Mycobacterium tuberculosis Requires the Outer Membrane Lipid Phthiocerol Dimycocerosate for Starvation-Induced Antibiotic Tolerance

ABSTRACT Tolerance of Mycobacterium tuberculosis to antibiotics contributes to the long duration of tuberculosis (TB) treatment and the emergence of drug-resistant strains. M. tuberculosis drug tolerance is induced by nutrient restriction, but the genetic determinants that promote antibiotic tolerance triggered by nutrient limitation have not been comprehensively identified. Here, we show that M. tuberculosis requires production of the outer membrane lipid phthiocerol dimycocerosate (PDIM) to tolerate antibiotics under nutrient-limited conditions. We developed an arrayed transposon (Tn) mutant library in M. tuberculosis Erdman and used orthogonal pooling and transposon sequencing (Tn-seq) to map the locations of individual mutants in the library. We screened a subset of the library (~1,000 mutants) by Tn-seq and identified 32 and 102 Tn mutants with altered tolerance to antibiotics under stationary-phase and phosphate-starved conditions, respectively. Two mutants recovered from the arrayed library, ppgK::Tn and clpS::Tn, showed increased susceptibility to two different drug combinations under both nutrient-limited conditions, but their phenotypes were not complemented by the Tn-disrupted gene. Whole-genome sequencing revealed single nucleotide polymorphisms in both the ppgK::Tn and clpS::Tn mutants that prevented PDIM production. Complementation of the clpS::Tn ppsD Q291* mutant with ppsD restored PDIM production and antibiotic tolerance, demonstrating that loss of PDIM sensitized M. tuberculosis to antibiotics. Our data suggest that drugs targeting production of PDIM, a critical M. tuberculosis virulence determinant, have the potential to enhance the efficacy of existing antibiotics, thereby shortening TB treatment and limiting development of drug resistance. IMPORTANCE Mycobacterium tuberculosis causes 10 million cases of active TB disease and over 1 million deaths worldwide each year. TB treatment is complex, requiring at least 6 months of therapy with a combination of antibiotics. One factor that contributes to the length of TB treatment is M. tuberculosis phenotypic antibiotic tolerance, which allows the bacteria to survive prolonged drug exposure even in the absence of genetic mutations causing drug resistance. Here, we report a genetic screen to identify M. tuberculosis genes that promote drug tolerance during nutrient starvation. Our study revealed the outer membrane lipid phthiocerol dimycocerosate (PDIM) as a key determinant of M. tuberculosis antibiotic tolerance triggered by nutrient starvation. Our study implicates PDIM synthesis as a potential target for development of new TB drugs that would sensitize M. tuberculosis to existing antibiotics to shorten TB treatment.

virulence determinant (33), our findings suggest that PDIM synthesis is an attractive target for development of new drugs that would both decrease virulence and sensitize M. tuberculosis to existing antibiotics.

RESULTS
Construction of an arrayed and sequence-mapped M. tuberculosis Erdman transposon mutant library. To identify M. tuberculosis determinants of drug tolerance in nutrient starvation, we planned to screen transposon (Tn) mutants for those with defects surviving drug exposure. Since a large percentage of the population is killed by antibiotic treatment, we expected our screen to have an inherent bottleneck that would cause stochastic loss of individual Tn mutant strains. To overcome this bottleneck, we screened defined pools of Tn mutants, which we created as an arrayed library. To enable recovery of auxotrophs, Tn mutants were selected on a nutrient-rich medium, MtbYM, that contains additional carbon and nitrogen sources, vitamins, and cofactors compared to the standard Middlebrook 7H9 medium (34). Approximately 8,000 M. tuberculosis Erdman Tn mutants were arrayed in 80 racks, each with 96 barcoded tubes. Tn mutant pools for experiments were created by combining all ;96 Tn mutants in a rack.
To facilitate recovery of Tn mutants of interest, we used orthogonal pooling and Tnseq to map the location of Tn mutants in the library. The library was divided into two sets of 40 racks. For each set of 40 racks, pools were created of all mutants in each row (rows A to H, 8 pools, each with 12 Â 40 = 480 mutants), all mutants in each column (columns 1 to 12, 12 pools, each with 8 Â 40 = 320 mutants), and all mutants in each rack (racks 1 to 40 or racks 41 to 80, 40 pools, each with 96 mutants) to generate 60 pooled samples per set of racks for Tn-seq. For Tn mutants with no sibling clones in the set of racks, sequence reads corresponding to the Tn insertion site appear in equal abundance in one rack pool, one column pool, and one row pool. For racks 1 to 40, we used two mapping methods: the heuristic Straight Three strategy (35) and the probabilistic Knockout Sudoku algorithm (36). We found good agreement between these mapping methods, with a larger percentage of Tn mutants mapped by Knockout Sudoku (see Table S1 in the supplemental material). Mutants in racks 41 to 80 were mapped only by Knockout Sudoku (Table S1).
