Uncovering resistance pathways to first- and last-line antibiotics in Mycobacterium tuberculosis populations Authors: Prapaporn Srilohasin, Jasmine M. Williams, Aidan P. Tay, Brodie F. Gillieatt, Daniel R. Pascoe, Ram P. Maharjan, Angkana Chaipraset, Pravech Ajawatanawong, Amy K. Cain Published: 2026 ABSTRACT AbstractTuberculosis (TB), caused by Mycobacterium tuberculosis, is one of the world’s deadliest diseases, currently responsible for ~1.5 million deaths per year and rising. Recently, rifampicin-resistant M. tuberculosis was designated as a critical priority pathogen status by the World Health Organization. However, few controlled laboratory studies are available that systematically assess the molecular drivers of antibiotic resistance development in TB. In this study, we paired laboratory-directed evolution and population-level deep-sequencing approaches to map the evolutionary pathways taken by M. tuberculosis to develop resistance to first- and last-line therapies (rifampicin and linezolid) and then characterized de novo resistance mutation occurrence over time. We demonstrated that the majority of M. tuberculosis populations readily acquire mutations in genes commonly found in rifampicin- and linezolid-resistant clinical isolates (rpoB and rplC). However, we also identified mutations in six genes, mostly present in subpopulations (17–41%) and not previously linked to rifampicin or linezolid resistance, including four associated with rifampicin resistance (Rv0052, ppsD, ppsE and mptC) and two associated with linezolid resistance (glpK and echA12). The ppsD, glpK and mptC mutations were also identified in published individual sequencing reads of antibiotic-resistant clinical isolates. Further investigation of the identified resistance determinants ppsD/E established that mutations in these genes appear to mediate resistance across multiple species, with an Escherichia coli mutant of the ortholog (fabF), representing a shared domain featured in PpsD and PpsE, phenotypically displaying increased antibiotic tolerance to low-level rifampicin. This study highlights the power of using controlled laboratory studies to uncover minority variants in populations of M. tuberculosis. These outcomes will lead to improved diagnosis of antibiotic resistance emergence in TB, to optimize management and treatment of TB infections, and ultimately to minimize patient deaths. Impact Statement With the emergence of antibiotic resistance in the world’s deadliest disease, tuberculosis, studies investigating the mechanisms of resistance in Mycobacterium tuberculosis are essential to minimize patient deaths. However, due to the difficulty of lab-based studies on M. tuberculosis, most studies to date focus on analysing resistant strains arising in the clinic and identifying their primary mutations. To date, no controlled laboratory evolution utilizing multi-step increasing exposure to antibiotics, combined with population-level sequencing, has been used to uncover progressive acquisition of resistance determinants in whole populations. In this study, we applied a novel directed evolution approach that induces resistance to the first- and last-line antibiotics, rifampicin and linezolid, in populations of M. tuberculosis and used whole-genome sequencing to monitor development of resistance mutations across the population over time. We uncover novel mutations not yet linked to antimicrobial resistance, often found at low frequency in the resistant population and would otherwise go undetected in traditional resistance studies of M. tuberculosis. Additionally, we demonstrate that identified gene(s) (ppsD/E) confer low-level rifampicin resistance in an additional species – highlighting the importance of this method to studying complex diseases like TB, but also broader implications for other critical antibiotic-resistant bacterial species. Data Summary Population-level sequencing data of the TB-resistant passages (Figs 1 and 2) can be found in ENA under the study accession number: PRJEB82621. Individual accessions are as follows: ERR13962878 ERR13962879, ERR13962880, ERR13962881, ERR13962882, ERR13962883, ERR13962884, ERR13962885, ERR13962886, ERR13962887 and ERR13962888. Data, scripts and bioinformatics pipeline used in the analysis of genomic variants and resistance phenotypes can be accessed via GitHub at https://github.com/jazmia/Variant_Analysis. The authors confirm all supporting data, code and protocols have been provided within the article or through supplementary data files. Introduction The emergence of antibiotic resistance in Mycobacterium tuberculosis, the bacterium that causes tuberculosis (TB), poses one of the greatest public health threats in history. The current TB pandemic, where active and latent infections of M. tuberculosis are estimated to occur in a quarter of the world’s population and cause ~1.5 million deaths each year, represents the world’s leading cause of death from a single infectious agent [12]. Further, this deadly disease is becoming increasingly difficult to treat owing to the global rise of antibiotic resistance [13]. Standard treatment of TB involves a 6-month course of a combination of three to four anti-TB drugs, yet even this involvement regimen only yields an 85% cure rate [2]. Lengthy treatments are required due to the ability of TB to enter a dormant (non-replicating) state where, alongside other innate resistance mechanisms, M. tuberculosis is able to resist antibiotics [4]. However, prolonged treatments provide continuous drug exposure and increase the risk of patient non-adherence, thereby creating a perfect storm of selective pressures for antibiotic-resistant M. tuberculosis to emerge [5]. M. tuberculosis can develop resistance to a wide spectrum of antibiotics, ranging from single resistance to first-line drugs, to extensively drug-resistant against multiple and last-line therapies [2]. Unlike other critical priority pathogens like the Enterobacteriaceae, which horizontally exchange resistance determinants, resistance in M. tuberculosis is primarily mediated via single-point mutations or homologous recombination under antibiotic selective pressure [56]. Due to the de novo nature of resistance mutation development in M. tuberculosis, populations can be heterogeneous, comprised of sensitive and resistant subpopulations that are otherwise isogenic [7]. Multiple mutations may be observed in a single population as resistance mutations can come at a fitness cost, often requiring