Mycobacterium tuberculosis partitions the Krebs cycle under iron starvation.

Introduction

Over the past decade, mounting evidence has indicated a clear correlation between central metabolism and antibiotic resistance. It has been demonstrated that clinically relevant mutations in genes encoding core metabolic enzymes are associated with antibiotic resistance ([Lopatkin et al., 2021](#ref-Lopatkin et al., 2021)) and tolerance ([Hicks et al., 2018](#ref-Hicks et al., 2018)). Furthermore, secondary or ‘collateral’ effects of antibiotic action involve alterations to central carbon metabolism (CCM) ([Wang et al., 2019](#ref-Wang et al., 2019);[Kawai et al., 2019](#ref-Kawai et al., 2019)) and specific metabolic states affect their efficacy ([Allison et al., 2011](#ref-Allison et al., 2011);[Lim et al., 2021](#ref-Lim et al., 2021);[Lopatkin et al., 2019](#ref-Lopatkin et al., 2019);[Schrader et al., 2021](#ref-Schrader et al., 2021)).

Mycobacterium tuberculosis(Mtb),the causative agent of tuberculosis (TB), has been recently included in the Bacterial Priority Pathogen List of the World Health Organization ([WHO, 2024](#ref-WHO, 2024)) due to its intrinsic resistance to antibiotics and the increased spreading of multi-drug-resistant strains. Since the 1960s, only three new drugs have been approved for the treatment of TB ([Murray et al., 2015](#ref-Murray et al., 2015);[Zheng and Av-Gay, 2016](#ref-Zheng and Av-Gay, 2016);[Dartois and Dick, 2024](#ref-Dartois and Dick, 2024)), and resistance to these new antibiotics is increasing. It is therefore critical that new medicines are developed to treat this infection. For more efficient antibiotic discovery, we must gain a more comprehensive understanding of the physiology and metabolism ofMtb. This knowledge will help identify novel vulnerabilities and inform the design of more effective therapeutic strategies.

During infection, pathogenic bacteria are subject to nutritional immunity, a process in which the host alters metal availability to intoxicate or starve the invading pathogens ([Murdoch and Skaar, 2022](#ref-Murdoch and Skaar, 2022)). Metals are essential micronutrients that play structural, signalling, or catalytic roles in most cellular processes. Many enzymes of central metabolism depend on metals for their activity; however, the effects of nutritional immunity on CCM have not been extensively investigated ([Serafini, 2021](#ref-Serafini, 2021)). Experimental evidence indicates thatMtbis exposed to iron starvation during infection ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Wells et al., 2013](#ref-Wells et al., 2013);[Fang et al., 2015](#ref-Fang et al., 2015);[Madigan et al., 2015](#ref-Madigan et al., 2015);[Pisu et al., 2020](#ref-Pisu et al., 2020)). This study aims to examine the relationship between iron homeostasis and CCM, a complex and understudied aspect ofMtbmetabolism.Mtbexposed to prolonged and severe iron starvation exhibits a reduction in growth rate until replication is arrested; yet, cells remain viable ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)). Based on the publicly available transcriptomic and proteomic data of H37RvMtbstrain grown under iron limitation (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Wong et al., 1999](#ref-Wong et al., 1999);[Serafini et al., 2013](#ref-Serafini et al., 2013)), we have recently proposed a model that explains how this pathogen might adapt its core of CCM to limited iron availability ([Serafini, 2021](#ref-Serafini, 2021)). We proposed thatMtb: (i) reduces the activity of the Krebs cycle due to the compromised activity of its iron-dependent enzymes, which consequently reduces the production of FADH2, NADH, and ATP; (ii) activates an iron-independent pathway, the glyoxylate shunt, to maintain succinate and malate production; and (iii) secretes succinate to maintain an energised cell membrane, bypassing the reduced ATP synthase activity.

The aim of the present study is to test these predictions, thereby clarifying the effect of iron starvation onMtbmetabolism. We monitored metabolite levels using liquid chromatography coupled to mass spectrometry and employed13C isotope tracing to identify the metabolic pathways active under iron starvation. These experiments were conducted under the same growth conditions used in the more comprehensive transcriptomic study ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)), which, like the other published studies on iron limitation ([Wong et al., 1999](#ref-Wong et al., 1999);[Serafini et al., 2013](#ref-Serafini et al., 2013)), included two carbon sources in the growth medium, glucose and glycerol. To circumvent genomic background-dependent effects, the work was performed in twoMtbstrains with different metabolic capacities, H37Rv and Erdman ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Marrero et al., 2010](#ref-Marrero et al., 2010)).

