Modifying TIMER to generate a slow-folding DsRed derivative for optimal use in quickly-dividing bacteria

It is now well appreciated that members of pathogenic bacterial populations exhibit heterogeneity in growth rates and metabolic activity, and it is known this can impact the ability to eliminate all members of the bacterial population during antibiotic treatment. It remains unclear which pathways promote slowed bacterial growth within host tissues, primarily because it has been difficult to identify and isolate slow growing bacteria from host tissues for downstream analyses. To overcome this limitation, we have developed a novel variant of TIMER, a slow-folding fluorescent protein, named DsRed42, to identify subsets of slowly dividing bacteria within host tissues. The original TIMER folds too slowly for fluorescence accumulation in quickly replicating bacterial species (Escherichia coli, Yersinia pseudotuberculosis), however DsRed42 accumulates red fluorescence in late stationary phase cultures of E. coli and Y. pseudotuberculosis. We show DsRed42 signal also accumulates during exposure to sources of nitric oxide (NO), suggesting DsRed42 signal detects growth-arrested bacterial cells. In a mouse model of Y. pseudotuberculosis deep tissue infection, DsRed42 signal was detected, and primarily accumulates in bacteria expressing markers of stationary phase growth. There was no significant overlap between DsRed42 signal and NO-exposed subpopulations of bacteria within host tissues, suggesting NO stress was transient, allowing bacteria to recover from this stress and resume replication. This novel DsRed42 variant represents a tool that will enable additional studies of slow-growing subpopulations of bacteria, specifically within bacterial species that quickly divide.


Introduction
Identifying differences in the growth rates of individual cells can provide significant information in many biological systems. Quickened growth rates can indicate transformative properties in mammalian cells, indicate that specific B cell populations are expanding in response to infection or vaccination, or suggest distinct differentiation programs are occurring within cell populations during the developmental stages of an organism. For populations of bacterial cells, rapidly dividing bacteria are generally more metabolically active and thereby more susceptible to antibiotic treatments, whereas slowly dividing cells have decreased metabolic activity and are less susceptible to antibiotic treatment [1][2][3].
Many research groups have begun to develop tools to differentiate between faster and slower dividing cells, which enable downstream studies to identify additional differences within subpopulations. Many of these tools, such as slow-folding fluorescent proteins, were initially developed to study eukaryotic systems [4,5]. However, several of these approaches have been adapted to study bacterial populations, primarily to better understand the pathways that lead to the formation of slow-growing subsets of bacteria with reduced antibiotic susceptibility. Some bacterial tools to detect differential cell division rates include: fluorescence dilution or plasmid loss approaches [6][7][8]; detection of DNA replication machinery [9,10]; and differential labeling of the cell envelope to detect newly synthesized macromolecules [11,12]; which can all be used in combination with fluorescence microscopy and live imaging techniques. It has been difficult to apply these techniques to identify slow growing bacteria in the context of mammalian infection models, since many infection models are not amenable to live imaging platforms, and instead utilize endpoint analyses. Bacterial population numbers can also contract and expand dramatically during infection [13,14], which can dilute fluorescent signals or plasmids very early after inoculation and result in a complete loss of signal.
Slow-folding fluorescent proteins may offer an alternative method to identify slow growing cells within bacterial populations. TIMER is a slow-folding derivative of the tetrameric fluorescent protein, DsRed, that has been successfully used to detect differential replication rates of Salmonella enterica serovar Typhimurium (S. Typhimurium) during growth in the spleen [15]. Wild-type (WT) DsRed generates both green and red monomers during fluorescence maturation; green monomers fold more quickly than red monomers, which require an extra amidation reaction during folding [16]. The green monomeric species is also less stable, resulting in incorporation of very few (~10%) green monomers into tetrameric DsRed, and predominantly red signal [16]. In contrast, TIMER contains a S197T mutation that stabilizes the green monomeric form [4,17]. When TIMER is initially expressed, this results in an early accumulation of predominantly green monomers and green fluorescent signal. With continued translation and folding, red monomeric species are incorporated into newly formed tetramers, which generates mixed 'orange' tetramers containing both green and red monomers [17]. In fastgrowing cells, cell division occurs before red monomers fold, and green signal predominates.
In slow-growing cells, red monomers are incorporated before cell division occurs; therefore mixed green and red tetramers, can be detected. To increase the fluorescent signal intensity for use in bacteria, TIMER bac (S4T, S197T relative to DsRed) was generated [15]. The accumulation of red fluorescent signal requires TIMER bac protein folding prior to cell division [15], and so the utility of this tool is dependent on the cell division rate of the cells of interest. Red TIMER bac signal emerges within S. Typhimurium dividing approximately 0.16 times/hour, or one doubling every 6 hours [15]. However, bacterial species such as Escherichia coli or Yersinia pseudotuberculosis can divide as quickly as 1.5-2 times/hour, or one doubling every [30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45] minutes, in culture [18,19]. When we began this study, it remained unclear if either TIMER or TIMER bac could be used to identify slow growing cells within bacterial populations that divide this rapidly, although our results and others suggest TIMER bac could be used [15,20,21].
Yersinia pseudotuberculosis in an enteric human pathogen that can escape the intestinal tract and replicate within deep tissue sites to form clonal extracellular clusters of bacteria called microcolonies (or pyogranulomas) [22][23][24]. Multiple immune cell types are recruited to the site of infection; however, bacteria continue to replicate, aided by their large arsenal of virulence factors [25][26][27]. Bacteria at the periphery of microcolonies respond to neutrophil contact by expressing high levels of the anti-phagocytic type-III secretion system and its associated effector proteins [24]. Peripheral bacteria also respond to and detoxify the diffusible antimicrobial gas, nitric oxide (NO), which is produced by recruited monocytes circumscribing the neutrophil layer [24,28]. NO detoxification occurs through expression of hmp, a nitric oxide dioxygenase, which converts antimicrobial NO to the innocuous NO 3 [29]. It remains unclear if the protective expression of hmp by peripheral cells comes at a fitness cost, and if this stress promotes the formation of a slow-growing subset of bacteria at the periphery of microcolonies. The preferential survival of hmp-expressing cells during antibiotic treatment supported this hypothesis [30], however our results here will suggest that NO exposure occurs transiently within host tissues, and that Hmp + cells retain the ability to divide.
