Introduction

A key goal of antibiotic therapy is, in conjunction with the immune system, to eradicate the infecting bacteria. While many common bacterial infections respond rapidly to antibiotics, with 1-2 weeks of therapy sufficient to achieve high cure rates, there are also infections where bacterial clearance is slow and often incomplete^{1, 2}. This challenge is exemplified by mycobacterial infections. Fully-susceptible Mtb requires multiple antibiotics for four months or longer^{3, 4}, and infections by non-tuberculous mycobacteria (NTM) such as Mycobacterium abscessus (Mabs) are often even more difficult to eradicate; Mabs often requires treatment for 12-18 months, and even then has a 50% relapse rate^{5, 6}.

While the ability of mycobacteria to escape antibiotic-mediated killing is multifactorial, the phenomenon of antibiotic tolerance is likely an important contributor^7–10. Studies dating from the 1940s noted that when a population of susceptible bacteria were exposed to a bactericidal antibiotic such as penicillin, the majority of the population died within a few hours, but that a small sub-population of cells remained viable for days¹¹. Importantly, these antibiotic-tolerant cells referred to as “persister cells” had not acquired a mutation conferring heritable antibiotic resistance, and do not grow in the presence of the antibiotic. Rather, they had entered into a readily-reversable phenotypic state where, despite antibiotic-mediated inhibition of critical processes, they are able to survive^12–14. In addition, a number of physiologic stresses increase antibiotic tolerance in a population, as bacterial cell death is markedly slowed by stresses such as nutrient deprivation or acidic pH^{10, 11, 15, 16}. Notably, these same stresses are encountered in the lysosome of an activated immune cell¹⁷, and studies of pathogens isolated from activated macrophages indeed show a strong immune-mediated increase antibiotic tolerance^{9, 18}. Thus, paradoxically, the immune system can actually impede bacterial eradication by antibiotics.

Antibiotic tolerance has been studied extensively in model systems such as Escherichia coli, which has provided important insights, but also highlighted uncertainties of current models. Several different regulators of antibiotic tolerance have been identified in E. coli including the HipBA toxin-antitoxin system¹⁹, guanosine pentaphosphate ((p)ppGpp) synthesis by RelA/SpoT enzymes^{20, 21}, and the activity of Lon protease²⁰. In each of these models, the postulated mechanism is to arrest cell division and render the process targeted by antibiotics non-essential. However, important questions remain unanswered. It is unclear how cells remain viable when critical processes such as transcription or translation are blocked by antibiotics, as well as how the process is regulated and induced by stress. Even the mechanism of cell death following antibiotic exposure itself has been a matter of debate - originally antibiotics were presumed to kill bacteria as a direct result of inhibition of their target molecule, such as β-lactam antibiotics disrupting cell wall integrity, directly leading to mechanical cell lysis²². However, a number of studies, largely from E. coli, have demonstrated that in addition to the initial target inhibition, which may be bacteriostatic, bactericidal antibiotics can also cause secondary lethal ROS accumulation, leading to cell death^{14, 23–28}.

Mycobacterial persister cells are particularly resilient, as Mycobacterium smegmatis (Msmeg) and Mycobacterium tuberculosis (Mtb) persisters can endure many weeks of antibiotic exposure, and stress-induced antibiotic tolerance develops readily^{10, 14, 29}. Mabs is a species of rapidly-growing mycobacteria, with a doubling time of ∼4h in rich media³⁰, and while often environmental, it causes opportunistic infections in patients with structural lung disease such as cystic fibrosis. It is also among the most difficult of all bacterial pathogens to treat, because in addition to forming persister cells, it is also intrinsically resistant to many classes of antibiotics, leaving few treatment options⁵. This leads to the use of antibiotics with greater toxicity to patients, and a need to use these agents for prolonged periods to prevent relapse. Thus, identifying the genes that Mabs persister cells rely on for survival, as well as the genes involved in the induction and maintenance of stress-induced antibiotic tolerance might highlight pathways that could be targeted therapeutically to eliminate antibiotic-tolerant cells.

Previous genetic screens have studied antibiotic responses in mycobacteria, with some evaluating heritable resistance, and others investigating tolerance. Several studies of resistance have successfully used either transposon mutagenesis with insertion site sequencing (Tn-Seq) or CRISPR-based transcriptional repression with high-throughput sequencing of guide RNAs (CRISPRi) to identify genes promoting growth in sub-inhibitory concentrations of antibiotic. These studies have provided insights such as highlighting the importance of cell membrane permeability as a mechanism controlling antibiotic penetration into the cytoplasm^31–33.

Persister formation has proven challenging to study, likely because the low frequency of persisters leads to population bottlenecks that confound genetic analysis. Although screens in Mtb have been conducted in macrophages and mice, and genes such as glpK and cinA identified, overall the number of mutants isolated in these screens has been low^34–35. There has been one effective in vitro Tn-Seq study examining antibiotic tolerance in Mtb exposed to starvation and rifampin that isolated over 100 mutants²⁷, demonstrating the feasibility of genetic screening in this context. However, whether these phenotypes seen with rifampin in Mtb extend to other mycobacteria and other antibiotics remain to be determined.

