell population is unchanged.

Video 15

Download asset

Although the presence of cells that neither grow nor divide during the pre-exposure period in the MCMA device is evident in the post-late stationary phase cell populations, such non-growing cells were also present in the post-exponential and post-early stationary phase cell populations, albeit at lower frequencies. Based on this observation, we next asked how the frequencies of non-growing cells immediately before Amp exposure depend on cultivation history and whether these non-growing fractions produce persisters (non-growing persisters) at comparable frequencies across conditions (Figure 7A).

Figure 7

Download asset Open asset

To estimate the fractions of non-growing cells in the populations during the pre-exposure period in the MCMA device, we performed single-cell time-lapse measurements with post-exponential, post-early stationary, and post-late stationary phase cells using the 100× objective. Here, non-growing cells were defined as cells that neither grew nor divided during the 1.5 h period immediately before the Amp exposure in the MCMA device. The estimated fractions of non-growing cells in the MG1655 cell populations immediately before the drug exposure were (2.9±1.5)×10⁻⁴ (${\displaystyle n}$ = 13650), (2.0±1.1)×10⁻³ (${\displaystyle n}$ = 1511), and 0.61±0.02 (${\displaystyle n}$ = 521) for the post-exponential, post-early stationary, and post-late stationary phase conditions, respectively (horizontal axis in Figure 7B). Those in the MF1 cell populations were (1.6±0.9)×10⁻⁴ (${\displaystyle n}$ = 19122), (8.7±2.9)×10⁻³ (${\displaystyle n}$ = 1031), and 0.41±0.02 (${\displaystyle n}$ = 469) for the post-exponential, post-early stationary, and post-late stationary phase conditions, respectively (horizontal axis in Figure 7C). Therefore, the frequency of non-growing cells in the populations increased significantly with the time spent under stationary phase conditions.

We next asked whether these non-growing cell fractions with different cultivation histories have persister cells at comparable frequencies (Figure 7A). If the probability of non-growing cells being persisters (${\displaystyle {\displaystyle s}}$) is independent of the conditions, the frequency of non-growing persisters in the population (${\displaystyle f}$) should follow

where P_n is the fraction of non-growing cells in the population immediately before the drug exposure. Since all persister cells in the post-late stationary phase cell populations for which we identified their pre-exposure growth trait were non-growing persisters (Figure 3A), we estimated ${\displaystyle s}$ as ${\displaystyle f^{(late)}/P_n^{(late)}=(8.4\pm 2.2)\times {10}^{-2}}$ for MG1655, where ${\displaystyle f^{(late)}}$ is the frequency of persister cells in the post-late stationary phase cell populations estimated from the killing curve at 6 h under Amp exposure (Figure 6B), and ${\displaystyle P_n^{(\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{e})}}$ is the frequency of non-growing cells in the post-late stationary phase cell populations estimated above.

In Figure 7B, the diagonal line shows the frequency of non-growing persister cells at different non-growing cell frequencies (${\displaystyle P_n}$) predicted by Equation 1. We also show in Figure 7B the frequencies of persister cells (non-growing persisters + growing persisters) in the post-exponential phase and post-early stationary phase cell populations estimated from their killing curves (Figure 6B). The result shows that the frequencies of persister cells under these conditions are lower than the predicted frequencies of non-growing persisters. Since the persisters in the post-early stationary phase cell populations contained growing persisters, albeit at a low frequency, and all the persisters detected in the post-exponential phase cell populations were growing persisters (Figure 3A), the true frequencies of non-growing persisters under these conditions should be lower than the points in Figure 7B, especially for the post-exponential phase condition. Therefore, the non-growing cells in the post-exponential phase and post-early stationary phase cell populations are less likely to be non-growing persisters than those in the late-stationary phase cell populations.