Here, we sought to generate a new TIMER variant to detect differential growth rates of Y. pseudotuberculosis (Yptb) within host tissues using both site-directed and random mutagenesis. This variant, DsRed 42 , accumulates red fluorescent signal late in stationary phase Yptb, allowing us to detect slow-growing subsets of bacterial cells during growth in culture and within host tissues. DsRed 42 lacks green fluorescent signal, thus it is no longer a TIMER protein, but this allows us to compare red fluorescence accumulation to green fluorescent stress reporter constructs and ask if stressed cells have a heightened DsRed 42 expression. Using these constructs together, we show that NO stress is sufficient to promote DsRed 42 signal accumulation in culture; however, NO exposure within host tissues appears to be transient, and does not cause prolonged slowed growth. Instead, we detected a separate population of DsRed 42 + bacteria within host tissues, which also have markers of stationary phase cells (dps expression). DsRed 42 will be a very useful tool for future studies by enabling isolation of slow growing subsets of bacteria from host tissues for downstream analyses.

Site-directed and random mutagenesis were used to generate a novel TIMER variant, termed DsRed 42
To test the hypothesis that responses to reactive nitrogen species (RNS) were sufficient to slow the growth of stressed cells, we sought to develop tools to specifically detect slow-growing subpopulations of bacteria within Y. pseudotuberculosis microcolonies. We focused our efforts on TIMER, since this fluorescent protein had been successfully used to identify and isolate slowgrowing populations of S. Typhimurium from the mouse spleen [5,15]. TIMER was inserted into the high copy vector, pUC18, downstream of the IPTG-inducible P tac promoter, and transformed into E. coli. E. coli cultures were grown 24 hours (h) in the presence of IPTG to induce TIMER expression, and fluorescence was detected by plate reader using 480nm excitation and emission detection from 500nm to 600nm to detect the incorporation of both green and red monomers into TIMER tetramers [15]. Fluorescence was not detected in exponential phase cells (4h growth), and faint green signal could be detected from bacteria late in stationary phase (24h growth). However, this green signal remained below the background signal from non-fluorescent E. coli, and red TIMER signal did not accumulate (Fig 1A and 1B). We then generated a TIMER I161T T197A variant, which should lack green fluorescent signal and accumulate red signal more quickly due to quickened folding kinetics [1]. Because TIMER I161T T197A lacks green fluorescence, this is no longer a TIMER protein, but we have left the TIMER name here to depict it was derived from TIMER. We were hoping faster folding would allow us to detect red TIMER I161T T197A fluorescence accumulation in post-exponential or stationary phase cells, and eliminating the green monomeric species would allow us to use green fluorescent stress reporters alongside TIMER I161T T197A . However, TIMER I161T T197A also did not have detectable red fluorescent signal after 24 h growth with IPTG induction (Fig 1A and 1B). Additionally, we quantified expression of TIMER I161T T197A in a low copy vector, pMMB67EH, which we typically use for reporters in our mouse model of infection. We did not observe detectable fluorescence in stationary phase (24h growth) E. coli or Y. pseudotuberculosis (Yptb), in the presence or absence of IPTG induction (S1 Fig). Site-directed and random mutagenesis were used to generate a novel TIMER variant, termed DsRed 42 . A) Clones were grown in LB in the presence and absence of IPTG for 24 hours, shaking at 37˚C. Fluorescence was detected with a spectral curve with excitation at 480nm and emission detected between 500nm-600nm. Relative fluorescence units (RFUs) represent raw values with media only background subtracted. TIMER and DsRed clones are shown alongside the non-fluorescent E. coli parent strain. Data represents mean of three biological replicates. B) Optical density of the indicated strains was measured by absorbance (A 600nm ) at the indicated timepoints (hours, h). C) Gating strategy for fluorescenceactivated cell sorting. FSC, SSC, and pulse width were used to select the bacterial population (within R1 and R2 gates), and cells within gate R3 were defined as red fluorescent, and were collected. D) We hypothesized this lack of red fluorescence accumulation was because the folding of red monomeric TIMER I161T T197A was still too slow, relative to the doubling times of Yptb and E. coli. Therefore, we used the TIMER I161T T197A construct as a template for random mutagenesis, to specifically isolate variants that have red fluorescence accumulation in stationary phase cells. Since these variants have only red fluorescence, they will be referred to as DsRed variants from here on. Variants were inserted into pUC18, transformed into E. coli, and transformants were grown for 24h with IPTG. Non-fluorescent E. coli cultures were grown in parallel to define the threshold for fluorescence detection during cell sorting. Forward scatter (FSC), side scatter (SSC) and pulse width detection were used to identify individual bacterial cells, and we collected bacteria with detectable 572nm emission (gate R3) (Fig 1C). Collected bacteria (approximately 200 cells) were plated for single colonies, and 192 colonies were screened for red fluorescence during growth in the presence of IPTG by plate reader. 26 clones had detectable red fluorescence in stationary phase (24h growth) compared to exponential phase (4h growth); one clone, DsRed 42 , is shown as a representative (Fig 1A and 1B). Red fluorescence emerged between 16h and 20h of growth in the presence of IPTG (S2A- S2C Fig); some red fluorescence was also detected without IPTG induction (Figs 1A and S2D), likely due to the high copy number of pUC18. The 26 clones were sequenced, and their DsRed sequences represented three different 'types', numbered based on the well number of the first observed clone. Three of the sequenced clones were the DsRed 42 type, with a T22T synonymous mutation (ACC-> ACA) that is predicted to increase translation efficiency [31]; one clone was the DsRed 48 type, with a S69T mutation in the central helix that likely impacts chromophore maturation, since it is adjacent to chromophore residues 66-68 [11]; and the other 22 clones were the DsRed 56 type, which contain a TIMER I161T T197A gene with a single adenosine insertion upstream of the start codon, likely resulting in increased transcript levels (Fig 1D). Of these three clone types, DsRed 42 and DsRed 56 had high red fluorescent signal in stationary phase, but the signal from DsRed 48 was dim in comparison (S2D Fig). Because of its bright signal and confirmed mutation within the gene, we selected the DsRed 42 clone for additional experiments.