compensatory mutations to maintain fitness, while additional mutations may boost resistance levels [8]. Thus, understanding this complex interplay among selective pressures, population resistance acquisition and fitness is crucial for addressing M. tuberculosis resistance and may uncover novel or low-level compensatory mutations supporting overall resistance [12]. Rifampicin-resistant M. tuberculosis has recently been classified by the World Health Organization (WHO) as one of only four ‘critical’ priority pathogens in urgent need of new antimicrobials [12]. Rifampicin (RIF) is the most potent and commonly used first-line drug to treat TB [29]. It kills M. tuberculosis by binding to the β-subunit of the RNA polymerase, inhibiting transcription [910]. However, in 2023, ~400,000 TB cases were multidrug-resistant or rifampicin-resistant, rendering treatment ineffective [2]. The vast majority of rifampicin resistance is conferred by SNPs in rpoB (~95%), altering the target β-subunit of the RNA polymerase to prevent binding [611 12]. Mutations in rpoA and rpoC, which encode the RNA polymerase α- and β’-subunits, respectively, are observed to co-occur in approximately one-third of rpoB mutants as compensatory mutations [1213]. Linezolid (LZD) is an oxazolidinone antibiotic that has been repurposed for the treatment of TB and is currently recommended by WHO as a last-line treatment to treat multidrug-resistant/rifampicin-resistant infections [214]. LZD kills M. tuberculosis by binding to the 23S rRNA of the 50S ribosomal subunit, thereby inhibiting translation [14]. The primary genes associated with linezolid resistance are rplC and rrl, which encode the ribosomal protein L3 and the 23S rRNA, respectively, and SNPs in these genes prevent antibiotic binding to these targets [614]. rplC and rrl are estimated to represent 36.8 and 10.5% of linezolid resistance cases, respectively. Therefore, there is a need to expand the extremely limited catalogue of linezolid resistance mutations [1215 16]. Owing to the global importance of ensuring viable TB treatment options for this prevalent and severe infectious disease, we assessed the spectrum of mutations that arise at a population level, as M. tuberculosis evolves resistance to first- and last-line anti-TB drugs. Most studies to date focus on analysing resistant strains arising in the clinic and identifying their primary mutations (e.g. rpoB, rplC), although a handful of studies have used a comprehensive metadata approach or generated resistant isolates via single-step exposure to uncover novel mutations [1718]. To date, no controlled laboratory evolution utilizing multi-step progressive exposure to increasing concentrations of antibiotics, combined with population-level sequencing approaches, has been used to uncover low-level resistance variants. To address this, we performed the first controlled laboratory evolution of M. tuberculosis communities under stepwise exposure to rifampicin and linezolid, coupled with population-level whole-genome sequencing (WGS), adapting previously developed methodologies to M. tuberculosis [19]. From this, we identified the drivers of resistance development in M. tuberculosis in a controlled, stepwise manner, assessing genetic variability within populations over time. The outcomes of this study will inform future treatment strategies to minimize the development of resistance and ensure successful treatment, and guide development of much-needed new TB drugs and combinations. Methods M. tuberculosis strain and minimum inhibitory concentration M. tuberculosis H37Rv (ATCC 27294) was used for this study. The minimum inhibitory concentration (MIC) for rifampicin (Sigma-Aldrich) and linezolid (MedChemExpress) was determined across a range of 0.039–5 µg ml−1, in line with the standardized EUCAST methodology used in clinical settings [20]. Antibiotic dilutions were prepared in Middlebrook 7H9 (M7H9, BD Difco) supplemented with 10% oleic acid-albumin-dextrose-catalase (BD Difco) and 0.2% glycerol. Briefly, 100 µl of antibiotic dilution and 100 µl of inoculum were prepared in a U-shaped 96-well polystyrene microtitre plate (Corning) with a final inoculum of 105 c.f.u. ml−1. Plates were incubated at 36±1 °C for 7 days. MICs were determined using the resazurin microtitre assay [21], which relies on an oxidation-reduction indicator. After the incubation period, 30 µl of resazurin solution was added to each well and incubated overnight at 37 °C. Colour change was then observed, with a shift from blue to pink indicating bacterial growth due to the reduction of resazurin. MICs were defined as the lowest concentration of the drug that prevented this colour shift, indicating inhibition of bacterial growth. Directed evolution of antibiotic-resistant M. tuberculosis mutants Our previously described directed evolution methodologies [19] were adapted for use on M. tuberculosis H37Rv ATCC 27294. M. tuberculosis H37Rv mutants were selected through culturing 10 µl of a 1 ml M7H9 liquid culture (106 cells) on antibiotic-supplemented M7H9 agar plates, in triplicate. Plates were supplemented with stepwise increases in antibiotic concentrations, starting at 0.25× their MICs. When full population growth was achieved (~2 months per passage), the populations were re-passaged, with mutants collected from the highest concentration plate, resuspended in 1 ml of M7H9 broth, and used to inoculate new plates with a stepwise antibiotic gradient – thereby ‘training’ the population to increase resistance. M. tuberculosis was cultured in non-selective media as a growth control for directed evolution experiments. For both antibiotics, concentrations started at 0.25× MIC with two-fold increases, reaching 64× MIC for rifampicin and 4× MIC for linezolid (with one 1.5-fold increase). Additionally, mutants from each passage were isolated, purified and stored in 15% glycerol at −80 °C for further investigation. Although we report resistance phenotypes on solid media relative to MIC determined in liquid media to align with clinical MICs determined by standardized EUCAST methodologies, we also acknowledge that there can be variations between liquid and agar MIC values [2022]. DNA extraction, library preparation and WGS Isolates were sub-cultured on Löwenstein–Jensen medium and incubated at 37 °C for 4 weeks. Several colonies of M. tuberculosis were collected for DNA extraction following the cetyltrimethylammonium bromide protocol and resuspended in nuclease-free water [23]. DNA