Observation of non-dormant persister cells reveals diverse modes of survival in antibiotic persistence

  1. Miki Umetani
  2. Miho Fujisawa
  3. Reiko Okura
  4. Takashi Nozoe
  5. Shoichi Suenaga
  6. Hidenori Nakaoka
  7. Edo Kussell
  8. Yuichi Wakamoto

  • Curated by eLife

    eLife logo

    eLife assessment

    The work is interesting in its characterization of a large number of antibiotic persisters from a wild-type strain. Previous work was typically limited to directly observe either high persister strains or a smaller number of wt persisters. Therefore, it sheds new light on the elusive non-dormant persisters present in exponentially growing cultures and should help resolve previous conflicting observations.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Bacterial persistence is a phenomenon in which a small fraction of isogenic bacterial cells survives a lethal dose of antibiotics. It is generally assumed that persistence is caused by cells with inactive growth generated prior to drug exposure. However, evidence from direct observation is scarce due to the extremely low frequencies of persisters. Here, we visualize the responses of more than 10 6 individual cells of wildtype Escherichia coli to lethal doses of antibiotics, sampling cells from different growth phases and culture media. We show that preexisting growth-arrested dormant persisters constitute only minor fractions of persistent cell lineages in populations sampled from exponential phase. Non-dormant persisters survive with radical morphological changes in response to drug exposure, including L-form-like morphologies or filamentation. Incubating cells under stationary phase conditions increases both the frequency and the probability of survival of dormant cells. While dormant cells in late stationary phase express a general stress response regulator, RpoS, at high levels, persistent cell lineages show lower levels of RpoS among the dormant cells. These results demonstrate that diverse survival pathways coexist within bacterial populations to achieve persistence and that persistence does not necessarily require dormant cells.

Article activity feed

  1. eLife

    Author response:

    Reviewer #1 (Public Review):

    The work of Umetani et al. monitors the death of about 100,000 cells caused by lethal antibiotic treatments in a microfluidic device. They observe that the surviving bacteria are either in a dormant or in a non-dormant state prior to the antibiotic treatment. They then study the relative abundances of these different persister cells when varying the physiological state of the culture. In agreement with previous observations, they observe that late stationary phase cultures harbor a high number of dormant persister cells and that this number goes down as the culture is more exponential but remains non-zero, suggesting that cultures at the exponential phase contain different types of persister bacteria. These results were qualitatively similar in a rich and poor medium. Further characterization of the growing persister bacteria shows that they often form Lforms, have low RpoS-mcherry expression levels and grow only slightly more slowly than the non-persister bacteria. Taken together, these results draw a detailed view of persister bacteria and the way they may survive extensive antibiotic treatments. However, in order to represent a substantial advance on previous knowledge, a deeper analysis of the persister bacteria should be done.

    We thank the reviewer for suggesting the addition of more detailed analyses of persister cells. As we wrote in our response to Essential Revision 1, we now include a new section titled “Response of growing persisters to Amp exposure is heterogeneous” (Page 11-12) and present the results of the detailed analyses of single-cell dynamics of growth and cell morphology over the course of the pre-exposure, exposure, and post-exposure periods (Fig. 2D and H, Fig. 4B and D, Fig. 4 – figure supplement 1 and 2, Fig. 5B and D, Fig. 5 – figure supplement 1, Fig. 8B and D, and Figure 8 – figure supplement 1). The new results characterize differential responses to Amp treatment among growing persister cells (Fig. 4A-D, Fig. 4 – figure supplement 1, Fig. 4 – figure supplement 2A, Fig. 5A-D, and Fig. 5 – figure supplement 1), comparable division rates of MG1655 between non-surviving cells and persister cells growing prior to antibiotic treatments (Fig. 4E and Fig. 8E), except for the post-exponential phase cell populations of MF1 to Amp treatment in the LB medium and the post-exponential phase cell populations of MG1655 to Amp treatment in the M9 medium (Fig. 4 – figure supplement 2B and Fig. 5E) and the presence of persister cells to CPFX that avoid filamentation after the treatment (Fig. 8C and D, and Fig. 8 – figure supplement 1). We believe that these new analyses would provide new insights into the diverse dynamics and survival modes of antibiotic persistence at the single-cell level and represent important contributions to the field.