We then sought to move DsRed 42 into the low copy vector pMMB67EH, which is stably maintained during mouse infection [24,28]. Insertion of DsRed 42 into the pMMB67EH multiple cloning site (MCS) did not result in detectable fluorescence despite the upstream P tac promoter, so we also inserted P hmsT , the promoter for the diguanylate cyclase hmsT, upstream of DsRed 42 . Our Y. pseudotuberculosis strain (IP2666) naturally contains a disrupted Fur binding site in P hmsT , which should result in low, constitutive transcription of genes downstream of P hmsT [32,33]. P hmsT was sufficient to promote the accumulation of red DsRed 42 monomers in stationary phase E. coli (24h growth), in the absence of IPTG induction (Fig 1E).
After generating DsRed 42 , we also tested another TIMER variant, TIMER bac [15]. This variant had high fluorescence in Y. pseudotuberculosis, but expression from a low-copy vector (pSC101) resulted in significant growth inhibition, and overnight cultures only reached 2.5 x 10 7 CFU compared to 1 x 10 10 CFU for WT Y. pseudotuberculosis strains. For future studies, we could modify this TIMER bac construct by changing the promoter or vector, or modify the ribosomal binding site or start codon to reduce translation initiation, to generate a construct with strong fluorescence that does not inhibit growth [34][35][36]. This TIMER bac variant folds into green and red monomeric species, which provides an important internal control for levels of expression, however the presence of the green monomeric species means we would not be able to assess stress levels with green fluorescent reporters alongside TIMER bac . Additional reporters could be generated with blue fluorescent protein variants to use alongside TIMER bac [20,37], and mCherry-derived TIMER variants also exist with emission that transitions from blue to red fluorescence [38], however we chose to move forward with DsRed 42 and green fluorescent reporters for this study.

Y. pseudotuberculosis
The P hmsT ::DsRed 42 construct was then transformed into WT Y. pseudotuberculosis (Yptb) to determine if DsRed 42 signal accumulation could be detected in stationary phase Yptb. Initially, we utilized IPTG induction, and detected fluorescent signal accumulation using a plate reader alongside bacterial growth, based on absorbance. Bacterial cultures were in stationary phase between 16h (hours) and 24h growth; bacterial numbers were similar between the strains, although DsRed 42 had a slightly higher absorbance value after 24h of growth (Fig 2A). A significant increase in DsRed 42 signal was detected after 16h of growth in the presence of IPTG (Fig 2B). DsRed 42 signal continued to increase between 20h and 24h of growth in the presence of IPTG, confirming DsRed 42 signal accumulates in late stationary phase Yptb. We hoped to utilize green fluorescent reporters alongside our DsRed 42 construct, so it was important to determine if the DsRed 42 construct would produce interfering fluorescence across green wavelengths. Despite high red signal accumulation, no significant green fluorescence was detected ( Fig 2C). We then asked whether DsRed 42 signal would accumulate in the absence of IPTGinduction, since we would not be inducing with IPTG in our mouse model of deep tissue infection, and found that DsRed 42 signal could be detected after 24h growth in LB (Fig 2D). We also confirmed that cells lacked green fluorescent signal at this timepoint ( Fig 2E).
Plate reader detection represents fluorescent values averaged across all cells within a well of a 96-well plate. To determine if there was variability in DsRed 42 signal accumulation within individual bacterial cells in these samples, we used fluorescence microscopy to detect fluorescence at the single cell-level. There was a significant increase in the mean DsRed 42 signal after 16h, 20h, and 24h growth relative to non-fluorescent WT cells, but this was most apparent at 24h growth (Fig 2F). The observed variability across the bacterial population suggests that subsets of cells are truly slow-growing, DsRed 42 + , at each of these timepoints, while cells lacking DsRed 42 signal accumulation may remain more metabolically active. We also confirmed that cells lacked green fluorescent signal after 24h growth, again suggesting that GFP reporters can be used alongside TIMER 42 (Fig 2G). Collectively, these data indicate that DsRed 42 signal accumulates late in stationary phase Yptb, between 16h and 24h of culture, and that the DsRed 42 construct lacks green fluorescent signal. Many DsRed variants exist, including DsRed-Express2, and it remains unclear if DsRed 42 has advantages relative to this DsRed variant or others [39]. DsRed-Express2 is bright, photostable, and fast folding, and does retain green monomeric signal [39]; DsRed 42 is relatively dim, likely similar to DsRed, but lacks green fluorescence, which could be an advantage of DsRed 42 .

DsRed 42 is detected within the host spleen at 72h p.i.
To quantify the relative DsRed 42 signal accumulation within individual bacterial cells, we transformed DsRed 42 -expressing Yptb (P hmsT ::DsRed 42 ) with a plasmid that constitutivelyexpresses GFP [24,28]. To determine when DsRed 42 signal accumulated during bacterial growth in the spleen, we intravenously infected C57BL/6 mice with this GFP + DsRed 42 strain and harvested spleens at 48h and 72h post-inoculation (p.i.) to detect DsRed 42 signal. DsRed 42 signal was not detected in microcolonies at 48h p.i., but was detectable at 72h p.i. (Fig 3A). To determine if DsRed 42 signal accumulated in a specific spatial location within microcolonies, we compared the ratio of DsRed 42 /GFP signal intensity at the geometric centroid and periphery of individual microcolonies (Fig 3B). There was not a significant difference in the relative DsRed 42 signal at the centroid or periphery, suggesting DsRed 42 accumulation does not occur in a specific spatial location across microcolonies. We also compared the relative DsRed 42 signal at the centroid and periphery of each microcolony, and found a slight positive correlation in the two signals (Fig 3C). This shows that higher centroid signal correlates with higher periphery signal, and further suggests that on average, the DsRed 42 signal isn't specifically enriched at either the centroid or the periphery. However, DsRed 42 signal accumulation also did not appear uniform across microcolonies, which could suggest that specific subsets of cells may be transitioning to a slowed growth phenotype.