Results

Severe growth defects accompany iron limitation

As demonstrated previously ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)),Mtbexposed to severe iron starvation enters a non-replicative state that can persist for months. This severe iron starvation was achieved by subculturing bacteria in the absence of a source of Fe3+, followed by the addition of the Fe3+chelator deferoxamine (DFO), to trap any residual ion. Equivalent conditions were used in our laboratory to reproduce the growth arrest phenotype in Erdman and H37Rv (Figure 1A–D). As expected, the absence of a Fe3+source (0 μM FeCl3) resulted in a growth slowdown, while the addition of DFO (0 μM FeCl3+DFO) led to growth arrest (Figure 1A and B). In the latter, Erdman exhibited an apparent decline in viability over several weeks (Figure 1D) compared to H37Rv (Figure 1C). This decline could be attributed to a loss of viability, or more pronounced clumping of the cells in Erdman cultures (data not shown), that resulted in a reduction in the number of colony-forming units (CFUs), or to the cells entering an unculturable state ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)).

*Survival to Fe3+starvation inM. tuberculosis(Mtb) H37Rv and Erdman strains.Cells were exposed to 50 μM FeCl3(HI: high iron), 0 μM FeCl3(LI: low iron), or 0 μM FeCl3+DFO (deferoxamine [DFO]). Growth was monitored for 3 weeks by measuring OD600(A, B), and survival was monitored for more than 4 weeks by colony-forming unit (CFU)/mL (C–F). The charts show one experiment representative of two to three independent experiments. The CFU/mL charts show average and average deviation of two technical replicates of one independent experiment. (G, H) ATP levels and growth (OD600) after 1, 2, 3, and 8 days in DFO. ATP levels were calculated as µM of ATP molecules in about 107cells (0.1 optical density at 600 nm). The data are the average and standard deviation of three independent experiments and three technical replicates each. (I) NADH/NAD+ ratio detected after 3 days of exposure to DFO. The data are the average and standard deviation of two independent experiments and two technical replicates. The p-values were calculated against the HI condition. =p-value<0.05.

Previous studies have shown that replication-arrest-inducing stress conditions such as hypoxia and nutrient starvation ([Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013);[Rao et al., 2008](#ref-Rao et al., 2008);[Gengenbacher et al., 2010](#ref-Gengenbacher et al., 2010)) reduce ATP production and alter the NADH/NAD+ratio. Transcriptomics data shows that upon Fe3+depletion,Mtbdownregulates the operon encoding ATP synthase and the type I NADH dehydrogenase (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)). We examined the levels of ATP and the NADH/NAD+ratio in H37Rv and Erdman strains. Contrary to expectations, ATP levels (Figure 1G and H,Figure 1—figure supplement 1A and B) were not reduced upon Fe³+depletion, remaining comparable between HI and DFO conditions in H37Rv over 17 days. Strikingly, in the Erdman strain, ATP levels tended to be higher under DFO conditions relative to HI. In contrast, we did see a significant increase in the NADH/NAD+ratio upon Fe3+depletion compared to the presence of sufficient Fe3+(Figure 1I,Figure 1—figure supplement 1C and D), indicating an accumulation of NADH. The concordant results between the two wild-type strains and the higher levels of ATP in Erdman strain in DFO condition support the hypothesis that the loss of viability in Erdman is not due to cell death and indicates that inMtbFe3+starvation causes a metabolic and physiological stress that is distinct from that observed in hypoxia and nutrient starvation ([Rao et al., 2008](#ref-Rao et al., 2008);[Gengenbacher et al., 2010](#ref-Gengenbacher et al., 2010);[Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013)).

Slowdown of the Krebs cycle activity

Most genes encoding enzymes involved in the Krebs cycle exhibit a reduction in their expression levels over a 2-week period in Fe3+-starved H37Rv (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)). This finding indicates that the activity of this pathway may be reduced in such a condition. The levels of Krebs cycle intermediates were examined following an 8-day exposure to Fe3+-replete conditions (50 μM FeCl3, termed HI henceforth) and Fe3+-limiting conditions (0 μM FeCl3and 0 μM FeCl3+DFO, termed LI and DFO, respectively, henceforth). As the metabolomic analysis demonstrated highly comparable outcomes in the LI and DFO conditions, the subsequent sections will focus on the comparison between HI and DFO. Please refer to supplementary data (SI Appendix, Section I) for a detailed analysis in the LI condition.