We performed the same analysis with MF1, referring to the frequencies of surviving cells of the killing curves at 4 h under Amp exposure (Figure 6—figure supplement 1C) and to the frequencies of non-growing cells in the cell populations estimated by single-cell time-lapse observation. We used the frequencies of surviving cells at 4 h instead of 6 h, considering that the decay rate shift of the killing curves of MF1 occurs earlier than that of MG1655 (Figure 6B and Figure 6—figure supplement 1C). The probability of non-growing cells being persisters was estimated to be ${\displaystyle s=f^{(\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{e})}/P_n^{(\mathrm{l}\mathrm{a}\mathrm{t}\mathrm{e})}}$ = (1.3±0.6)×10⁻². The measured frequency of persister cells in the post-early stationary phase cell populations was slightly lower than the predicted frequency of non-growing persisters (Figure 7C). On the other hand, the measured frequency of persister cells in the post-exponential phase cell populations was approximately equal to the predicted frequency of non-growing persisters in this condition (Figure 7C). However, since the majority of persister cells in the post-exponential phase cell populations were growing persisters (Figure 3—figure supplement 1), the actual frequency of non-growing persisters in this condition should be lower than the predicted frequency. Therefore, similar to MG1655, non-growing cells in the post-exponential and post-early stationary phase cell populations of MF1 are less likely to be persisters than those in the post-late stationary phase cell populations.

Taken together, these analyses suggest that prolonged incubation under stationary phase conditions increases the persister fraction of non-growing cells. This conclusion is consistent with a previous study by Luidalepp, et al. in which the number of non-growing cells was estimated by the number of non-lysed cells in a batch culture incubated for 3 h in Amp-containing fresh media after transfer from cultures of varying lengths of stationary phase (Luidalepp et al., 2011).

Having observed that the persisters against Amp in the post-exponential phase cell populations of MG1655 and MF1 were predominantly growing persisters (Figure 3, Figure 3—figure supplement 1), we next investigated persistence against another antibiotic drug, ciprofloxacin (CPFX). While Amp is a ${\displaystyle {\displaystyle \beta }}$-lactam antibiotic targeting cell walls, CPFX is a fluoroquinolone antibiotic targeting DNA gyrase. As shown previously in Figure 1—figure supplement 3B, post-exponential phase cell populations of MG1655 and MF1 in M9 minimal medium exposed to 1 µg/mL CPFX (32×MIC, Figure 1—figure supplement 2B) exhibited persistent killing curves, confirming the presence of persistence against CPFX. The frequencies of persisters evaluated at the 6 h time point were 4.5×10⁻⁵ (3.2×10⁻⁵–6.4×10⁻⁵) for MG1655 and 1.4×10⁻⁴ (8.9×10⁻⁵–2.2×10⁻⁴) for MF1 (Figure 1—figure supplement 3B).

To characterize the dynamics of individual persister cells in the populations, we conducted time-lapse experiments using the MCMA device (Figure 8A–D, Figure 8—figure supplement 1, Videos 16 and 17). We loaded exponentially growing cells in M9 medium into the MCMA device, allowed them to continue growing in the device by flowing fresh medium for 4 h, exposed them to 1 µg/mL CPFX for 6 hr, and monitored the re-proliferation of surviving cells (Figure 1C and D). We detected 32 persister cells that regrew after CPFX exposure among 2.0×10⁵ drug-exposed cells for MG1655 (Figure 3) and 148 persister cells among 8.3×10⁵ drug-exposed cells for MF1 (Figure 3—figure supplement 1). We found that all persister cells for which we were able to identify the pre-exposure growth trait were growing persisters (Figure 3, Figure 3—figure supplement 1, Videos 16 and 17). Therefore, as observed for the Amp exposure, growing persisters were predominant in the post-exponential phase cell populations exposed to CPFX.

Figure 8 with 2 supplements see all

Download asset Open asset

Video 16

Download asset

Video 17

Download asset

We also analyzed persister cells of MG1655 against CPFX in the post-late stationary phase cell populations in M9. Unlike the result for Amp, the post-late stationary phase cell populations did not produce significantly more persisters against CPFX than the post-exponential phase cell populations in batch cultures (Figure 8—figure supplement 2). We loaded cells in the late stationary phase (24 h time point in the growth curve in Figure 1—figure supplement 4B) into the MCMA device and allowed the cells to regrow from the stationary phase condition by changing the media around the cells from the stationary phase conditioned medium to fresh M9 medium at the beginning of the time-lapse measurement (Figure 1C and D). After 4 h of pre-exposure incubation in the fresh medium, we exposed the cells to 1 µg/mL CPFX for 6 h. We then allowed them to regrow by switching the flowing medium back to the drug-free fresh medium and monitored the re-proliferation.