Here, we study antibiotic tolerance in Mabs and describe the results of a genome-wide Tn-seq screens seeking to identify the genes required for both spontaneous persister cell survival as well as starvation-induced antibiotic tolerance following exposure to translation-inhibiting antibiotics. We identified several discrete processes contributing to survival and observed a prominent role for ROS detoxifying factors such as the catalase-peroxidase enzyme KatG, which contributed to both spontaneous persister survival and starvation-induced antibiotic tolerance. Consistent with the protection conferred by KatG, we found that endogenous ROS accumulated following antibiotic exposure, and that the removal of oxygen significantly impaired bacterial killing. Taken together these findings support a model in mycobacteria where the lethality of translation-inhibiting antibiotics is amplified by a secondary accumulation of toxic ROS, and that survival requires active detoxification systems.

Results

Starvation-induced antibiotic tolerance in mycobacteria

We first sought to develop conditions suitable for genetic analysis of persister cell survival and stress-induced antibiotic tolerance mycobacteria. Genetic screens examining persister cell physiology face two inherent obstacles. First, these cells are rare in unstressed bacterial populations, and antibiotic-mediated cell death creates population bottlenecks that obscure mutant phenotypes. Second, most mycobacterial populations contain spontaneous drug-resistant mutants that can expand if the population is exposed to a single antibiotic. To overcome these obstacles, we sought to establish large-scale, high-density culture conditions to prevent genetic bottlenecks, and used multiple antibiotics to suppress spontaneous drug-resistant mutants. We began by assessing the feasibility of this approach using wild-type Msmeg. We exposed the cells to either the combination of rifampin, isoniazid, and ethambutol (RIF/INH/EMB) used to treat Mtb, or to the combination of tigecycline and linezolid (TIG/LZD), two translation-inhibiting antibiotics frequently used to treat Mabs^{6, 36}, and empirically determined the minimum inhibitory concentrations (MICs) and minimum bactericidal concentrations (MBC) for each antibiotic under the high-density culture conditions that would be needed for genetic analysis of persister cells. Both antibiotic combinations reduced the bacterial population >1000-fold within 72h (Figure 1A). We then evaluated both spontaneous persister formation and stress-induced tolerance under these conditions in Msmeg. We compared logarithmically growing (mid-log) cultures in 7H9 rich media to cultures starved for 2 days in phosphate buffered saline (PBS) prior to addition of antibiotics. Consistent with expectations, we found a marked increase in antibiotic tolerance in starved cultures, with a 100-fold increase in survival following TIG/LZD exposure and a 10,000-fold increase following RIF/INH/EMB exposure (Figure 1A).

Figure 1.

We next examined two species of pathogenic mycobacteria under these conditions to similarly assess starvation-induced antibiotic tolerance^37–40. We again compared cells starved in PBS to logarithmically growing cells in 7H9 and found that cultures of wild-type Mabs (ATCC 19977) and Mtb (Erdman) also displayed dramatic increases antibiotic tolerance in nutrient-deprived cultures (Figure 1B,C). Notably, for Mabs and Msmeg, the development of tolerance required an adaptation period of several days under starvation conditions, as survival was dramatically impaired if cells were shifted immediately into nutrient-deficient conditions with antibiotics, suggesting that a regulated process needed to be completed. Surprisingly, Mtb tolerance developed rapidly without pre-adaptation suggesting that this organism might have additional response pathways enabling more rapid adaptation.

Identification of pathways needed for antibiotic tolerance in Mabs

We used these conditions to carry out Tn-Seq screens in Mabs to identify genes necessary for both the survival of spontaneous persister cells and for starvation-induced antibiotic tolerance. We conducted the screen using a Mabs Himar1 Tn library comprised of ∼55,000 mutations across ∼91,000 possible TA insertion sites covering all 4,992 non-essential Mabs genes in strain ATCC 19977³³. To study spontaneous persister cells, cultures were maintained in continuous log-phase in 7H9 rich media for 48h prior to antibiotic exposure and then exposed to TIG/LZD for 6 days (Figure 2A) a point at which spontaneous persister cells comprise the majority of the population (Figure 1C). To study starvation-induced tolerance, cultures were starved in PBS for 48h prior to antibiotic exposure in PBS. Following antibiotic treatment, cells were then washed and resuspended in antibiotic-free liquid media to recover and passaged 1:100 three times in continuous log-phase to expand surviving cells. We then isolated genomic DNA, sequenced the Tn insertion sites, and used TRANSIT software⁴¹ to quantify the abundance of each Tn mutant across different conditions to identify genes with statistically-significant differences in distribution. We identified 277 Mabs genes required for surviving TIG/LZD exposure in rich media, 271 genes required for survival during starvation and 362 genes required to survive the combined exposure to antibiotics and starvation (Log₂ fold-change > 0.5 and Benjamini–Hochberg adjusted p-value (p-adj.) ≤ 0.05; Figure 2B-E). Of the genes required for survival, ∼60% were required in both nutrient-replete and starvation states, although condition-specific determinants were also seen (Figure 2F). As expected, we identified genes with already-established functions in antibiotic responses, including MAB_2752 and MAB_2753 which are both homologs of known antibiotic transporters in Mtb, as well as tetracycline-responsive transcription factors like MAB_4687 and MAB_0314c (Table S1), indicating an ability of these Tn-Seq conditions to identify physiologically relevant genes known to mitigate antibiotic stress.