quality was assessed by Nanodrop and gel electrophoresis, and quantity was measured by Nanodrop and Qubit Fluorometer (Table S1, available in the online Supplementary Material). Whole-genome paired-end Illumina sequencing was performed by the Australian Institute for Microbiology and Infection, University of Technology Sydney. DNA was prepped using the Nextera Flex kit before sequencing on a NovaSeq 2×150 PE sequencer, generating >2 million reads to >40× coverage (Table S1). Individual sequencing reads were deposited in the European Nucleotide Archive under accession number PRJEB82621 (https://www.ebi.ac.uk/ena/browser/view/PRJEB82621). Identification of genomic variants and their function Clinical pipelines, including the WHO TB resistance database [12], comprehensive antibiotic resistance database (CARD) TB mutations [24], Mykrobe [25] v0.10.0, TBProfiler [2627] v6.2.2, TBPortals Genomic Analysis Pipeline [28], DrPRG [29] v0.1.1 and tbtAMR [30] v1.0.3, were used as an initial assessment of known resistance alleles. However, for the detection of novel resistance alleles, genomic variants were detected as previously described [19] by comparing aligned sequence reads of resistant isolates to the parental strain. Sequencing data were assessed for quality using FastQC [31] v0.12.1 and trimmed accordingly using Trimmomatic [32] v0.39 (Table S1). Reads for the parental strain were aligned to the M. tuberculosis H37Rv reference genome (GenBank: CP110619), while reads for each mutant strain were aligned to the consensus genome sequence of the parental strain. Reads were aligned using BWA [33] v0.7.17 (r1188) and consensus genome sequences for each strain were produced using SAMtools [34] v1.19.2. BCFtools [34] v1.19 was then used to identify single-nucleotide variants and short insertions/deletions (indels) in each mutant strain from their mapped reads. Mutations were also identified with Snippy [35] v4.6.0 pipeline using the trimmed reads and the H37Rv reference genome. Additionally, all mutations identified were assessed for accuracy by visualizing BAM files in IGV [36] v2.16.0 for rifampicin- and linezolid-resistant populations (Figs S1 and S2, respectively). ORFs in consensus genomes for the parental and mutant strains were annotated using Prokka [37] v1.14.6, with mutations that induced a change in the amino acid sequence considered mutations of interest in this study. Proteins of interest were confirmed using blastp v2.16.0 against the non-redundant protein sequence database, with hypothetical proteins further analysed for potential function using BioCyc [38] v4 and InterPro [39]. The existence of specific mutations in previously published M. tuberculosis sequences was searched against the blastp non-redundant protein sequence database, blastn v2.17.0 core nucleotide database and unassembled FASTA reads from published clinical M. tuberculosis isolates deposited on the Sequence Read Archive. Protein domains were predicted by InterPro and the relevant domain containing the mutation was aligned with equivalent domains from published whole-genome sequences using muscle [40], and a phylogenetic maximum likelihood tree was constructed by mega11 [41] using complete sequences with 100 bootstrap replications. Growth curves and membrane permeability assays M. tuberculosis genes associated with rifampicin and linezolid resistance were assessed for resistance phenotypes using single-gene knockout mutants of Escherichia coli. Methods are described in further detail in the Supplementary Methods. Briefly, E. coli K12 BW25113 ΔfabF (Table S2), an identified ortholog to M. tuberculosis ppsD/E, was assessed for resistance phenotypes using the broth microdilution method [42]. The MIC was determined across a range of 2.5–60 µg ml−1 of rifampicin, while growth was assessed over a finer range of 2–20 µg ml−1 of rifampicin. The possible mechanism underlying the tolerance phenotype of ΔfabF was further investigated by assessing relative membrane permeability through flow cytometric assessment of fluorescence, corresponding to SYTOX Green (Thermofisher, S7020) uptake, using a BD FACSymphony™ A1 Cell Analyzer, conducted by the Australian Genome Foundry (Macquarie University, Sydney, Australia). In total, ~50,000 single-cell events per replicate were recorded from triplicate cultures of E. coli K12 WT and ΔfabF with treated samples cultured in 8 µg ml−1 rifampicin for 18 hrs, relative to a dead cell control (Fig. S3). Results and discussion Directed evolution of rifampicin- and linezolid-resistant M. tuberculosis populations To assess the acquisition of rifampicin and linezolid resistances in M. tuberculosis, we conducted directed evolution experiments over six successive passages (each lasting ~8 weeks) using the antibiotic-sensitive strain H37Rv (the parental strain) with no prior exposure to either drug (Fig. 1). In the first round of selection on rifampicin (passage 1), the strain H37Rv was able to form colonies on plates containing up to 2× the MIC (Fig. 1b). After four additional rounds of successive re-passaging with increasing rifampicin concentrations (over the time period of ~1 year), M. tuberculosis was able to grow on plates containing up to 64× the MIC (Fig. 1b). This pattern of growth suggests that M. tuberculosis populations can readily evolve rifampicin resistance in the laboratory using this directed evolution methodology [19]. Directed evolution of antibiotic resistance in M. tuberculosis H37Rv. (a) Workflow of directed evolution to evolve rifampicin- or linezolid-resistant M. tuberculosis populations. Evolved resistant populations for (b) rifampicin, with an estimated 64-fold MIC increase by the 5th round of selection, and (c) linezolid, with an estimated four-fold MIC increase by the 6th round of selection. * Indicates MIC relative to those determined in liquid microdilution in accordance with EUCAST methodology, determining MIC as 0.16 µg ml−1 for rifampicin and 2.5 µg ml−1 for linezolid. † No antibiotic (No Ab). Darker colour fill indicates highest concentration of antibiotic supporting growth for each passage. ‘Y’ indicates growth, ‘N’ indicates no growth and ‘NA’ indicates not analysed. Bold face type ‘Y’ indicates samples used in subsequent experimental steps, where colonies from ‘Y*’ were sequentially transferred to plates with higher antibiotic concentrations in subsequent selection rounds, and colonies from ‘Y#’ were used for WGS. In contrast, during the first round of selection on linezolid, the