HflX is a GTPase that controls hypoxia-induced replication arrest in slow-growing mycobacteria

Significance

GTPases play essential roles in living organisms, functioning as molecular switches or timers in many fundamental cellular processes. HflX is a GTPase that is highly conserved in the entire prokaryotic kingdom and is involved in the stress response of bacteria. We show here that under hypoxic stress, by modulating the cell’s translational activity, HflX controls mycobacteria growth rate and entry into nonreplicating state. Our work assigns a physiological role to HflX in pathogenic mycobacteria, implicating this factor in one of the most elusive but critical phenomena of replication arrest, which is responsible for the worldwide TB drug resistance crisis and a major roadblock to successful shorter TB therapies.

Abstract

GTPase high frequency of lysogenization X (HflX) is highly conserved in prokaryotes and acts as a ribosome-splitting factor as part of the heat shock response in Escherichia coli. Here we report that HflX produced by slow-growing Mycobacterium bovis bacillus Calmette–Guérin (BCG) is a GTPase that plays a critical role in the pathogen’s transition to a nonreplicating, drug-tolerant state in response to hypoxia. Indeed, HflX-deficient M. bovis BCG (KO) replicated markedly faster in the microaerophilic phase of a hypoxia model that resulted in premature entry into dormancy. The KO mutant displayed hallmarks of nonreplicating mycobacteria, including phenotypic drug resistance, altered morphology, low intracellular ATP levels, and overexpression of Dormancy (Dos) regulon proteins. Mice nasally infected with HflX KO mutant displayed increased bacterial burden in the lungs, spleen, and lymph nodes during the chronic phase of infection, consistent with the higher replication rate observed in vitro in microaerophilic conditions. Unlike fast growing mycobacteria, M. bovis BCG HlfX was not involved in antibiotic resistance under aerobic growth. Proteomics, pull-down, and ribo-sequencing approaches supported that mycobacterial HflX is a ribosome-binding protein that controls translational activity of the cell. With HflX fully conserved between M. bovis BCG and M. tuberculosis, our work provides further insights into the molecular mechanisms deployed by pathogenic mycobacteria to adapt to their hypoxic microenvironment.

GTP-binding proteins are found across all living kingdoms and are involved in the regulation of many cellular processes. Among which, the small GTPases act as molecular switches that are active or “on” when binding GTP and inactive or “off” when binding GDP. Universally conserved prokaryotic GTPases (ucpGTPases) are a core group of GTPases that are conserved in most prokaryotes, hinting at a critical function in biology (1), although the actual physiological role of most of them has remained elusive. ucpGTPases are characterized by the presence of highly conserved motifs or domains, including the phosphate-loop (P-loop) within the G-domain, a characteristic site where GTP binding and hydrolysis occur (2).

High frequency of lysogenization X (HflX) protein belongs to the superfamily of the Obg-HflX–like ucpGTPases, class of Translation Factors (TRAFAC), which have been described to participate in protein translation, likely by providing energy for protein synthesis, or by facilitating the recycling of factors involved in translation (1). Recent work has shown that Escherichia coli HflX acts as a ribosome-splitting factor under heat shock stress, whereby it binds to and splits stalled 70S ribosomal subunits (3, 4). In Staphylococcus aureus, HflX binds to and dissociates hibernating 100S ribosomes (homodimeric 70S) into 50S and 30S subunits, thereby recycling the pool of ribosomes for new rounds of translation during the stationary phase (5). A recent study has reported a similar ability for HflX expressed by fast growing mycobacteria species Mycobacterium abscessus and Mycobacterium smegmatis to bind to and split ribosomal subunits (6).

In Mycobacterium tuberculosis (Mtb), responsible for human tuberculosis (TB), hflX has been categorized as a nonessential gene using a transposon site hybridization (TraSH) library (7). In a number of transcriptomics studies, hflX gene expression was found to be influenced by a variety of stressors ranging from antibiotics and chemical exposure to environmental stresses such as nutrient starvation (8–12). Consistently, we previously reported that hflX is overexpressed in Mtb exposed to the hostile lysosomal environment of macrophages (13). Transcription of hflX was found to be linked with whiB7, a transcriptional regulator that controls intrinsic antibiotic resistance and redox homeostasis in Mtb (9, 14). Exposure to antibiotics targeting the ribosomal complex including streptomycin, erythromycin, tetracycline, and pristinamycin was found to induce both whiB7 and hflX expression (15). Consistently, absence of HflX in fast growing mycobacteria species M. abscessus and M. smegmatis increased antibiotic resistance to macrolide–lincosamide antibiotics (6).

In this study, we investigated the physiological role of HflX in slow-growing mycobacteria Micobacterium bovis bacillus Calmette–Guérin. We report that M. bovis bacillus Calmette–Guérin (BCG) HflX, which is 100% conserved with Mtb HflX, is a GTPase that is involved in response to hypoxia-induced nonreplicating state that allows tubercle bacilli to persist inside their host for extended periods and become phenotypically antibiotic-resistant (16). We provide experimental evidence that HflX interacts with ribosomal subunits and plays a master regulatory role in protein translation during transition to hypoxia.

Results

Mycobacterial HflX Is Involved in Adaptation to Hypoxia.