Figure 2.

To identify other cellular processes necessary for survival we performed pathway enrichment analysis on the set of genes identified by Tn-Seq. We used the DAVID⁴² analysis tool to perform systematic queries of the KEGG, GO, and Uniprot databases and identify over-represented processes and pathways. Interestingly, although cells were exposed to translation-inhibiting antibiotics, and no exogenous oxidative or nitrosative stress was applied, we identified a number of factors needed to combat these stresses in spontaneous persister cells. This included bfrB (bactoferritin), ahpE (peroxiredoxin) and katG (catalase/peroxidase) as well as 5 components of the bacterial proteasome pathway, known to mediate resistance to nitrosative stress in Mtb⁴³ (Figure 2G, Table S2). We also identified multiple members of DNA-damage response pathways including recF, recG, uvrA, uvrB and uvrC. Examining genes required for starvation-induced tolerance, a number of the same pathways were again seen, and the mutant with the greatest survival defect in this context was mntH, a redox-regulated Mn²⁺/Zn²⁺ transporter implicated in peroxide resistance in other organisms^44–45.

To independently confirm a role in antibiotic tolerance for a set of genes from diverse pathways that were identified by Tn-Seq we selected a set of genes required for survival, representing several of the functional pathways identified, and used oligonucleotide-mediated recombineering (ORBIT)⁴⁶ to disrupt their open reading frames. The initial genes selected were pafA (proteasome pathway), katG (catalase-peroxidase), recR (DNA repair), blaR (β-lactam sensing), and MAB_1456c (cobalamin synthesis). To control for non-specific effects of antibiotic selection during the recombineering process, we created a control strain using ORBIT to target a non-coding intergenic region downstream of a redundant tRNA gene (MAB_t5030c). We then individually screened each of these mutants to determine if they displayed deficits in survival by exposing cells to TIG/LZD, either in rich 7H9 media or under starvation conditions, as had been done in the Tn-Seq screen. For ΔkatG we detected clear defects in survival as soon as 3 days after antibiotic exposure in either rich media or under starvation conditions, corroborating the results of our Tn-Seq analysis (Figure 3A). We observed similar, albeit smaller, defects in the ΔpafA, ΔMAB_1456c, and ΔblaR, mutants under the conditions predicted by the screen, whereas we saw no defect in the ΔrecR mutant (Figure 3B-E). We also noted that several mutants displayed delays in growth recovery after antibiotic removal at times before survival defects were noted (Figure 3 Supplement). To further confirm the role of katG and pafA, and exclude off-target effects of recombineering, we performed genetic complementation analysis by restoring a wild-type copy of each gene into the respective ΔpafA and ΔkatG mutants. In each case, we integrated a single copy of the wild-type gene, under the control of its endogenous promoter, into the genome at the L5 attB site (hereafter pafA+, katG+ strains), and constructed isogenic control strains with an empty vector integrated at the same site (hereafter pafA-, katG-strains). We confirmed expression of the re-introduced copy of each gene by RT-qPCR in the pafA+, and katG+ strains, and found expression within roughly 2-fold of endogenous wild-type levels (Figure 4A,D). We then challenged these strains with TIG/LZD as before. In rich media, where the ΔkatG mutants have a moderate survival defect, the katG+ strain had roughly a 50-fold increase in viable cells relative to the katG-strain. We then exposed cells to antibiotics under starvation conditions, where the ΔkatG mutant phenotype is more severe. Under these conditions the katG-cells succumbed rapidly between 3d and 10d after antibiotic exposure, with a 1,000-fold decrease in viable cells relative to control cells, whereas the katG+ strain showed a near-complete restoration of antibiotic tolerance (Figure 4B). We analogously examined complementation of ΔpafA mutants, and although the phenotype of the ΔpafA mutant is less severe overall than a ΔkatG mutant, we saw a similar restoration of survival in pafA+ cells relative to pafA-cells (Figure 4E). We next evaluated whether the pafA-, and katG-strains were overall more sensitive to the growth inhibitory effects of TIG/LZD, or whether they had specific defects in survival above the mean bactericidal concentration. We performed MIC determination for TIG and LZD individually for each strain, comparing the katG+/katG- and pafA+/pafA- strains. We found that that the MICs for each of these strains were unchanged, demonstrating that these mutants were not more readily inhibited by these antibiotics. Instead, they have more rapid kinetics of cell death at bactericidal concentrations, consistent with a specific defect in antibiotic tolerance, and supporting a model whereby an initial growth-arresting inhibition of the direct antibiotic target can be uncoupled mechanistically from a distinct cell-death step (Figure 4C,F), as has been seen with other antibiotic classes²⁶.