parental strain showed population growth only on plates containing sub-inhibitory concentrations (up to 0.25× the MIC). After five additional rounds of selection, colonies were able to grow on plates with concentrations up to 4× the MIC (Fig. 1c), indicating that M. tuberculosis had evolved to increase linezolid resistance, albeit at a slower rate than rifampicin resistance. These results demonstrate that serial passaging under stepwise increasing antibiotic pressure can select for populations with enhanced rifampicin and linezolid resistance, providing a controlled framework from which we can investigate the genetic basis and evolutionary dynamics of antibiotic resistance in M. tuberculosis. Identification of mutations in rifampicin- and linezolid-resistant M. tuberculosis populations To identify the genetic determinants conferring the observed stepwise increases in resistance, WGS was performed on the parental strain H37Rv, and five rifampicin-resistant and five linezolid-resistant evolved populations (indicated in Fig. 1, represented by #), with the total number of SNPs identified outlined in Table S3. As expected, our SNP analysis identified mutations with 98–100% allele frequency (AF) in genes commonly implicated in TB clinical resistance, rpoB (RIF) and rplC (LZD) (Table 1 and Fig. 2a), which are known to prevent rifampicin and linezolid from binding to their respective targets [612]. For rifampicin, the mutation in rpoB occurred inside the rifampicin resistance-determining region and was observed to arise at 4× MIC (Fig. 2a and c) [12]. For linezolid, mutation in rplC was observed in all linezolid-resistant mutants, even those at sub-inhibitory concentrations at full penetrance (100% AF) (Fig. 2b and d). This suggests that linezolid resistance can arise at lower concentrations relative to the MIC, highlighting the potential for rapid and stable selection under subtherapeutic exposure. These resistance mutations could also be detected using all standard clinical variant pipelines for resistance mutation identification in M. tuberculosis (CARD TB Mutations [24], TBProfiler [2627], Mykrobe [25], TBPortals [28], DrPRG [29] and tbtAMR [30], using default parameters; Table S4), except rplC for linezolid resistance, which was not detected via Mykrobe or CARD. Table 1.Genomic changes in annotated genes detected for evolved rifampicin and linezolid populations compared to parental M. tuberculosis H37Rv. Outline of function and previous association with antibiotic resistance (AbR)AntibioticGene(locus ID)FunctionMIC*Mutation(position in gene)Amino acid change†(position in protein)Previous links with AbR‡Reference RifampicinrpoB(Rv0667)Transcription4–64×C → T(1,333)H445YConfers RIF resistance[6]Rv0052(Rv0052)Transcriptional regulatory protein (putative)0–2×GCC → GCCC(186)M63Dfs151XNone ppsD(Rv2934)Lipid biosynthesis0–2×GCCCCCCC → GCCCCCC (2,655)H888Tfs941XppsA-E confer low level and boost resistance to pyrazinamide, upregulated in RIF-resistant rpoB mutants[4344]ppsE(Rv2935)0–2×CGGGGG → CGGGGGG (1,281)T430Yfs141XmptC(Rv2181)Glycolipid synthesis (putative)4–64×T → C(962)V321AUpregulated in response to isoniazid and rifampicin[4950] LinezolidrplC(Rv0701)Translation0.25–4×T → C(460)C154RConfers LZD resistance[6]glpK(Rv3696c)Glycerol metabolism0.5–2×GGGGGGGT → GGGGGGGGT (573)V192Cfs61XSupports broad antibiotic resistance, unable to confer resistance alone[53]echA12(Rv1472)Cholesterol metabolism (putative)4×G → A(715)G239RFlorfenicol resistance[5459]*Indicates MIC relative to those determined in liquid microdilution in accordance with EUCAST methodology, determining MIC as 0.16 µg ml−1 for rifampicin and 2.5 µg ml−1 for linezolid, where mutation was detected by variant calling pipelines, excluding detection only by visual assessment.†Amino acid change where ‘X’ indicates a stop codon and ‘fs’ indicates a frameshift mutation.‡Antibiotic resistance (AbR). Genomic changes in annotated genes detected for evolved rifampicin and linezolid populations compared to parental M. tuberculosis H37Rv. Outline of function and previous association with antibiotic resistance (AbR) *Indicates MIC relative to those determined in liquid microdilution in accordance with EUCAST methodology, determining MIC as 0.16 µg ml−1 for rifampicin and 2.5 µg ml−1 for linezolid, where mutation was detected by variant calling pipelines, excluding detection only by visual assessment. †Amino acid change where ‘X’ indicates a stop codon and ‘fs’ indicates a frameshift mutation. ‡Antibiotic resistance (AbR). Variant AFs of identified genes within M. tuberculosis populations, giving stepwise resistance phenotypes of increasing MIC. Visualized acquisition of variants (%) to (a) rifampicin and (b) linezolid. Exact number of reads with the variant present out of the total reads, with percentage below (%), for (c) rifampicin- and (d) linezolid-resistant populations as identified by the variant calling pipeline used in this study (see methods). * MIC represents the relative MIC to those determined in liquid media. Numbers in brackets represent the concentration (μg ml−1). † Mutations in superscript represent changes in the amino acid sequence. ‘X’ indicates a stop codon and ‘fs’ indicates a frameshift mutation. ‡ No antibiotic (No Ab). § Indicates presence determined by manual assessment of read outputs, but not detected by variant calling pipelines. However, in the passages where these core resistance mutations in rpoB (passage 3) and rplC (passage 1) had been acquired, the respective M. tuberculosis populations were only able to grow on plates up to 4× MIC for rifampicin and 0.25× MIC for linezolid, indicating that additional mutations are necessary to gain the further resistance seen in following passages (Figs 1b, c and 2). Our SNP analysis also identified eight mutations in the resistant populations compared to the parental H37Rv strain that were outside of the catalogue of known resistance mutations: five in rifampicin-resistant populations and three in linezolid-resistant populations (Table 1). Additionally, of the clinical TB resistance variant identification pipelines also used, only tbtAMR and TBProfiler identified low-frequency variants, including variants not detected by the focal pipeline used in this study (Tables S4 and S5). In the rifampicin selection regime, we also detected subpopulations carrying mutations within the genes Rv0052 (21% AF), predicted to encode a transcriptional regulatory protein, and ppsD (41% AF) and ppsE (17% AF), encoding related polyketide synthases (Fig. 2a and c) [643 44]. Interestingly, these mutations were already present in the parental strain population (passage 1), albeit at low frequency, indicating standing genetic variation prior to antibiotic exposure (Fig. 2c). Their AFs fluctuate during rifampicin exposure, increasing initially but declining following the emergence of the core rpoB mutation. As the rpoB mutation appears to arise in a different genetic background, the observed frequency shifts likely reflect changes in the relative abundance of distinct subpopulations rather than reversion within a single lineage. This pattern is consistent with heteroresistance [45] and clonal interference and suggests that these variants may contribute to transient stress adaptation or antibiotic tolerance. These genes have not previously been identified as rifampicin resistance mutations in the majority of the TB resistance databases, including the WHO TB resistance database [12], TBProfiler, Mykrobe, TBPortals or DrPRG databases. However, CARD indicated that alternative mutations to ppsD can confer high-level resistance to pyrazinamide, another TB treatment, with ppsD/E mutants also been shown to confer low-level resistance to pyrazinamide (Table 1) [44]. In addition to the alleles described above, we detected an mptC allele carrying a substitution mutation at high frequency (96% AF) at passage 3 (4× MIC), coinciding with the initial sweep of the population by the rpoB H445Y allele (Fig. 2a and c). Interestingly, the mptC AF subsequently decreased to ~50% and remained stable at this level under higher rifampicin concentrations, whereas the rpoB H445Y allele reached fixation (100%). The co-occurrence of these mutations is noteworthy and may suggest a functional interaction, potentially indicative of a compensatory or adaptive role. However, the rpoB H445Y mutation alone has been reported to confer resistance up to 100 µg ml−1 of rifampicin [46]. Additionally, the equivalent rpoB mutation in E. coli (H526Y) has been shown to confer increased fitness under environmental stress compared to the wild-type and other rifampicin resistance mutations, suggesting relatively low fitness costs and therefore a limited requirement for compensatory adaptation [1747 48]. The mptC gene is predicted to encode an alpha-(1-2)-phosphatidylinositol mannoside mannosyltransferase and has not been previously linked to rifampicin resistance (Table 1), although the functionally related mptA gene [49] has been associated with isoniazid resistance [50]. While additional low-frequency mutations were detected by tbtAMR and TBProfiler in ettA (2× MIC) and embB (L172R at 2× MIC and L172R and S174R at 64×, Tables S4 and S5), the coverage at ettA was unknown, while the coverage at embB was much lower than that for the rest of the genome (35× compared to >100× coverage), and thus these mutations were not considered for further investigation. Further increases in linezolid resistance were associated with additional mutations that co-occurred with the rplC mutation. Mutations in glpK, encoding a glycerol kinase, were observed between 0.5× and 2× the MIC (passages 1–4), initially arising with the acquisition of the core rplC mutation (100% AF) and progressively increasing in prevalence (19–67% AF) and also detected by TBProfiler (Fig. 2b and d). A secondary glpK mutation (W322R, Tables S4 and S5) was also detected by both tbtAMR and TBProfiler, however, with a frequency <15% and the much higher prevalence of the other glpK mutation, it was not considered for further investigation. Although mutations in glpK have previously been observed in rifampicin-, isoniazid-, ethambutol-, streptomycin-, pyrazinamide- [5152], levofloxacin- [12] and moxifloxacin-resistant isolates [53], they are considered not to be associated with resistance and have not been observed in linezolid-resistant isolates in the WHO TB resistance database [12]. However, glpK knockout mutants have also been demonstrated to show drug tolerance to rifampicin, isoniazid and moxifloxacin [53]. At the highest level of linezolid resistance (4× MIC, passage 6), glpK mutation prevalence in the population drastically decreased to 2% AF and was replaced by a mutation in the completely novel resistance gene echA12 (94% AF), encoding a probable enoyl-CoA hydratase (Fig. 2b and d) [54]. Both mutations appear to have an additive effect to complement the core rplC mutation, with echA12 taking over the role of glpK at the highest level of linezolid resistance assessed. Our variant calling analysis has revealed several potential new mutations and genes associated with rifampicin and linezolid resistance in M. tuberculosis, many of which co-occur with common resistance in M. tuberculosis populations, namely rpoB and rplC. Although tbtAMR and TBProfiler appear to be the only clinical variant calling pipelines that demonstrated the capacity to identify minority variants in M. tuberculosis populations, demonstrating the capacity [ettA and embB (RIF) and glpK (LZD) mutations]. As demonstrated by the inability of M. tuberculosis populations with core resistance mutations being unable to gain further resistance without the acquisition of additional mutations, most of which did not occur at full penetrance, the detection of minority variants is an important aspect of understanding resistance in M. tuberculosis. These findings highlight the limitations of standard TB resistance variant pipelines and even widely used generalized variant calling pipelines, including Snippy, for detection of novel, intrinsic or low-level mutations [29]. While only the general variant calling pipelines used in this study were able to identify novel mutations (Table S4), alterations to more user-friendly, existing pipelines with capacity for low-frequency variant identification, such as tbtAMR and TBProfiler, to output high-confidence resistance and resistance-associated mutations may inform the clinical management of resistant TB cases. Accruing more data on variants associated with resistance can compile those associated with resistance and also provide a priority list of those that should be followed up with functional validation to become high-confidence. This will support continuously tracking novel resistance mechanisms in M. tuberculosis, informing clinical management of resistant TB cases. To further strengthen that the variants identified in this study are due to antibiotic selection, sequencing could be