Bacterial strains, culture conditions, and chemicals

Mycobacterium smegmatis, including drug-sensitive and drug-resistant strains, along with their CRISPRi strains including ItreS^SM and IotsA^SM were cultured at 37 °C in Middlebrook 7H9 broth (m7H9) (Difco) or on Middlebrook 7H10 agar (m7H10) (Difco). The media were supplemented with 0.04% Tyloxapol (for planktonic growth in m7H9 only), 0.5 g L⁻¹ BSA (Fraction V), 0.2% glycerol, 0.2% dextrose, and 0.085% NaCl. Kanamycin (50 µg/mL) was used to select CRISPRi mutants. Mycobacterium tuberculosis strains including HN878, Erdman, H37Rv, and CDC1551, along their corresponding CRISPRi strains (ItreS^HN, ItreS^Erd, ItreS^CDC, and ItreS^Rv) were cultured in a biosafety level 3 (BSL-3) facility. The TB clinical isolates, including drug-sensitive (DS), rifampicin single-resistant (RSR), MDR, extensively drug-resistant (XDR), and totally drug-resistant (TDR) TB clinical isolates, were isolated from sputum samples of patients with active pulmonary TB at the National Masan Hospital (NMH), South Korea. All procedures adhered to relevant ethical regulations, and the isolates were collected as part of a prospective observational cohort study (ClinicalTrials.gov NCT00341601) conducted between 2005 and 2008. The study was approved by the ethics review boards of both NMH and the National Institute of Allergy and Infectious Diseases (NIAID), with all participants providing written informed consent.

When appropriate, cultures were supplemented with 10 mM sodium butyrate, 20 mM trehalose, 200 µM Validamycin A (ValA), 200 ng/mL anhydrotetracycline (ATc), or varying MIC of rifampicin (RIF), isoniazid (INH), bedaquiline (BDQ), or d-cycloserine (DCS). These compounds were purchased from Sigma and Advanced ChemBlocks Inc. Experiments involving CRISPRi mutants were conducted with or without ATc; untreated cultures served as controls.

Metabolite extraction and LC-MS analysis

M. smegmatis– or Mtb-laden filters were generated and incubated at 37 °C for 5 days to reach mid-log phase of growth²¹. To prepare for filter culture-based metabolomics, cultures on agar-supported filters were treated with trehalose and/or ValA. M. smegmatis or Mtb-laden filters were metabolically quenched by immersion in a precooled mixture of acetonitrile:methanol:H₂O (40:40:20, v:v:v) at −40 °C. Metabolites were extracted via mechanical lysis using 0.1-mm zirconia beads in a Precellys tissue homogenizer for 4 min at 6000 rpm, repeated twice under continuous cooling at or below 2 °C. The lysates were clarified by centrifugation and filtered through a 0.22-µm Spin-X column. The residual protein content was measured with a BCA protein assay kit (Thermo Scientific) to normalize metabolite levels to cell biomass.

Extracted metabolites were separated using a Cogent Diamond Hydride type C column (gradient 3) with mobile phase comprising solvent A (ddH₂O with 0.2% formic acid) and solvent B (acetonitrile with 0.2% formic acid). An Agilent 6230 TOF mass spectrometer (MS) was coupled to an Agilent 1290 Liquid Chromatography (LC) system. Dynamic mass axis calibration was maintained via continuous infusion of a reference mass solution through an isocratic pump with a 100:1 splitter. This setup achieved mass errors of ~5 ppm and a mass resolution between 10,000 to 25,000 over the m/z range of 62–966, with a dynamic range of 5 log₁₀. Ions were identified based on unique accurate mass and retention time identifiers corresponding to expected isotopomer distributions. Data processing was conducted using Agilent Qualitative Analysis B.07.00 and Profinder B.07.00 software (Agilent Technologies), with a mass tolerance of <0.005 Da. Clustered heatmaps, hierarchical clustering, principal component analysis (PCA), and pathway enrichment analysis were performed using MetaboAnalyst (ver. 6.0). All metabolomics data represent the average of at least two independent triplicates.

