Mycobacterium smegmatis mc²155 (ATCC 700084) was plated on 7H10 agar supplemented with 10% Middlebrook Oleic Albumin Dextrose Catalase (OADC) growth supplement and 0.2% glycerol. After picking three separate colonies, the mycobacterial strain was grown in 10 mL of 7H9 media supplemented with 10% albumin D-glucose-salt supplement (ADS), 0.2% glycerol, 0.05% Tween 80 at 37°C in a shaker. M. smegmatis was passaged twice before treatment of mid-log growing cells. M. bovis bacille Calmette Guérin (BCG) Pasteur strain 1172P2 was cultured from isogenic freezer stocks in 50 mL conical tubes then passaged into roller bottles with supplemented 7H9 media (Chionh et al., 2016). For nutrient-deprived (i.e., starved) cultures, M. smegmatis and BCG were grown to OD₆₀₀ 0.8–1.0, washed twice in PBS (137 mM NaCl, 2.7 mM KCl, 10 mM NaH₂PO₄, 1.8 mM KH₂PO₄, pH 7.4), and re-suspended to an OD₆₀₀ ∼1.0 (BCG) or 0.1 (M. smegmatis) in PBS plus 0.05% Tyloxapol (non-metabolizable detergent to prevent clumping) in roller bottles. CFUs for BCG were determined from 10-fold serial dilution plating on 7H10 at 37°C for 14–21 days. Single colonies of E. coli DH5α or K-12 BW25113 were grown overnight in LB broth, diluted 1:100 in fresh LB as a subculture and grown to OD₆₀₀ 0.4–0.6 to use in studies. CFUs for E. coli and M. smegmatis was determined from 10-fold serial dilution plating on LB incubated overnight (E. coli) or 2 days (M. smegmatis) at 37°C. Knock-out mutants for sodA and sodB in the E. coli K-12 BW25113 background were obtained from the E. coli Keio Knockout Collection (GE Dharmacon).

Antibiotics were subjected to twofold serial dilutions in 7H9 media or PBS in 96-well plates containing 100 μL per well. For H₂O₂ and menadione treatments, reagents were added to each well of a 96-well plate at concentrations twofold higher than the final exposure condition. Bacterial cultures were diluted to an OD 0.1 (M. smegmatis and E. coli) or 0.2 (BCG) and 100 μL was added to each well containing antibiotic, H₂O₂, menadione, media or PBS, for a total treatment volume of 200 μL. The plates were incubated for indicated times at 37°C. For menadione and H₂O₂ treatments, cultures were washed once and resuspended in PBS, while antibiotic-treated cells were diluted in fresh media without washing. For the PBS versus 10% 7H9/LB, 20 uL of culture was transferred directly (no wash) into either 180 uL of 100% PBS or PBS + 10% LB/7H9 in the presence of 0.5 uM CellROX (a 10x dilution). After 30 min cells were fixed by the addition of paraformaldehyde (PFA; 4%) and analyzed by flow cytometry. Antibiotic-tolerant M. smegmatis were generated by diluting mid-log growing cells to OD 0.1 and aliquoting 100 μL to individual wells containing an antibiotic at a range of concentrations in a 96-well plate. Cultures were covered and incubated at 37°C with humidity for 1 or 5 days prior to analysis.