In the prokaryotic kingdom, HflX is widely distributed and conserved across species (1, 17).The amino acid sequence of Mtb HflX is 100% and 84.5% identical to M. bovis BCG HflX and Mycobacterium leprae HflX, respectively, while it shares about 45% identity within the GTPase catalytic site of E. coli HflX, including the P-loop, Switch I-II, and G1-G5 domains (SI Appendix, Fig. S1A). A three-dimensional computational model was constructed by adopting a previously described strategy (18). The Phyre2.0 modeling platform was employed to compare the predicted structure of Mtb HflX with that of E. coli HflX, whose crystal structure is available (Protein Data Bank entry: 5ADY) (3). A high degree of homology was observed visually and from the low rmsd scores, which supports the conservation of HflX structure and function between these two evolutionarily distant prokaryotes (SI Appendix, Fig. S1B).

To study the role of HflX in slow-growing mycobacteria, a M. bovis BCG HflX null mutant (∆hflX) and its complemented strain (∆hflX::phflX) were constructed (SI Appendix, Fig. S2A). RT-PCR revealed undetectable levels of hflX mRNA in M. bovis BCG ∆hflX, while the complemented strain displayed parental levels of hflx mRNA (SI Appendix, Fig. S2B). Comparable growth kinetics were observed between wild-type (WT), ∆hflX, and ∆hflX::phflX strains when cultured at 37 °C in standard 7H9 medium (SI Appendix, Fig. S2C), supporting that HflX is nonessential for in vitro growth in rich, aerated (normoxia) culture conditions.

We then assessed whether, similar to its E. coli counterpart (3, 19), mycobacterial HflX plays a role during heat shock. However, a comparable number of colony-forming units (CFUs) were obtained upon heat shock of the WT, ∆hflX, and complemented strains (Fig. 1A). Furthermore, codon-optimized M. bovis BCG hflX or homologous E. coli hflX were expressed in ∆hflx E. coli under the control of an arabinose-inducible promoter (E. coli ∆hflX::pBCGhflX and E. coli ∆hflX::hflX). The hflX mRNA levels in both strains were comparable and about 100 times higher than the endogenous level measured in WT E. coli (SI Appendix, Fig. S2D). Upon heat shock, expression of homologous hflX partially restored parental survival (60% as compared with WT), while codon-optimized M. bovis BCG hflX did not confer protection to ∆hflX E. coli (Fig. 1B). Thus together, these observations support that HflX is unlikely to be involved in the heat shock response in mycobacteria.

Fig. 1.

Phenotypic characterization of M. bovis BCG ΔhflX. (A) CFU counts from M. bovis BCG WT, ΔhflX, and complemented strains before and after heat shock (55 °C for 10 min). Data are expressed as the mean ± SD from three independent experiments (n = 3). (B) CFU counts from E. coli WT, ΔhflX, and ΔhflX complemented with homologous HflX or with codon-optimized M. bovis BCG/Mtb HflX after heat shock, and with (+) or without arabinose in the culture medium. Data are expressed as a percentage of survival relative to heat-treated WT. Data are expressed as the mean ± SD from three independent experiments (n = 3). One-way ANOVA with Bonferroni posttest. **P < 0.01, ***P < 0.001. ns, not significant. (C) OD₆₀₀ of M. bovis BCG WT, ΔhflX, and complemented strains in the gradual hypoxia Wayne model. Full decoloration of methylene blue indicator for ΔhflX (red arrow), or WT and complemented strains (blue arrow) is indicated. Data are expressed as the mean ± SD from four independent experiments (n = 4). Two-way ANOVA with Bonferroni posttest compared with WT. ***P < 0.001. (D) CFU counts for M. bovis BCG WT, ΔhflX, and complemented strains grown in the gradual hypoxia Wayne model. Data are expressed as the mean ± SD from four independent experiments (n = 4). One-way ANOVA with Bonferroni posttest compared with M. bovis BCG WT. *P < 0.05. (E) Representative images of M. bovis BCG WT, ΔhflX, and complemented strains obtained by scanning electron microscopy on days 0, 8, and 17 of the Wayne model. (Scale bar, 1 µm.). (F) Changes (in %) in size of M. bovis BCG WT, ΔhflX, and complemented mycobacteria on days 8 and 17 of the Wayne model, relative to day 0. Average sizes calculated based on ∼20 bacteria were counted for each strain. Two-way ANOVA with Bonferroni posttest. **P < 0.01, ns: not significant. (G) CFU counts from C57BL/6 mice infected with M. bovis BCG WT, ΔhflX, and complemented strains. Organs were harvested at weeks 8, 12, and 16 postinfection. One representative of two independent experiments is shown. Individual mouse data are plotted. Two-way ANOVA with Bonferroni posttest. *P < 0.05; ***P < 0.001.

To probe for a possible physiological role of mycobacterial HflX during adaptation to other stresses, M. bovis BCG WT, ∆hflX, and complemented strains were grown under various conditions, including macrophage infection, nutrient starvation (Loebel in vitro model) (20), and gradual oxygen depletion (21). No significant difference among the three strains was observed under nutrient starvation or during macrophage infection (SI Appendix, Fig. S3 A and B). In the gradual oxygen depletion model (i.e., Wayne model), however, differences were observed. In this in vitro model, mycobacterial growth is characterized by two stages, namely, the nonreplicating persistence stage 1 (NRP-1) or microaerophilic stage, during which mycobacteria actively replicate while oxygen gets progressively depleted in the sealed tube, eventually reaching 1.0% O₂ saturation; and the nonreplicating persistence stage 2 (NRP-2), characterized by an oxygen tension below 0.06%, and where mycobacteria have entered a nonreplicating state (21). In this model, M. bovis BCG ∆hflX was found to replicate faster than the WT and complemented strains during the microaerophilic phase between days 3 and 8, as evidenced by significantly higher the optical density at 600nm (OD_600nm) values and higher CFU counts (Fig. 1 C and D). From day 8 onward, OD_600nm values plateaued, indicating that M .bovis BCG ∆hflX has stopped replicating and has reached NRP-2 by day 14 (as indicated by complete decolorization of methylene blue indicator), which is significantly earlier than the WT and complemented strains that reached NRP-2 by days 18 and 17, respectively (Fig. 1 C and D). This observation thus pointed at a role for HflX in controlling growth rate during the microaerophilic phase and entry of mycobacteria into nonreplicating state.