Isotope tracing analysis using ¹³C₁₂ trehalose

To investigate trehalose metabolism in DS- and DR- TB clinical isolates, we performed isotope tracing experiments using fully ¹³C labeled trehalose (¹³C₁₂ trehalose), purchased from Cambridge Isotope Laboratory (CIL). Nine TB clinical isolates were randomly selected, three DS- and six DR- strains, and cultured in m7H9 until mid-log phase. The cultures were then transferred to fresh m7H9 containing 20 mM trehalose, composed of 20% ¹³C₁₂ trehalose and 80% ¹²C unlabeled trehalose. After 1 day, the cultures were harvested, washed with PBS, and rapidly quenched by immersion into a precooled mixture of acetonitrile/methanol/H₂O (40:40:20) at −40 °C. The extent of isotopic labeling in key metabolites such as glucose 6-phosphate (glycolysis), sedoheptulose 7 phosphate (PPP), and malate (TCA cycle) was quantified by dividing the summed peak height of all labeled isotopologue species by the total peak heights of both labeled and unlabeled isotopologues, expressed as a percentage. To correct for naturally occurring ¹³C isotopologues (i.e., [M + 1] and [M + 2]), label-specific ion counts were adjusted accordingly.

qRT-PCR analysis

M. smegmatis or Mtb strains were cultured in m7H9 until reaching mid-log phase. Bacilli were harvested by adding an equal volume of guanidine thiocyanate buffer. Total RNA was extracted using TRIzol reagent and the PureLink RNA Mini Kit (Invitrogen), following the manufacturer’s instructions. Genomic DNA contamination was eliminated using the Turbo DNA-free kit (Invitrogen). cDNA was synthesized from 500 ng of RNA using the iScript cDNA Synthesis Kit (Bio-Rad). Quantitative PCR was performed on a C1000 Thermal Cycler (Bio-Rad). Primers and probes were designed using the PrimerQuest^TM Tool (Integrated DNA Technologies), with sequences provided in Supplementary data file 2. Gene expression levels were quantified by calculating the ΔΔCt values, normalized to the housekeeping gene sigA. Fold changes were expressed as log₂ values relative to control samples.

Lipid extraction and TLC analysis

TDM and TMM from mycobacterial bacilli were prepared as previously described¹³. Bacilli were transferred to a 15 mL amber glass bottle and incubated overnight with 3 mL of chloroform: methanol (2:1, v:v) to sterilize the bacteria and extract total lipids. Subsequently, 10 mL acetone was added, and the mixture was incubated for 24 h at − 80 °C. After centrifugation, the lipid-containing supernatant was decanted. The lipids were then resuspended in 1 mL of chloroform: methanol (2:1, v:v), and equal amounts from each condition were loaded onto thin-layer chromatography (TLC) plates. TMM and TDM were resolved using TLC in a solvent system of chloroform: methanol: H₂O (90:10:1, v:v:v). Lipids were visualized by spraying the plates with 1% molybdophosphoric acid in ethanol, followed by charring.

CRISPRi knockdown generation

Single guide RNA (sgRNA) sequences were designed to target the 3’-end of the non-template strand within the open reading frame of the target genes. Each sgRNA consists of ~20 nucleotides located upstream of an effective protospacer adjacent motif (PAM) sequence. The PLJR962 M. smegmatis CRISPRi backbone plasmid was amplified in E. coli, selected with kanamycin (50 µg/mL), and subsequently digested with BsmBI restriction enzymes (NEB). The digested plasmid was then purified. Designed oligonucleotide primers (see Supplementary Table 1) were annealed and ligated into the BsmBI-digested plasmid. Competent M. smegmatis cells were prepared by washing mid-log phase cultures multiple times with ice-cold 15% glycerol. The recombinant plasmid was introduced into these competent cells via electroporation using a Pulse Controller II and Gene Pulser II (BioRad). Transformed cultures were grown to mid-log phase and plated on m7H10 containing 50 µg/mL kanamycin to select for successful mutants. Selected colonies were regrown in m7H9 with kanamycin and incubated with ATc (200 ng/mL) for at least 1 day to induce target gene repression. Knockdown of the target genes was confirmed by qRT-PCR.