    Reviewer #2 (Public Review):

    The main question asked by Umenati et al. is whether persister cells to ampicillin arise preferentially from dormant, non-dividing cells or from cells that are actively growing before antibiotic exposure. The authors tracked persister cells generated from populations at different growth phases and culture media using a microfluidic device coupled to fluorescence microscopy, which is a challenge due to the low frequency of these persister cells. One of the main conclusions is that the majority of persisters arising in exponentially-growing populations originated from actively-dividing cells before the antibiotic treatment, reinforcing the idea that dormancy is not a prerequisite for persister formation. The authors made use of a fluorescent reporter monitoring RpoS activity (RpoS-mCherry fusion) and observed that RpoS levels in these persister cells were low. In the few lineages that exhibited no growth before the ampicillin treatment, RpoS levels were low as well, indicating that RpoS is not a predictive marker for persistence. By performing the same experiment with early and late stationary phase cultures, the authors observed that the proportion of persister cells that originated from dormant cells before the ampicillin treatment is significantly increased under these conditions. In the late stationary phase condition, dormant cells were expressing high levels of RpoS. The authors suggested that RpoS-mCherry proteins form aggregates which were suggested by the authors to be a characteristic of 'deep dormancy'. These cells were mostly unable to restart growth after the antibiotic removal while others with the lowest levels of RpoS tended to be persister. Confirming that these cells indeed contain protein aggregates as well as determining the physiological state of these cells appears to be crucial.

    We thank reviewer #2 for pointing out the critical issue with the RpoS-mCherry fusion that we used to quantify RpoS expression levels in single cells in the original manuscript. As explained in our reply to the comments below, we performed a suggested experiment and confirmed that the RpoS function was impaired by tagging it with mCherry. To resolve this issue, we repeated almost all the experiments using the wild-type strain MG1655 and confirmed the reproducibility of the main results (Fig. 3, Fig. 3 – figure supplement 1, and Fig. 7). Due to this change of the main strain used in this study, we removed the results on the correlation between RpoS expression and the persistence trait in the revised manuscript because it may not reflect the relationship of intact RpoS. However, we decided to still keep and show some of the results with the MF1 strain, such as the population killing curves and the survival mode analyses, because they also provide insight into the role of RpoS in antibiotic persistence. In particular, we found both beneficial and detrimental effects of RpoS on antibiotic persistence, depending on culture conditions and duration of antibiotic treatment (Fig. 1 – figure supplement 3 and Fig. 6 – figure supplement 1). Therefore, we have included these results and related discussions in the revised manuscript.

    Reviewer #3 (Public Review):

    In their manuscript, Umetani, et al. address the question of the origin of persister bacteria using single-cell approaches. Persistence refers to a physiological state where bacteria are less sensitive to antibiotherapy, although they have not acquired a resistance mutation; importantly, the concept of persistence has been refined in the past decade to distinguish it from tolerance where bacteria are only transiently insensitive. Since persister cells are very rare in growing populations (typically 1e-5 or 1e-6), it is very challenging to observe them directly. It had been proposed that individual cells surviving antibiotics are not growing at the start of the treatment, but recent studies (nicely reviewed in the introduction) where persister bacteria were observed directly do not support this link. Following a similar line, the authors nonetheless still aim at "investigating whether non-growing cells are predominantly responsible for bacterial persistence". Based on new experimental data, they claim the contrary that most surviving cells were "actively growing before drug exposure" and that their work "reveals diverse survival pathways underlying antibiotic persistence".

    We thank the reviewer for this helpful comment, which suggested to us that some revisions in our Introduction would better place our study in the context of previous understanding of antibiotic persistence. As mentioned in our response to Essential Revision 4 and the second comment of Reviewer 1's Recommendations for the authors, we have modified the Introduction to more appropriately place our study in the context of the field.

    The main strengths of the manuscript are in my opinion:

    - To report on direct observation of E. coli persisters to ampicillin (200µg/mL) in 5 different growth media (typically 20 persisters or more per condition, one condition with 12 only), which constitutes without a doubt an experimental tour de force.