Changes in cell morphology have been reported previously for nonreplicating mycobacteria grown under hypoxic conditions (22, 23). Analysis by scanning electron microscopy revealed 43% reduction in size for M. bovis BCG ∆hflX harvested at day 8 compared with its size at day 0, while the average size of WT and complemented strains at day 8 was comparable to their respective size measured at day 0 (Fig. 1 E and F). At day 21, the average size of WT and complemented strains decreased significantly compared with that measured at day 0, reaching a size that was similar to that measured with ∆hflX bacilli (Fig. 1 E and F). These observations therefore further supported that M. bovis BCG ∆hflX displayed a nonreplicative phenotype earlier than the WT and complemented strains.

The role of mycobacterial HflX was next investigated during infection in mammalian host, where oxygen saturation ranges between 1 and 14% depending on the organ, thereby likely exposing mycobacteria to microaerophilic environments (24). Upon intratracheal infection, the number of CFUs recovered at weeks 2 and 4 from the lungs, spleen, and lymph nodes of mice infected with WT, ∆hflX, and complemented strains were mostly comparable (SI Appendix, Fig. S3C). In contrast, from week 8 onward, which corresponds to the chronic phase of infection triggered by the host adaptive immunity (25), the bacterial loads measured in these organs were consistently higher in mice infected with ∆hflX compared with mice infected with WT and complemented strains (Fig. 1G). The higher bacterial loads recovered during the chronic phase of infection in mice were consistent with the higher replication rate observed with M. bovis BCG ∆hflX during the microaerophilic phase of the Wayne model. This observation may suggest that during the chronic phase of infection in mice, mycobacteria are exposed to microaerophilic environments, possibly within granuloma-like structures, which further supports a role for HflX in the physiological response of mycobacteria to low oxygen tension environments that are encountered during the course of infection in the human host.

Absence of HflX Impairs the Energetic Status of Hypoxic Mycobacteria and Leads to Phenotypic Drug Resistance.

We previously reported significantly lower intracellular ATP levels in hypoxic, nonreplicating mycobacteria (26). Here, while comparable intracellular ATP levels were measured at day 0 for the WT, ∆hflX, and complemented strains, significantly lower ATP levels were measured in M. bovis BCG ∆hflX at all the subsequent time points, except at day 21 (Fig. 2A). Furthermore, the membrane potential (Δψ) of M. bovis BCG ∆hflX was increased by 17–46% compared with WT and complemented strains (Fig. 2B), indicating that the plasma membrane of ∆hflX is hyperpolarized during growth in the Wayne model. Of note, addition of proton ionophore carbonyl cyanide m-chlorophenyl hydrazine (CCCP) caused a slight reduction only of RFU, presumably reflecting an incomplete depolarized state of the membrane. Altogether, our data indicated that under gradual oxygen depletion, M. bovis BCG ∆hflX displays significantly lower ATP levels compared with WT and complemented strains that may result from the high energy needs to sustain the faster replication rate observed with the mutant during the microaerophilic phase. The membrane hyperpolarization observed may reflect a compensatory mechanism to maintain the proton-motive force (PMF) for de novo ATP synthesis (26).

Fig. 2.

Energetic status, drug susceptibility, and expression of the dos regulon in M. bovis BCG ΔhflX. (A) Intracellular ATP level in M. bovis BCG WT, ΔhflX, and complemented strains grown in the Wayne model. Data are expressed as the mean of relative luminescence units (RLU) per CFU ± SD from three independent experiments (n = 3). One-way ANOVA with Bonferroni posttest compared with M. bovis BCG WT. *P < 0.05, **P < 0.01, ***P < 0.001, ns: not significant. (B) Membrane potential of M. bovis BCG WT, ΔhflX, and complemented strains grown in the Wayne model. CCCP was added to cultures as positive control (+). Data are expressed as the mean of relative fluorescence units (RFU) ratio ± SD of technical triplicates. One representative of two independent experiments is shown. Two-way ANOVA with Bonferroni posttest compared with M. bovis BCG WT. *P < 0.05; ***P < 0.001. (C) CFU counts from M. bovis BCG WT, ΔhflX, and complemented strains on day 8 or 17 of the Wayne model treated with various drugs as indicated. Data are expressed as the mean ± SD from three independent experiments (n = 3). Two-way ANOVA with Bonferroni posttest compared with drug-free counterpart (CTRL). *P < 0.05, **P < 0.01, ***P < 0.001. ns, not significant. (D) Relative gene expression of a subset of dos regulon genes measured by qRT-PCR in M. bovis BCG ΔhflX compared with WT on days 0, 4, 8, and 17 of the Wayne model. Data are expressed as the mean ± SD of three technical replicates. One representative of two independent experiments is shown. One-way ANOVA with Bonferroni posttest compared with WT counterpart. *P < 0.05, ns, not significant.