Figure 3.

Figure 4.

Reactive oxygen contributes to antibiotic lethality in Mabs

We next investigated the role of KatG and reactive oxygen in antibiotic tolerance more broadly. We began by assessing whether KatG conferred protection from other antibiotics with diverse mechanisms of action, selecting antibiotics that are used clinically for mycobacterial infections. Because katG- mutants showed the greatest defects in starvation-induced tolerance, we analyzed survival of katG+ and katG- strains in starvation-adapted cultures exposed to a panel of different antibiotics. Because both TIG and LZD both act by inhibiting translation, we began by exposing cells to either TIG or LZD alone. As expected, the degree of bacterial killing was significantly less with either agent alone than when they are added in combination. Upon exposure to either of these antibiotics the katG- cells died more rapidly than katG+ cells, though we note that the final proportion of persister cells in the population was unchanged in katG- cells (Figure 5A). When we exposed cells to rifabutin, an RNA polymerase inhibitor, we saw a similar effect, with a 100-fold loss of viability in katG- cells relative to the katG+ cells (Figure 5B). In contrast, when we exposed cultures to either levofloxacin (topoisomerase inhibitor) or cefoxitin (β-lactam inhibitor of peptidoglycan cross-linking), katG had little to no effect on cell viability (Figure 5C-D). Thus, the role of KatG is context-dependent, suggesting that some antibiotics generate oxidative stress that can be ameliorated by KatG while others do not.

Figure 5.

The identification of katG as essential for cells to survive exposure to TIG/LZD suggests that ROS are present and causing damage. Although TIG/LZD are translation inhibitors that do not directly generate ROS we evaluated whether they might nonetheless be triggering ROS accumulation as a secondary effect. We examined ROS levels in control Mabs using the ROS indicator dye cellROX, that is retained in cells when it becomes oxidized⁴⁷. At baseline, during log-phase growth in rich media < 2% of cells had ROS accumulation (Figure 5E). We saw a moderate increase in ROS accumulation in starved cultures, with roughly 7% of the population cellROX+. However, when cells were exposed to antibiotics we saw a dramatic accumulation of ROS, with 57% of cells becoming cellROX+ when exposed to TIG/LZD in rich media and 33% of PBS-starved cells becoming cellROX+ when exposed to TIG/LZD. Taken together, these data indicate that translation inhibition does indeed have important downstream effects on cellular redox balance, with ROS accumulation that could be contributing to the lethal effects of antibiotics.

We next tested whether ROS were contributing to cell death by reducing ROS production and assessing the impact on cell viability. A well-established system for studying hypoxia in mycobacteria is the Wayne Model of gradual-onset hypoxia, whereby low-density cultures are inoculated in sealed vessels with minimal headspace. As the culture slowly grows, the soluble oxygen is consumed, resulting in the slow onset of hypoxia over several days, a process that can be monitored by the decolorization of Methylene Blue dye in the media⁴⁸. Under aerobic conditions in rich media, we observed the expected rapid killing of Mabs over the first 5 days with the combination of TIG/LZD, with more rapid loss of viability in KatG- cells. However, under hypoxic conditions, where ROS production is suppressed, we saw much slower bacterial killing. Importantly, under hypoxic conditions katG- cells no longer had a survival defect relative to katG+ cells, supporting the hypothesis that translation-inhibiting antibiotics also cause secondary accumulation of lethal ROS in antibiotic-treated cells that need to be detoxified by KatG (Figure 5F).

We also evaluated whether other methods of alleviating ROS damage might enhance survival. The iron chelator 2,2’-dipyridyl has been shown in other contexts to reduce ROS-mediated damage by suppressing the reaction of H₂0₂ with Fe²⁺ that generates highly oxidizing hydroxyl radicals (Fenton reaction), and which has been shown to mitigate oxidative damage in other bacteria following antibiotic exposure^{23, 26, 49}. As seen in other bacteria, we find that in Mabs, 2,2’- dipyridy does indeed improve bacterial survival following exposure to bactericidal translation inhibitors, further supporting a role in ROS in cell death following translation inhibition (Figure 5G). We also assessed whether free radical scavengers like thiourea and 2,2,6,6- tetramethylpiperidine-1-oxyl (TEMPO) ameliorated antibiotic toxicity, although similar antioxidants had previously been shown in Mtb to increase respiration and paradoxically to increase ROS generation²⁷. When we treated Mabs simultaneously with TIG/LZD in combination with either thiourea or TEMPO we did not observe a restoration of antibiotic tolerance, and, similar to observations in Mtb, actually observed increased bacterial cell death (Figure 5 Supplement).