For flow cytometry staining, 20 μL volumes of treated cells were added to 180 μL of 0.5 μM CellROX Green (ThermoFisher), 5 mM CTC (ThermoFisher), or 10 ng/mL CFDA (ThermoFisher) in PBS in 96-well V-bottom plates and incubated for 30 min (1 h for CTC) at 37°C. For menadione and H₂O₂ treatments, CellROX Green was included in the PBS during treatment (60 min). After incubation, 50 μL of DAPI (5 μg/mL) in 4% PFA-PBS was added to the cells for >10 min at 25°C prior to analysis. Samples were analyzed within 2 h on a custom LSR II flow cytometer (BD Biosciences) with CellROX Green, CFDA, and CTC excited by an argon laser and detected with a 530/30 nm band-pass emission filter or 670 LP emission filter for CTC. DAPI was excited with a UV laser and detected with a 450/50 nm band-pass emission filter. Samples were analyzed using a HTS fluidics system directly from the 96-well V-bottom staining plate. The flow rate was set to 3.0 μL/s with a 150 μL injection volume, 100 μL mixing volume, 250 μL/s mixing speed, five mixes, and a wash volume of 800 μL. For each sample 50,000–200,000 events were collected when possible. For antibiotic treatments, only samples with >10,000 DAPI⁺ cells were analyzed. Data was collected by Diva (BD Biosciences) and analyzed with FlowJo v10.0.6 (Tree Star, Inc.). Gating was set using unstained samples for the bacterial population by forward-scatter (FSC; correlates with cell size) and side-scatter (SSC; correlates with cell internal granularity) of light and to determine background fluorescence. Single DAPI-stained cells were used to gate the DNA-positive population, while single CellROX Green-, CFDA-, or CTC-stained cells were used to determine the dye-positive population. The single-color stains were also used for compensation, as needed, between the dye signals. A Biotek Synergy 4 microplate reader was used for fluorescence measurements with excitation of CellROX Green using a 485/20 nm filter and emission captured at 528/20 nm. Statistical analyses were performed with Prism GraphPad Software 6.

The availability of fluorescent dyes for the detection and quantification of reactive chemical species in cells has fostered the development of single-cell, multi-parameter, high-throughput analyses in microbiology and antibiotic pharmacology, including both multi-well plate-based population assays and flow cytometry assays. However, factors such as dye uptake efficiency, dye efflux, changes in cell morphology, and background fluorescence complicate the application of these fluorescent dye-based assays to bacteria and confound the interpretation of observed changes in fluorescence. To achieve assay precision and accuracy and ensure credible and biologically meaningful results, we developed a four-step gating scheme (Figures 1A,B) to select for: (1) particle size based on FSC and SSC as light passes through the cells, (2) DNA content based on DAPI staining (or other DNA-binding dyes), (3) the presence of dye based on fluorescence above unstained and DAPI-stained particles, and (4) dye reactivity above baseline or basal fluorescence. Application of this two-dye, multi-gating flow cytometry method distinguishes bacteria from debris by selecting for size and DNA content, which also distinguishes single cells from dividing cells and individual bacterial species in mixed cultures. Additionally, greater sensitivity and specificity for ROS and metabolic activity is achieved by removing contributions from autofluorescence and by selecting only cells that have taken up the dye, with sub-gating of those cells for highly fluorescent activated dye. Here, we illustrate this scheme with CellROX^hi gating for endogenous ROS quantification (Figure 1B) and then demonstrate the versatility of the assay with examples using CTC and CFDA dyes for cellular metabolism (Figures 1C,D).

Step 1 in the analysis scheme (Figures 1A,B) defines the bacterial population of interest based on FSC and SSC properties. In addition, Step 1 gating can be adjusted for populations in which a significant portion exhibits changes in cell morphology. By considering increases in fluorescence observed in cells with elevated SSC or FSC post-treatment, which can be performed by two-dimensional gating such as SSC versus fluorescence, these cells can be selected for analysis if they pass the subsequent gates for non-specific background fluorescence (to be considered Dye⁺) and then for desired activation of the fluorescent dye. Each bacterial cell has morphological and biological properties that may be uniquely altered in response to stress or as part of its life cycle. Gating (i.e., restricting) SSC and FSC in Step 1 allows consistent selection for cell size, shape, complexity, and composition. For example, during logarithmic growth, SSC and FSC profiles of rod-shaped filamentous E. coli differ from those of M. smegmatis bacilli (Figure 1E). Overall, both bacilli and filamentous bacteria have higher FSC signals during cell division. At the same time, population heterogeneity in cell size or length and cellular macromolecular composition can be detected by FSC and SSC, respectively. For M. smegmatis, heterogeneity is likely due to bacterial replication on the basis of time rather than cell size (Aldridge et al., 2012). Depending upon cytometer settings, non-coccoid bacteria passing through the light source at different angles will be detected as lower or higher FSC/SSC signals than those observed for uniform spheres. Additionally, stresses such as antibiotic exposures can alter cell light scatter by inhibiting replication or altering cell morphology, such as by causing filamentation or by inhibiting septation (Figure 1E) (Renggli et al., 2013; Paulander et al., 2014; Saint-Ruf et al., 2016). These differences in morphology and composition detected by FSC and SSC should be considered when interpreting bacterial populations during an analysis since these changes can significantly alter the observed fluorescence.