    - To aim at bridging the population level and the single-cell level by measuring relevant variables for each and analyzing them jointly.

    - To demonstrate that in most conditions a large fraction of surviving cells was actively growing before drug exposure.

    In addition, although it is well-known that E. coli doesn't need to maintain its rod shape for surviving and dividing, I found very remarkable in their data the extent to which morphology can be affected in persister cells and their progeny, since this really challenges our understanding of E. coli's "lifestyle" (these swimming amoeba-like cells in Supp Video 11 are mind-blowing!).

    We are grateful to the reviewer for the articulation of the strength of this study.

    Unfortunately, these positive aspects are counter-balanced by several shortcomings in the way experiments are analyzed and interpreted, which I explain below. Moreover, the manuscript is written in a way that makes it very hard to find important information on how experiments are done and is likely to leave the reader with an impression of confusion about what the main findings actually are.

    We thank the reviewer for pointing out these important issues regarding the original manuscript. Please see our replies below regarding how we corresponded to each specific comment to resolve the issue. To make the experimental methods and procedures more accessible and interpretable, we have added more explanations of the experimental details to the Results and Methods sections. Furthermore, since we understood that some of the confusions came from the insufficient explanation of the preculture procedures for the microfluidic experiments, we have modified the schematic illustration of the method shown in Fig. S1 in the original manuscript and moved it as the first main figure in the revised manuscript (Fig. 1C and D). We have also added an illustration that explains the cultivation procedures for the batch culture experiments as Fig.

    6A.

    My major concerns are the following:

    (1) The main interpretation framework proposed by the authors is to assess whether cells not growing before drug exposure (so-called "dormant") are more or less likely to survive the treatment than growing ones ("non-dormant"). Fig 2A and Fig 3G show the main conclusions of the article from this perspective, that growing cells can survive the treatment and that the fraction of persisters in a given condition is not explained by the fraction of "dormant" cells, respectively. With this analysis, the authors essentially assume that "dormant" cells are of the same type in their different conditions, which ignores the progress in this field over the last decade (Balaban et al. 2019). I argue on the contrary that the observation of "diverse modes of survival in antibiotic persistence" is expected from their experimental design. In particular, the sensitivity of E. coli to beta-lactams such as ampicillin is expected to be much lower during the lag out of the stationary phase, a phenomenon which has been coined "tolerance"; hence in the Late Stationary condition, two subpopulations coexist for which different response to ampicillin is expected. I propose steps toward a more compelling interpretation of the experimental data. Should this point be taken seriously by the authors, it, unfortunately, implies a major rewriting of the article, including its title.

    We thank the reviewer for bringing to our attention the point that may have caused confusion in the original manuscript.

    The primary purpose of this manuscript was not to assess whether non-growing cells prior to drug exposure are more or less likely to survive treatment than growing cells. Rather, we wanted to examine how different persister cell dynamics emerge at the single-cell level depending on previous cultivation history, growth media, and antibiotic types. We believe that this point is clearer in the revised manuscript with the newly added single-cell dynamics data (Fig. 2D, 2H, 4B, 4D, Fig. 4 – figure supplement 1 and 2A, Fig. 5B, 5D, Fig. 5 – figure supplement 1, Fig. 8B, 8D, and Fig. 8 – figure supplement 1).

    We also did not mean to imply that "dormant cells" were of the same type under different conditions, as we were aware of the diversity of cellular states of non-growing cells, as well as the reduced sensitivity of cells to antibiotics during the lag out of stationary phase. We believe that one of the reasons this point may have been unclear is that in the previous version we had referred to all cells that were not growing prior to antibiotic treatment as "dormant cells", a term that is often used in a more restricted way to refer to cells under prolonged growth arrest. Therefore, in the revised manuscript, we have avoided the term "dormant cells" and instead simply referred to these as "non-growing cells". Accordingly, we have changed the title of the paper from "Observation of non-dormant persister cells reveals diverse modes of survival in antibiotic persistence" to "Observation of persister cell histories reveals diverse modes of survival in antibiotic persistence".