Incomplete penetrance of katG phenotype among Mabs strains

To test whether ROS accumulation was an effect occurring more broadly across different Mabs strains we obtained two clinical strains, exposed them to TIG/LZD in 7H9 media, and measured ROS accumulation with cellROX as above. We found that similar to ATCC 19977, both clinical Mabs strains had elevated ROS levels following translation inhibition (Figure 6A), suggesting that this is a conserved process in Mabs. Next, we tested the role of KatG in these Mabs clinical strains. We used ORBIT to disrupt the katG locus and evaluated the ability of these ΔkatG clinical strains to survive exposure to TIG/LZD under both stressed and un-stressed conditions. Unlike ROS accumulation, where the responses across strains were consistent, we saw a variable dependency on katG. Clinical strain-1 behaved differently overall, with no appreciable starvation-induced antibiotic tolerance, and no contribution of katG to survival. In contrast, for clinical strain-2, katG contributed significantly to starvation-induced antibiotic tolerance, behaving similarly to the ATCC 19977 reference strain. However, unlike the reference strain, katG was not required for survival of clinical strain-2 when exposed to antibiotics in 7H9 media (Figure 6B). Thus, although antibiotic-induced ROS accumulation was observed across all three Mabs strains, the ΔkatG phenotype displays incomplete penetrance, suggesting that in some Mabs strains alternative pathways exist that are able to compensate for the loss of katG.

Figure 6.

In summary, the results of these studies point to an important effect of ROS in amplifying the lethality of transcription and translation-inhibiting antibiotics in Mabs. Through genetic analysis we identified a number of ROS detoxification factors, including KatG, as necessary for survival in this context. This suggested that antibiotics induced an oxidative state in cells, and direct measurement of ROS following antibiotic exposure indicated that this was indeed the case. Further supporting the toxic effects of ROS in this context, we found that removal of oxygen both slowed bacterial killing and rendered KatG dispensable. Taken together, these results suggest that in Mabs antibiotic lethality is accelerated by ROS accumulation, and that survival requires active detoxification systems.

Discussion

Pathways necessary for antibiotic tolerance in Mabs

The phenomenon of antibiotic tolerance has been recognized for decades and has been observed in a broad array of bacterial species, but without a uniform underlying mechanism - suggesting that different pathways may play roles in different physiologic contexts. For example, the pathways identified in one bacterial species may not contribute to tolerance in another. In addition to E. coli, relA plays a role in tolerance in Pseudomonas aeruginosa⁵³, Staphylococcus aureus⁵⁴, and Mtb⁵⁵. However, its role is not universal. Deletion of relA had no effect on antibiotic tolerance in Msmeg⁵⁶, and in our Tn-Seq analysis, Mabs relA Tn mutants had no survival defect. However, a prior study of the Mabs relA mutant demonstrated that this strain still synthesizes (p)ppGpp⁵⁷, suggesting potential genetic redundancy. Even within a single species there can be differences in the critical survival mechanisms depending on the context. In E. coli, RelA contributes to persister viability following exposure to β-lactams, but not quinolones⁵³, and in our study we find KatG to be essential for tolerance to transcription and translation inhibitors but not to β-lactams or quinolones.

Mechanisms of antibiotic lethality

Our findings strongly support the idea that ROS can be a significant contributor to antibiotic lethality in Mabs and contribute to the growing evidence that this is conserved across mycobacteria. In Mabs, Mtb and Msmeg, other groups have also observed reduced antibiotic-mediated killing in hypoxic conditions^{25, 37}. In addition, in Mtb, exposure to rifampin also generates ROS^{19, 60}, and katG contributes to survival in rifampin-treated cells²⁶.

Exactly how transcription or translation blockade leads to increased ROS levels is also not known. In principle, any of several derangements could lead to ROS accumulation. One of the major sources of cellular ROS is oxidative phosphorylation, as hydrogen peroxide and superoxide are natural byproducts. Thus, increased ROS generation by oxidative phosphorylation is an attractive hypothesis. Alternatively, particularly under starvation conditions, it is possible that antioxidants and ROS scavengers may become depleted, creating a more oxidizing environment. Our Tn-Seq analysis provides additional insight on this. We noted a small class of Tn mutants that were paradoxically protected from antibiotic lethality (Figure 2B). Prominent among this class of mutants were several independent components of the NADH dehydrogenase complex. Also known as Complex I of the electron transport chain, it is one of the key entry points for electrons into the oxidative phosphorylation pathway. The observation that mutants lacking this complex are protected suggests that decreasing flux through oxidative phosphorylation, with a concomitant decrease in ROS generation, may enhance survival during antibiotic exposure. A mechanistic understanding of how blockade of either transcription or translation leads to deranged oxygen utilization is an unresolved question that will require further study.