Step 2 ensures that only cells with intact DNA are included in the analysis. Small non-viable particles such as cell debris and dust particles also generate FSC and SSC signals. Therefore, DNA content staining is used to differentiate bacteria from particles. This staining step further enables the assessment of several biologically relevant parameters including detection of doubling cells, cell aggregates, and dead cells with fragmented DNA (Nebe-von-Caron et al., 2000). The gating also excludes artifacts from the analysis, which may arise from an abundance of dead cell debris in studies of bacterial persistence and dormancy, as pursued here with M. smegmatis. For example, subsequent experiments (illustrated in Figures 4, 5) show that >90% of the mycobacteria were killed by antibiotics before a drug-resistant population was established, with the debris potentially confounding flow cytometry analysis.

Step 3 accounts for dye-based background fluorescence and helps to distinguish cells that take up dye molecules from those that do not, thus increasing the accuracy and rigor of the ROS analysis. Here, the relatively weak but detectable fluorescence of unactivated dye is exploited to select all presumptive dye-containing cells by virtue of their fluorescence above that of unstained cells. This fluorescence of unactivated CellROX Green is apparent as a 10-fold increase in fluorescence in PBS solutions above the PBS background (28 versus 254 f.u. under conditions used in the flow studies) and by a >1.5-fold increase in fluorescence when CellROX Green is added to heat-killed or PFA-fixed cells (depending upon fluorimeter settings and cell concentration). It is important to note that flow cytometry cannot be used to accurately attribute the source of increased fluorescence (above unstained controls) to specific changes in cellular morphology, composition and metabolism, steady-state ROS generation, or increased dye uptake. However, using this gate for dye uptake increases the accuracy for ROS-specific events in the assay. In the absence of this step, all cells from Steps 1 and 2 will be analyzed regardless of dye uptake status. Only after gating for size, DNA, and dye presence is it possible to accurately identify cells in which ROS has potentially activated the dye molecules to fluoresce above background levels and to quantify the fluorescence intensities of these cells.

It is important to point out that Step 3 also accounts for changes in background fluorescence from a variety of sources. For example, in our hands, the exponential growth in the presence of LB or 7H9 cell culture medium during staining of E. coli and M. smegmatis, respectively, increased background signals by up to 20-fold. Increases in fluorescence were also observed in PFA-fixed and heat-killed bacteria (Figure 1F). As dead cells do not have ROS-generating metabolic processes, such fluorescence increases indicate that ungated fluorescent signals cannot be entirely attributed to the presence of ROS. In these cases, variation in the background fluorescence makes it difficult to determine whether the summed fluorescence increment, above that observed in unstained cells, is due to activated dye that reacted with ROS produced in metabolically active cells, by unreacted dye taken up by cells, or by fluorescent intracellular compounds, such as flavins (Benson et al., 1979). The Dye^+/- gating in Step 3 minimizes this issue by pre-selecting cells that possess fluorescence greater than unstained cells, which includes cells that have internalized dye molecules regardless of their activation state.

In Step 4, the degree of dye activation, presumably by ROS, is quantified, here as “CellROX^hi.” Step 3 selects for cells with fluorescence above untreated or control levels, with those with higher levels of fluorescence (CellROX^hi) are deemed to contain elevated levels of the dye-reactive chemicals, presumably ROS; the identity of the reactive chemicals as specific forms of ROS must be defined by appropriate control experiments. Heat-killed or fixed cells that take up dye, but do not produce ROS or other reactive chemicals, are used as gating controls to set the threshold for CellROX^hi. Alternatively, ROS production above basal levels or production observed in a control population (e.g., log growing, wild-type, or untreated cells) may also be used to set the CellROX^hi threshold. With the prior gating steps, the normalizing denominator for the CellROX^hi population is the sum total of cells with detectable fluorescence rather than all bacteria analyzed in the sample.