Our arrayed library contains 11,189 total Tn insertions at 6,842 unique locations in the M. tuberculosis Erdman genome. These include 1,323 unique insertions in intergenic regions and 5,519 unique insertions within 2,328 of the 4,302 annotated M. tuberculosis Erdman open reading frames (ORFs) (;54% coverage). The library contains Tn insertions in 1,975 of the 3,102 ORFs previously described as being nonessential for growth of M. tuberculosis H37Rv in MtbYM rich medium (34). Tn insertions were distributed evenly throughout the M. tuberculosis Erdman genome (Fig. S1A). We confidently mapped the location of 6,917 Tn mutants (61.8%) by Knockout Sudoku (Table S1), which is comparable to the ;68% mapping confidence reported for a similar Enterococcus faecalis Tn mutant library (35). Most tubes (3,879) have a single Tn insertion mapped (Fig. S1B).
Tn-seq identifies starvation-induced drug tolerance determinants. To identify nutrient-limited conditions that reproducibly increase M. tuberculosis antibiotic tolerance, we either grew cultures to stationary phase (a general nutrient limitation) or starved the cultures of inorganic phosphate (P i , a defined nutrient limitation). We used combinations of two drugs, each with different modes of action, to prevent emergence of drug-resistant mutants: ciprofloxacin and isoniazid (CIP1INH) or rifampicin and isoniazid (RIF1INH). Each combination included a bactericidal drug (RIF or CIP) and INH at a bacteriostatic low dose to promote isolation of persister variants (37). We compared wild-type (WT) M. tuberculosis Erdman drug tolerance under these conditions between Middlebrook 7H9 and MtbYM rich media. The rate at which M. tuberculosis was killed by antibiotics was decreased in MtbYM compared to 7H9 medium under both stationary-phase and P i -starved conditions and with both the CIP1INH and RIF1INH drug combinations ( Fig. 1A and B). These data suggest that the additional nutrient sources in MtbYM induce higher M. tuberculosis drug tolerance.
To identify M. tuberculosis determinants of starvation-induced drug tolerance, we screened low-complexity pools of Tn mutants in P i -starved or stationary-phase MtbYM cultures treated with CIP1INH or RIF1INH and identified Tn mutants with altered fitness by Tn-seq. We note that our screen cannot distinguish between mutations that alter intrinsic drug resistance, drug tolerance, or formation of persister variants, any of which would alter Tn mutant fitness upon drug exposure. Briefly, Tn mutant pools were grown to stationary phase (7 days in MtbYM) or P i starved (72 h in P i -free MtbYM). Each starved culture was plated on MtbYM agar prior to drug exposure as an input control and then split into triplicate no-drug control, CIP1INH-treated, or RIF1INH-treated experimental groups. Cultures were incubated for 9 days before plating on MtbYM agar to recover surviving Tn mutants. Tn mutant abundance under each experimental condition (input, no-drug output, CIP1INH output, RIF1INH output) was determined by Tn-seq. In preliminary experiments, we determined that we could screen ;500 Tn mutants (pools of 5 racks) simultaneously without stochastic loss of mutants in individual biological replicates. We screened two pools, each with ;500 Tn mutants (racks 6 to 10 and racks 16 to 20), and obtained similar numbers of Tn-seq reads mapped to the M. tuberculosis genome under all experimental conditions (Table S2). To determine the fitness of Tn mutants, we used TnseqDiff, a parametric method that identifies conditionally essential genes between conditions based on Tn insertion-level data and that is compatible with low-density Tn libraries (38). We compared the normalized frequency of sequence reads at each Tn insertion site between the experimental conditions and the input (CIP1INH/input, RIF1INH/ input, or no-drug control/input) or between drug-treated conditions and the no-drug control (CIP1INH/control or RIF1INH/control) with TnseqDiff (38). Complete TnseqDiff analyses are available in Table S3. Using statistical significance cutoffs of >62 log 2 fold change and an adjusted P value of ,0.025, we identified 122 Tn insertions that exhibited differential fitness in one or more comparisons corresponding to 17 intergenic insertions and 92 unique ORFs disrupted (Table S4).
Under the P i starvation condition, we identified 102 Tn mutants with significantly altered fitness ( Fig. 2; Table S4). Of the 86 Tn mutants with significantly reduced fitness (negative fold change), 11 showed phenotypes in the no-drug control/input comparison (Fig. 2C), suggesting that these gene products are required for survival of P i starvation. These included Tn insertions in genes putatively involved in nucleotide metabolism or transport (purN, pyrR, mkl), central metabolism (pckA), cell division (ftsX), and stress responses (htpX, uvrB) ( Fig. 2C; Table S4). We identified 76 mutants that showed significantly reduced fitness in the CIP1INH/input comparison ( Fig. 2A) and 11 mutants with significantly reduced fitness in the RIF1INH/input comparison (Fig. 2B). Of these, five Tn mutants (three of which were in ORFs) exhibited reduced fitness under both drug treatment conditions, but not the no-drug control ( Fig. 2D drug treatment condition to the no-drug control at day 9. Only four Tn mutants were identified that met our statistical significance criteria in these comparisons (Table S4; Fig. S2).