DsRed 42 signal accumulates after stationary phase markers
To determine if there are specific subsets of Yptb that exhibit DsRed 42 accumulation and slowed growth, we compared DsRed 42 signal to expression from several different fluorescent reporters. We began by testing DsRed 42 alongside two different dps reporter constructs, which should also detect slow-growing subsets of cells. dps expression occurs when bacteria transition into stationary phase, and dps gene expression has been used extensively as a marker of slowed growth [40]. However, Dps protein levels are regulated at the transcriptional, post-transcriptional, and translational levels by multiple pathways, suggesting that a dps transcriptional reporter alone may not be sufficient to identify slow growing cells, necessitating a tool like DsRed 42 [41]. We generated two dps reporters: P dps ::gfp and P dps ::gfp-ssrA, where the dps promoter is fused to stable and destabilized gfp, respectively [42,43]. This destabilized version of gfp has a half-life of approximately 37 minutes in culture, compared to approximately 147 minutes for stable gfp, indicating the destabilized GFP signal will depict recent transcriptional changes (S3 Fig). dps reporters were transformed into DsRed 42 -expressing Yptb (DsRed 42 P dps ::gfp or DsRed 42 P dps ::gfp-ssrA), individual colonies were inoculated into 96 well plates, and fluorescent signals were detected over time. Expression of the stable dps reporter was first detected after 5h growth, and remained relatively steady for about 12h, until we saw a gradual increase in signal (Fig 4A). The signal increase after 5h growth clearly occurred prior to stationary phase, based on absorbance data (Fig 4B). In contrast, the destabilized dps reporter had very little fluorescence until 20h growth, which was similar to the pattern with DsRed 42 signal accumulation in the same strain background (Fig 4A). 20h growth clearly represented stationary phase based on absorbance data (Fig 4B and 4E). DsRed 42 signal accumulation was very similar in both strain backgrounds; only DsRed 42 signal in DsRed 42 P dps ::gfp-ssrA is shown to directly compare the two signals in this strain. These data suggested that the stable https://doi.org/10.1371/journal.ppat.1009284.g004 P dps ::gfp reporter signal accumulates prior to the stationary phase transition, and that the destabilized P dps ::gfp-ssrA reporter is more accurately detecting slow growing cells.
There was a significant increase in both P dps ::gfp and DsRed 42 signal after 16h growth, a timepoint that represented stationary phase (Fig 4B), as detected by fluorescence microscopy (Fig  4C), however there was no correlation between these two signals within individual bacteria (Fig  4D), likely due to the early accumulation of stable GFP prior to the stationary phase transition. However, with the DsRed 42 P dps ::gfp-ssrA strain, significant GFP signal was not detected until 20h growth by fluorescence microscopy, while DsRed 42 was detected at both 16h and 20h growth ( Fig  4F). There was a significant correlation between P dps ::gfp-ssrA and DsRed 42 signal at 20h growth, suggesting these reporters are detecting the same subsets of cells, and are behaving similarly at this late timepoint (Fig 4G). For additional information about statistical analyses for correlations, see Materials and Methods below. Collectively, this indicates that both dps and DsRed 42 can be used to identify slow growing cells. However, dps signal detection is heavily dependent on the stability of GFP, and an early increase in gfp transcripts prior to the transition into stationary phase can lead to prolonged signal detection. DsRed 42 signal accumulation is detected well after this point, and likely means that DsRed 42 can be used to detect cells much later in stationary phase.

DsRed 42 signal accumulation correlates with hmp expression during exposure to acidified nitrite
We then sought to determine if the response to reactive nitrogen species (RNS) is sufficient to slow the growth of cells and promote DsRed 42 signal accumulation, since we observe bacterial responses to this stress during growth in the spleen [24,28]. To determine if RNS were sufficient to promote DsRed 42 signal accumulation, we first used the acidified nitrite method of generating nitric oxide (NO), where NaNO 2 is added to acidified LB (pH 5.5) [24,28,44]. The 2.5mM NaNO 2 concentration was known to inhibit growth and promote high levels of hmp expression, whereas 0.25mM represents a sub-inhibitory concentration [28]. hmp and DsRed 42 fluorescence were detected within individual cells by fluorescence microscopy. At 2h post-treatment, all concentrations of NaNO 2 promoted hmp expression, and there was significantly higher hmp expression with the highest concentration, 2.5mM (Fig 5A). DsRed 42 signal accumulation was also detected following exposure to 1mM and 2.5mM NaNO 2 , but DsRed 42 signal was highest with 1mM NaNO 2 (Fig 5B). DsRed 42 and hmp signals were correlated during 1mM NaNO 2 treatment, but 2.5mM NaNO 2 resulted in a significant negative correlation, suggesting this high concentration was impacting bacterial viability due to high levels of stress (Fig 5C). At 4h post-treatment, the highest hmp reporter expression was detected with the 1mM intermediate dose, and the hmp signal had decreased with the 2.5mM dose, again suggestive of high levels of stress (Fig 5D). DsRed 42 signal was detected with all three doses, and was highest with 2.5mM (Fig 5E). However, there was an interesting divergence in the 1mM population, where both DsRed 42 + and DsRed 42 cells were observed, suggesting that some cells may have recovered from stress and resumed growth resulting in a loss of DsRed 42 signal (Fig 5E).
We observed a significant positive correlation between DsRed 42 and hmp signals with the low 0.25mM and intermediate 1mM doses at this timepoint, but no correlation with the highest 2.5mM dose (Fig 5F). We believe this decrease in hmp signal and lack of correlation between hmp and DsRed 42 signals with the 2.5mM dose could suggest additional impacts of this stress, potentially on either transcriptional activity or transcript stability within these cells, resulting a more pronounced impact on hmp reporter signal and less impact on DsRed 42 signal, which is predominantly impacted by protein folding. Collectively, these results show that acidified nitrite is sufficient to promote DsRed 42 signal accumulation, and suggest DsRed 42 can be used to detect growth-arrested cells.

NO stress is sufficient to promote DsRed 42 signal accumulation
The results with acidified nitrite suggest that RNS are sufficient to promote DsRed 42 signal accumulation, but to truly show nitric oxide (NO) exposure is sufficient, we performed experiments with the DETA-NONOate NO donor compound. Cultures of the DsRed 42 P hmp ::gfp and DsRed 42 P hmp ::gfp-ssrA strains were exposed to 2.5mM DETA-NONOate for 4h, and fluorescence microscopy was used to detect fluorescent signals within individual bacterial cells. The 2.5mM dose was chosen for these experiments based on our previous studies, which showed this dose is sufficient to promote hmp reporter expression [24,28,45]. We also confirmed that 2.5mM DETA-NONOate significantly inhibited growth of the DsRed 42 P hmp ::gfp strain, which would allow us to determine if growth inhibition by the NO donor was sufficient to promote DsRed 42 signal accumulation (Fig 6A). Treatment with the NO donor compound resulted in high levels of hmp reporter expression (Fig 6B), and DsRed 42 signal accumulation (Fig 6C).