The general applicability and utility of the gating scheme is further illustrated in BCG for reductive respiratory activities using CTC and for cellular esterase activity using CFDA-AM (Figures 1B,C). Similar considerations were made for FSC and SSC (Step 1), DNA counterstaining to differentiate cells from debris (Step 2), dye update and retention (Step 3) and dye activation above background fluorescence (Step 4). It is important to note here that CTC becomes insoluble upon reaction, which results in strong fluorescence from activated dye molecules retained within the cells, while the fluorescent carboxyfluoroscein product of CFDA-AM is negatively charged and thus retained within cells longer than fluorescein. This positive attribute is similar to the DNA-binding activity of ROS-activated CellRox Green.

In addition to quantifying ROS, identifying the ROS species generated during stress is a crucial step in establishing the precise molecular mechanisms underlying the stress response. The challenge here is that the reactivity of individual fluorescent probes for specific ROS has not been clearly defined under biological conditions and many probes react with a variety of oxidants and free radical species. We chose to use CellROX Green as an ROS probe due to its unique feature of having high fluorescence quantum yields only upon ROS-activation and binding to dsDNA (Invitrogen, 2012). This feature obviates artifacts due to altered membrane integrity resulting from stress or cell state, and enhances accumulation of ROS-activated dye to maximize fluorescence signal intensity above background and autofluorescence in cells – critical to setting the Dye⁺ gate in Step 3. However, for this probe especially, there is a lack of clarity in the scientific community about its intracellular ROS specificity (ResearchGate, 2013; Choi et al., 2015), and, to our knowledge, there is no consensus on its reactivity in bacterial cells. CellROX Green is a propriety photostable dye, which, according to the manufacturer, exhibits ROS reactivity that complements HPF. Unlike HPF, CellROX Green reacts with hydroxyl radicals, O₂^•– and t-butylhydroperoxide in vitro, but not with H₂O₂, NO, ONOO^-, or HOCl (Invitrogen, 2010). However, CellROX Green has recently been claimed to react almost exclusively with O₂^•– in vivo (Choi et al., 2015). Hence, we sought to confirm the identity of the intracellular ROS detected by CellROX Green in bacteria using two well-studied ROS generators: (1) menadione, a quinone that undergoes catalytic one-electron redox cycling to reduce molecular oxygen (O₂) to O₂^•– (t_1/2 ∼ 1–15 min) (Hassan and Fridovich, 1979; Iyanagi, 1990), and (2) H₂O₂ that reacts with redox-active metals to produce extremely short-lived ferryl or hydroxyl radicals (t_1/2 ∼ 10^-9 s) under specific conditions (Imlay et al., 1988; Imlay, 2013, 2015; Bataineh et al., 2016).

CellROX Green reproducibly detected of intracellular ROS when incubated for 1 h with live bacteria in the presence of menadione in PBS (Figure 2). While menadione can induce CellROX Green-reactive species by indirect mechanisms, such as alterations of cell metabolism that lead to ROS production (Marinho et al., 2014), we observed no shift in FSC, SSC, or fluorescence signals in either the DAPI or CellROX Green channels (Supplementary Figures S1A,B), which ruled out menadione-induced changes in cell biochemistry or morphology that could cause increased non-specific fluorescence or light scattering. Using the four-step gating protocol to minimize signals due to menadione-induced cell permeability changes, non-cytotoxic menadione exposures (Supplementary Figure S2A) generated CellROX Green-detectable ROS in both E. coli and M. smegmatis in a dose-dependent manner (Figures 2A–E). Although ROS levels were not elevated in all menadione-exposed cells in an isogenic population when compared with untreated cells, both the cellular fluorescence intensity and number of cells with detectable ROS increased in a dose-dependent manner (Figures 2A,C,E).