We detected 218 persister cells of MG1655 that re-proliferated upon the removal of CPFX and found that all persister cells for which we could identify the pre-exposure growth trait (170 of the 218 persister cells) were growing persisters (Figure 3 and Video 18). Therefore, growing persisters were dominant in both post-exponential and post-late stationary phase cell populations exposed to CPFX in M9 medium. Notably, a previous study (Goormaghtigh and Van Melderen, 2019) detected only non-growing persister cells against ofloxacin when cells were sampled from the stationary phase and allowed to grow in a microfluidic device for 7 h in the flow of a fresh medium (MOPS medium) before drug exposure. Our results show that the presence of growing persisters is not limited to the cell populations placed under well-controlled exponential growth conditions for prolonged periods, depending on the antibiotic type.

Video 18

Download asset

We also analyzed in detail the growth and division dynamics of the persister cells in the post-exponential phase cell populations during the pre-exposure, exposure, and post-exposure periods (Figure 8A–D, Figure 8—figure supplement 1, Videos 16 and 17). Unlike Amp, the CPFX exposure did not lyse cells but arrested their growth and division, including those of the persister cells (Figure 8A–D, Figure 8—figure supplement 1, Videos 16 and 17). After the removal of CPFX, the persister cells resumed growth and many of them became filamentous (Figure 8A–D, Figure 8—figure supplement 1, Videos 16 and 17). The filamentous cells restarted cell divisions and produced normal-sized, and normally growing and dividing, progeny cells. The observed post-exposure filamentation dynamics of persister cells are consistent with what was reported for E. coli persisters against another fluoroquinolone antibiotic, ofloxacin (Goormaghtigh and Van Melderen, 2019). On the other hand, we found another type of persister cells that resumed normal growth and division without showing prominent filamentation (Figure 8C and D, Figure 8—figure supplement 1). Therefore, extensive filamentation after CPFX treatment is not necessarily required to restore normal growth and division in persister cells.

In addition, comparison of the pre-exposure division rate showed no significant difference between growing persisters of MG1655 (0.94±0.06 doublings/h) and non-surviving cells (0.86±0.03 doublings/h; ${\displaystyle p}$ = 0.12, Wilcoxon rank sum test) (Figure 8E). Thus, MG1655 persisters to CPFX grow as fast as non-surviving cells prior to drug exposure. This result is again consistent with the results observed for the persistence of post-exponential phase E. coli cell populations against ofloxacin (Goormaghtigh and Van Melderen, 2019).

The E. coli strains MG1655 and MF1 (MG1655 rpoS-mcherry/pUA66-P_rpsL-gfp) were mainly used in this study. pUA66-P_rpsL-gfp is a low-copy plasmid taken from the E. coli promoter library (Zaslaver et al., 2006). In the propidium iodide (PI) staining experiment, we used the MG1655/pUA66-P_rpsL-gfp strain to examine the intactness of cellular states by GFP fluorescence in addition to susceptibility to PI staining. In the experiment testing the susceptibility of the MF1 strain to oxidative stress, we also used the MG1655 ΔrpoS strain.

For constructing the MF1 strain, a DNA fragment containing mcherry and kanamycin-resistance genes was amplified by PCR with the template pLmCherryK6Δterm and the primers rpoS-mcherry-F (CAAACGCAGG GGCTGAATAT CGAAGCGCTG TTCCGCGAGA GTGATTTTAT GGTGAGCAAG GGCGAGGA) and rpoS-mcherry-R (CAGCCTCGCT TGAGACTGGC CTTTCTGACA GATGCTTACA TTCCGGGGAT CCGTCGACC) and integrated downstream of the rpoS locus on the chromosome of BW25113 strain by ${\displaystyle \lambda }$-Red recombination (Datsenko and Wanner, 2000) to express the RpoS-mCherry fusion protein from the native locus. The rpoS-mcherry gene and kanamycin-resistance gene flanked with two FRT sequences were transferred to MG1655 by P1 transduction. Kanamycin-resistance gene was removed by transforming pCP20 plasmid. The pCP20 plasmid was removed from the host cells by culturing them at 37°C overnight.

To construct the MG1655 ΔrpoS strain, the FRT-kanamycin resistance gene-FRT cassette in the JW5437 strain (BW25113 ΔrpoS, a strain in the Keio collection [Baba et al., 2006]) was transferred into MG1655 by P1 transduction. Removal of the kanamycin resistance gene and subsequent exclusion of the pCP20 plasmid from the cell was performed as for the construction of the MF1 strain.

We cultured the cells with either Luria-Bertani (LB) broth (Difco) or M9 minimal medium (Difco) supplemented with 1/2 MEM amino acids solution (SIGMA) and 0.2% (w/v) glucose. All cultivation was done at 37°C.