The P hmp ::gfp strain was used here as a control to detect the amount of red fluorescence in NOtreated bacteria that contained GFP but lacked DsRed 42 , and we confirmed that DsRed 42 signal was significantly above the baseline fluorescence of this control strain (Fig 6C). We detected significantly higher fluorescent signals with the DsRed 42 P hmp ::gfp strain, however all strains had significant increased hmp reporter expression, and both DsRed 42 strains had red fluorescence above baseline in response to NO. DsRed 42 and hmp signals were significantly correlated for both the stable reporter strain (Fig 6D) and the unstable reporter strain (Fig 6E), showing that NO is sufficient to promote DsRed 42 signal accumulation. We also performed qRT-PCR with the DsRed 42 P hmp ::gfp strain to confirm that DsRed 42 signal accumulation in response to NO was not due to an increase in DsRed 42 transcript levels, and to verify that hmsT was not increasing in transcript levels during NO exposure, since this promoter was used to drive DsRed 42 expression. DsRed 42 P hmp ::gfp cells were exposed to 2.5mM DETA-NONOate for 4h and transcripts were detected by qPCR. DsRed 42 and hmsT transcript levels were not impacted by NO, and high levels of both hmp and gfp transcripts were detected, consistent with DsRed 42 accumulation as a result of protein folding, and gfp expression as a result of transcription downstream of P hmp (Fig 6F). We also quantified dps transcript levels to determine if these increased in response to NO, and found there was no change in dps expression in response to NO (Fig 6F). These results provide additional insight into the mechanism of NO-mediated stress, and suggest the response to NO is sufficient to arrest cell division, but not promote a stationary phase transition. This is consistent with published results showing NO disrupts cell division by arresting assembly of the FtsZ cytokinetic ring [46,47].

DsRed 42 signal correlates with dps expression in vivo
We then asked whether there was a correlation between DsRed 42 signal accumulation and either dps or hmp reporter expression during growth within host tissues. C57BL/6 mice were inoculated intravenously with either the DsRed 42 P dps ::gfp or DsRed 42 P hmp ::gfp strains, and spleens were harvested at 72h p.i. to visualize reporter expression by fluorescence microscopy. At 72h p.i., some microcolonies had detectable DsRed 42  We hypothesized that larger microcolonies may run out of nutrients, which then slows growth, and results in DsRed 42 signal accumulation. We compared the areas of microcolonies that had or lacked DsRed 42 signal, and found that generally the DsRed 42 + microcolonies were indeed larger (Fig 7A). We then asked whether dps or hmp reporter signal correlated with DsRed 42 signal accumulation within these tissues. We first compared the ratio of dps and DsRed 42 signal at the periphery and centroid of microcolonies and found the signal intensity of each of these reporters looked very similar (Fig 7B). In comparison with DsRed 42 , hmp reporter signal was significantly enriched at the periphery (Fig 7B). This hmp reporter signal pattern was consistent with our previously published results [24,28,45] but suggested that NO stress was not significantly slowing bacterial growth within host tissues, based on DsRed 42 signal accumulation in some, but not all, peripheral cells. Representative images indicate there was substantial overlap between dps and DsRed 42 signals, and that hmp and DsRed 42 are detecting distinct subsets of bacteria (Fig 7C). To further analyze the relationship between these signals, we compared the DsRed 42 and GFP signals at microcolony centroids (Fig 7D) and microcolony peripheries (Fig  7E). For each individual microcolony spatial location, the relative DsRed 42 and GFP signals are plotted on the Y-and X-axis respectively, to determine if there is a linkage between the two values by linear regression (see Materials and Methods below). At both the centroid and periphery, we found a significant positive correlation between DsRed 42 and dps reporter signal, however we found a significant negative correlation between DsRed 42 and hmp reporter signal. This supports the conclusion that dps and DsRed 42 are marking the same subsets of bacteria within host tissues, but suggests that hmp and DsRed 42 are detecting distinct subsets. This suggests that at 72h p.i., there is a subset of slow-growing Yptb at the interior of microcolonies, and also suggests that NO stress may not be sufficient to cause prolonged growth arrest within host tissues. It is also possible that changes in oxygen tension levels or DsRed 42 expression levels in vivo could be impacting fluorescence accumulation, and this should be explored in future studies.

DsRed 42 signal dissipates after initial responses to NO, suggesting recovery occurs in vivo
We were surprised to find that NO stress did not correlate with DsRed 42 signal accumulation within host tissues, and hypothesized this could be due to temporal differences in when bacteria were exposed to NO and when tissues were collected. We believe that bacteria respond to NO much earlier during infection (approximately 48h p.i., [24,30]), and it was possible that bacteria were no longer actively responding to stress. We returned to the acidified nitrite method of introducing NO stress, since this likely results in an initial burst of NO, and then slower NO generation [44]. We exposed the DsRed 42 P hmp ::gfp cells to the indicated concentrations of NaNO 2 in LB pH5.5, and measured absorbance (A 600nm ) and reporter signal by fluorescence, in a plate reader over time. Consistent with the predicted initial burst of NO release, we initially observed growth arrest after treatment followed by slowed growth (Fig 8A). There was a significant increase in hmp reporter signal after 1h treatment that continued to increase over the course of 8h (Fig 8B). However, DsRed 42 signal accumulation was consistent with an initial arrest of the cell population followed by recovery and resumed growth, based on significantly higher DsRed 42 signal accumulation with 1mM treatment relative to 0.25mM treatment at 2h and 4h post-treatment, and a return to baseline levels 6h after the initial exposure to acidified nitrite (Fig 8C). Consistent with this DsRed 42 result, growth also appeared to resume at these timepoints in cultures, based on absorbance (Fig 8A).