Unlike menadione treatment, no increases in the proportion of Dye⁺ or CellROX^hi cells were detected with H₂O₂ treatment in PBS, even at bactericidal concentrations (Figure 2 and Supplementary Figure S2B). H₂O₂ exposures did not change SSC, FSC, or fluorescent profiles of either bacterial species, which suggests minimal contributions by non-specific fluorescence or light scattering (Supplementary Figures S1C,D). CellROX Green is highly sensitive to hydroxyl radicals arising from Fe²⁺/H₂O₂-mediated Fenton chemistry, as demonstrated with hydroxyl radicals generated in reactions of ferrous perchlorate (II) with H₂O₂ (Invitrogen, 2010, 2012, 2016). So our results suggest that H₂O₂-derived hydroxyl radicals and other H₂O₂-derived ROS do not reach levels detectable by CellROX Green in H₂O₂-treated E. coli or M. smegmatis. This observation stands in contrast to previous observations in human cells treated with H₂O₂ (Segard et al., 2013). However, the fluorescence attributed to CellROX dye in the studies of Segard et al. (2013) is possibly due to autofluorescence or morphologic changes in FSC/SSC caused by extreme protein aggregation in these studies. Changes in autofluorescence and morphology, and not dye activation, similarly accounted for increased fluorescence observed in antibiotic-treated E. coli (Renggli et al., 2013; Paulander et al., 2014). Our findings with superoxide-generating menadione and hydroxyl radical-producing H₂O₂ support the conclusion that bacterial exposure to even high levels of H₂O₂ does not produce hydroxyl radicals at a level detectable by activation of CellROX Green and the conclusion by Choi et al. (2015) that CellROX Green is a sensitive detector of O₂^•– in bacteria.

Here, we tested the ability of CellROX Green to detect changes in O₂^•– levels in cells in different states of respiration and growth. A major source of O₂^•– in respiring cells involves reduction of O₂ by electrons leaking from the bacterial respiratory chain from NADH dehydrogenase and cytochrome b and at the ubiquinone sites. In E. coli, this respiration-dependent O₂^•– production varies according to nutrient levels, at least at early times following changes in nutrient sources and levels (Gonzalez-Flecha and Demple, 1995; Turrens, 2003; Korshunov and Imlay, 2006; Brynildsen et al., 2013; Dwyer et al., 2014; Adolfsen and Brynildsen, 2015; Lobritz et al., 2015). To assess the effects of active respiration on basal O₂^•– levels, we performed CellROX Green staining after transferring actively dividing E. coli and M. smegmatis into either 10% LB or 7H9 media, respectively, in PBS or into 100% PBS, both after washing twice in their respective media. After 30 min in the absence of LB or 7H9, in which cells have reduced respiration due to acute nutrient deprivation, both E. coli and M. smegmatis had very low levels of activated CellROX Green fluorescence compared to the background signal from unstained cells (Figures 3A,B). However, >95% of both cell types in 10% media/90% PBS showed evidence of dye uptake and a uniform several order-of-magnitude increase in the CellROX^hi fluorscence (Figures 3A,B). CellROX Green activation observed in cells in the presence of LB or 7H9 media is likely attributable to higher basal metabolism in the actively dividing cells, which leads to respiration-derived O₂^•– (Gonzalez-Flecha and Demple, 1995; Turrens, 2003; Korshunov and Imlay, 2006; Brynildsen et al., 2013; Dwyer et al., 2014; Adolfsen and Brynildsen, 2015; Lobritz et al., 2015).

To verify that metabolically derived O₂^•– is the predominant ROS species reacting with CellROX Green in these actively respiring cells, we quantified dye activation in E. coli knockout strains lacking SODs sodA and sodB and grown in LB to different states of cell density (early log, late log, and stationary phases). Cytosolic O₂^•– is rapidly converted to H₂O₂ by the SODs at diffusion-limited rates (k_cat/K_M ∼7 × 10⁹ M^-1 s^-1) (Berg et al., 2002). As shown in Figure 3B, there is a significant increase in CellROX^hi cells among the strains lacking SOD compared to wild-type E. coli. Further, by accounting for differences in cell numbers at early, mid-log and stationary cultures by adjusting for turbidity pre-treatment and events collected post-data acquisition (Figure 3B), we recapitulated previous observations that metabolic O₂^•– generation correlates with growth stage-dependent cellular metabolic activity rather than overall culture viability (Yamashoji et al., 2001). These results provide further support for the conclusion that CellROX Green activation in bacteria is predominantly due to reactions with O₂^•–.