ROS accumulation is not universal following exposure to bactericidal antibiotics. While antibiotic-induced ROS is well-documented, under some conditions, such as higher concentrations of antibiotic, multiple studies have also observed antibiotic lethality without ROS accumulation^{23, 58–61}. Similarly, prior studies have found that under certain conditions, E. coli mutants lacking catalase have defects in antibiotic tolerance^62–64, whereas under other conditions they do not^{23, 65}. In Mabs the role of katG was also not uniform, as it had no impact on survival following exposure to levofloxacin or cefoxitin. Additional studies will be needed to determine whether levofloxacin and cefoxitin kill without generating ROS, or whether ROS is generated but effectively detoxified in the absence of katG. This latter possibility is suggested by our findings in Mabs clinical strains. While we saw ROS accumulation in both strains after TIG/LZD exposure, the role of katG was variable between the strains. This suggests that compensatory pathways likely exist and that in Mabs they can overcome the loss of katG.

Therapeutic implications

Mabs infections are particularly challenging to treat, with relapse rates over 50%³. Our results highlight several bacterial processes, such as the bacterial proteasome and ROS detoxification, which might be targeted therapeutically to reduce the survival of antibiotic tolerant bacteria in patients with Mabs infection. Agents targeting these processes might not have any intrinsic antimicrobial activity alone, but might act to disrupt the unique physiology required to survive antibiotics. This would represent a new therapeutic class of “persistence inhibitors” that might act synergistically with traditional antibiotics to eliminate the subpopulation of cells that would otherwise remain viable despite prolonged antibiotic treatment in patients with Mabs and other chronic infections^{6, 27}.

Limitations

Tn-Seq has inherent drawbacks, including an inability to identify mutants in essential genes, or in cases of genetic redundancy. Thus, there are likely genes needed for antibiotic tolerance in Mabs that were not identified in this study. In addition, we studied the response to a single class of antibiotic, focusing on the translation inhibitors often used to treat Mabs infections, and we studied only spontaneous persister cells and starvation-induced antibiotic tolerance. It is likely that examining other antibiotics, with different mechanisms of action, or different stresses that induce tolerance would identify additional genes contributing to survival and would allow identification of core pathways that might be shared in differing physiologic contexts of antibiotic and stress^{25, 27}.

Materials and methods

Key Reagent Table

Bacterial strains and culture conditions

Mabs ATCC 19977, clinical Mabs strains and Msmeg (MC2 155) were grown in BD Middlebrook 7H9 media (liquid) or 7H10 media (solid) supplemented with 0.5% glycerol (Sigma) and 0.2% Tween-80 (Fisher) but without any OADC supplementation except for transformations. Sacramento clinical isolates were obtained from the Sacramento County Department of Public Health Mycobacteriology Laboratory. Confirmation of clinical isolates as Mabs was performed by amplifying the 16s rRNA locus and Sanger sequencing (maybe reference new data table?). Mtb (Erdman) was grown in 7H9 (liquid) or 7H10 (solid) supplemented with 0.5% glycerol, 0.1% Tween-80, and 10% OADC (BD). All cultures were grown at 37°C with gentle shaking. Except for specific hypoxia conditions, all liquid cultures were grown with 90% container headspace or using a gas permeable cap to ensure culture oxygenation. PBS starvation was achieved by washing OD0.5-1.0 Mabs 1X in DPBS (-Ca/Mg, Gibco) and resuspending in DPBS at OD=1, supplemented with 0.1% tyloxapol (Sigma).

Mabs antibiotic experiments

For PBS starvation experiments, stocks of Mabs were grown for 48 hours in 7H9 passaging continuously in log phase, then PBS starved or passaged in log phase for an additional 48h. Log phase or PBS starved Mabs were then resuspended in antibiotic containing media at OD1.0. For experiments with hypoxia, Mabs in mid-log aerobic growth was adjusted to OD0.001 in media with 1.5μg/ml methylene blue and added to a rubber septum sealed glass vial with 50% headspace. Methylene blue discoloration was observed at 3d and antibiotics were added at 5d. For thiourea and TEMPO experiments, these were added to the cultures at the time of antibiotic administration. For bipyridyl experiments, 62.5μM bipyridyl was added to the cultures 2h prior to antibiotic administration. We empirically determined the half-life of each antibiotic in 7H9 media at 37-deg and for those with half-lives shorter than the experiment, supplemented cultures to with additional antibiotic to maintain the concentration of active antibiotic. Antibiotics were used at the following concentrations: tigecycline (Chem-Impex) at 10μg/ml (8 fold above MIC, re-administered every 3 days), linezolid (Chem-Impex) at 100μg/ml (20 fold above MIC), levofloxacin (Sigma) at 40μg/ml (8 fold above MIC), cefoxitin (Chem Impex) at 80μg/ml (8 fold above MIC, re-administered every 3 days), and rifabutin (Cayman) at 40μg/ml (4 fold above MIC). After antibiotic administration, colony forming units over time were measured by plating on 7H10 solid media. For experiments where growth recovery time in liquid media was quantified, 100μl of sample was removed at d6 after antibiotic administration and washed 2X in antibiotic-free media. The samples were resuspended in 5ml of antibiotic free media, and OD620 measurements were taken with a FilterMax F3 plate reader (Molecular Devices) until maximum cell density (OD of ∼5.0) was reached.