In H37Rv, a consistent increase, exceeding a 100-fold change (FC), in the intracellular levels of pyruvate and α-ketoglutarate was observed in DFO compared to HI (Figure 2A). The accumulation of these two metabolites is in line with the downregulation of theaceE,dlaT,andlpdCgenes (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)), which encode subunits of the pyruvate dehydrogenase (PDH) and the α-ketoglutarate dehydrogenase (KDH) complexes ([Tian et al., 2005](#ref-Tian et al., 2005);[Shi and Ehrt, 2006](#ref-Shi and Ehrt, 2006);[Maksymiuk et al., 2015](#ref-Maksymiuk et al., 2015)). This accumulation is consistent with the existence of two stalling points at which the activity of the Krebs cycle is impeded, one before the cycle (PDH) and one at the KDH level. It is likely that the blockage at the level of pyruvate results in a slowdown in the flux through glycolysis. The downregulation of glycolytic genes (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)) lends support to this hypothesis. A smaller increase in the levels of (iso)citrate was observed, with a 2- to 4-FC in two independent experiments (Figure 2A). This increase may be attributed to two potential mechanisms: (i) citrate accumulation resulting from the reduced activity of the iron-dependent aconitase (Acn), which is also known to be downregulated ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Wong et al., 1999](#ref-Wong et al., 1999);[Serafini et al., 2013](#ref-Serafini et al., 2013)); and (ii) isocitrate accumulation due to blockage of the Krebs cycle at the KDH level. The major fold increase (∼100-fold) in αKG compared to (iso)citrate is consistent with the irreversibility of the isocitrate dehydrogenase reaction, which drives carbon flux from isocitrate towards αKG but not in the reverse direction. As a result, isocitrate cannot accumulate to the same extent. Further, the intracellular threshold of these two metabolites may differ, leading the cells to divert (iso)citrate into other metabolic pathways and thereby limiting its accumulation.

*Intracellular and extracellular levels of metabolites in H37Rv.Cells were exposed to 50 μM FeCl3(HI: high iron), 0 μM FeCl3(LI: low iron), or 0 μM FeCl3+DFO (deferoxamine [DFO]). The analysed metabolites are shown in black in the schematic pathways. (A) Intracellular polar metabolites levels at 8 days; the y-axis is shown on Log10scale. Tick labels (0–5) represent the exponents of 10 (100–105); values are reported in arbitrary units. (B) Extracellular polar metabolites levels at 1, 3, and 8 days. The plots show the normalised levels of metabolites. The data represent the average and the standard deviation of four biological replicates from an independent experiment, representative of two independent experiments. The y-axis is shown on Log10scale; values are reported in arbitrary units. (C) Viability of H37Rv from one independent experiment. Cells were grown in liquid medium in HI and DFO conditions and in the presence of 2 mM of succinate. Aliquots of cells were collected after 0, 3, and 8 days and diluted to the same final OD600; 5 μL of 10-fold serial dilutions were plated on 7H10. Growth was recorded after 19–25 days. The p-values were calculated against the HI condition and independently for the two experiments; the highest p-value was reported. n.s.f.=non-significant fold change; the observed trend change was different between independent experiments; n.s.=non-significant, p-value>0.05; n.d.=non-detected; *=p-value<0.05; *=p-value<0.01. Ac-CoA: acetyl-CoA; CIT: citrate; FUM: fumarate; α-KG: α-ketoglutarate; ISO: isocitrate; (ISO)CIT: isocitrate and citrate; MAL: malate; OAA: oxaloacetate; PYR: pyruvate; SUCC: succinate; SUCC-CoA: succinyl-CoA.

In H37Rv, the succinate pool exhibited a statistically significant decrease in DFO, with intensity varying from 0.3 to 0.8 FC in two independent experiments (Figure 2A). No relevant alterations were observed in the levels of fumarate and malate. These results seem to contrast with the observation that genes encoding iron-dependent enzymes involved in the synthesis of fumarate and malate (fumarate reductase/Frd, fumarase/Fum, and succinate dehydrogenase/Sdh) are downregulated under Fe3+starvation (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)) and therefore their activity should be lower in such conditions. Surprisingly, the genes encoding the α-ketoglutarate ferredoxin-oxidoreductase KorAB complex ([Baughn et al., 2009](#ref-Baughn et al., 2009)) are induced ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)), suggesting that despite its iron-dependent nature, its activity is necessary to produce succinyl-CoA and then succinate from α-ketoglutarate.

As observed in H37Rv, in the Erdman strain (Figure 2—figure supplement 1A), there is a substantial intracellular accumulation of pyruvate and α-ketoglutarate (>10 FC) and a more modest accumulation of (iso)citrate (1.2–2.8 FC, in four independent experiments) in DFO. No major changes were observed in the levels of malate, fumarate, and succinate.

In conclusion, when experiencing Fe3+starvation,Mtbslows down the activity of the Krebs cycle, and pyruvate conversion to acetyl-CoA and α-ketoglutarate conversion into succinyl-CoA may represent two distinct checkpoints in this process.