Under the P i starvation condition, we observed several genes or pathways with multiple independent Tn insertions that exhibited significantly altered fitness upon drug treatment (Table 1). Mutants significantly impaired for survival of CIP1INH treatment during P i starvation included two with Tn insertions in genes required for production of the outer membrane lipid phthiocerol dimycocerosate (PDIM; ppsC, fadD28) (Table 1). PDIM limits permeability of the M. tuberculosis outer membrane to small hydrophilic nutrients, including glucose and glycerol (39,40), and may also restrict diffusion of certain antibiotics, such as the glycopeptide vancomycin (41). We also identified two independent Tn insertions each in Erdman_0220 (rv0194), which encodes an ATP-binding cassette (ABC)-type efflux pump previously implicated in intrinsic resistance of Mycobacterium bovis BCG to multiple drugs, including ampicillin (42), and Erdman_3338 (rv3049), which encodes a putative monooxygenase ( Fig. 2A; Table 1).
In P i starvation, we also identified 16 mutants with significantly increased fitness (positive fold change). These included five independent Tn insertions in phoPR, which encodes a two-component system that responds to acid stress (43)(44)(45). Although the phoPR mutants exhibited increased fitness in both drug combinations ( Fig. 2A and B), they were also more abundant in the no-drug control/input comparison (Fig. 2C), suggesting that PhoPR normally impairs survival of P i limitation, rather than specifically altering antibiotic susceptibility. We identified eight additional Tn mutants with increased abundance in both the CIP1INH/ input and RIF1INH/input comparisons ( Fig. 2A, B, and E). These included four independent Tn insertions in the mce1 locus ( Fig. 2A and B; Table 1), which encodes a fatty acid transporter (46), suggesting Mce1 contributes to drug susceptibility. Under the stationary-phase condition, we identified 32 Tn mutants with significantly altered fitness. In comparisons with the input control, three Tn mutants decreased in relative abundance (negative fold change) and 25 Tn mutants increased in relative abundance (positive fold change) ( Fig. 3; Table S4). Only two Tn mutants (pe_pgrs31::Tn and an intergenic insertion 59 of rv3796) exhibited decreased fitness in drug-treated stationary-phase cultures (Fig. 3B). In comparisons of drug-treated cultures to the no-drug control, Tn mutants in phoP and Erdman_3938 (rv3553) showed significantly decreased abundance ( Fig. S3; Table S4). However, these mutants also increased in relative abundance in the no-drug control/input comparison (Fig. 3C), suggesting that these mutants have a survival advantage in stationary phase.
Under the stationary-phase condition, we identified several genes or pathways with multiple independent Tn insertions that exhibited increased fitness in drug-treated stationary-phase cultures ( Table 2). These included five independent Tn insertions in the mce1 locus ( Fig. 3A and B; Table 2). The mce1 mutants were all overrepresented in RIF1INHtreated treated cultures compared to either the input or no-drug controls ( Fig. 3A and B;  Fig. S3). We also identified multiple independent Tn insertions in mpa, sigB, the nuo operon, ppe32, and Erdman_2155 (rv1957) that caused increased fitness in drug-treated stationary-phase cultures ( Fig. 3A and B; Table 2). Rv1957 is a SecB-like chaperone of the antitoxin HigA1 (47). In the absence of Rv1957, HigA1 is degraded by the ClpXP protease (48), freeing the HigB1 toxin to degrade mRNA and tmRNA (49). Mutation of rv1957 is expected to limit bacterial replication via HigB1 toxin activation, thereby enhancing drug tolerance. Mpa is the ATPase of the mycobacterial proteasome, which degrades proteins that are posttranslationally modified with the prokaryotic ubiquitin-like protein Pup (50). While the proteasome itself has not been directly implicated in mycobacterial drug tolerance, several toxins and antitoxins of TA systems are modified by Pup and may be stabilized in mutants lacking proteasome activity (51,52).
Retesting confirms increased drug susceptibility of ppgK::Tn and clpS::Tn mutants. To validate the phenotypes observed in our screen, we determined the sensitivity of individual Tn mutants to antibiotics in monoculture. We selected only Tn mutants that exhibited significantly reduced fitness in the drug-treated versus input TnseqDiff analyses to characterize pathways that, when inhibited, would sensitize M. tuberculosis to existing antibiotics. We focused on genes that were not previously implicated in mycobacterial drug tolerance. The three Tn mutants we selected (rv0457c::Tn, ppgK::Tn, and clpS::Tn) had relatively severe phenotypes based on the TnseqDiff fold change, had the Tn insertion in the middle of the ORF, and were identified only in the P i starvation screen ( Table 3). Each of these genes was identified by only a single Tn insertion (Table S4). rv0457c::Tn and ppgK::Tn were also the only two mutants that exhibited significantly reduced fitness in the RIF1INH-treated versus no-drug control comparison ( Fig. S2; Table S4). Each mutant was recovered from the arrayed library, and the Tn insertion site was confirmed by PCR and sequencing.
The rv0457c::Tn mutant exhibited specific susceptibility to RIF1INH in the P i starvation screen (Table 3). rv0457c encodes a prolyl oligopeptidase (53) and is located immediately 59 of the mazE1-mazF1 operon that encodes a TA system. MazF toxins were previously implicated in M. tuberculosis drug tolerance (54). The rv0457c::Tn mutant displayed a subtle but statistically significant increase in sensitivity to RIF1INH, but not CIP1INH, under P i starvation conditions ( Fig. 4A and B). As the rv0457c::Tn mutant did not exhibit strong phenotypes upon retesting, it was not pursued further.