conducted on an untreated passaging control to eliminate variants that can occur during the multiple passaging due to time, ageing, prolonged in vitro culture or genetic drift. Although this was not done due to initial experimental design, five out of the six novel mutations identified have previous links to antibiotic resistance (Table 1). Additionally, sequencing additional colonies at each passage may have improved the representation of variants identified through attempting to capture more variants or differing frequencies. However, the additional samples require highly restricted resources and a significant time investment (>1 year), often not generating enough resistant M. tuberculosis colonies at each passage to sequence and use for further passages, so it was not feasible for the current investigation. Novel rifampicin- and linezolid-resistant mutations While core resistance mutations, rpoB (RIF) and rplC (LZD), were identified in most samples (Fig. 2), they alone do not explain the continued stepwise increases in resistance over each passage. The community-level sequencing approach revealed six additional mutations, with five observed as minority subpopulations, that had not been previously linked to rifampicin or linezolid clinical resistance [12]. This allowed the identification of four novel rifampicin resistance mutations (Rv0052, ppsD, ppsE, mptC) and two mutations with a novel association to linezolid resistance (glpK and echA12). A similar stepwise fixation of spontaneous mutations in distinct genes occurs for resistance to bedaquiline in M. tuberculosis, both in vitro [8] and within TB patients [55]. Notably, pre-existing resistance to bedaquiline has been associated with TB treatment failure [55]. Substantial increases in the MIC of bedaquiline are thought to arise initially from mutations in the efflux pump regulator gene Rv0678, followed by subsequent mutations in atpE, the bedaquiline target gene, during sequential passages under bedaquiline selection pressure [8]. Among the mutations in the current study, only the glpK mutation is represented in the National Center for Biotechnology Information (NCBI) core nucleotide database (Table S6). However, further investigation of unassembled WGS FASTA reads from clinical, RIF-profiled, TB isolates available in the NCBI database revealed that the ppsD, glpK and mptC resistance mutations identified in this study could also be observed in some rifampicin-resistant clinical isolates (Table 2). These reads may represent mutant subpopulations but remain undetected in a clinical context using standard resistance detection analysis due to low penetrance. Furthermore, when comparing the variability within rifampicin resistance in a strain set of published, clinical M. tuberculosis, clustering of the consensus protein domains from strains with a resistance phenotype does not occur for any protein except RpoB, which clustered into two major clades (Fig. S4). This mutation distribution in consensus sequences, for the proteins other than RpoB, may suggest that these mutations have low penetrance and are consequently lost in the consensus sequence or are likely to occur sporadically and are not obviously attributed to a single event followed by clonal expansion or a combination of both hypotheses. There is an insufficient number of WGS data of linezolid-resistant isolates to perform a similar analysis due to a paucity in phenotypic screening for linezolid resistance [56]. Table 2.Unassembled FASTA reads of published clinical M. tuberculosis containing rifampicin- or linezolid-associated mutations detected in evolved population H37Rv. Proportion of sequencing reads (%) with mutations compared to total reads of each sequence. MDR=resistant to rifampicin and isoniazid. Pre-XDR=MDR plus resistance to a fluoroquinoloneRaw read accessionStrainGeneMutation(position in gene)Amino acid change*(position in protein)Read % with mutationResistance phenotypeSRR1082818925474 glpK GGGGGGGT → GGGGGGGGT (573)V192Cfs61X10.81pre-XDRSRR1082819329366 glpK GGGGGGGT → GGGGGGGGT (573)V192Cfs61X2.33pre-XDRSRR1082862137032 ppsD GCCCCCCC → GCCCCCC (2,655)H888Tfs941X0.52pre-XDRSRR1082864140543 rplC T → C (460)C154R0.41pre-XDR glpK GGGGGGGT → GGGGGGGGT (573)V192Cfs61X1.2SRR1166227840868 rpoB C → T (1,333)H445Y100pre-XDR ppsD GCCCCCCC → GCCCCCC (2,655)H888Tfs941X1.03 mptC T → C (962)V321A0.44SRR8186761TCDC3 ppsD GCCCCCCC → GCCCCCC (2,655)H888Tfs941X0.36MDR glpK GGGGGGGT → GGGGGGGGT (573)V192Cfs61X0.83*Amino acid change where ‘X’ indicates a stop codon and ‘fs’ indicates a frameshift mutation. Unassembled FASTA reads of published clinical M. tuberculosis containing rifampicin- or linezolid-associated mutations detected in evolved population H37Rv. Proportion of sequencing reads (%) with mutations compared to total reads of each sequence. MDR=resistant to rifampicin and isoniazid. Pre-XDR=MDR plus resistance to a fluoroquinolone *Amino acid change where ‘X’ indicates a stop codon and ‘fs’ indicates a frameshift mutation. While there is very limited characterization of Rv0052, the amino acid sequence of Rv0052 was matched with 100% identity and query coverage to a DJ-1/PfpI family protein (WP_003912418.1) and indicated to be involved with regulation of DNA-templated transcription in Uniport (I6Y6S3). This mutation is also present at a low frequency in the parental H37Rv strain but not present in any higher-level resistance mutants, suggesting that the mutation is only effective in lower-level concentrations and is selected against when higher levels of rifampicin are present. Further experimentation will be required to define the exact role of this mutation in rifampicin resistance. Both ppsD and ppsE encode subunits of a type I polyketide synthase complex involved in the biosynthesis of phthiocerol dimycocerosate, which are lipids located in the outer mycobacterial cell wall, implicated in virulence and required for in vivo growth [4344]. Mutants with low-level rifampicin resistance have low-frequency ppsD mutations, with a restorative mutation AF of 41%, while the parental strain sequence structure, which results in a truncated protein that annotates as two separate ppsD genes, has an AF of 59% in these low-level rifampicin-resistant mutants. The two separate PpsD proteins lack the polyketide dehydratase domain, suggesting a loss of function. Similarly, the ppsE mutation, also observed at this level of resistance, results in a truncated protein, with the loss of the acyl-synthase