NO exposure has recently been shown to arrest bacterial cell division through disruption of cytokinesis machinery, and so it was expected that DsRed 42 signal would accumulate in growth-arrested, NO-exposed, bacteria [47]. The lack of correlation between the two signals during growth in tissue was unexpected, and suggested that NO stress may have occurred much earlier during infection than the 72h timepoint. To test this hypothesis, we generated constructs that contained combinations of stable and unstable fluorescent protein genes with expression driven by the P hmp promoter. P hmp ::mCherry::gfp was a control construct generated with stable mCherry and stable GFP proteins, and we confirmed these signals overlapped during growth in tissue (Fig 8D). The P hmp ::mCherry-ssrA::gfp construct contained destabilized mCherry [30] and stable GFP, to allow us to detect bacteria that had responded to NO as some point during infection (GFP + ) compared to bacteria that were actively responding to NO when the tissue was harvested and fixed (mCherry + ). Using this reporter, we found very few bacterial cells were actively responding to NO when tissue was harvested (Fig 8D), suggesting that other cells may have responded to NO, recovered, and resumed growth, explaining why they no longer exhibited DsRed 42 signal accumulation. Instead, DsRed 42 signal accumulation is clearly marking an interior subset of bacteria, that are likely experiencing nutrient limitation at this late timepoint post-inoculation.

Discussion
In this study, we sought to modify TIMER to detect slow-growing bacteria within host tissues, with a goal of generating a tool to enable future studies of slow-growing bacterial populations. The DsRed 42 variant has been tested in E. coli and Y. pseudotuberculosis; the signal accumulates late in stationary phase in both of these quickly-replicating Gram-negative organisms. DsRed 42 signal was detected within host tissues, and is also visible within cultured individual  42 and WT strains were grown in LB pH5.5 in 96 well plates at 37˚C with shaking, with the indicated concentrations of NaNO 2 , for the indicated timepoints. A) Optical density of the P hmp ::gfp DsRed 42 strain was measured by absorbance (A 600nm ) over time (hours, hrs), in a plate reader. B) hmp reporter expression (GFP signal) was detected over time by plate reader. Fold change is relative to WT non-fluorescent samples grown under the same conditions. C) DsRed 42 signal was detected over time by plate reader. Fold change is relative to WT non-fluorescent samples grown under the same conditions. D) C57BL/6 mice were infected intravenously with strains containing either the P hmp ::mCherry::gfp or P hmp ::mCherry-ssrA::gfp reporter constructs, and spleens were harvested for fluorescence microscopy at 72h p.i. Representative images are shown. Nine biological replicates are shown for plate reader data with mean and standard deviation. Statistics: A)-C): Two-way ANOVA with Tukey's multiple comparison test; ���� p < .0001, ��� p < .001, �� p < .01, � p < .05, n.s.: not-significant. Statistics shown represent: A) and C) pair-wise comparisons between 0.25mM and 1mM treatment conditions; B) comparison between 1mM treatment and untreated cells.
https://doi.org/10.1371/journal.ppat.1009284.g008 cells. However, the fold change quantification we used here indicate that the signal is sometimes dim, so non-fluorescent cells need to be included as a control for all assays to define DsRed 42 + cells. DsRed 42 could potentially be used directly in other Gram-negative species, or used as a template for random mutagenesis to select an optimized DsRed 42 variant in another system. Additional rounds of random mutagenesis and sorting could also be used to select for a brighter DsRed 42 variant, as done with other DsRed variants [39,48,49], and again, this may be needed to apply DsRed 42 to other bacterial systems. We believe the signal generated with DsRed 42 is sufficient to isolate expressing cells from host tissues, and plan to analyze these cells in future experiments to better understand the pathways promoting slowed bacterial growth. Additional recent applications of fluorescence dilution approaches could also be applied in our model to isolate stressed, slow-growing subsets of bacteria [50][51][52]. Fluorescence signal accumulation requires folding prior to cell division events, and so additional variants that fold more quickly could lead to increased signal accumulation. We have shown here that DsRed 42 signal accumulation in response to NO is likely due to protein folding during an arrest in cell division, and not changes in transcript level, based on similar DsRed 42 transcript levels after NO donor addition (Fig 6F). However, it does remain somewhat unclear why the T22T synonymous mutation (ACC➔ACA) is sufficient to promote DsRed 42 signal accumulation in stationary phase cells. We initially thought this change may result in use of a more favorable threonine codon (ACA), but the original threonine codon (ACC) is used more frequently across the Y. pseudotuberculosis genome [53]. Instead, we believe the ACA codon may increase the stability of DsRed 42 mRNA, leading to increased levels of translation of the DsRed 42 variant relative to TIMER I161T T197A , based on the predicted influence of an ACC➔ ACA change in E. coli [31]. These results in E. coli suggest that protein levels correlate better with mRNA concentrations than genomic codon-usage frequency, and that specific codons (including ACA) at the 5' end of mRNA molecules could have a significant impact on mRNA stability and subsequent protein levels. Based on the predictions that codons ending in 'A' are generally more favorable and lead to increased protein expression, we postulate that this likely underlies the increased signal accumulation with DsRed 42 .
Recently, we have shown that NO-exposed Hmp + cells preferentially survive doxycycline treatment in our mouse model of infection, which suggested the Hmp + cells may represent a slow growing subpopulation of bacteria [30]. Our in vitro results were consistent with this, where exposure to NO was sufficient for DsRed 42 signal accumulation. Our results are also consistent with a recently published study which showed that NO causes a collapse of proton motive force (PMF) and collapse of the divisome complex, thus resulting in cell division arrest [47]. However, within host tissues, bacterial cells with DsRed 42 signal accumulation were largely Hmp -, and instead expressed the stationary phase marker, Dps. We believe this surprising result may be a reflection of the transient bacterial exposure to NO during infection. Consistent with this idea, we found that only a few bacterial cells express the destabilized hmp reporter at 72h p.i. (Fig 8D). The half-life of this destabilized mCherry protein is approximately 35 minutes in culture [30], and assuming the turnover rate is similar within host tissues, this marks a very recent exposure to NO. This mosaic of expression suggests that while many cumulative cells were exposed to NO and responded with hmp expression, this process is transient and actively occurring in only a few cells at a time.