5 µL of the glycerol stock cell suspension was inoculated into 2 mL of LB or M9 minimal medium containing 40 µg/mL of kanamycin (Kan), and we cultured the cells at 37°C for 15 h (LB) or 16 h (M9) with shaking. Kan was added to the culture media to avoid the loss of the GFP-expressing plasmid (pUA66-P_rpsL-gfp) during the pre-culture of MF1. We diluted this pre-culture cell suspension with fresh LB or M9 medium (without Kan) to the cell density corresponding to the optical density at 600 nm (OD₆₀₀) being 0.005 (LB) or 0.01 (M9). The total volume of the diluted cell cultures was adjusted to 50 mL. After the dilution, we started culturing cells at 37°C with shaking. We sampled 100 µL of the cell cultures every 1 h and measured OD₆₀₀ with a spectrophotometer (UV-1800, Shimadzu). We repeated the measurements four times for MG1655 in LB, three times for MG1655 in M9, five times for MF1 in LB, and five times for MF1 in M9. The mean and standard error of the logarithm of the OD₆₀₀ were calculated for each sampling time point and plotted as the point and error bar in Figure 1—figure supplement 4. In Figure 1—figure supplement 1A and B, we show only the growth curves from one experiment for each condition, in which we measured RpoS-mCherry fluorescence of individual cells in addition to the optical density of the cell cultures.

An overview of the sample preparation for the killing curve assay is shown schematically in Figure 6—figure supplement 1. To measure killing curves in batch cultures, we first prepared 50 mL cell suspensions of OD₆₀₀ = 0.005 (LB) or OD₆₀₀ = 0.01 (M9) following the same protocol as for growth curve measurements (first pre-culture). We incubated the diluted cell suspensions at 37°C with shaking (second pre-culture) and sampled a portion of the cell cultures at a pre-determined time point (2 h for the exponential phase in LB; 8 h for the early stationary phase in LB; 24 h for the late stationary phase in LB; 4 h for the exponential phase in M9; and 24 h for the late stationary phase in M9). To match the pre-culturing conditions between the killing curve assays and the single-cell measurements, we diluted the extracted cell suspension with the fresh LB or M9 medium to the cell density with which the OD₆₀₀ of the diluted cell suspension would reach 0.05 in the pre-culture of 2 h (LB) or 4 h (M9). The volume of the cell cultures was adjusted to 50 mL. After the pre-culture, we added the antibiotics (Amp or CPFX) to the cell culture and incubated it at 37°C with shaking. The concentrations of the antibiotics were adjusted to 200 µg/mL for Amp and 1 µg/mL for CPFX. We extracted 50 µL or 500 µL of the cell suspensions at the pre-determined time points to estimate the number of viable cells in the cellular populations.

To avoid underestimating the viable cells due to non-culturability on solid agar (Gelman et al., 2012; Nosho et al., 2018), we estimated the viable cell number by a limiting dilution method. We conducted serial dilution of the extracted cell suspension (10× dilution in each dilution step) in 5 mL. We sampled 200 µL of each diluted cell suspension into a well in a 96-well plate. For each dilution condition, cell suspensions were mostly dispensed into 12 wells. For later time points of the post-exponential and post-early stationary phase cell populations, cell suspensions were dispensed into 21, 24, or 36 wells to detect low-frequency surviving cells. We incubated the 96-well plates at 37°C for 36 h (LB) or 48 h (M9), and the turbidity of each well was inspected.

We found one or two dilution conditions for each sample taken from different time points where both turbid and non-turbid wells coexist among the dispensed 12 or more wells. Assuming Poissonian allocation of viable cells, the proportion of non-turbid wells, ${\displaystyle x_0}$, is equal to ${\displaystyle e^{-dy{N}_v}}$, where ${\displaystyle N_v}$ is the original concentration of viable cells, ${\displaystyle d}$ is the dilution level, and ${\displaystyle y}$ (= 0.2 mL) is the volume; then, ${\displaystyle N_v=(yd{)}^{-1}\ln (1/x_0)}$. Thus, we estimated the concentration of viable cells in the cell population as ${\displaystyle N_v=(5/d)\ln (W/k)}$ cells/mL, where ${\displaystyle {\displaystyle W}}$ is the number of dispenced total wells, and ${\displaystyle k}$ is the number of non-turbid wells. For example, if we found three non-turbid and nine turbid wells for the cell suspension with the dilution of 10⁻⁶, we could estimate the viable cell density as ${\displaystyle N_v=5\times {10}^6\times \ln (12/3)}$ cells/mL ${\displaystyle =7\times {10}^6}$ cells/mL. When the coexistence of turbid and non-turbid wells was found in the two dilution conditions, we estimated the viable cell density of this time point by the geometric mean of the two estimated values.