domain likely leading to loss of function. The restorative mutation of ppsD is consistent with the upregulation of polyketide synthase genes as a known mechanism to compensate for rpoB mutation [43]. Notably, frameshift mutations in ppsA-E have been linked to low-level pyrazinamide resistance and found to boost pyrazinamide resistance conferred by panD in M. tuberculosis [44]. However, like Rv0052, the ppsD mutation is present in the parental H37Rv, while higher resistance mutants do not feature this mutation. The mptC gene encodes an enzyme involved in the synthesis of mannose-capped lipoarabinomannan, a glycolipid found in the cell wall of M. tuberculosis, where it functions as a virulence factor, supports survival and helps evade host defence mechanisms [4950]. mptC is upregulated in rifampicin- and isoniazid-resistant M. tuberculosis, suggesting that mptC protects the mycobacterial cell [50]. It is uncertain what effect the conservative missense mutation has on MptC functionality, although the knockout of mptC is predicted to provide a growth advantage in 7H9 and 7H10 with 10% oleic acid-albumin-dextrose-catalase media [57]. This growth advantage, combined with the high AF relative to the other mutations associated with RIF resistance (Fig. 2a and c), could support a hypothesis that mptC mutations in particular may improve fitness, which is impaired by rpoB mutation [58]. However, further investigation is required to validate mptC in the context of rifampicin resistance. Additionally, low-frequency mutations found in linezolid-resistant mutants include glpK and echA12. glpK mutations were observed in linezolid-resistant strains between 0.5× and 2× MIC, increasing in frequency as antibiotic concentration increased (Fig. 2b and d). This exact frameshift mutation in glpK has previously been detected in clinical M. tuberculosis isolates (Table S6), including in isolates from in vivo murine infection models treated with moxifloxacin [53]. Consistent with our findings, these glpK mutants were phase variants, associated with an antibiotic-tolerant phenotype. GlpK is a glycerol kinase, so glpK frameshift mutants are likely unable to metabolize glycerol and therefore activate a stress response, supporting antibiotic resistance. This previous study also assessed single knockout mutants of M. tuberculosis H37Rv with a deletion of glpK and demonstrated increased percentage growth in low levels of rifampicin, but they were also found to have the same MIC as the wild-type, indicating that glpK mutations are unable to confer high levels of resistance without additional mutations [53]. The echA12 is predicted to encode an enoyl-CoA hydratase, by sequence similarity, with other M. tuberculosis enoyl-CoA hydratases involved in lipid membrane metabolism and cholesterol metabolism [5459 60]. While not previously observed in linezolid resistance, laboratory-evolved florfenicol-resistant Mycobacterium smegmatis mutants feature the same SNP in echA12 (G239R) [60]. The mutated echA12 is subsequently overexpressed, resulting in increased resistance to florfenicol [60]. Despite the low sequencing coverage, the high frequency of this echA12 mutation suggests it is a high-level resistance mutation. Additionally, mutants possessing this mutation have a large decrease in the prevalence of the glpK mutation seen in prior stages of resistance, indicating that at this stage of resistance, echA12 supersedes glpK in providing a fitness benefit. Out of the eight resistance mutations identified in this study, half were frameshift mutations (Rv0052, ppsD/E, glpK) occurring in mononucleotide repeats, a known mechanism of phase variation in M. tuberculosis [53]. Phase variation is an adaptive mechanism that allows genes to be rapidly switched ‘on’ and ‘off’ by genotypic changes [53]. This includes the insertions/deletions observed in this study in mid-level resistant populations, indicating that M. tuberculosis is heavily reliant on this adaptive mechanism to mediate the acquisition of higher levels of resistance. The minority variants identified in this study also appear to share commonalities within their respective antibiotic resistance groupings. With the exception of Rv0052, rifampicin mutations observed in ppsD/ppsE and mptC are all genes linked to mycobacterial cell wall components [61]. On the other hand, genes linked to linezolid, glpK and echA12, are both linked to alternative carbon source metabolism [5359]. These commonalities represent distinct antibiotic-specific alternative resistance strategies observed to be employed by heteroresistant M. tuberculosis populations in this study. While we have been unable to perform a direct functional validation of each novel mutation, stepwise tracking allows the connection of phenotypic resistance progression with genotypic evolution [62]. Therefore, the temporal association of these novel mutations with increasing MIC supports a likely role in resistance enhancement. Thus, providing a short list of key genes associated with resistance for further functional validation. Novel rifampicin tolerance conferred by ppsD/E ortholog fabF Having identified several new genes, whose mutations are potentially associated with rifampicin and linezolid resistance in M. tuberculosis, we next assessed whether disruption to these genes would allow cells to tolerate antibiotics. Owing to laboratory restrictions and the difficulty surrounding genetically modifying M. tuberculosis, we opted to conduct a preliminary test using mutants of gene orthologs in E. coli. E. coli was chosen as a model strain due to its ease of genetic manipulation and previous use in anti-mycobacterial drug discovery, due to the similar structure of its outer membrane to Mycobacterium, despite differences in membrane composition [1863]. Of the six M. tuberculosis genes we had identified as potentially associated with rifampicin or linezolid resistance (not in the WHO TB resistance database), we further investigated ppsD/E due to previous links with pyrazinamide resistance. Further, both ppsD/E displayed a disruptive frameshift mutation that can be assessed with a knockout mutant and had a conserved, non-essential ortholog in E. coli. For both ppsD/ppsE, a single ortholog, with a conserved domain, fabF, was identified as an appropriate E. coli K12 ortholog. fabF and ppsD/E are similarly involved in the elongation of fatty acyl chains during lipid and type II fatty acid biosynthesis, crucial to the production of cell membranes