The ppgK::Tn mutant exhibited the highest sensitivity to RIF1INH under P i -starved conditions in our screen (Fig. 2B) and was also susceptible to the CIP1INH combination, though it did not reach our statistical significance cutoffs (Table 3). ppgK encodes the dominant glucokinase in M. tuberculosis (55), catalyzing phosphorylation of glucose with a preference for polyphosphate as the phosphodonor (56). The ppgK::Tn mutant was significantly more susceptible to both CIP1INH and RIF1INH in P i -free MtbYM medium ( Fig. 4C and D). However, the MICs (MIC 90 s) for all three drugs were similar between the ppgK::Tn and WT strains, suggesting that the ppgK::Tn mutant has altered antibiotic tolerance (Table 4). We attempted to complement these phenotypes by providing ppgK in trans using a construct similar to that previously reported to complement a DppgK mutant (55). Quantitative reverse transcription-PCR (qRT-PCR) confirmed ppgK transcription from the pMV-ppgK vector (data not shown), but complementation did not increase the tolerance of the ppgK::Tn mutant to either drug combination ( Fig. 4C and D). These data suggest that the ppgK::Tn mutant harbors a secondary mutation, unlinked to the Tn, that causes increased susceptibility to antibiotics.
The clpS::Tn (rv1331::Tn) mutant exhibited high susceptibility to CIP1INH, but not RIF1INH, under P i -starved conditions in our screen (Table 3). clpS encodes an adaptor for the M. tuberculosis ClpC1P1P2 (ClpCP) protease. ClpS promotes ClpCP degradation of proteins with destabilizing N-terminal residues (the N-end rule) and inhibits degradation of SsrA-tagged proteins derived from translationally stalled ribosomes (57,58). M. tuberculosis ClpCP was implicated in drug tolerance because it degrades antitoxins from several classes of TA systems (58). Loss of the ClpS adaptor may therefore alter the stability of certain ClpCP protease substrates that influence drug tolerance. When tested in monoculture, the clpS::Tn mutant was highly susceptible to both CIP1INH and RIF1INH in P i -free MtbYM medium ( Fig. 4E and F). While the clpS::Tn mutant had a similar MIC 90 for both CIP and INH as that of the WT control, the MIC 90 for RIF was reduced 4-to 8-fold, suggesting that the clpS::Tn mutant has reduced intrinsic resistance to RIF (Table 4). We attempted to complement the clpS::Tn mutant by providing clpS in trans. We observed clpS transcript from pMV-clpS by qRT-PCR (data not shown), but the complemented strain remained susceptible to both drug combinations ( Fig. 4E and F). As clpS is carried at the 59 end of a putative operon, we considered the possibility that the Tn insertion was polar on expression of downstream genes. A cosmid covering the complete clpS region also failed to complement the clpS::Tn mutant phenotype ( Fig. 4E and F). These data suggest that the clpS::Tn mutant also has a secondary mutation that increases susceptibility to antibiotics. Tn mutants with defects in drug tolerance harbor secondary mutations that disrupt production of PDIM and cause decreased drug tolerance. Since neither the ppgK::Tn nor clpS::Tn mutant phenotype could be complemented, we sought to identify secondary mutations responsible for their drug susceptibility phenotypes. We conducted whole-genome resequencing on rv0457c::Tn, ppgK::Tn, and clpS::Tn mutants and our WT M. tuberculosis Erdman strain. This sequencing confirmed the predicted Tn insertion sites in each strain and demonstrated that each strain harbored a single Tn, ruling out the possibility that a secondary Tn insertion was responsible for their phenotypes ( Fig. S4A to C). In both the ppgK::Tn and clpS::Tn mutants, we identified nonsynonymous single nucleotide polymorphisms (SNPs) in genes required for production  of the lipid phthiocerol dimycocerosate (PDIM). No mutations in genes required for PDIM biosynthesis were identified in the WT Erdman or rv0457c::Tn strain. The ppgK::Tn strain had a G-to-C mutation at position 2360 in ppsE (Fig. S4D), which encodes a polyketide synthase required for production of the phthiocerol chain of PDIM (59). This SNP is predicted to cause a W787S amino acid substitution in PpsE that may alter PpsE activity and PDIM production. The clpS::Tn strain had a C-to-T mutation at position 655 in ppsD (Fig. S4E), which is predicted to introduce a premature amber stop codon at position 219 in PpsD. As ppsD also encodes a polyketide synthase required for production of the phthiocerol component of PDIM (59), the ppsD Q219* mutation is predicted to completely block PDIM production. Excluding highly repetitive sequences, such as the pe and ppe genes, which are difficult to resolve by short-read sequencing, these were the only nonsynonymous SNPs identified in the ppgK::Tn and clpS::Tn strains.