We repeated the assay three or more times for each culture condition. The mean and standard error of the logarithm of the estimated viable cell density for each sampling time were calculated and shown as the point and error bar in Figure 1—figure supplement 3, Figure 6B, Figure 6—figure supplement 1, and Figure 8—figure supplement 2.

5 µL of the glycerol stock of the E. coli strain was inoculated into 2 mL of LB or M9 medium and incubated at 37°C with shaking for 15 h (LB) or 16 h (M9). We added Kan at the concentration of 40 µg/mL when we cultured the MF1 strain from the glycerol stock to avoid the loss of the GFP-expressing plasmid (pUA66-P_rpsL-gfp) during the pre-culture. The cell cultures were diluted with a fresh LB or M9 medium to the cell density corresponding to OD₆₀₀ = 0.005 (LB) or OD₆₀₀ = 0.01 (M9). The diluted cell cultures of 2 mL were incubated at 37°C with shaking for 2.5 h (LB) or 3 h (M9) to let the cell cultures enter exponential phases. The cell cultures were again diluted with a fresh medium to the cell density corresponding to OD₆₀₀ = 0.0002. We mixed 100 µL of the diluted cell suspension with 100 µL of the medium containing Amp or CPFX and dispensed it into a well of the 96-well plate. We prepared 12 antibiotic concentration conditions with 2-fold serial differences (Amp: 2⁻²–2⁸ µg/mL (+0 µg/mL); CPFX: 2⁻¹⁰–2⁰ µg/mL (+0 µg/mL)). We used three wells for each drug concentration condition. The cell cultures in the 96-well plates were incubated at 37°C with shaking for 23 h. We measured the optical density at 595 nm (OD₅₉₅) of each well using a plate reader (FilterMax F5, Molecular Devices) and calculated the geometric mean of the three wells as the optical density of the cell culture in each drug concentration. We repeated the experiment three times and determined the MIC as the drug concentration above which the geometric mean of OD₅₉₅ among the three replicate experiments were below 0.001.

We followed the same protocol as in ‘Growth curve measurements’ above to prepare cell cultures of MF1 for measuring the growth curves and RpoS-mCherry fluorescence level transitions shown in Figure 1—figure supplement 1A and B. In addition to measuring OD₆₀₀, we collected 1 µL of cell suspension every hour and placed it between a PBS block solidified with 1.5% agarose and a coverslip.

We acquired brightfield, RpoS-mCherry fluorescence, and GFP fluorescence images using a microscope. The optical configuration and settings of the microscope used for this measurement were the same as those used for the time-lapse measurement described in detail below (see ‘Time-lapse observation with the MCMA microfluidic device’). We used Fiji (Schindelin et al., 2012) to find the outlines of individual cells and quantify their RpoS-mCherry fluorescence levels.

We used the MG1655, MG1655 ΔrpoS, and MF1 strains in this test. 5 µL of the glycerol stock of each strain was inoculated into 2 mL of M9 medium and incubated overnight at 37°C with shaking. Kan was added to the culture of MF1 during this pre-culture to avoid the loss of the GFP-expressing plasmid (pUA66-P_rpsL-gfp). The optical density of each overnight culture was measured, and the culture was diluted in 20 mL of fresh M9 medium to a cell density corresponding to OD₆₀₀ = 0.01. The culture was incubated for 4 h with shaking. 2 mL of cell suspension was taken into each of two microtubes per strain, and the cells were spun down by centrifugation. The cell pellet was resuspended in 2 mL of sterile water, and the cell suspension was transferred to a glass test tube. To one of the tubes, 4 µL of 30% H₂O₂ was added to adjust the final concentration of H₂O₂ to 20 mM. An equal volume of sterile water was added to the other tube (control). These cell cultures were incubated at 37°C for 20 min with shaking. 500 µL of cell suspension was transferred from each test tube to a microtube. The cells were centrifuged, and the cell pellet was resuspended in 500 µL of M9 medium. The cell suspension was serially diluted in multiple steps with a 5-fold dilution in each step. 100 µL of each cell suspension was plated on an LB agar plate and incubated overnight at 37°C.