    To further address these points, we have improved the description of the experimental procedures for the single-cell measurements (see the reviewer's next comment as well). The nongrowing persisters of the MF1 strain found in the post-exponential phase cell populations must be of a different type than those found in the post-early and post-late stationary phase cell populations due to the experimental design. All early and late stationary phase cells were maintained in a non-growing state by flowing conditioned media prepared from the early and late stationary phase cultures until the start of the time-lapse measurements. Thus, aside from potential physiological heterogeneity, the non-growing cells prior to drug treatment are all long lagging cells. On the other hand, for the post-exponential phase condition, we maintained exponential growth conditions during the period from the start of the second pre-culture to the start of antibiotic treatment, including the period during sample preparation for time-lapse measurements. Given the exponential dilution by growth of cell populations, the non-growing persisters are unlikely to be long lagging cells (see our response to Reviewer 2's third comment in "Recommendations for the authors"). We now describe these experimental procedures in more detail in the Results section (L161-178, L287-297). In addition, we discuss the diversity of cellular states of both non-growing and growing cells in Discussion, citing literature (L545-557).

    (2) The way the authors describe their experiments with bacteria in the stationary phase is very problematic. For instance, they write that they "sampled cells from early and late stationary phases (...) and exposed them to 200 μg/mL of Amp in both batch and single-cell cultures." For any reader in a hurry (hence skipping methods and/or supplementary figure), this leads to believe that bacteria sampled in the stationary phase were exposed to the drug right away (either by adding the drug to the stationary phase sample, or more classically by transferring cells to fresh media with antibiotics). However, it turns out that, after sampling and loading in the microfluidic device, bacteria are grown 2 h in LB (or 4 h in M9) - I don't know what to think of such a blatant omission. The names chosen for each condition should reflect their most important aspects, here "stationary" is simply not appropriate - maybe something like "post early stationary" instead. In any case, I believe that this point highlights further the misconception pointed out in 1 and implies that the average reader will be at best confused, and probably misled.

    We again thank the reviewer for pointing out the insufficient explanation of the method for the single-cell measurements and the helpful recommendation regarding our nomenclature for different conditions. As mentioned above, we now present the previous supplementary figure that schematically explains the experimental procedure as the first main figure to clarify how we prepared the cells loaded into the microfluidic device for single-cell measurements (Fig. 1C and D). Also, following the reviewer's suggestion, we now refer to the conditions as "post-exponential phase," "post-early stationary phase," and "post-late stationary phase" in the revised manuscript.

    We included a 2-hour (or 4-hour in M9) cultivation period in fresh medium in batch cultures for measuring killing curves to make the cultivation conditions prior to antibiotic treatment as similar as possible between batch and microfluidic experiments. We have clarified the presence of preexposure cultivation of post-early stationary and post-late stationary phase cell populations in the fresh medium before treating them with antibiotics (L264-269, Fig. 6A), so that readers can more easily recognize the experimental conditions.

    (3) Figures 4 and 5 are of very minor significance, and the methodology used in Fig 4 is questionable. The authors measure the abundance of an Rpos-mCherry translational fusion because its "high expression has been suggested to predict persistence". The rationale for this (that an RpoS-mCherry fusion would be a proxy for intracellular ppGpp levels, and in turn predict persistence) has never been firmly established, and the standards used in the article where this reporter was introduced (Maisonneuve, Castro-Camargo, and Gerdes 2013) are notoriously low (which eventually led to its retraction) - I don't know what to think of the fact that the authors cite a review by this group rather than their retracted article. While transcriptional fusions of promoters regulated by RpoS have been proposed to measure its regulatory activity (Patange et al. 2018), the combination of self-regulation and complex post-translational regulation of rpoS makes the physical meaning of the reporter used here completely unclear. Moreover, this translational fusion is introduced without doing any of the necessary controls to demonstrate that the activity of RpoS is not impaired by the addition of the fluorescent protein. Fig 5 simply reports the existence of persisters to ciprofloxacin growing before the treatment. This might be a new observation but it is not unexpected given that a similar observation has been made with a similar drug, ofloxacin (Goormaghtigh and van Melderen 2019), as pointed out in the introduction. There is no further quantitative claim on this.