There is emerging and intriguing evidence for multiple roles for ROS in the bacterial response to antibiotic exposure (Nguyen et al., 2011; Grant et al., 2012; Dwyer et al., 2015). However, methods using ROS-reactive dyes to identify and quantify ROS in antibiotic-treated bacteria are prone to artifacts in the absence of rigorous controls. For example, drug-induced changes in cell morphology can increase background fluorescence instead of activating the dye (Kohanski et al., 2007; Renggli et al., 2013; Paulander et al., 2014), the use of ROS-reactive dyes in batch culture methods does not inform about ROS inside cells (Albesa et al., 2004), and there is disagreement about the ability of ROS-reactive dyes to gain entrance into bacteria such as Pseudomonas aeruginosa and M. smegmatis (Nguyen et al., 2011; Grant et al., 2012). In all of these cases, rigorous flow cytometric gating to control for dye uptake, cell morphology and background fluorescence could minimize these problems. Here, we illustrate this idea using CellROX Green to assess the relationship between ROS production and cytotoxicity in M. smegmatis treated with mechanistically distinct antibiotics known to be effective against mycobacteria: ethambutol (cell wall synthesis inhibitor), streptomycin (binds to 30S ribosome subunit), rifampicin (binds to DNA-dependent RNA polymerase), isoniazid (inhibits mycolic acid synthesis), kanamycin (binds to 30S ribosome subunit), and norfloxacin (topoisomerase inhibitor). M. smegmatis cytotoxicity dose-response curves for these antibiotics are shown in Figure 4A.

We first assessed the effects of antibiotic exposures on FSC, SSC and background fluorescence in M. smegmatis. As shown in Supplementary Figure S3, while there were no consistent changes in SSC or increased background fluorescence (in absence of dye), there were time-, dose- and antibiotic-dependent shifts to lower SSC in treated bacteria. Given the recognized link between SSC shift and bacterial cell cycle progression (Walberg et al., 1997a,b), elongation (Renggli et al., 2013), and the uniquely asymmetric cell division of rod-shaped mycobacteria (Kieser and Rubin, 2014), these results are consistent with a delayed cell elongation in M. smegmatis populations hours after the initial drug exposure.

After accounting for morphological changes in and background fluorescence, we next assessed antibiotic-induced O₂^•– generation in M. smegmatis using CellROX Green. As shown in Figures 4B,C, antibiotics did not consistently cause O₂^•– generation in M. smegmatis. While ethambutol, rifampicin, and isoniazid increased CellROX^hi levels after 24 h of treatment (Figure 4B) and as early as 3 h after treatment (Figure 4C), norfloxacin and kanamycin treatments produced minimal or no dye-detectable ROS in response to 24 h treatment (Figure 4B). For streptomycin, no dye-detectable ROS was observed after 3 h of exposure (Figure 4C), while too many cells were killed at 24 h for accurate measurements.

Quantitative relationships between CellROX activation and cytotoxicity caused by the antibiotics were observed to be inconsistent in three studies. First, treatment with ethambutol produced dose-dependent killing while >95% of the cells produced dye-detectable ROS across all tested doses (Figures 4A,B). We then assessed antibiotic-induced toxicity by co-staining with propidium iodide (PI), a dye that is excluded from viable cells but enters cells with damaged membranes, and by survival as colony-forming units (CFUs). Based on flow cytometry contour plots, such as those in Supplementary Figures S4A–C, uptake of PI and CellROX varied among antibiotic exposures. However, similar proportions of cells were stained with PI in both the total population (Figure 4E) and the CellROX^hi population (Figure 4F) at 24 h post-exposure. In a third set of studies, the inconsistency of ROS formation and cell toxicity was more pronounced in CFU cell death assays for both E. coli and M. smegmatis (Supplementary Figure S4D). Together, these data suggest that the mechanisms by which antibiotics cause increased O₂^•– production are unique to each antibiotic, and elevated steady-state O₂^•– levels alone are not causative or predictive of cell death.

Finally, we assessed the formation of O₂^•– in M. smegmatis clones induced to be resistant to isoniazid and rifampicin, two antibiotics that caused significant O₂^•– production (Figures 4B,C). Antibiotic-tolerant bacteria were generated by culturing M. smegmatis in varying concentrations of antibiotics for 5 days. Despite several logs of killing by 20–30 μg/mL isoniazid and rifampicin at 24 h (Figure 4A), 5 days of growth in the presence of the antibiotics led to a drug-resistant population (Figure 4E). Compared to M. smegmatis treated for 24 h, the antibiotic-tolerant bacteria displayed greatly reduced O₂^•– production following antibiotic exposures (Figure 4F). Furthermore, there were fewer PI⁺ bacteria (i.e., bacteria with damaged membranes) and no positive correlation between PI staining and antibiotic concentration at 5 days was observed, in contrast with the same treatment after 24 h (Figures 4E,G). The mechanisms by which bacteria became resistant to the antibiotics was not established.