Furthermore, phenotypic drug resistance was investigated, a phenomenon that has been well described for nonreplicating mycobacteria and believed to explain the prolonged chemotherapy necessary to achieve sterility in TB patients (27). The minimum inhibitory concentrations (MIC) of various antimycobacterial drugs with different mechanisms of action, namely, bedaquiline (BDQ), isoniazid (INH), streptomycin (STM), rifampicin (RIF), chloramphenicol (CM), and ethambutol (ETB), were found comparable between WT, ∆hflX, and complemented strains grown under aerobic conditions (Dataset S1). In contrast, at day 8 of the Wayne model, M. bovis BCG ∆hflX was found highly resistant to BDQ, INH, and STM, compared with day 8 cultures of WT and complemented strains (Fig. 2C). At a later time point (day 17), when all three strains have reached NRP-2, they were all found resistant to BDQ, INH, and STM but remained susceptible to RIF (Fig. 2C), consistent with previous studies reporting that RIF is very effective at killing nonreplicating mycobacteria (28). The lack of killing efficacy observed with BDQ, previously shown to kill both actively replicating and nonreplicating mycobacteria (29), may be explained by the fact that this drug exerts a delayed killing (30) and that 5-d incubation may not be sufficient to observe significant killing of nonreplicating mycobacteria. Together, the drug resistance profile observed further supported that in the Wayne model, M. bovis BCG ∆hflX enters the nonreplicating stage earlier than its WT counterpart.

Finally, the relative expression of the dos regulon was also determined in M. bovis BCG ∆hflX compared with WT during growth in the Wayne model. The 48-member dos regulon, controlled by the two-component system DosS/T-DosR, has been shown to mediate mycobacteria transition to a nonreplicating state in response to various stresses, including low oxygen tension (31, 32). Results indicated increased transcriptional activity of dosR and hspX genes at day 8 in M. bovis BCG ∆hflX compared with WT, but not at the other time points, and not for the other dos genes surveyed (dosS/T, pfkB, acg, and tgs1) (Fig. 2D). This finding supported that HflX may have a limited regulatory role on the transcriptional activity of the dos regulon.

M. bovis BCG ∆hflX Displays Global Proteomics Changes.

Since HflX in other microorganisms has been shown to modulate translational activity through its ribosome-splitting activity, we employed tandem mass tag mass spectrometry (TMT-MS) to compare the proteomic profile between M. bovis BCG ∆hflX and its WT counterpart, when grown in the Wayne model. At day 0, corresponding to the aerobic phase for both strains, a total of 122 proteins were differentially expressed between both strains, with 66 underrepresented and 56 overrepresented proteins in ∆hflX compared with WT (Fig. 3 A and D). Gene ontology analysis indicated a significant enrichment among down-regulated proteins involved in lipid biosynthesis and metabolism; cell wall components biogenesis and assembly; valine, leucine, and isoleucine biosynthesis; protein folding; and response to copper ion (Fig. 3D and Dataset S1).

Fig. 3.

Differential protein expression in M. bovis BCG ∆hflX in response to hypoxia. (A–C) Volcano plot of differentially expressed proteins in M. bovis BCG ΔhflX compared with WT on day 0 (A, normoxia); day 8 WT vs. day 4 ∆hflX (B, exponential growth phase); and day 17 (C, NRP-2) of the Wayne model. (D–F) Gene ontology analysis (biological functions) of the differentially expressed proteins in M. bovis BCG ΔhflX compared with WT on day 0 (D, normoxia); day 8 WT vs. day 4 ∆hflX (E, exponential growth phase); and day 17 (F, NRP-2) of the Wayne model with false discovery rate (FDR) <0.05. Abbreviations: PDIM, phthiocerol dimycocerosate; ESAT-6, early secreted antigenic target-6.

To study protein content during the exponential growth phase of the Wayne model, we took into consideration the differential growth rate observed between both strains, and therefore, M. bovis BCG ∆hflX culture was harvested at day 4, whereas WT culture was harvested at day 8. A total of 137 proteins were found to be differentially expressed, with 68 underrepresented and 69 overrepresented proteins in ∆hflX compared with WT (Fig. 3 B and E and Dataset S1). Among the down-regulated hits, significant enrichment was seen in proteins involved in ESAT-6–like secretion pathway, sulfur compound metabolism, and Ubl conjugation. Among the up-regulated proteins, 11 belong to the Dos regulon (Fig. 3E and Dataset S1, in bold), among which M. bovis BCG_0112 (Rv0079) encodes a dormancy-associated translation inhibitor. Other pathways and processes involved in metabolism (lipid metabolism; propanoate metabolism; and valine, leucine, and isoleucine degradation), DNA replication, signaling (two-component systems), and universal stress proteins were also found overrepresented in M. bovis BCG ∆hflX compared with WT during the exponential replication phase. The up-regulation of DNA replication process (DnaA, DnaN, DnaE1, and PolA) correlated with the greater replication rate observed with M. bovis BCG ∆hflX during this phase. Furthermore and importantly, caseinolytic protease subunit X (ClpX) and SsrA-binding protein (SmpB) were found up-regulated in M. bovis BCG ∆hflX. These proteins are part of protein quality control and play an important role in rescuing stalled ribosomes mediated by trans-translation and maintaining protein homeostasis (33–35). This latter observation could suggest that absence of HflX resulted in accumulation of stalled ribosomes whose interrupted translation products need to be degraded.