    We thank the reviewer for pointing out the issue of the RpoS-mCherry fusion. As we mentioned in our response to Essential Revision 2 and also to the comment from reviewer #2, we have tested the sensitivity of this fluorescent reporter strain to oxidative stress and confirmed that it is as sensitive as the rpoS strain (Fig. 1 – figure supplement 1C). Therefore, the RpoS function seems to be defective in this strain, as now explained in Results (L69-79). After confirming the problem with the RpoS-mCherry fusion, we removed all analyses and related arguments that relied on the RpoS expression level (previous Figure 4). In addition, we repeated almost all the experiments with the original MG1655 strain to confirm that the observed results are not specific to the problematic reporter strain.

    Regarding the experiments with CPFX, we have added a more detailed analysis of single cell dynamics and found that, contrary to the reported results for ofloxacin, not all persistent cells show filamentation after drug withdrawal (Fig. 8C and D, Fig. 8 – figure supplement 1). In addition, we performed new microfluidic experiments in which we treated post-late stationary phase cells with CPFX (Fig. 3). In contrast to the Amp treatment result and the previous study that reported the persistence of post-stationary phase cell populations to ofloxacin (ref. 20), all the persisters for which we identified the pre-exposure growth traits in this condition grew normally prior to CPFX treatment. These newly added analyses and experiments clarify the significance of the CPFX experiments.

    (4) The authors don't mention the dead volume nor the speed of media exchange in their device. Hopefully, it is short compared to the duration of the treatment; however, it is challenging to remove all antibiotics after the treatment and only 1e-3 or 1e-4 of the treatment concentration is already susceptible to affecting regrowth in fresh media. If this is described in another article, it would be worth adding a comment in the main text.

    We thank the reviewer for bringing up this important point. We have added the perfusion chamber volume and medium flow rate information in the Methods section (L809-817).

    In the study in which two of the authors participated, the medium exchange rate across the semipermeable membrane was evaluated in a similar device with similar microchamber dimensions (ref. 26). There, we confirmed that the medium exchange was completed within 5 min, which is much shorter than the period of antibiotic treatment and post-antibiotic treatment periods for observing regrowth. We have also included this information in the main text with the reference (L58-63).

    Despite the relatively high medium exchange rate, we cannot formally exclude the possibility that a small amount of antibiotic may remain in the device, e.g. due to non-specific adsorption on the internal surface of the microchambers. In such cases, the residual antibiotics may influence the physiological states of the cells and the regrowth kinetics in the post-exposure periods, as suggested by the reviewer. However, the frequencies of persister cells in the cell populations in our single-cell measurements are comparable to those in the batch culture measurements. Therefore, the removal of antibiotic drugs in our device is at least as efficient as in the batch culture assay. To clarify this point, we have added a paragraph to the Discussion with a reference that reviews the influence of antibiotics at concentrations significantly lower than the MICs (L482-

    489).

    (5) Fig 2A supports the main finding that a significant fraction of bacteria surviving the treatment are growing before drug exposure, but it uses a poorly chosen representation.

    - In order to compare between conditions, one would like to see the fraction of each type in the population.

    - The current representation (of a fraction of each type among surviving cells) requires a side-byside comparison with a random sample (which will practically be equivalent to the fraction of each type among killed cells) in order to be informative.

    We have changed the style of the previous Fig. 2A to show the fraction of each type in the population instead of the fraction of each type among surviving cells (Fig. 3 and Fig. 3-figure supplement 1).

  2. eLife

    eLife assessment

    The work is interesting in its characterization of a large number of antibiotic persisters from a wild-type strain. Previous work was typically limited to directly observe either high persister strains or a smaller number of wt persisters. Therefore, it sheds new light on the elusive non-dormant persisters present in exponentially growing cultures and should help resolve previous conflicting observations.