To directly test whether the ppsD Q219* or ppsE W787S mutation blocks PDIM production by the clpS::Tn or ppgK::Tn strain, respectively, we analyzed PDIM production by an established radiolabeling method. Bacteria were labeled with [ 14 C]propionate, which is selectively incorporated into PDIM, and the PDIM (DIM A) and phthiodiolone dimycocerosate precursor (DIM B) were detected in apolar lipid extracts by thin-layer chromatography (TLC) (33,60). As expected, the clpS::Tn ppsD Q219* mutant did not produce any detectable PDIM (Fig. 5A, lane 2). The ppgK::Tn ppsE W787S mutant exhibited an intermediate PDIM production phenotype, with a 2.3-fold reduction in both DIM A and DIM B compared to the WT control (Fig. 5A, lane 4). These results suggest that the antibiotic susceptibility phenotypes of both mutants could be due to reduced PDIM production rather than the Tn insertion. These data also suggest that the intermediate drug susceptibility phenotypes of the ppgK::Tn ppsE W787S mutant could be caused by its intermediate level of PDIM production.
To determine if PDIM deficiency caused increased susceptibility of the clpS::Tn ppsD Q219* mutant to antibiotics, we complemented the ppsD Q219* mutation with ppsD on a plasmid. Complementation with ppsD fully restored PDIM production (Fig. 5A,  lane 3). We tested the sensitivity of the ppsD complemented strain to both CIP1INH and RIF1INH under P i starvation conditions and observed similar resistance to both drug combinations as that with the WT control ( Fig. 5B and C). These data demonstrate that loss of PDIM production, rather than loss of ClpS function, causes increased drug susceptibility of the clpS::Tn ppsD Q219* mutant. Since two of the Tn mutants that we identified in our P i starvation screen exhibited reduced fitness upon antibiotic treatment due to spontaneous mutations in the PDIM biosynthesis locus, we cannot exclude the possibility that other Tn mutants with reduced antibiotic tolerance harbor similar secondary mutations responsible for their phenotypes. PDIM-deficient mutants are hypersusceptible to antibiotics under stationaryphase and exponential-phase growth conditions. Although the clpS::Tn ppsD Q219* and ppgK::Tn ppsE W787S mutants were initially identified only in the P i -starved drug screen, we sought to determine if increased susceptibility to antibiotics was specific to this growth condition. We therefore tested susceptibility of both mutants to the CIP1INH and RIF1INH drug combinations in stationary-phase and exponential-phase cultures grown in MtbYM rich medium. In stationary phase, both the clpS::Tn ppsD Q219* and ppgK::Tn ppsE W787S mutants showed a significant decrease in tolerance to CIP1INH (Fig. 6A). Both mutants also exhibited decreased tolerance to RIF1INH in stationary phase, though this was not statistically significant (Fig. 6B). These data suggest that some Tn mutants with antibiotic tolerance phenotypes were not uncovered by our stationary-phase Tn-seq screen, perhaps due to the stringent statistical significance cutoffs we used. The decreased antibiotic tolerance of the clpS::Tn ppsD Q219* mutant in stationary phase was fully complemented by ppsD (Fig. 6D), confirming that PDIM deficiency also increases susceptibility to antibiotics in stationary phase. The clpS::Tn ppsD Q219* mutant also exhibited a modest, but statistically significant, decrease in antibiotic tolerance in exponential phase for the CIP1INH drug combination, which was partially complemented by ppsD (Fig. 6C and E). These data suggest that the clpS::Tn ppsD Q219* mutant also produces fewer stochastic persister variants due to loss of PDIM production.

DISCUSSION
Molecular mechanisms driving M. tuberculosis recalcitrance to antibiotics under nutrient starvation are poorly characterized. Here, using a Tn-seq screen, we identify PDIM production as a critical determinant of M. tuberculosis drug tolerance under nutrient-limited conditions. We identified two Tn mutants, clpS::Tn and ppgK::Tn, that were hypersusceptible to antibiotics. Both mutants harbored secondary mutations, unlinked to the Tn, that disrupted PDIM production. We restored PDIM production to the clpS:: Tn strain by complementing the ppsD Q291* mutation and showed that this restored normal drug tolerance, directly demonstrating that M. tuberculosis requires PDIM to tolerate antibiotic exposure. Loss of PDIM caused a decrease in the MIC 90 for RIF, demonstrating that PDIM contributes to intrinsic RIF resistance. However, PDIM-deficient strains also exhibited increased susceptibility to the CIP1INH combination despite no change in the MIC 90 for these drugs, indicating that PDIM also promotes antibiotic tolerance. The ppgK::Tn ppsE W787S mutant exhibited an intermediate drug tolerance phenotype that was associated with reduced, but not absent, PDIM production, suggesting that even partial inhibition of PDIM synthesis can sensitize M. tuberculosis to antibiotics.
There are at least two mechanisms by which PDIM could increase M. tuberculosis drug tolerance: decreasing the intracellular concentration of antibiotics by decreasing permeability of the outer membrane or altering the intracellular concentrations of central metabolites by functioning as a metabolic sink for propionate. PDIM decreases permeability of the M. tuberculosis outer membrane to small molecules, including glucose and glycerol (39,40), and may also restrict diffusion of some antibiotics. Indeed, PDIM is required for intrinsic resistance to vancomycin, likely by decreasing vancomycin access to its peptidoglycan target (41). PDIM has previously been implicated in drug tolerance in other mycobacteria. PDIM-deficient Mycobacterium bovis BCG exhibited increased susceptibility to RIF, with a 4fold decrease in MIC, but there was no change in susceptibility to INH or CIP (41). PDIM also increases antibiotic tolerance of Mycobacterium marinum, which was correlated with reduced outer membrane permeability (61,62). We observed that PDIM enhances M. tuberculosis drug tolerance particularly under nutrient-limited conditions that limit accumulation of RIF and fluoroquinolone antibiotics (32). Since efflux pump inhibitors did not reverse the drug tolerance triggered by nutrient starvation (32), it is tempting to speculate that PDIM decreases drug uptake under nutrient-limited conditions by limiting import of antibiotics through the outer membrane.