Luria-Delbrück fluctuation assay and analysis

The classical fluctuation assay was modified for this study⁴⁵. To generate single-cell suspensions of M. smegmatis wildtype and ItreS^SM, cultures were diluted to an OD₅₉₅ of 0.00005 in 200 µL within 96-well plates. The cultures were then incubated and allowed to grow until reaching an OD₅₉₅ of 0.7–1.0. From each of 60 randomly selected cultures, 100 µL was plated on m7H10 containing either 100 µg/mL RIF or 200 µg/mL INH. Spontaneous resistant colonies were counted after incubation for up to 10 days. To determine the total viable cell input, the remaining 100 µL of each culture was serially diluted and plated on antibiotic-free m7H10. Mutation rates were calculated using the Lea-Coulson method: m/Nt, where m is number of resistant colonies, and Nt is the total number of input cells⁹¹.

Co-culture competition assay

The plasmids pTE-OX-GCT5 and pGMEH-p38-mRFP (Addgene) were transformed into wildtype M. smegmatis and ItreS^SM, respectively, to generate wildtype::GFP and ItreS^SM::RFP strains. Using these strains, we evaluated their relative viability following cyclic exposure to bactericidal concentrations of RIF or D-cycloserine (DCS). Briefly, equal volumes of mid-log phase cultures of wildtype::GFP and ItreS^SM::RFP, each at an OD₅₉₅ of ~0.7, were mixed to create the initial untreated culture (G0). This mixture was then treated with 100 µg/mL RIF or DCS for 1 day, followed by washing with PBS and resuspension in antibiotic-free m7H9. The resulting culture, adjusted to an OD₅₉₅ of 0.05, was allowed to grow until reaching an OD₅₉₅ of ~0.7, referred to as G1 subculture. This cycle of treatment, washing, and regrowth was repeated sequentially until G5. Flow cytometry was employed to determine the relative abundance of wildtype::GFP and ItreS^SM::RFP during each generation (G0 to G5), allowing to track changes in strain viability over successive cycles of antibiotic exposure.

Spot assay

Cultures of Flux^RIF and Flux^INH at mid-log phase were diluted to an OD₅₉₅ of 0.1. Five 10-fold serial dilutions were then prepared from each culture, and 2 µL of each dilution were spotted onto m7H10 containing various concentrations of antibiotics: 8–100 µg/mL RIF, 4–32 µg/mL INH, or 0.0075–0.03 µg/mL BDQ. The plates were incubated at 37 °C until colonies became visible.

Whole genome sequencing of RIF resistant M. smegmatis

Genomic DNA was extracted from Flux^RIF #1-#10 bacilli and sequenced using the Illumina HiSeq platform (Novogen Corp). To ensure high-quality reads, raw sequences were trimmed with Trimmomatic (v0.39.2)⁹² using the parameters “SLIDINGWINDOW:4:20 MINLEN:50.” Variant analysis across the ten samples was performed using the Snippy pipeline (v4.6.0, default options; https://github.com/tseemann/snippy). The M. smegmatis reference genome (accession: NC_008596.1) was downloaded in GenBank format from NCBI and used for alignment. The pipeline involved: I. Mapping high-quality reads to the reference genome with BWA-MEM (v0.7.17-r1188), II. Calling variants with FreeBayes (v1.3.6)⁹³, III. Filtering variants with BCFtools (v1.18)⁹⁴, and IV. Annotating variant effects with SnpEff (v5.0e)⁹⁵. Given the stringent criteria of the Snippy pipeline in variant calling, we manually inspected the mapped bases at each position using SAMtools (v1.18) mpileup on the BAM files generated by BWA-MEM. The SnpEff results from all samples were organized by variant position and effect, then visualized using the ComplexUpset package (v1.3.3) in R. The reference bases and variant alleles were plotted with ggplot2 (v3.5.1), based on sequencing depth.