  3. eLife

    Reviewer #1 (Public Review):

    The work of Umetani et al. monitors the death of about 100,000 cells caused by lethal antibiotic treatments in a microfluidic device. They observe that the surviving bacteria are either in a dormant or in a non-dormant state prior to the antibiotic treatment. They then study the relative abundances of these different persister cells when varying the physiological state of the culture. In agreement with previous observations, they observe that late stationary phase cultures harbor a high number of dormant persister cells and that this number goes down as the culture is more exponential but remains non-zero, suggesting that cultures at the exponential phase contain different types of persister bacteria. These results were qualitatively similar in a rich and poor medium. Further characterization of the growing persister bacteria shows that they often form L-forms, have low RpoS-mcherry expression levels and grow only slightly more slowly than the non-persister bacteria. Taken together, these results draw a detailed view of persister bacteria and the way they may survive extensive antibiotic treatments. However, in order to represent a substantial advance on previous knowledge, a deeper analysis of the persister bacteria should be done.

  4. eLife

    Reviewer #2 (Public Review):

    The main question asked by Umenati et al. is whether persister cells to ampicillin arise preferentially from dormant, non-dividing cells or from cells that are actively growing before antibiotic exposure. The authors tracked persister cells generated from populations at different growth phases and culture media using a microfluidic device coupled to fluorescence microscopy, which is a challenge due to the low frequency of these persister cells. One of the main conclusions is that the majority of persisters arising in exponentially-growing populations originated from actively-dividing cells before the antibiotic treatment, reinforcing the idea that dormancy is not a prerequisite for persister formation. The authors made use of a fluorescent reporter monitoring RpoS activity (RpoS-mCherry fusion) and observed that RpoS levels in these persister cells were low. In the few lineages that exhibited no growth before the ampicillin treatment, RpoS levels were low as well, indicating that RpoS is not a predictive marker for persistence. By performing the same experiment with early and late stationary phase cultures, the authors observed that the proportion of persister cells that originated from dormant cells before the ampicillin treatment is significantly increased under these conditions. In the late stationary phase condition, dormant cells were expressing high levels of RpoS. The authors suggested that RpoS-mCherry proteins form aggregates which were suggested by the authors to be a characteristic of 'deep dormancy'. These cells were mostly unable to restart growth after the antibiotic removal while others with the lowest levels of RpoS tended to be persister. Confirming that these cells indeed contain protein aggregates as well as determining the physiological state of these cells appears to be crucial.

  5. eLife

    Reviewer #3 (Public Review):

    In their manuscript, Umetani, et al. address the question of the origin of persister bacteria using single-cell approaches. Persistence refers to a physiological state where bacteria are less sensitive to antibiotherapy, although they have not acquired a resistance mutation; importantly, the concept of persistence has been refined in the past decade to distinguish it from tolerance where bacteria are only transiently insensitive. Since persister cells are very rare in growing populations (typically 1e-5 or 1e-6), it is very challenging to observe them directly. It had been proposed that individual cells surviving antibiotics are not growing at the start of the treatment, but recent studies (nicely reviewed in the introduction) where persister bacteria were observed directly do not support this link. Following a similar line, the authors nonetheless still aim at "investigating whether non-growing cells are predominantly responsible for bacterial persistence". Based on new experimental data, they claim the contrary that most surviving cells were "actively growing before drug exposure" and that their work "reveals diverse survival pathways underlying antibiotic persistence".

    The main strengths of the manuscript are in my opinion:

    - To report on direct observation of E. coli persisters to ampicillin (200µg/mL) in 5 different growth media (typically 20 persisters or more per condition, one condition with 12 only), which constitutes without a doubt an experimental tour de force.

    - To aim at bridging the population level and the single-cell level by measuring relevant variables for each and analyzing them jointly.

    - To demonstrate that in most conditions a large fraction of surviving cells was actively growing before drug exposure.

    In addition, although it is well-known that E. coli doesn't need to maintain its rod shape for surviving and dividing, I found very remarkable in their data the extent to which morphology can be affected in persister cells and their progeny, since this really challenges our understanding of E. coli's "lifestyle" (these swimming amoeba-like cells in Supp Video 11 are mind-blowing!).

    Unfortunately, these positive aspects are counter-balanced by several shortcomings in the way experiments are analyzed and interpreted, which I explain below. Moreover, the manuscript is written in a way that makes it very hard to find important information on how experiments are done and is likely to leave the reader with an impression of confusion about what the main findings actually are.