Alternatively, PDIM could alter drug tolerance by effects on central metabolism. Synthesis of the long-chain branched fatty acids in PDIM requires the metabolite methylmalonyl coenzyme A (methylmalonyl-CoA), which is derived from propionate (59). During infection, M. tuberculosis catabolizes fatty acids and cholesterol, which serve as primary carbon sources, to propionate (63,64). Excess propionate stimulates increased production of PDIM with longer mycocerosic acid chain lengths both in vitro and during infection of macrophages or mice (63,65). PDIM was therefore proposed to act as a sink for propionyl-CoA, which can be toxic at high concentrations (63,66). PDIM-deficient strains may be more susceptible to antibiotics, particularly under growth conditions with fatty acids or cholesterol as carbon sources, due to the combined effects of the antibiotic and accumulation of toxic central metabolites. Growth in medium with either propionate or cholesterol as a carbon source increases the intrinsic resistance of M. tuberculosis to RIF (67). This increased RIF resistance was correlated with increased production and chain length of sulfolipid-1 (SL-1) (67), another branchedchain outer membrane lipid synthesized from propionate (65). PDIM production and chain length also increase with propionate as a carbon source, and PDIM is much more abundant than SL-1 (65), suggesting that PDIM could be primarily responsible for the carbon source-dependent increase in RIF resistance.
Connections between M. tuberculosis central metabolism and antibiotic tolerance have previously been reported. Genomic analysis of drug-resistant M. tuberculosis clinical isolates identified mutations in prpR, which encodes a transcriptional activator of PrpDC that catabolizes propionate through the methyl citrate cycle (68). Strains harboring prpR mutations exhibited increased drug tolerance specifically in medium with propionate as the sole carbon source (68). The authors of that study proposed that propionate accumulation in the prpR mutants limits antibiotic efficacy, but the propionate-dependent drug tolerance of prpR mutants could also simply be due to their slow growth with propionate as the sole carbon source (68). M. tuberculosis isocitrate lyase (ICL) is required for catabolism of both even-and odd-chain fatty acids and for tolerance to several different classes of antibiotics (69). The increased susceptibility of mutants lacking ICL activity was correlated with accumulation of tricarboxylic acid (TCA) cycle intermediates and with increased endogenous oxidative stress (69). However, ICL is also required for propionate catabolism (66), suggesting that accumulation of toxic propionate metabolites could also cause the hypersusceptibility of icl mutants to antibiotics.
Our results contrast with a previous study, in which selection for M. tuberculosis mutants with higher antibiotic persistence revealed multiple strains harboring spontaneous mutations in genes required for PDIM production (22). These PDIM-deficient mutants exhibited increased tolerance to multiple classes of antibiotics in exponential phase in the standard Middlebrook 7H9 medium, which contains glucose and glycerol as primary carbon sources (22). We may have observed decreased antibiotic tolerance of PDIM-deficient strains either because we used stationary-phase or P i -starved cultures or because we used MtbYM rich medium. MtbYM rich medium contains additional carbon sources, including branched-chain amino acids and pyruvate, which are catabolized to propionate, and vitamin B 12 , which activates production of methylmalonyl-CoA that is used for PDIM synthesis (70,71). It is unclear which in vitro growth medium more closely reflects the conditions M. tuberculosis experiences in the host or whether loss of PDIM would enhance drug susceptibility during lung infection. This question will be challenging to address because PDIM is also a critical M. tuberculosis virulence determinant that is required for resistance to innate immunity (33,60,72). We intend to explore the role of PDIM in antibiotic tolerance during lung infection in our future studies.
Our screen identified over 100 unique M. tuberculosis Tn insertion mutants with altered drug tolerance phenotypes, including several genes or pathways with multiple independent Tn insertions. Our results point to the importance of regulated protein degradation in M. tuberculosis drug tolerance. Loss of Rv1957, a chaperone of the HigA1 antitoxin, or loss of proteasome components (Mpa or Paf) caused increased drug tolerance, possibly due to stabilization of toxins that inhibit bacterial replication. Mutations in genes encoding the Mce1 system, which is required for uptake of fatty acids (46), also increased antibiotic tolerance. Mce1 may also function in uptake of antibiotics, such that loss of Mce1 reduces antibiotic import. Alternatively, loss of Mce1 function may reduce accumulation of fatty acid-derived metabolites that synergize with antibiotics by reducing fatty acid uptake. We identified two independent Tn insertions in sigB that increased drug tolerance. SigB is an alternative sigma factor that was reported to be required for mycobacterial tolerance to RIF and INH (73,74). Our results contrast with these studies, possibly due to our use of different growth media, and suggest that SigB can under certain conditions limit M. tuberculosis drug tolerance.