Last, at day 17, when both strains have reached the nonreplicating state (NRP-2), only 42 proteins displayed differential abundance, with 27 down-regulated and 15 up-regulated proteins in M. bovis BCG ∆hflX. Proteins involved in response to starvation and copper ion were overrepresented in the mutant, while proteins involved in sulfur compound metabolism, cysteine synthesis, and oxidoreductase activity were underrepresented in ∆hflX compared with WT (Fig. 3 C and F and Dataset S1).

Together, the distinct proteomic profiles between M. bovis BCG ∆hflX mutant and WT grown in the Wayne model, involving diverse pathways such as central metabolism, cell wall biogenesis, DNA replication, and protein degradation, suggested a master regulatory role of HflX in translational activity during hypoxic stress. The overrepresentation of Dos proteins in M. bovis BCG ∆hflX during the replication phase correlates with earlier entry into the nonreplicating phase.

Furthermore, we also compared the proteomic profile at day 8 in the Wayne model between M. bovis BCG ∆hflX and WT, probing for further evidence to support that ∆hflX mutant has progressed into the nonreplicating stage. Consistently, among the 151 proteins that were found to be overrepresented in M. bovis BCG ∆hflX, 24 belong to the Dos regulon (Dataset S1, in bold). Furthermore, enrichment in ribosomal subunits was observed (Dataset S1, highlighted in yellow), as well as proteins involved in the formation of 70S hibernating ribosomes (hpf, Rv0079) and 70S ribosomes quality control (YbeY) (36–38). Therefore, these observations further support the earlier entry of M. bovis BCG ∆hflX into the nonreplicating stage and suggest a role for mycobacterial HflX in modulating ribosome abundance inside the cell.

Mycobacterial HflX Is a GTPase and Interacts with Ribosomal Subunits.

E. coli HflX binds at the E-site of 70S bacterial ribosomes and induces split into 50S/30S ribosomal subunits upon GTP hydrolysis (4). Using a biochemical approach, a recent study also supported that HflX produced by M. abscessus and M. smegmatis is a ribosome-splitting factor (6). To investigate whether HflX expressed by slow growing mycobacteria M. bovis BCG interacts with ribosomal subunits, we conducted a cell-based pull-down experiment combined with LC/MS analysis, using an anti-HflX monoclonal antibody (SI Appendix, Fig. S4 A and B). Results showed that the pull-down fraction was enriched in ribosomal proteins, namely, S6 and S17 of the 30S ribosomal subunits and L27 and L30 of the 50S ribosomal subunits, thus supporting that HflX binds to ribosomes (Table 1). Interestingly, two proteins from the Dos regulon (Rv0572c and Rv2003) were also pulled down (Table 1).

Table 1.

Proteins identified from anti-HflX pull-down

Accession	Gene	Protein name	Mass (Da)	co-IP score (emPAI)
A0A0H3M750	hflX	GTPase HflX	53,467	3.47
A1KKF4	Pup	Prokaryotic ubiquitin-like protein Pup	6,940	0.78
A0A0H3M2V9	rpsQ	30S ribosomal protein S17 rpsQ	14,863	0.74
A1KGK3	rpmD	50S ribosomal protein L30 rpmD	7,342	0.73
A0A0G2Q9J0	Rv3291c	Probable transcriptional regulatory protein (probably asnC-family)	16,586	0.65
A1KLD7	rpmA	50S ribosomal protein L27 rpmA	8,963	0.57
A0A0H3M4W6	Rv0968	Uncharacterized protein	10,257	0.49
A0A0H3M700	Rv2309a	Uncharacterized protein	10,648	0.47
A1KEM1	rpsF	30S ribosomal protein S6 rpsF	10,928	0.46
A0A0H3M4Y6	Rv0991c	Conserved hypothetical serine rich protein	11,679	0.42
A0A0H3M2K2	Rv0572c*	Uncharacterized protein	12,826	0.38
A0A0H3M725	glyS	Glycine–tRNA ligase glyS	53,019	0.37
A0A0H3M6K7	argD	Acetylornithine aminotransferase argD	41,055	0.36
A0A0H3MGK2	Rv2857c	Probable short-chain type dehydrogenase/reductase	27,016	0.36
A0A0H3M2I4	Rv0546c	Uncharacterized protein	14,337	0.34
A0A0H3M2Q7	hadB	Uncharacterized protein	14,895	0.32
A0A0H3M568	Rv2003*	Uncharacterized protein	31,315	0.31

Bolded, ribosomal subunits. Da, daltons; co-IP, co-immunoprecipitation; emPAI, exponentially modified protein abundance index.

^*Member of the dos regulon.

Furthermore, mycobacterial HflX was characterized biochemically, in particular its ability to hydrolyze GTP. Codon-optimized M. bovis BCG/Mtb HflX was expressed in and purified from E. coli (SI Appendix, Fig. S4 C and D), and in vitro enzymatic assays were carried out. Significant GTPase activity in the presence of MgCl₂ was measured with purified mycobacterial HflX (Fig. 4A). Direct interaction between mycobacterial HflX and GTP hydrolysis product, GDP, was demonstrated by isothermal titration calorimetry (ITC), with a dissociation constant K_d at 1.89 µM (Fig. 4B). Furthermore and importantly, incubation with purified 70S ribosomes significantly increased the GTPase activity of mycobacterial HflX (Fig. 4C). Of note, limited ATPase activity was noticed for the recombinant protein (SI Appendix, Fig. S4E).

Fig. 4.