    My major concerns are the following:

    (1) The main interpretation framework proposed by the authors is to assess whether cells not growing before drug exposure (so-called "dormant") are more or less likely to survive the treatment than growing ones ("non-dormant"). Fig 2A and Fig 3G show the main conclusions of the article from this perspective, that growing cells can survive the treatment and that the fraction of persisters in a given condition is not explained by the fraction of "dormant" cells, respectively. With this analysis, the authors essentially assume that "dormant" cells are of the same type in their different conditions, which ignores the progress in this field over the last decade (Balaban et al. 2019). I argue on the contrary that the observation of "diverse modes of survival in antibiotic persistence" is expected from their experimental design. In particular, the sensitivity of E. coli to beta-lactams such as ampicillin is expected to be much lower during the lag out of the stationary phase, a phenomenon which has been coined "tolerance"; hence in the Late Stationary condition, two subpopulations coexist for which different response to ampicillin is expected. I propose steps toward a more compelling interpretation of the experimental data. Should this point be taken seriously by the authors, it, unfortunately, implies a major rewriting of the article, including its title.

    (2) The way the authors describe their experiments with bacteria in the stationary phase is very problematic. For instance, they write that they "sampled cells from early and late stationary phases (...) and exposed them to 200 μg/mL of Amp in both batch and single-cell cultures." For any reader in a hurry (hence skipping methods and/or supplementary figure), this leads to believe that bacteria sampled in the stationary phase were exposed to the drug right away (either by adding the drug to the stationary phase sample, or more classically by transferring cells to fresh media with antibiotics). However, it turns out that, after sampling and loading in the microfluidic device, bacteria are grown 2 h in LB (or 4 h in M9) - I don't know what to think of such a blatant omission. The names chosen for each condition should reflect their most important aspects, here "stationary" is simply not appropriate - maybe something like "post early stationary" instead. In any case, I believe that this point highlights further the misconception pointed out in 1 and implies that the average reader will be at best confused, and probably misled.

    (3) Figures 4 and 5 are of very minor significance, and the methodology used in Fig 4 is questionable. The authors measure the abundance of an Rpos-mCherry translational fusion because its "high expression has been suggested to predict persistence". The rationale for this (that an RpoS-mCherry fusion would be a proxy for intracellular ppGpp levels, and in turn predict persistence) has never been firmly established, and the standards used in the article where this reporter was introduced (Maisonneuve, Castro-Camargo, and Gerdes 2013) are notoriously low (which eventually led to its retraction) - I don't know what to think of the fact that the authors cite a review by this group rather than their retracted article. While transcriptional fusions of promoters regulated by RpoS have been proposed to measure its regulatory activity (Patange et al. 2018), the combination of self-regulation and complex post-translational regulation of rpoS makes the physical meaning of the reporter used here completely unclear. Moreover, this translational fusion is introduced without doing any of the necessary controls to demonstrate that the activity of RpoS is not impaired by the addition of the fluorescent protein. Fig 5 simply reports the existence of persisters to ciprofloxacin growing before the treatment. This might be a new observation but it is not unexpected given that a similar observation has been made with a similar drug, ofloxacin (Goormaghtigh and van Melderen 2019), as pointed out in the introduction. There is no further quantitative claim on this.

    (4) The authors don't mention the dead volume nor the speed of media exchange in their device. Hopefully, it is short compared to the duration of the treatment; however, it is challenging to remove all antibiotics after the treatment and only 1e-3 or 1e-4 of the treatment concentration is already susceptible to affecting regrowth in fresh media. If this is described in another article, it would be worth adding a comment in the main text.

    (5) Fig 2A supports the main finding that a significant fraction of bacteria surviving the treatment are growing before drug exposure, but it uses a poorly chosen representation.
    - In order to compare between conditions, one would like to see the fraction of each type in the population.
    - The current representation (of a fraction of each type among surviving cells) requires a side-by-side comparison with a random sample (which will practically be equivalent to the fraction of each type among killed cells) in order to be informative.

  6. Version published to 10.1101/2021.10.28.466227v2 on bioRxiv
  7. Version published to 10.1101/2021.10.28.466227v1 on bioRxiv