Following transient exposure to NO, bacteria need to repair NO-mediated damage to survive and resume replication [46,54]. We also know repair of Fe-S clusters promotes Y. pseudotuberculosis survival after NO exposure, and likely promotes continued bacterial growth after stress [28]. What remains unclear is why Hmp + cells have decreased antibiotic susceptibility, since this does not appear to be linked to slowed cell division rates. Recently, it has been shown that slowed metabolic activity is a better correlate for decreased antibiotic susceptibility [1], which could mean that Hmp + cells have decreased metabolic activity even after cell division has resumed. These results could also suggest that Hmp + cells may change in another way that promotes a tolerance phenotype, possibly by upregulation or downregulation of specific genes or pathways, or utilization of less energetically favorable metabolic pathways, that again lead to decreased metabolic rates. It would also be interesting to visualize the impact of antibiotic treatment on DsRed 42 -expressing populations, to see if DsRed 42 + Yersinia also have decreased antibiotic susceptibility, as you would predict based on results with Salmonella [15]. Consistent with this idea, doxycycline treatment at 72h p.i. with Y. pseudotuberculosis did not significantly prolong mouse survival, suggesting the DsRed 42 + cells may be very tolerant to antibiotics [30].
The in vitro studies we performed here were used to better characterize the conditions that lead to DsRed 42 signal accumulation, and compare DsRed 42 signal to other stress reporters. The results we obtained with NO were expected, based on the role of NO in causing cell division arrest. However, the results comparing DsRed 42 with the dps reporter were less clear, especially as we compared DsRed 42 and dps reporters in culture to the results we observed within host tissues. Based on the results in vitro, we believe that dps transcript levels begin to accumulate early in the transition into post-exponential and stationary phase. This early increase in transcripts leads to an early accumulation in the stable dps reporter signal, prior to the point where growth actually slows. In vitro results with the unstable dps reporter suggest there is a second wave of dps transcription late in stationary phase that correlates with slowed growth and DsRed 42 signal accumulation. Our in vivo results instead show overlap in the stable dps reporter signal and DsRed 42 signal; detection of both signals together lends confidence to the idea that these cells represent a slow-growing subset. However, there are many different conditions that could arise in vivo that may promote differential DsRed 42 signal accumulation, such as low oxygenation or in vivo differences in expression levels. The unstable dps reporter unfortunately could not be detected within host tissues, likely because the signal was so dim with this reporter construct in general (Fig 4F), and because the spleen has relatively high background green fluorescence. Dps protein was also enriched in TIMER bac+ Salmonella populations [15], again supporting the idea that the Dps + DsRed 42 + cells we detected within host tissues are truly slow-growing. Dps regulation is complex, and expression can be triggered by exposure to ROS and iron availability, in addition to RpoS-dependent stationary phase transitions and general nutrient deprivation [55][56][57]. DsRed 42 , in contrast, should indicate slowed growth and could be used to isolate bacterial subsets to confirm this slowed growth phenotype, and to better understand what stressors are promoting slowed growth. The combination of these two reporters together could be particularly informative in future experiments to determine the mechanisms underlying the formation of slow-growing subsets of bacteria within host tissues.

Ethics statement
All animal studies were approved by the Institutional Animal Care and Use Committee of Johns Hopkins University. This study does not include human participants or donors. Bacterial strains & growth conditions. The WT Y. pseudotuberculosis strain, IP2666, was used throughout [24,27]. For all mouse infection experiments, bacteria were grown overnight (16h) to post-exponential phase in 2xYT broth (LB, with 2x yeast extract and tryptone) at 26˚C with rotation. For in vitro experiments with hmp and DsRed 42 reporters, or dps and DsRed 42 reporters, bacteria were grown overnight (16h) to post-exponential phase in LB prior to experimental set-up. Overnight cultures of E. coli were grown in LB with 1mM IPTG at 37˚C with rotation for 24h prior to cell sorting. 1mM IPTG was used for induction throughout these experiments.
Site-directed and random mutagenesis. Site-directed mutagenesis was used to generate the TIMER I161T T197A variant from WT TIMER. Mutations were made sequentially by amplifying across the gene with primers containing the desired mutations. Briefly, TIMER was amplified in two pieces, with the mutation incorporated at the 3' end of the upstream piece, and 5' end of the downstream piece; overlap extension was used to join these homologous fragments [24]. We generated TIMER I161T first, then used this as a template to generate TIMER I161T T197A . For random mutagenesis, the TIMER I161T T197A gene was amplified by PCR to use as a template for random mutagenesis PCR reactions. TIMER I161T T197A was then amplified using the Agilent Gene Morph II kit for a single round of random mutagenesis and PCR products were purified (QIAGEN). These products were used as template DNA in an additional PCR reaction, where products were amplified to add restriction enzyme sites onto the 5' and 3' ends (10 cycles). PCR products were purified again before restriction digestion (EcoRI, BamHI), ligation into the lacZ gene of pUC18, and electroporation into XL1-BLUE E. coli. Transformants were plated onto IPTG/X-gal LB plates to ensure the majority of colonies contained an insert, based on white colony color. Transformants were then grown in culture +IPTG (1mM) for 24h, bacteria were pelleted, resuspended in PBS, and placed on ice during fluorescenceactivated cell sorting (MoFlo XDP cell sorter, Beckman Coulter). Bacteria with red fluorescence (approximately 200 cells) were collected and plated onto LB/Amp for single colonies. 192 colonies (two 96 well plates) were screened for red fluorescence after 24h growth in 200μl LB +IPTG at 37˚C with shaking. 26 colonies had clear red fluorescent signal, and pUC18 plasmids were isolated by miniprep (QIAGEN) from these colonies for sequencing.
Generation of reporter strains. Two of the fluorescent constructs in this study were previously described: the constitutive GFP and P hmp ::gfp (GFPmut3) [24,26]. The final DsRed 42 plasmid was constructed by amplifying the hmsT promoter, fusing this immediately upstream of DsRed 42 , and inserting this into pMMB67EH. The destabilized gfp was constructed by fusing an 11 amino acid ssrA tag, with AAV terminal amino acids, between the last coding amino acid and the stop codon [42,43]. This was inserted downstream of the P tac IPTG inducible promoter in pMMB207Δ267 [58] to quantify GFP half-life following IPTG induction. A stable gfp construct was generated in parallel as a control (S3 Fig). The P dps ::gfp and P dps ::gfp-ssrA reporters were generated by amplifying the dps promoter, fusing this to either stable or destabilized gfp by overlap extension PCR, and inserting this into pACYC184 [24]. The P hmp ::gfp-ssrA reporter was generated by fusing the hmp promoter to destabilized gfp, and was inserted into pACYC184 [24]. The tandem hmp reporter constructs were generated by first fusing either stable or destabilized mCherry to stable gfp; these tandems were inserted into pMMB207Δ267 first to quantify mCherry half-life [30]. The hmp promoter was then fused to each tandem, and these fragments were inserted into pACYC184. All plasmid transformations into Y. pseudotuberculosis were performed using a previously published protocol [59].  42 fluorescence was detected with 560nm excitation/610nm emission, and GFP fluorescence was detected with 480nm excitation/520nm emission, using a Synergy H1 microplate reader (Biotek). Fold change represents fluorescence in reporter strains relative to the average value for non-fluorescent WT Yptb. For E. coli experiments, TIMER signal was also detected with a spectral curve, with excitation at 480nm, and emission detected between 500nm-600nm.