By counting the number of colonies on the plate for the dilution condition with 30–300 colonies, we estimated the viable cell density in each cell culture. When we found 30–300 colonies for multiple dilution conditions, we adopted the mean of them as the estimated viable cell density. For each strain, the viable cell density of the cell culture exposed to 20 mM H₂O₂ was divided by that of the control cell culture. This relative frequency value is shown in Figure 1—figure supplement 1C.

We fabricated microchamber arrays on glass coverslips by wet etching. The dimensions of each microchamber were 62 µm (w) × 47 µm (h) × 0.8 µm (d) for the array used in the experiments with a 100× objective (The number of microchambers in the array was 1240), and 53 µm (w) × 33 µm (h) × 0.8 µm (d) for the array used in the experiments with a 40× objective (The number of microchambers in the array was 22,302).

We first fabricated the photomask for the microchamber array with mask blanks (CBL4006Du-AZP, CLEAN SURFACE TECHNOLOGY) using a laser drawing system (DDB-201-TW, Neoark or DLS-50, Nano System Solutions). The photoresist on mask blanks was developed in NMD-3 (Tokyo Ohka Kogyo). The uncovered chromium (Cr)-layer was removed in MPM-E350 (DNP Fine Chemicals), and the remaining photoresist was removed by acetone. Lastly, the slide was rinsed in MilliQ water and air-dried.

For creating microchambers based on the pattern on the photomask, we first coated a 1,000-angstrom Cr-layer on a clean coverslip (NEO Micro glass, No. 1., 24 mm × 60 mm, Matsunami) by evaporative deposition and AZP1350 (AZ Electronic Materials) by spin-coating on the Cr-layer. After the pre-bake (95°C, 90 sec), we exposed the photoresist layer on the coverslip to UV light through the photomask using a mask aligner (MA-20, Mikasa). We developed the photoresist in NMD-3 and removed the part of Cr-layer uncovered by the photoresist in MPM-E350. The coverslip was soaked in buffered hydrofluoric acid solution (110-BHF, Morita Kagaku Kogyo) for 14 min 20 sec at 23°C for glass etching. The etching reaction was stopped by soaking the coverslip in milliQ water. The remaining photoresist and the Cr-layer were removed by acetone and MPM-E350, respectively.

We used a polydimethylsiloxane (PDMS) pad with bubble trap groove to control and change the culture conditions around the cells in the microchamber array. We followed the same procedure as that reported in Yamauchi et al., 2022 to fabricate the PDMS pad.

To enclose the E. coli cells in the microchambers, we employed a method of attaching a cellulose membrane to a glass coverslip via biotin-streptavidin binding (Inoue et al., 2001). We chemically decorated glass coverslips with biotin and cellulose membranes with streptavidin. We followed the same protocol reported in Yamauchi et al., 2022 to prepare biotin-decorated coverslips and streptavidin-decorated cellulose membranes.

As described below (‘Sample preparation for time-lapse measurements’), we used the conditioned media to maintain the growth-suppressive conditions while preparing single-cell measurements of the cells sampled from the early and late stationary phases. We prepared 50 mL of the E. coli cell suspension at OD₆₀₀ = 0.005 in the LB medium or at OD₆₀₀ = 0.01 in the M9 medium following the same procedure of growth curve measurements and incubated the cell cultures at 37°C with shaking. At the pre-determined time (8 h for the early stationary phase condition and 24 h for the late stationary phase condition), we spun down the cells at 2,200×g for 12 min. The supernatant was again spun down at 2,200×g for 12 min, and its supernatant was filtered with a 0.2-µm syringe filter. We used this filtered medium as the conditioned medium.

To prepare the agarose blocks of conditioned media, we dissolved 2% (w/v) low-melt agarose in the conditioned media with a microwave and solidified it in a plastic dish. To use this for attaching a cellulose membrane to a coverslip, we excised a piece of the conditioned medium pad with a sterilized blade and placed it on the cellulose membrane.