Msmeg antibiotic experiments

Individual colonies were picked and grown for 48 hours in log phase before being PBS starved or passaged in log phase for 48h. Log phase or PBS starved Msmeg were then resuspended in antibiotic containing media at OD1.0. Antibiotics were used at the following concentrations: tigecycline (Chem-Impex) at 1.25μg/ml (8 fold above MIC, re-administered every 3 days), linezolid (Chem-Impex) at 2.5μg/ml (8 fold above MIC), rifampin (Sigma) at 32μg/ml (8 fold above MIC, re-administered every 6 days), isoniazid (Sigma) at 32μg/ml (8 fold above MIC, re-administered every 6 days), and ethambutol (Thermo) at 4μg/ml (8 fold above MIC, re-administered every 3 days). After antibiotic administration, colony forming units over time were measured.

Mtb antibiotic experiments

Freezer stocks of Mtb were thawed and grown for 5-7 days in log phase before being starved for 14d or longer. Non-starved control Mtb were thawed such that they were also grown for 5-7 days in log phase before experimental use. Log phase or PBS starved Mtb was then resuspended in antibiotic containing media and adjusted to OD1.0. Antibiotics were used at the following concentrations: rifampin (Sigma) at 0.1μg/ml (4 fold above MIC, re-administered every 6 days), isoniazid (Sigma) at 0.1μg/ml (4 fold above MIC, re-administered every 6 days), and ethambutol (Thermo) at 8μg/ml (4 fold above MIC, re-administered every 6 days). After antibiotic administration, colony forming units over time were measured.

Transposon insertion sequencing

The construction of this Himar1 transposon Tn library has been described previously³³. Screening was performed by growing a freezer stock of the library for 2.5 days in log phase before 48-hour PBS starvation or further continuous log-phase growth. Samples were then resuspended in media containing either tigecycline/linezolid or an equal volume of DMSO solvent and incubated for 6 days, with a re-administration of tigecycline or matching DMSO on day 3. Cultures were aerated by culturing in a vented cap bottle with gentle agitation at 40 revolutions per minute throughout the experiment. The samples were then washed 2X in antibiotic free liquid media, resuspended in antibiotic free liquid media (10X the original culture volume), and grown until OD0.5-1.0.

Subsequently, the samples underwent three more rounds of 100-fold passaging in liquid media to amplify surviving bacteria before the samples were collected in Trizol (Invitrogen). A sample taken at the time of the commencement of PBS starvation was collected in Trizol and used as the input control. Three independent trials of this experiment were submitted to the UC Davis DNA Technologies Core, where Tn insertion site flanking sequences were amplified as described previously³³ and sequenced on an Element Biosciences AVITI. Sequence reads were mapped to the ATCC 19977 genome and analyzed using TRANSIT software with the following parameters: 0% of N/C termini ignored, 10,000 samples, TTR normalization, LOESS correction, include sites with all zeros, site restricted resampling. Genes with significant changes were defined as those with adjusted p-value (p-adj.) <0.05 and log₂ fold change >0.5. P-adj. was calculated using the Benjamini-Hochberg correction.

Pathway enrichment analysis

To improve gene annotation, Mabs orthologs to Mtb genes were identified. Mabs genes were first converted into protein sequences using Mycobrowser, and protein sequences were then used to perform reciprocal BLASTp searches. Mabs genes and Mtb genes that mapped to each other using independent one-way BLASTp searches with a maximum e-value cutoff of 0.1 were considered orthologs. For pathway analysis, gene lists (Mtb orthologs) were then imported into the DAVID knowledgebase⁴² and pathway enrichment analysis performed for Gene Ontology biological process, Uniprot keyword and KEGG databases with statistical analysis Fisher’s exact test and nominal p-value reported.