for the respective hosts [6164] (Table S7). A targeted transposon insertion mutant was used for further investigation of its rifampicin resistance phenotype (Table S2). Using MIC and concentration gradient growth assays, we observed that ΔfabF displayed increased growth between the range 8 and 14 µg ml−1, as compared to the wild-type (Fig. 3a and Table S8). This indicates that mutations in fabF do result in an increased rifampicin tolerance phenotype but not a high-level resistance phenotype per se and may explain why these mutations were only observed in the lower concentrations of rifampicin (Fig. 2a and c). Due to the role of fabF in the production of cell membrane lipids, to investigate the mechanism by which the ΔfabF mutant mediates this rifampicin tolerance, we performed cell membrane permeabilization assays using SYTOX green uptake as assessed by FACS, on the mutant and wild-type. After treatment with 8 µg ml−1 rifampicin, we observed a significant decrease in the permeability of the WT compared to the mutant, but no change was observed without rifampicin treatment (Fig. 3b). Thus, this could suggest that ΔfabF mutants were more tolerant to rifampicin exposure due to a less permeable membrane than the wild-type. Surprisingly, rifampicin tolerance mediated by fabF has not been reported in E. coli, despite being studied thoroughly [6566]. Given that fabF is an E. coli ortholog of ppsD/ppsE in M. tuberculosis, with ppsD/E similarly involved in lipid biosynthesis for the cell membrane [43], these results suggest that ppsD and/or ppsE may also play a role in rifampicin tolerance in M. tuberculosis; however, this still requires further functional validation in an M. tuberculosis strain. Characteristics of E. coli K12 and its ΔfabF mutant in response to rifampicin. (a) Growth dynamics (mean±se) of wild-type E. coli K12 and ΔfabF mutant across 2–20 µg ml−1 of rifampicin, with improved growth in the ΔfabF mutant between 8 and 14 µg ml−1 of rifampicin. (b) Membrane permeability of wild-type E. coli K12 and ΔfabF mutant, where the ΔfabF mutant shows decreased membrane permeability relative to the wild-type after rifampicin exposure (18 h growth in 8 µg ml−1). Differences between wild-type and ΔfabF mutant were statistically analysed using a t-test, where median fluorescence intensity difference was used for fluorescence data. P value: * P < 0.05, ** P < 0.01, *** P < 0.001 and **** P < 0.0001. Rifampicin tolerance was observed in E. coli ΔfabF by both improved growth in the ΔfabF mutant between 8 and 14 µg ml−1 of rifampicin and decreased permeabilization following rifampicin exposure (Fig. 3b). We propose that rifampicin entry into the cell is limited by these mutations that alter the fatty acid composition of the cell membrane. Reduced efficiency in long-chain (>12C atoms) fatty acid elongation decreases the uptake of rifampicin [6768] and other lipophilic antibiotics [68] into cells. This can confer resistance to lipophilic antibiotics; however, it may also confer a slight reduction of fitness, which is why the native gene is retained and upregulated in rpoB mutants [43]. Due to the complexity of M. tuberculosis cell envelopes, this E. coli model has some limitations. Despite the analogous involvement in fatty acid elongation of PpsD and PpsE in M. tuberculosis and FabF in E. coli, these enzymes share 26.42 and 29.97% amino acid identity. This limited similarity is due to the entirety of FabF (413AA), representing a ketosynthase family 3 domain profile that is only one of the domains featured in PpsD/E (1827AA and 1488AA) (Table S7). Differences in protein structure, and the requirements for both PpsD and PpsE in M. tuberculosis, are likely due to membrane differences between M. tuberculosis and E. coli. Although limited in amino acid sequence homology, ΔfabF phenotypic results are comparable to ppsD and ppsE mutations in M. tuberculosis. Additionally, with their role in cell membrane production, type II fatty acid biosynthesis pathway is of key interest for drug targets [69]. Thus, ppsD and ppsE may represent potential drug targets for M. tuberculosis, in particular due to the necessity of full function for higher levels of resistance [43]. Further, the observed ΔfabF tolerance phenotype in E. coli indicates that modulation of the cell membrane composition is an important antibiotic tolerance strategy conserved across species, highlighting the broader implications of finding and testing low-level resistance from population-level sequencing in multiple species. Implications for TB therapeutics and conclusions Due to rifampicin and linezolid targeting consecutive steps in protein synthesis, they have been investigated for combination therapies. These studies demonstrate the bacteriostatic effect of these combined antibiotics on rifampicin susceptible isolates, which could prevent the emergence of rifampicin resistance [1070]. However, there is the potential for rifampicin- and linezolid-resistant mutants arising from combinatorial therapy, in addition to linezolid-resistant mutants seen in cases of extensively drug-resistant M. tuberculosis [210]. These newly identified minority variants could represent novel molecular targets as helper drugs to increase the effectiveness of either the single or combinational therapies [7071]. In conclusion, this was the first study to apply in-depth directed evolution methodology to M. tuberculosis, which allowed for the identification of population stepwise mutations supporting rifampicin and linezolid resistance. Here, variations in five genes were identified as being associated with rifampicin resistance, while three genes were identified for linezolid resistance. As expected, the major resistance mutations rpoB and rplC, conferring resistance to rifampicin and linezolid, respectively, were observed. However, an additional six genes were observed with novel mutations associated with resistance, with five presenting as subpopulations. Novel ppsD, glpK and mptC mutants were also identified in individual sequencing reads of antibiotic-resistant clinical isolates. Through utilization of E. coli as a model organism, ppsD and ppsE or orthologue mutations were phenotypically linked to low-level rifampicin resistance for the first time in both M. tuberculosis and E. coli. Thus, an antibiotic tolerance strategy common to multiple critically antibiotic-resistant bacterial species is identified as a clinically important target for further study. Supplementary material