An overview of the sample preparation for the killing curve assay is shown schematically in Figure 1C. To prepare E. coli cells for the time-lapse single-cell observation, 5 µL of the glycerol stock of the E. coli cells was inoculated into 2 mL LB or M9 medium and incubated at 37 °C with shaking for 15 h (LB) or 16 h (M9) (first pre-culture). The cell culture was diluted to OD₆₀₀ = 0.005 (in the case of LB medium), OD₆₀₀ = 5×10⁻⁶ (in the case of the 1000-fold more diluted condition in LB medium, Figure 3—figure supplement 1) or OD₆₀₀ = 0.01 (in the case of M9 medium) in a 50 mL fresh medium and again incubated at 37°C with shaking. We sampled cells from the cell culture at the pre-determined time points (2∼2.5 h for the post-exponential phase condition in LB; 6 h for the post-exponential phase condition grown from the diluted cell suspension; 8 h for the post-early stationary phase condition in LB; 24 h for the post-late stationary phase condition in LB; 4 h for the post-exponential phase condition in M9; and 24 h for the post-late stationary phase condition in M9; second pre-culture).

For the post-exponential phase conditions (both LB and M9), 1 µL of cell suspension in the second pre-culture was sampled from the culture and placed on the microchamber array. We put a streptavidin-decorated cellulose membrane on the cell suspension and removed an excess suspension with a clean filter paper. A small piece of an LB or M9 agarose block (2% (w/v) agarose) was placed on the cellulose membrane and incubated at 37°C for 15 min to allow the membrane to adhere to the coverslip surface tightly. After removing the agarose block, a PDMS pad was attached to the coverslip via a Frame-Seal Incubation Chamber (9 mm×9 mm, 25 µL internal volume; or 15 mm×15 mm, 65 µL internal volume, Bio-Rad). We connected the medium inlet and outlet of the PDMS pad to silicone tubes and immediately started to flow LB or M9 medium at a rate of 32 mL/h until the Frame-Seal Incubation Chambers and tubes were filled with the medium, and then at 2 mL/h using syringe pumps (NE-1000, New Era Pump Systems). After registering the microchamber positions, time-lapse measurements were started and continued as described in the following section.

For the early and late stationary phase samples, we followed the same protocol except that we used the conditioned medium agarose pads made from the cell cultures of the corresponding growth phases to attach the cellulose membrane. We also flowed the liquid conditioned medium while registering the microchamber positions to maintain the growth-suppressive conditions. We switched the flowing media from the conditioned media to the fresh medium at the start of the time-lapse measurements to arrange the same duration of regrowth before the Amp exposure across the replicate experiments.

We used Nikon Ti-E microscopes equipped with ORCA-flash camera (Hamamatsu Photonics) for the time-lapse observations. For the time-lapse observations with a 40× objective (Plan Apo λ, 40×/NA0.95), we acquired only bright-field images to monitor many microchamber positions simultaneously. For the time-lapse observations with a 100× objective (Plan Apo λ, 100×/NA1.45, oil immersion), we acquired bright-field for MG1655 and MF1, RpoS-mCherry fluorescence, and GFP fluorescence images for MF1. We used an LED illuminator (MCWHL2, Thorlabs) for fluorescence excitation. We maintained the temperature around the microscope stage at 37°C with a thermostat chamber (TIZHB, Tokai Hit).

The microscope was controlled from a PC using Micromanager (Edelstein et al., 2014). The time-lapse intervals were 3 min in the measurements with the LB medium and 6 min in the measurements with the M9 medium.

In the time-lapse measurements of the cells in the LB medium, we first allowed the cells to grow in the microchambers flowing fresh medium at a rate of 2 mL/h for either 1.5 h (the cells from the exponential phase) or 2 h (the cells from the early and late stationary phases). We then changed the flowing media to the one containing 200 µg/mL of Amp rapidly at an increased flow rate of 32 mL/h for 3.75 min and then observed the cells flowing drug-containing medium at a rate of 2 mL/h for 6 h for MG1655 and 3.5 h for MF1. We quickly changed the flowing media back to the fresh medium without drug at an increased flow rate of 32 mL/h for 3.75 min and allowed the viable cells to re-proliferate by flowing the drug-free fresh medium at a rate of 2 mL/h. We typically observed cells for 12 h after antibiotic exposure. If cells begin to grow rapidly after treatment and their progeny grow into adjacent chambers, we stopped observing these chambers earlier than 12 h.