We next assessed O₂^•– generation in mycobacteria in the well-established non-replicating and antibiotic-tolerant (i.e., persistent) physiological state caused by nutrient deprivation (Xie et al., 2005; Gengenbacher et al., 2010). Interestingly, starved bacteria consistently produced significantly higher levels of endogenous O₂^•– than log-growing, nutrient-replete cells (Figures 5A–C). This was not a quirk of M. smegmatis as we observed the same phenomenon in slow-growing M. bovis BCG when starved up to 22 days (Figures 5A–C). Given this starvation-induced increase in basal O₂^•– production and the observation that increased ROS levels render some bacteria susceptible to killing by specific antibiotics (Brynildsen et al., 2013; Dwyer et al., 2014; Adolfsen and Brynildsen, 2015; Belenky et al., 2015), we quantified antibiotic sensitivity of log-growing or starved (4, 10, 20 days) BCG with the front-line TB antibiotics isoniazid, rifampicin, ethambutol, or streptomycin. As shown in Figures 5D–G, starved mycobacteria were tolerant to all four antibiotics compared to log-growing bacilli in spite of elevated basal O₂^•– production. This idiosyncratic behavior of starvation-induced persisters led us to investigate the effect of antibiotic exposure on O₂^•– levels in these cells. After accounting for elevated basal O₂^•– caused by starvation, we found that generation of O₂^•– in antibiotic-stressed, starved BCG was idiosyncratic to the antibiotic. While higher than basal levels of O₂^•– were observed in log-growing BCG for isoniazid and ethambutol (Figures 5J,K), as observed for M. smegmatis (Figures 4B,C), elevated O₂^•– generation above the basal level was not observed in starved BCG treated with streptomycin, isoniazid, or ethambutol (Figures 5I–K), which agrees with other studies on antibiotic-related ROS in persisters (Nguyen et al., 2011; Grant et al., 2012).

A possible explanation for the lack of antibiotic-related endogenous O₂^•– induction in starvation-induced persisters is metabolic quiescence. To test for metabolic activity in starved cells, we performed staining of starved BCG with CTC, CFDA, and DAPI. Enhanced respiratory activity was observed in a sub-population of cells, as evident in the increase of the CTC^hi sub-population during starvation and the reversal of CTC activation by incubation in nutrient-replete medium (Figure 5L and Supplementary Figure S5A). Starvation also increased cellular esterase activity as shown by a CFDA^hi sub-population (Figure 5M and Supplementary Figure S4A), which again is reversed by transfer to nutrient-replete medium. Together, these observations eliminate metabolic stasis as the cause of the muted ROS production in the response of starved BCG antibiotic exposure.

Another possible explanation for reduced O₂^•– generation in antibiotic-exposed non-replicating mycobacteria is the active exclusion of drugs through efflux pumps. Indeed, drug efflux pumps are often up-regulated in drug-tolerant cells (Sarathy et al., 2013; Maisonneuve and Gerdes, 2014). However, we observed that rifampicin treatment induced O₂^•– generation in a dose-dependent manner in both log and starved BCG (Figure 5H), in spite of starvation-induced up-regulation of transcription of a known rifampicin efflux pump, BCG_1316c (M. tuberculosis Rv1258c homolog) (Adams et al., 2011) (Supplementary Figure S5B). Even with enhanced O₂^•– production above an already high basal level, rifampicin treatment did not result in cell death (Figure 5D), indicating that starved BCG is refractory to O₂^•–-mediated killing, unlike log-growing E. coli (Brynildsen et al., 2013; Dwyer et al., 2014; Adolfsen and Brynildsen, 2015; Belenky et al., 2015). These findings demonstrate that antibiotic-induced O₂^•– production and associated killing is both antibiotic- and cell state-dependent.