RRDR sequencing of lab-made RIF resistant M. smegmatis

The rpoB gene sequences of the Flux^RIF mutants were compared to the reference rpoB gene sequences for wildtype M. smegmatis⁹⁶. ApE plasmid-editing software was utilized to identify mutations. Flux^RIF strains were streaked onto antibiotic-free LB agar plates and incubated overnight at 37 °C to isolate single colonies for sequencing. The RRDR (RIF Resistance Determining Region) of the rpoB gene was amplified and sequenced using Sanger Sequencing (Quintara Biosciences) using the following primers: Msmeg-rpoB-fwd (5’-gctgatccagaaccagatcc-3’) and Msmeg-rpoB-rev (5’-gatgacaccggtcttgtcg-3’).

Membrane bioenergetics—membrane potential, NAD/NADH, and energy charge

For membrane potential (ΔΨm) measurement, cultures were grown in m7H9 to mid-log phase and concentrated to an OD₅₉₅ of ~1.0 in fresh m7H9. Cultures were stained with 15 µM DiOC₂ and incubated at 37 °C for 40 min. Following incubation, cultures were washed with PBS to remove excess dye. As a positive control for membrane depolarization, one culture was treated with 5 µM of the carbonyl-cyanide 3-chlorophenylhydrazone (CCCP; Invitrogen), while PBS served as the vehicle control. The assay was performed in black, clear-bottom 96-well plates (Costar). Fluorescence was measured using a SpectraMax M4 spectrofluorimeter (Molecular Devices), recording green fluorescence (excitation 488 nm, emission 530 nm) and shifts to red fluorescence (excitation 488 nm, emission 610 nm). ΔΨm was calculated as the ratio of red to green fluorescence, with each condition measured in triplicate.

M. smegmatis-laden filters generated for the metabolomics profiling were used to measure intrabacterial ATP and NADH/NAD levels. ATP concentrations were determined using the BacTiter Glo Microbial Cell Viability Assay kit (Promega), following the manufacturer’s instructions. NAD and NADH levels were measured with the FluroNAD/NADH detection kit (Cell Technology), also according to the provided instructions. Bacterial metabolism was rapidly quenched by immersing the filters into the respective assay buffer or reagent, ensuring immediate stabilization of metabolic states. ADP and AMP levels were quantified using LC-MS metabolomics, with calibration curves generated from chemical standards spiked into homologous mycobacterial extracts to correct for matrix-associated ion suppression effects.

RMR-tre labeling and flow cytometry

RMR-tre was synthesized as previously described and characterized by nuclear magnetic resonance (NMR) spectroscopy⁵³. Wildtype M. smegmatis, ItreS^SM, pTreS^SM, Flux^RIF, or ItreS^Flux cultures in mid-log phase were treated with 1X MIC of RIF for 1 day. Following treatment, cultures were stained with the fluorogenic dye RMR-tre at a final concentration of 10 µM and incubated at 37 °C for 1 h. After incubation, the cultures were analyzed using a flow cytometer (Attune NxT, Thermo Fisher Scientific). The reported values represent the gated cell fractions. Data were exported from the flow cytometer and analyzed using FlowJo software (BD Biosciences). Error bars indicate the standard deviation from biological replicates.