GTPase activity of recombinant mycobacterial HflX. (A) Quantification of inorganic phosphate (IPO₄) released over time in the presence of GTP and purified recombinant Mtb HflX, with or without MgCl₂. Data are expressed as the mean ± SD of three technical replicates. One representative of three independent experiments is shown. Two-way ANOVA with Bonferroni posttest. ***P < 0.001, ns, not significant. (B) Binding of HflX to GDP assessed by ITC. Binding curve was obtained by integrating the DP signal. One representative of two independent experiments is shown. (C) Differential release of IPO₄ at 24 h in the presence of GTP, recombinant M. bovis BCG/Mtb HflX, MgCl₂; and with or without mycobacterial ribosomes 70S. Data are expressed as the mean ± SD from two independent experiments (n = 4) and as fold change compared with GTP control. One-way ANOVA with Bonferroni posttest. **P < 0.01, ***P < 0.001. (D) Quantification of IPO₄ released over time in the presence of GTP and MgCl₂ with purified recombinant M. bovis BCG/Mtb HflX or GTPase-deficient HflX (AAY). Data are expressed as the mean ± SD of three technical replicates. One representative of three independent experiments is shown. Two-way ANOVA with Bonferroni posttest. *P < 0.05, ***P < 0.001, ns: not significant. (E) OD₆₀₀ of WT and GTPase-deficient M. bovis BCG strains grown in 7H9 (normoxia). Data are expressed as the mean ± SD from two independent experiments (n = 2). (F) OD₆₀₀ of WT and GTPase-deficient M. bovis BCG strains grown in the Wayne model. Data are expressed as the mean ± SD from two independent experiments (n = 2).

A recombinant HflX mutant harboring a triple amino acid substitution (AAY) in the predicted GTPase catalytic site was also produced and purified from E .coli (SI Appendix, Fig. S4 C and D). The GTPase-deficient HflX recombinant protein was found unable to either hydrolyze GTP (Fig. 4D) or bind to GDP (SI Appendix, Fig. S4F). Furthermore, a M. bovis BCG strain expressing GTPase-deficient HflX was generated. Interestingly, while this strain displayed no growth defect under normoxic growth (Fig. 4E), it was severely impaired in the Wayne model (Fig. 4F), thus indicating that production of GTPase-deficient HflX is detrimental to the mycobacterial cell under hypoxic conditions.

Together, these data establish that mycobacterial HflX is a GTPase with minimal ATPase activity. Its enhanced GTPase activity in the presence of 70S ribosomes further supports the ability of HflX to interact with ribosomal subunits.

Differential Ribosome Footprint in Bacillus Calmette–Guérin ∆hflX.

The total protein content per bacteria, expressed as pg/CFU, was drastically lower in M. bovis BCG ΔhflX at day 8 compared with WT, whereas protein content was comparable between both strains at days 0 and 17 (Fig. 5A). While this observation seems counterintuitive when considering the greater replication rate in M. bovis BCG ΔhflX, it may be due to the significantly smaller size of ΔhflX cells compared with WT (Fig. 1E).

Fig. 5.

Translational activity in M. bovis BCG ∆hflX. (A) Quantification of total protein content/bacterial cell in M. bovis BCG ΔhflX and WT at days 0, 8, and 17 of the Wayne model. Data are expressed as the mean ± SD from two independent experiments (n = 2). Two-way ANOVA with Bonferroni posttest. *P < 0.05, ns, not significant. (B) Ribosome-sequencing data showing the percentage of mapped reads to respective regions of M. bovis BCG ΔhflX compared with WT on day 8 of the Wayne model. CDS: coding sequence; Other: 5′ and 3′ untranslated regions (UTRs). Data are expressed as the mean ± SD from two independent experiments (n = 4). Two-way ANOVA with Bonferroni posttest. *P < 0.05. (C) Distance from the stop codon stop of 15- to 26-nt long RPF in M. bovis BCG ∆hflX and WT strains. Negative values indicate that RPF are located before the stop codon, whereas positive values indicate that RPF are found after the stop codon. (D) Volcano plot of differential TE between M. bovis BCG ΔhflX and WT on day 8 of the Wayne model. Highlighted in a lighter shade of gray are genes with log2TE > 1 or log2TE < 1. False discovery rate <0.05. (E and F) Gene ontology and KEGG pathways analysis of the differentially translated genes in M. bovis BCG ΔhflX compared with WT on day 8 of the Wayne model. Enriched pathways (E, increased TE; F, decreased TE) were selected where false discovery rate was <0.05.

We next investigated by ribo-sequencing the translational activity in both strains at day 8 in the Wayne model. Ribo-sequencing allows us to identify ribosomally protected footprints (RPF), covering both coding mRNA sequence (CDS) and canonical noncoding RNAs (ncRNAs) such as tRNAs and rRNAs, thereby providing a snapshot of the translational activity at the point of investigation (39).

The 15- to 40-nucleotide (nt)-long fragments were selected for Ribo-seq analysis which covers all RNA species that were either protected by ribosomes (RPF of the CDS) or bound to ribosomes (transfer RNAs [tRNAs] and ribosomal RNAs [rRNAs]). Interestingly, there was a significant enrichment in tRNA in the Ribo-seq fraction from M. bovis BCG ΔhflX compared with WT (Fig. 5B), while the percentage of reads mapped to rRNA, CDS, and 5′ and 3′ untranslated region (Others) were generally higher in WT (Fig. 5B). Since total RNA-seq data indicated that the percentage of reads mapped to the respective regions (tRNA, rRNA, CDS, and Other) was similar between both strains (SI Appendix, Fig. S5A), this observation suggested enriched tRNA pool and lower translational activity of the other RNA species in M. bovis BCG ΔhflX, the latter being consistent with the mutant’s nonreplicating status at the time point studied.