In the time-lapse measurements of the cells in the M9 medium, we allowed the cells to grow in the microchambers for 4 h flowing a fresh M9 medium. We then switched the flowing media to the one containing 200 µg/mL Amp or 1 µg/mL CPFX and observed the cells under the drug-exposed conditions for 6 h. We again switched the flowing media back to the M9 medium containing no drug and allowed the viable cells to regrow. We typically observed cells for 24 h after antibiotic exposure. As in the LB condition, if cells begin to grow rapidly after treatment and their progeny grow into adjacent chambers, we stopped observing these chambers earlier than 24 h.

Based on the protocol described in the previous two sections, the upper bound for the fraction of long lagging cells in the post-exponential phase cell populations at the time of Amp exposure in the single-cell measurements can be estimated. In this protocol, cell populations enter stationary phase in the first pre-culture (Figure 1C). This stationary phase cell culture is diluted in fresh medium to a cell density corresponding to OD₆₀₀ = 0.005 in the case of LB condition. For the experiment with MF1, we typically incubated this cell culture for 2.5 h at 37°C with shaking for the post-exponential phase condition. We maintained growth conditions while loading cells into the MCMA device by placing LB agarose block at 37°C and while registering microchamber positions for time-lapse measurements by flowing fresh LB medium into the device at 37°C. This preparation step before the time-lapse measurements took approximately 1.5 h. The pre-exposure period in the time-lapse measurements (Figure 1D) was 1.5 h. In total, cells were placed under exponential growth conditions for approximately 5.5 h prior to Amp exposure. Consequently, the lagging cell fraction present at the beginning of the second pre-culture should have reduced its relative frequency in the cell population to ${\displaystyle 2^{-16}\approx {10}^{-5}}$ assuming a mean doubling time of 20 min. This sets the upper bound for the fraction of lagging cells. In addition, not all lagging cells can produce persister cells. Therefore, the probability of finding non-growing persister cells coming from the lagging cell fraction in our experimental setup is low.

To further eliminate such long lagging cell fractions, we also performed experiments in which the second pre-culture was started from a 1000-fold diluted inoculum and the duration of the second pre-culture was extended by 3.5 h to allow the cell population to reach the same cell density in the exponential phase as in the standard post-exponential phase cell populations described above. This change in the pre-culture procedure reduces the frequency of long lagging cells by a factor of 1000 at the time of Amp treatment, and the probability of detecting non-growing persisters from long-lagging cells becomes virtually zero in our experimental setup.

We used a custom Fiji (Schindelin et al., 2012) macro to analyze the time-lapse images acquired with the 40× objective or the 100× objective. Cell outlines were manually detected and represented as polygons. The macro assigns the cell lineages based on the segmented cell image sequences. We corrected the errors in the segmentation and lineage assignments manually. The cell size (area), circularity, and the connections in the cell lineages were exported into a text file for the data analysis.

We opened the images as a stack to find the persister cell lineages in acquired the time-lapse images. From the end time point, we played back the stacked images to detect the microchambers with the surviving cells that stably re-proliferated after the drug removal. We tracked back the persistent cell lineages and determined the existence of growth and cell divisions before the drug exposure.

To estimate the number of drug-exposed cells, we randomly selected 20 microchambers and counted the number of cells in each microchamber. We multiplied the averaged number of the cells in a microchamber by the number of the microchambers tracked in the time-lapse measurement. We used this number as the estimate of the total number of the drug-exposed cells in each experiment (Figure 3D, Figure 3—figure supplement 1).

We found that many non-growing cells in the post-late stationary phase cell populations remained intact during the Amp exposure but did not regrow after the drug removal. We examined the viability of the non-growing cells flowing the media containing 1 µg/mL PI in the time-lapse measurements. In these experiments, we used the MG1655 pUA66-P_rpsL-gfp strain. The preparation of the cells for the time-lapse measurements was the same as that used for the cells sampled from the late stationary phase, except that we flowed the media containing PI from the time point of adding Amp.

E. coli cell strains and vectors used in this study are available from the authors upon request.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We acknowledge Nathalie Balaban for comments on the initial version of the manuscript; Shoji Takeuchi for his support of the microfabrication facility; and members of the Wakamoto lab for discussion. This work was supported by JST CREST Grant Number JPMJCR1927 (YW); JST ERATO Grant Number JPMJER1902 (YW); NIH Grant Number R01-GM097356 (EK); and Japan Society for the Promotion of Science KAKENHI Grant Number 17H06389, 19H03216, and 24H00552 (YW).

© 2025, Umetani, Fujisawa et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.