Mathematical modeling

The classical Luria-Delbrück fluctuation assay was adapted for this study (Fig. S6A). To generate single-lineage bacilli of M. smegmatis wildtype and ItreS^SM, cultures were diluted to an OD₅₉₅ of 0.00005 in 200 µL and dispensed into 96-well plates. Cultures were incubated until reaching an OD₅₉₅ of 0.7–1.0. From 60 randomly selected wells, 100 µL was plated on m7H10 containing 100 µg/mL RIF, and the number of spontaneous RIF-resistant colonies was counted after 10 days. To estimate the total viable input cell number, the remaining 100 µL of each culture was serially diluted and plated on antibiotic-free m7H10. To infer the reversible switching rates between drug-sensitive and drug-tolerant states (Fig. 5A), we utilized recent mathematical models that relate fluctuation assay data to these rates⁶⁰. The calculation is based on the fluctuation assay results, where single-cell clones are subjected to lethal stress, and the clone-to-clone variation in surviving bacilli is used to estimate switching rates. This method has successfully been applied to decipher reversible phenotypic transitions between drug-sensitive and drug-tolerant states in both bacterial pathogens and cancer cells^61,97,98. The basic mathematical model encompasses the following ingredients:

• Single cells exist in two phenotypic states: drug-sensitive and drug-tolerant, and reversibly switch between them with certain kinetic rates (Fig. 5A).

• These transitions occur spontaneously prior to drug exposure.

• Cells proliferate at a given rate, assumed equal in both states.

• During division, both daughter cells inherit the mother’s cell state just before division.

A detailed formulation for this stochastic population dynamics model, which describes clonal expansion with state switching, has been recently reported⁶⁰. Assuming colonies expand over \(T\) generations, the variability in the number of tolerant bacilli, measured by the squared coefficient of variation (standard deviation divided by the mean), \(C{V}_{{N}_{T}}^{2}\), is given by:

where \(T\) is the normalized expansion time (in bacterial doubling times), \({T}_{{on}}\) is the average duration (also normalized to bacterial doubling times) that a single bacillus spends in the drug-tolerant state, and \(f\) is the fraction of persisters^59,60. Here, \(C{V}_{{N}_{T}}^{2}\) is the model-predicted squared coefficient of variation of the drug-tolerant population across clones. For a fixed \(f\), smaller \({T}_{{on}}\) (faster switching between cell states) yields reduced inter-clonal fluctuations, while larger \({T}_{{on}}\) enhances these fluctuations^59,60.

During extended antibiotic exposure, each drug-tolerant bacillus irreversibly transitions to a drug-resistant state with a small probability \(p\ll 1\). Given the number of drug-tolerant bacilli \({N}_{T}\), the number of drug-resistant colonies \({N}_{R}\) follows a binomial distribution. The clone-to-clone variation in \({N}_{R}\) can be approximated as:

where \({\bar{N}}_{R}\) is the average number of drug-resistant colonies and \(C{V}_{{N}_{R}}\) denotes the coefficient of variation of \({N}_{R}\). Using equation (2), the coefficient of variation of \({N}_{R}\) allows estimation of \({{CV}}_{{N}_{T}}^{2}\):

Combining this with equation (1), we can estimate \({T}_{{on}}\). Applying this analysis to the fluctuation assay data for \({N}_{R}\) across 60 single-cell lineages, we obtained:

for the wildtype, where the \(\pm\) denoted the 95% confidence interval estimated by bootstrapping. Using this value in (1), with \(T=30\) (representing 30 bacterial divisions before plating on RIF-containing plates) and an average frequency of drug-tolerant persisters of \(f={10}^{-3}\), we estimate the transient heritability of the wildtype persister state as:

For ItreS^SM, the number of colonies surviving antibiotic treatment was roughly six-fold lower than in wildtype (Figs. 2A, 5B right). Assuming the same mutation probability \(p\) for both genotypes, this reduction likely results from around six-fold decrease in persister-formation rate in ItreS^SM following short-term treatment (Fig. 5B left), while \({T}_{{on}}\) remains unchanged. The fluctuation assay data for ItreS^SM yields:

and, with \(f=\frac{{10}^{-3}}{6}\), the estimated \({T}_{{on}}\) from (1) is:

This represents approximately a 20% decrease compared to wildtype, suggesting that ItreS^SM persisters are less stable and revert to a drug-sensitive state more rapidly. Overall, the primary reason for the six-fold reduction in drug-resistant colonies appears to be a corresponding decrease in persister formation rates in ItreS^SM relative to wildtype.