Gene deletion and complementation

Knockout strains were generated using ORBITT⁴⁸. Briefly, Mabs was transformed with the kanamycin-resistant ORBIT recombineering plasmid pkm444. 20ml Mabs at OD0.5-1.0 was washed 2X in 10% glycerol and resuspended in 200ul 10% glycerol. 500ng plasmid was added and electroporated at 2.5kV in 0.2cm cuvettes. The bacteria were allowed to recover overnight before plating on 150μg/ml kanamycin plates. Clones were selected and regrown in liquid media supplemented with 150μg/ml kanamycin and 10% OADC (BD) to OD0.5-1.0. For recombineering, the pkm444-Mabs was grown to mid-log, then diluted to OD0.1 and 200mM glycine (Fisher) was added to the media. 16 hours later, 500mM sucrose (Sigma) and 500ng/ml anhydrotetracycline (Cayman) were added and incubated for an additional 4 hours. Subsequently, the Mabs was washed 2X in ice cold 10% glycerol + 500mM sucrose. 200ul of 10X concentrated Mabs was then electroporated with 600ng of the zeocin-R ORBIT payload pkm496 plasmid and 2μg of targeting oligonucleotide (Table S3) at 2.5kV in 0.2cm cuvettes. The Mabs was then allowed to recover overnight in liquid media with 10% OADC and 500ng/ml anhydrotetracycline before being plated on 150μg/ml zeocin plates. Mutants were then selected and screened for gene deletion by PCR amplification and Sanger sequencing. For genetic complementation, the endogenous loci including promoter and terminator sequences were amplified by PCR and cloned into the EcoRV site of pmv306 with kanamycin resistance⁶⁶. In the case of katG, the upstream gene furA was also included in the complementation construct to achieve optimal katG expression.

MIC determination

Two-fold serial dilutions of antibiotics were prepared in a 96 well plate in 100ul volume. 100ul of 2X bacteria were added (for Mabs: used a final OD of 0.001, Msmeg: OD0.001, Mtb: OD0.01), making a final volume of 200μl. The plates were incubated until there was visible growth in the no antibiotic control well. At this time, the bacteria were transferred to a new plate with 20μl of 40% paraformaldehyde and OD620 measurements were taken with a FilterMax F3 plate reader (Molecular Devices).

Flow cytometry

OD1 Mab was stained with cellROX green (Invitrogen) at a final concentration of 5μM for 1hr at 37C. The cells were then washed in PBS and resuspended in PBS with 4% paraformaldehyde and 5μg/ml DAPI (Sigma). The samples were run on an LSRII flow cytometer (BD). Fluorophores were excited with the 405nm (DAPI) and 488nm (cellROX) lasers. Detection was performed using the 450/50 (505LP) filter for DAPI and a 525/50 (555LP) filter for cellROX. Data were analyzed with FlowJo software (BD).

DNA/RNA Purification

Samples were resuspended in 5 volumes of Trizol, and bead beat with 0.1mm zirconia beads (Biospec) 6x2min at 4°C in a Mini-Beadbeater-16 (Biospec). Chloroform was added and RNA in the aqueous phase removed. For DNA isolation, a second RNA extraction was performed with 0.8M guanidine thiocyanate and 0.5M guanidine hydrochloride, 60mM Acetate pH 5.2, 1mM EDTA. DNA was then isolated with back-extraction buffer (4M Guanidine Thiocyanate, 50mM Sodium Citrate, 1M Tris base (without pH adjustment ∼pH 11) and DNA purified using a PureLink RNA Mini Kit (Invitrogen).

RT-qPCR

RNA was purified using PureLink RNA Mini Kit per manufacturer’s instructions. The samples were DNAseI (NEB) treated for 15min/37°C before stopping the reaction by adding 3.5mM EDTA and heating for 10min/75°C. cDNA was synthesized from 500ng total RNA using random hexamers and Maxima H minus reverse transcriptase (Thermo). No reverse-transcription controls were also included and used to confirm the lack of genomic DNA-driven amplification. qPCR reactions used Taq polymerase (NEB) and EvaGreen (Biotium) and were run on Biorad CFX Opus 96 Real-Time PCR System. Melt curves were included for each sample to confirm uniform amplicon identity between samples. Gene-specific amplification was quantified by comparison to a standard curve generated from 3-fold serial dilutions of a control sample, then normalized to 16S rRNA within each sample.

Data availability

Numerical source data for all figures has been deposited at Dryad https://doi.org/10.5061/dryad.4xgxd25pv

Acknowledgements

We would like to thank Nick Campbell-Kruger for his technical insights on identifying Mabs/Mtb orthologs, Jonathan Van Dyke for his assistance with flow cytometry analysis, Emily Kumimoto and Siranoosh Ashtari for their assistance with Tn-Seq library preparation and sequencing, and Jessie Li and Bradley Jenner for their assistance with Tn-seq bioinformatics. We would also like to thank Caroline Dominic at the Sacramento County Department of Public Health for providing clinical isolates of Mabs for analysis.

Additional information

Funding

Pew Biomedical Fellowship BHP

NIH RO1 1R01AI144149 BHP

NIH R01 1R01AI143722 SAS

NIH Shared Instrumentation Grant 1S10OD010786-01 DNA Technologies and Expression

Analysis Core, UC Davis Genome Center

NCI Cancer Center Support Grant P30CA093373 Flow Cytometry Shared Resource, UC Davis

Additional files

Supplemental Table 1. Tn-Seq results.

Supplemental Table 2. DAVID functional pathway analysis.

Supplemental Table 3. Oligonucleotide sequences.

Supplemental Table 4. 16s sequencing data of clinical strains.