Comparable read length and read length distribution were seen between bacillus M. bovis BCG ΔhflX and WT across the various RNA species (SI Appendix, Fig. S5 B and C). Of note, the mean RPF read length mapped to CDS in ΔhflX (23 nt) and WT (21 nt) was substantially shorter than a recent study that reported mean RPF read lengths between 24 and 27 nt in bacteria (40). The read length distribution in CDS RPF displayed a left-skewed pattern with the highest fraction of read length concentrated at 15–18 nt and a second smaller fraction of 24- to 26-nt RPF reads (SI Appendix, Fig. S5C). Metagene analysis revealed that the bulk of shorter RPFs (15–18 nt) generally showed a significant 3-periodicity and an overrepresentation in coding regions (5′ CDS and 3′ CDS) for both WT and ΔhflX strains (SI Appendix, Fig. S5D). Interestingly, significant accumulation of 3′ UTRs in these short RPFs was seen in ΔhflX compared with WT. A metagene plot mapped the accumulation of these 3′ UTRs at 20–50 nt downstream of the stop codon, whereas WT displayed an increase in density of short RPFs predominantly upstream of the stop codon (Fig. 5C). Short RPF fragments and enrichment of 3′ UTRs have been associated with the presence of stalled ribosomes due to impairment of ribosome recycling factors (41, 42).

Furthermore, the overall expression rate of each gene was calculated based on read density and expressed as reads per kilobase per million. With the selection of CDS RPF length range between 27 and 37 nt which should correspond to actively translating ribosomes, we were able to reliably quantify translation rates for 3,886 genes, representing 99.8% of the annotated M. bovis BCG translatome. Plotting of gene expression rates for RPF and RNA revealed a high degree of correlation between translatome and transcriptome in M. bovis BCG ΔhflX (Spearman’s r = 0.85) and in WT (Spearman’s r = 0.79) (SI Appendix, Fig. S5E). Furthermore, the translation efficiency (TE) of 1,280 CDS was found to be significantly different between ΔhflX and WT, among which 835 had a lower TE in ΔhflX, representing ∼22% of the total translatome (Fig. 5D). Interestingly, genes coding for ribosomal subunits were among those with the greatest TE increase in M. bovis BCG ΔhflX (Fig. 5E). This observation could suggest a role for HflX in regulating the translation of ribosomal subunits. Genes involved in metabolic pathways including carbon metabolism, citrate cycle, and oxidative phosphorylation were also found to have greater TE in M. bovis BCG ΔhflX (Fig. 5E). This may represent a feedback response to the lower ATP pool measured in ΔhflX. Among the genes with significantly lower TE in M. bovis BCG ΔhflX, the PPE family genes, genes involved in response to stimuli, and two-component systems were enriched (Fig. 5F).

Together, this in-depth ribo-seq analysis has revealed significant differences in translational activity between M. bovis BCG ΔhflX and WT cultures grown under hypoxic conditions. While some of the differences observed may be explained by the differential replicating status between both strains at the time point of study, some may result from the lack of HflX in M. bovis BCG ΔhflX mutant and suggest that, in line with the known function of HflX in other microorganisms, mycobacterial HflX modulates the translational activity through its interactions with ribosomal subunits.

Discussion

Mtb can survive for decades in a nonreplicating state, causing a clinically asymptomatic, noninfectious form of the disease that is known as latent TB infection (LTBI) (16). It is estimated that about one third of the world’s population has LTBI, providing a large reservoir for reactivation to active, contagious disease. The ability of dormant Mtb to exhibit a form of noninheritable resistance to most of the currently available anti-TB drugs (i.e., phenotypic drug resistance) explains the long treatment regimen needed to achieve sterilization and has impeded the efforts in TB elimination. During latent infection, nonreplicating Mtb bacilli localize within granuloma, an organized structure of immune cells intended to constrain the infection (43, 44). The hypoxic microenvironment of granuloma is believed to trigger replication arrest in pathogenic mycobacteria (45). The dormancy survival regulon, i.e., dos regulon, is regulated by the two-component system Dos S/T and DosR and comprises 48 genes, which are essential for hypoxic survival (31). However, the molecular mechanisms involved in the hypoxic response and replication arrest of pathogenic mycobacteria have remained elusive. Our present work has identified the highly conserved GTPase HflX as a mycobacterial factor that plays an important role in the pathogen’s response to its hypoxic environment. Our findings are in line with previous reports on the role of HflX in stress adaptation in other distantly related microorganisms (3, 5). However, instead of heat shock, mycobacterial HflX responded specifically to oxygen limitation, a physiologically relevant stress encountered by mycobacteria in their host environment. Our data support that HflX controls the replication rate in slow growing pathogenic mycobacteria, thereby controlling entry into nonreplicating state. We propose that mycobacterial HflX exerts this function through modulation of the translational activity. This hypothesis is supported by several lines of experimental evidence. First, extensive proteomic changes were seen in M. bovis BCG ΔhflX during the microaerophilic phase of the Wayne model compared with WT. Second, we demonstrated that M. bovis BCG/Mtb HflX is a GTPase whose activity is significantly enhanced in the presence of 70S ribosomes. Third, we showed that HflX is a ribosome-interacting protein, as evidenced by the enrichment in ribosomal subunits in the pull-down fraction (Table 1). Although the biochemical function and mechanism of mycobacterial HflX remains to be demonstrated, these findings are in line with the ribosome-splitting activity of HflX described in E. coli and S. aureus (3, 5), as well as in fast growing mycobacteria species (nontuberculous mycobacteria [NTM]) M. abscessus and M. smegmatis (6). Interestingly, while the latter study reported that HflX-deficient M. smegmatis and M. abscessus displayed resistance to macrolide–lincosamide, we did not observe any drug resistance phenotype with M. bovis BCG ΔhflX mutant grown in normoxia, including macrolides such as erythromycin (Dataset S1). Macrolide–lincosamide has been used effectively to treat NTM infections (46, 47). However, mycobacteria from the Mtb complex (which includes M. tuberculosis, M. bovis BCG, and others, but not M. smegmatis or M. abscessus) have been found to be intrinsically resistant to macrolides due to the presence of Erm methyltransferase (ErmMT) that confers resistance to macrolide–lincosamide–streptogramin by methylation of 23S rRNA (48).