Antibiotic permeability (RIF uptake)

M. smegmatis wildtype, Flux^RIF, or Flux^INH in mid-log phase were incubated with 1X MIC of RIF at 37 °C. Bacteria were harvested at 0, 2, 4, 24, and 48 h and CFUs were determined by serial dilution plating on m7H10. The cell-free supernatant was collected by filtering through a 0.22 µm filter. RIF was extracted by adding a precooled solution of LC-MS grade acetonitrile: methanol: H₂O (40:40:20) −40 °C. RIF detection and quantification were performed by LC-MS, as previously described^21,23,99. The intrabacterial RIF concentration was calculated as [RIF]_{drug only} – [RIF]_filtrate, with three biological replicates per group.

EtBr permeability assay

M. smegmatis wildtype and drug-resistant strains in mid-log phase were cultured in m7H9 until reaching an OD₅₉₅ of 0.7. Cultures were centrifuged at 13,000 rpm for 3 min, the supernatant discarded, and the pellets washed with PBS. The OD₅₉₅ was adjusted to 0.4, and glucose was added to a final concentration of 0.4%. Ethidium bromide (EtBr) was added at 8 mg/mL, and 100 µL aliquots were transferred into each well of black, clear-bottom 96-well plates (Costar). Fluorescence was measured every 60 s for 60 min using a SpectraMax M5 spectrofluorometer (excitation 530, emission 595 nm). This assay assesses EtBr uptake as an indicator of cell membrane permeability.

SDS susceptibility test

The susceptibilities of wildtype and Flux^INH were evaluated following a method adapted from Banaei et al.¹⁰⁰. Briefly, cultures in mid-log phase were diluted with growth medium to an O.D₅₉₅ of 0.01 and incubated with 0.01% SDS in triplicate. At 0 and 8 h of incubation, CFUs were monitored by plating on m7H10 to assess viability.

MPN (most probable number) assay to detect total persister bacilli

The MPN assay was performed as previously reported⁷³. Briefly, HN878, H37Rv, Erdman, and CDC1551 strains in mid-log phase were grown in m7H9, harvested by centrifugation at 5000 rpm for 8 min, and washed twice with PBS. The cultures were then centrifuged at 800 rpm for 8 min to generate a single-cell suspension. The supernatant was transferred to a separate tube and diluted with PBS to achieve an OD₅₉₅ of 0.1. CFU assay was used for the inoculum quantification. A total of 20 mL of the single-cell suspension was prepared for each condition, transferred to a vented flask, and incubated at 37 °C for 2 weeks. Starved cultures were split into two 8 mL cultures in vented flasks and treated with either 100 µg/mL RIF with or without 200 µM Validamycin A (ValA). These cultures were incubated at 37 °C for 5 days. After incubation, cultures were harvested by centrifugation at 5000 rpm for 8 min, washed with PBS, and resuspended in 1 mL PBS. For the MPN assay, 15 µL of this suspension was diluted into 135 µL of m7H9 in 96-well plates. Ten-fold serial dilutions were carried out across the wells, which were then incubated at 37 °C. OD₅₉₅ was measured after 3 and 5 weeks to determine bacterial growth. The remaining culture from the assay was also used for CFU enumeration, with dilutions ranging from 10^-5 and 10^-1. ItreS^HN and ItreS^CDC, treated with or without ATc, were included in the assay.

Statistical analysis

All data were analyzed using Prism (v10.0; GraphPad Software). Statistical significance was determined using Student’s unpaired t-test with Welch’s correction or one-way ANOVA with Bonferroni post-test correction. P values less than 0.05 were considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.