Our results highlight several advantages of screening low-complexity Tn mutant pools made from an arrayed library. These include identification of Tn mutants with robust phenotypes from selection conditions with strict bottlenecks, efficient recovery of individual Tn mutants, and reproducibility of mutant phenotypes upon individual retesting. However, our results also uncovered one drawback of this method: the potential for recovery of Tn mutants with secondary mutations that alter the phenotype of interest. In standard Tn-seq screens that use high-complexity Tn mutant libraries, secondary mutations are less likely to influence identification of genes that significantly impact fitness due to the presence of multiple independent Tn insertions in each gene. Our screen identified numerous Tn mutants with decreased drug tolerance, but it is possible that many of these strains harbor spontaneous secondary mutations causing loss of PDIM production, similar to the clpS::Tn and ppgK::Tn mutants. Distinguishing whether the decreased drug tolerance of these mutants is due to the Tn insertion or to loss of PDIM will require recovery and individual retesting of these Tn mutants, which can be efficiently done from our arrayed Tn mutant library.
Overall, our results demonstrate that M. tuberculosis requires PDIM for drug tolerance under nutrient starvation conditions in vitro. As PDIM is also a critical virulence determinant that is required to counteract host immune pressures, our results suggest that inhibitors of PDIM production could synergize with both host-imposed stress and existing antibiotics to kill M. tuberculosis more efficiently. This could dramatically shorten TB treatment times and prevent emergence of new drug-resistant strains. PDIM biosynthesis is a complex process, requiring multiple polyketide synthases, fatty acyl ligases, and thioesterases, several of which have already been explored as potential drug targets (59,75,76). It will be critical to determine whether PDIM deficiency also increases M. tuberculosis antibiotic susceptibility during infection to further support development of new inhibitors targeting PDIM production, which we intend to pursue in our future studies.
Creation and mapping of an M. tuberculosis Erdman arrayed transposon mutant library. Transposon (Tn) mutagenesis of wild-type M. tuberculosis Erdman was performed by transduction with the mycobacteriophage phAE159 carrying the Himar1 Tn as previously described (77). Wild-type bacteria were grown to mid-exponential phase (optical density at 600 nm [OD 600 ] of 0.4 to 0.6) in 7H9 broth, washed and resuspended in MP buffer (50 mM Tris, pH 7.6, 150 mM NaCl, 10 mM MgCl 2 , 2 mM CaCl 2 ), and then transduced with mycobacteriophage at a multiplicity of infection (MOI) of at least 20:1 for 3 h at 40°C. Phage adsorption was stopped with stop buffer (MP buffer with 60 mM sodium citrate and 0.6% Tween 80), and transduced cells were plated on MtbYM agar (pH 6.6) with Kan at a density of 100 to 200 colonies per plate. Plates were incubated at 37°C with 5% CO 2 for at least 3 weeks. Approximately 8,000 individual Tn mutant colonies were picked from plates into 600 mL of MtbYM broth in 1-mL V-bottom Matrix screw-cap tubes in a 96-well rack (Thermo Scientific) and incubated with shaking at 37°C for 2 weeks, until turbid.
Tn mutants were orthogonally pooled using the Straight Three strategy and sequenced, as previously described (35). Briefly, the Tn library was pooled in 2 groups of 40 racks each (racks 1 to 40 and racks 41 to 80). For each rack, a small volume of culture was removed from each tube and combined appropriately to form 8 row pools (rows A to H), 12 column pools (columns 1 to 12), and a rack pool. Each individual rack pool was aliquoted into one sample for sequencing and nine 1-mL aliquots for experimental use, which were stored at 280°C with glycerol at a 15% final concentration. After pooling, glycerol was added at a 15% final concentration to each Tn mutant culture and racks were stored at 280°C. For each group of 40 racks, analogous row and column pools were pooled from all 40 plates to generate 8 row and 12 column samples. These were multiplexed with the 40 individual rack pools for a total of 60 samples for Tn-seq. Genomic DNA (gDNA) was extracted from each row, column, and rack pool using the cetyltrimethylammonium bromide (CTAB)-lysozyme method (78) and submitted to the University of Minnesota Genomics Center (UMGC) for library creation and Tn-seq as described below. Tn mutants associated with the reads were traced back to their rack location in the arrayed Tn library using two approaches: Straight Three (35) and Knockout Sudoku (36).