The hallmarks of nonreplicating bacilli include low energy profile and global protein synthesis down-regulation (49–51). In E. coli, global translation shutdown is associated with ribosome dimerization into a 100S ribosome species, which is translationally inactive when conditions are not favorable for bacterial growth (52). Under hypoxic stress, Mtb 70S ribosomes do not dimerize into 100S but associate with hibernating promoting factor (HPF) and ribosome-associated factor during hypoxia (RafH) into a stable complex (37, 38, 53). Ribosome stabilization is a strategy deployed by bacteria for stress management, so when cellular conditions become favorable, the hibernating ribosomes get disassembled and quickly recycled for new rounds of translation (52). The plasticity of hibernating ribosome disassembly has been proposed to play an essential role in the TB disease reactivation process (38, 54). It has been shown that E. coli and S. aureus HflX rescued stalled ribosomes and hibernating 100S ribosomes by splitting them into 50S and 30S subunits, allowing translation to resume (3, 5). We found an increased abundance in HPF and RafH proteins and reduced total protein content in M. bovis BCG ΔhflX under oxygen limitation (day 8 of the Wayne model). Furthermore, our ribo-seq data indicated enrichment of 3′ UTR sequences across short RPFs in nonreplicating M. bovis BCG ΔhflX, supporting ribosome stalling. These observations thus suggested that under hypoxia, HflX controls the amount of ribosomal subunits available for translation by splitting hibernating and/or stalled ribosomes after the stop codon, thereby controlling the overall translational activity, hence entry into the nonreplicating state. Of note, 3′ UTRs were recently found to regulate the decay and translation initiation in their mRNAs and have thus been proposed as a new class of posttranscriptional regulatory elements (55).

The proposed ribosome-splitting activity of mycobacterial HflX to free stalled ribosomes during hypoxic growth, however, does not explain the faster replication rate observed with M. bovis BCG ΔhflX during the microaerophilic phase of the Wayne model and the increased bacterial burden in the chronic phase of infection in mice. We propose here that in pathogenic mycobacteria, HflX could play other regulatory roles such as modulating ribosome biogenesis; this hypothesis is supported by the up-regulation of translation efficiency of 30S and 50S ribosomal subunits seen in M. bovis BCG ΔhflX. Several proteins from the TRAFAC family have been previously reported to regulate ribosome biogenesis and assembly, which has been shown to impact chromosome segregation, cell division initiation, and eventually growth rate (56–58). Alternatively, one could speculate that HflX may be able to interact with actively translating ribosomes, thereby modulating the overall translational activity, hence controlling bacterial growth rate. The observation that a M. bovis BCG mutant expressing GTPase-deficient HflX is severely impaired under hypoxia may support this idea, whereby GTPase-deficient HflX would bind to actively translating ribosomes, but would be unable to split them, thereby freezing the entire translation machinery.

Overall, our work ascribes a physiological function to the highly conserved HflX GTPase in slow-growing pathogenic mycobacteria and provides further insights into the mechanisms by which this pathogen adapts to its environment. Such fundamental knowledge may help design alternative strategies to accelerate or potentiate the killing efficacy of current TB drugs.

Materials and Methods

Referenced details of the materials and methods, including plasmids, strains, oligonucleotides, bacterial culture conditions, macrophage infection assay, and mice experiments are provided in SI Appendix. Assays measuring intracellular ATP, membrane potential, and GTPase activity followed previously described methods. Cloning, RNA extraction, complementary DNA (cDNA) synthesis, and qRT-PCR followed the manufacturers’ instructions. ITC consisted of injecting the ligands (GDP, GTP, GMP-PNP) into a chamber containing purified HflX protein and analyzing the thermodynamics using MicroCal PEAQ-ITC software as per the manufacturer’s instructions. Monoclonal anti-HflX antibody used for immunoprecipitation experiments was obtained upon immunizing mice with purified mycobacterial HflX and generating and screening hybridomas as described previously (59). Proteomics was performed using Tandem Mass Tag spectrometry where the bacterial samples were subjected to protein extraction and trypsin digestion before labeling with isobaric mass tags according to the manufacturer’s recommendations. Labeled samples were fractionated by reverse-phase high-performance liquid chromatography (HPLC) and analyzed by liquid chromatography with tandem mass spectrometry (LC-MS/MS). Samples for Ribo-seq analysis were prepared in-house, and Ribo-seq analysis was performed by TB-SEQ, Inc.

Data Availability

All study data are included in the article and/or supporting information.