Fluorescence microscopy: Bacteria. To visualize individual bacterial cells, 100μl of each sample was taken from 96 well plates and pelleted, resuspended in 100μl 4% PFA, and incubated at 4˚C for 30 minutes for fixation. PFA was removed and bacteria were resuspended in 25μl PBS prior to imaging. Agarose pads were prepared to immobilize bacteria for imaging, by solidifying a thin layer of 25μl 1% agarose in PBS between a microscope slide and coverslip. Once solidified, coverslips were removed, bacteria (8μl) were added, coverslips were replaced, and bacteria were imaged with the 63x oil immersion objective, using a Zeiss Axio Observer. Z1 (Zeiss) fluorescent microscope with Colibri.2 LED light source and an ORCA-R 2 digital CCD camera (Hamamatsu). Volocity image analysis software was used to specifically select individual bacterial cells and quantify the fluorescent signal associated with each cell.
Murine model of systemic infection. Six to 8-week old female C57BL/6 mice were obtained from Jackson Laboratories (Bar Harbor, ME). All animal studies were approved by the Institutional Animal Care and Use Committee of Johns Hopkins University. Mice were injected intravenously into the tail vein with 10 3 bacteria for all experiments [24,28,30]. At the indicated timepoints post-inoculation (p.i.), spleens were removed and processed. Intact tissue was fixed for fluorescence microscopy. Tissue was homogenized to quantify CFUs.
Fluorescence microscopy: Host tissues. Spleens were harvested and immediately fixed in 4% PFA in PBS for either 3 hours at room temperature or 24 hours at 4˚C. Tissues were frozen-embedded in Sub Xero freezing media (Mercedes Medical) and cut by cryostat microtome (Microm HM 505e) into 10μm sections. To visualize reporters, sections were thawed in PBS, stained with Hoechst at a 1:10,000 dilution, washed in PBS, and coverslips were mounted using ProLong Gold (Life Technologies). In Fig 3, neutrophils were also detected by immunofluorescence microscopy; sections were thawed, permeabilized with ice-cold 100% methanol and blocked in 2% BSA in PBS. Neutrophils were detected using a rat anti-mouse Ly6G unconjugated antibody (BD) and goat anti-rat Alexa 350-conjugated secondary antibody (Invitrogen). Sections were incubated with primary antibodies, washed, incubated with secondary antibodies, washed, and coverslips were mounted with ProLong Gold. Tissue was imaged as described above, with an Apotome.2 (Zeiss) for optical sectioning.
qRT-PCR to detect bacterial transcripts in broth-grown cultures. Bacterial cells were grown in the presence or absence of DETA-NONOate for 4 hours, pelleted, resuspended in Buffer RLT (QIAGEN) + ß-mercaptoethanol, and RNA was isolated using the RNeasy kit (QIAGEN) according to manufacturer's protocol. DNA contamination was eliminated using the DNA-free kit (Ambion). RNA was reverse transcribed using M-MLV reverse transcriptase (Invitrogen), in the presence of the RNase inhibitor, RnaseOUT (Invitrogen). Approximately 30 ng cDNA was used as a template in reactions with 0.5 μM of forward and reverse primers [24] and SYBR Green (Applied Biosystems). Control samples were prepared that lacked M-MLV, to confirm DNA was eliminated from samples. Reactions were carried out using the StepOnePlus Real-Time PCR system, and relative comparisons were obtained using the ΔΔC T or 2 -ΔCt method (Applied Biosystems). Kits were used according to manufacturers' protocols.
Image analysis. Volocity image analysis software was used to quantify microcolony areas and fluorescence as previously described [24]. Image J was used to quantify the signal intensity of each channel at the centroid and periphery of each microcolony, to generate relative signal intensities of fluorescent reporters, as previously described [24,30]. Thresholding was used to define the total area and centroid of each microcolony, and 0.01 pixel 2 squares were selected to calculate values. Peripheral measurements depict bacteria in contact with host cells [24,30].
Statistical analyses for correlations. To determine if fluorescent signals were correlated, we performed linear regression analysis and are depicting best-fit lines. R 2 values are shown to depict whether there is a linear correlation between the two fluorescent signals, however strength of correlation was assessed based on whether the slope of the best fit line was significantly non-zero. Low R 2 values for many of the correlations are due to significant differences in the magnitude of fluorescent signals; DsRed 42 signal is low relative to transcriptional fluorescent reporters. Additional statistical analyses were performed as described in figure legends. contain the indicated DsRed variants, expressed from the high copy pUC18 plasmid. Strains were grown for the indicated times at 37˚C in 96 well plates with shaking. Fluorescence was detected with a spectral curve with excitation at 480nm and emission detected between 500nm-600nm. Relative fluorescence units (RFUs) represent raw values with media only background subtracted, data represents mean of three biological replicates. A) Optical density of the indicated strains was measured by absorbance (A 600nm ) over time (hours, h), during incubation at 37˚C in a plate reader. B) The indicated strains were grown as described above, and optical density was measured by absorbance (A 600nm ) at the indicated timepoints (hours, h) C) DsRed variants were grown alongside the non-fluorescent WT E. coli strain in the presence of IPTG for the indicated times. E. coli +IPTG: 24h growth. D) DsRed variants grown alongside the non-fluorescent WT E. coli strain, in the presence or absence of IPTG for 24h. Vertical dotted lines: 520nm (green fluorescence) and 580nm (red fluorescence). Statistics: A)-B) Twoway ANOVA with Tukey's multiple comparison test; n.s.: not-significant.