Drug tolerance Tn-seq screen in stationary phase and P i starvation. Five frozen rack pools (1 mL each) generated during orthogonal pooling were inoculated in 250 mL MtbYM broth with Kan and grown at 37°C with aeration to mid-exponential phase (OD 600 of 0.4 to 0.6). A portion of the culture was removed to start a 250-mL P i -starved culture. Bacteria were washed twice in P i -free MtbYM broth, inoculated in P i -free MtbYM with Kan at an OD 600 of 0.1, and incubated at 37°C with aeration for 72 h. The remaining MtbYM culture was grown at 37°C with aeration for a total of 7 days to reach early stationary phase. We experimentally determined that at least 10 6 CFU of WT Erdman is recovered from a 12-mL culture after 9 days of drug treatment under either P i -free or stationary-phase conditions (Fig. 1). Therefore, as input controls, the P i -free or stationary-phase cultures were serially diluted and plated on MtbYM agar at a density of ;10 6 CFU/plate before addition of antibiotics. Cultures were then split into triplicate 12-mL antibiotic-treated (CIP1INH or RIF1INH) or untreated-control cultures and incubated with aeration at 37°C for 9 days. Antibiotic-treated bacteria were collected by centrifugation (3,720 Â g, 10 min), washed twice with an equal volume of PBS-T (Gibco phosphate-buffered saline [PBS], pH 7.4, with 0.05% Tween 80) to remove antibiotics, concentrated 100-fold in PBS-T, and plated on YM agar with Kan to recover at least 10 6 CFU. Untreated control cultures were serially diluted and plated at a density of ;10 6 CFU/plate. Plates were incubated at 37°C with 5% CO 2 until the biomass on the agar was confluent, up to 2 weeks. Confluent plates were flooded with 2 mL of GTE buffer (78) and gently scraped with a plastic 10-mL loop to loosen the biomass. Bacteria were collected by centrifugation (3,720 Â g, 10 min), and gDNA was extracted from cell pellets by the CTAB-lysozyme method (78) and cleaned using the genomic DNA Clean and Concentrator kit (Zymo) before submission to UMGC for Tn-seq library preparation and Illumina sequencing.
Tn-seq and data analysis. Transposon sequencing (Tn-seq) was performed as previously described (34). M. tuberculosis genomic DNA was fragmented with a Covaris S220 ultrasonicator, and a whole-genome library was prepared using the TruSeq Nano library preparation kit (Illumina). Library fragments containing Tn junctions were PCR amplified from the whole-genome library using the Tn-specific primer Mariner_1R_TnSeq_noMm and Illumina p7 primer (Table S6). The amplified products were uniquely indexed to allow sample pooling and multiplexed sequencing. Resulting Tn-seq libraries were sequenced on an Illumina 2500 high-output instrument in 125-bp paired-end output mode using v4 chemistry (Illumina). Sequencing reads were filtered to remove reads without the Tn sequence GGACTTATCAGCCAACCTGT. The 59 Illumina adaptor sequences were trimmed using BBDuk (https://sourceforge.net/projects/bbmap/). Each trimmed read was cut to 30 bases, and sequences not starting with TA were removed. Remaining reads were mapped to the M. tuberculosis Erdman genome (NC_020559.1) using HISAT2. Mapped reads were counted at each TA insertion site in the M. tuberculosis Erdman genome to generate read count tables for TnseqDiff analysis. TnseqDiff normalized the read counts using the default trimmed mean of M values (TMM) normalization method (79,80) and then determined conditional essentiality for each TA insertion site between experimental conditions (control/input, CIP1INH/input, RIF1INH/input, CIP1INH/control, RIF1INH/control). TnseqDiff calculated the fold change and corresponding two-sided P value for each TA insertion site (38). All P values were adjusted for multiple testing using the Benjamini-Hochberg procedure in TnseqDiff. The cutoff values for statistical significance were set at a fold change of >62 log 2 and an adjusted P value of ,0.025.
Recovery of Tn mutants from the arrayed Tn mutant library. Each Tn mutant individually retested was isolated from the tube corresponding to the Tn mutant location in the arrayed library by streaking for individual colonies on MtbYM agar containing Kan. Plates were incubated for at least 3 weeks at 37°C. Up to four individual colonies were picked and grown in 10 mL of MtbYM broth with Kan at 37°C with aeration until turbid. The Tn insertion site was confirmed by PCR using a gene-specific primer 59 or 39 of the TA site and a primer specific to the Tn Kan resistance cassette (Table S6) followed by Sanger sequencing.
Tn mutant complementation. Vectors for complementation of Tn mutants were made in the integrating plasmid pMV306hyg (Table S5). We generated pMV306hyg by replacing the aph Kan resistance marker in pMV306 (81) with a Hyg resistance cassette. pMV306 without aph was PCR amplified with primers pMV306_F and pMV306_R (Table S6), digested with SbfI and AflII, and ligated to a Hyg resistance cassette that was removed from pJT6a (82) by restriction with SblI and AflII. Each gene was PCR amplified along with ;300 bases 59 of the translation start site to include the native promoter (Table S6), cloned in pCR2.1 TOPO (Invitrogen) and sequenced, and then removed from pCR2.1 by restriction with XbaI and HindIII and ligated to XbaI-and HindIII-digested pMV306hyg. Cosmid 8C3.1 containing the genomic region rv1317 to rv1343 surrounding clpS was obtained from the lab of William R. Jacobs. The pMV361hyg-ppsD vector was generated by replacing the Kan resistance cassette in an existing pMV361-ppsD vector (60) with a Hyg resistance cassette by Gibson assembly. Primers pMV361.FOR/pMV361.REV and hyg_fwd_2/hyg_rev_2 were used to PCR amplify pMV361-ppsD without the Kan resistance cassette and the Hyg resistance cassette from pMV306hyg, respectively (Table S6). PCR products were assembled with New England Biolabs (NEB) HiFi Assembly master mix (New England Biolabs) followed by sequencing of ppsD and the Hyg resistance cassette. Tn mutants were electroporated with the corresponding complementation vector or cosmid as described previously (83). Transformants were selected on

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.