Substantial secretion of metabolic intermediates

The secretion of succinate has been linked to the maintenance of membrane potential in hypoxic growth-arrestedMtb([Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013)). It has been demonstrated that in such conditions the extracellular succinate levels increase and that the inhibition of its secretion (by supplementing the medium with succinate) causes alteration of the proton motive force and cell death ([Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013)). We hypothesised that succinate could have a similar role in DFO-treated cells ([Serafini, 2021](#ref-Serafini, 2021)). Levels of extracellular succinate were therefore monitored over a week period (1, 3, and 8 days). Unexpectedly, no change in succinate levels was observed in DFO, rather, a significant increase was noted in HI (Figure 2BandFigure 2—figure supplement 1B). Furthermore, the succinate levels were lower in DFO compared to the HI condition. SinceMtbcells remain viable in DFO (Figure 1C and D), it can be concluded that the secretion of succinate is not a crucial factor in maintaining cell survival and membrane potential in such conditions. This conclusion was confirmed by the observation that DFO-treated cells exposed to 2 mM succinate in order to inhibit its secretion ([Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013)) remained viable (Figure 2CandFigure 2—figure supplement 1C), in contrast to the results observed during hypoxia.

Surprisingly, higher levels of several Krebs cycle intermediates were found in the culture filtrate of DFO compared to HI cultures, with minor discrepancies between the H37Rv (Figure 2B) and Erdman (Figure 2—figure supplement 1B) strain. α-Ketoglutarate, (iso)citrate, malate, and pyruvate are secreted in substantial quantities, and their levels increase over time, with the exception of fumarate and succinate. The levels of pyruvate and α-ketoglutarate increase by more than 10-fold, while those of (iso)citrate rise by 2- to 10-fold under Fe3+starvation. The high levels of extracellular pyruvate, α-ketoglutarate, and (iso)citrate are consistent with the elevated intracellular levels. It is noteworthy that malate levels exhibit a substantial increase, approximately 10-fold, in DFO compared to HI conditions despite similar intracellular levels under the two conditions. This increase occurs after 1 day of DFO exposure in H37Rv (Figure 2B) and at a later stage in the Erdman strain (Figure 2—figure supplement 1B). The analysis of13C incorporation in extracellular metabolites is consistent with the altered flow of carbon through the Krebs cycle (Figure 2—figure supplement 2A–Hand SI Appendix, Section II).

The decrease in CFU/mL in Erdman (Figure 1D) raised the possibility of cell lysis that could explain the increase of extracellular malate despite unchanged intracellular levels in DFO compared to HI conditions. If increased lysis had occurred, we would expect an overall increase in extracellular metabolites. However, extracellular fumarate and succinate levels did not increase in DFO both in Erdman strain and in H37Rv (Figure 2BandFigure 2—figure supplement 1B). We also examined the levels of extracellular glutamate, an amino acid whose abundance increases over a week in HI. As for fumarate, extracellular glutamate levels remained unchanged or decreased in DFO (Figure 2—figure supplement 2I and J) in Erdman and H37Rv. Therefore, we conclude that the observed increase in abundance of specific metabolites is not due to cell lysis.

As the intracellular malate pool is constant under HI and DFO conditions, this increase in malate secretion suggested that this metabolite may be utilised for the maintenance of membrane potential. The viability of DFO-treated cells in the presence of malate was evaluated (Figure 2—figure supplements 1C and 3A), but no effect was observed on survival. The substantial secretion of several metabolites prompted the hypothesis that they may collectively contribute to the maintenance of the membrane potential. The experiment was then repeated by adding pyruvate, α-ketoglutarate, succinate, malate, and fumarate individually (Figure 2—figure supplement 1CandFigure 2—figure supplement 3A) and in combination, yet again we saw no loss of viability. These results suggest that the secretion of these metabolites under Fe3+starvation is unrelated to the maintenance of the membrane potential.

Partitioning of carbon flux into oxidative and reductive branches of Krebs cycle

The transcriptomic data from Fe3+-starved H37Rv ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)) shows upregulation of isocitrate lyase 1 (icl1/Rv04671) ([Höner Zu Bentrup et al., 1999](#ref-Höner Zu Bentrup et al., 1999)) and phosphoenolpyruvate carboxykinase A (pckA/Rv0211) ([Marrero et al., 2010](#ref-Marrero et al., 2010);[Machová et al., 2014](#ref-Machová et al., 2014)) genes (Source data 1). To verify the effective operation of these metabolic routes, cells were fed with13C3-glycerol, and the isotopic labelling profiles of multiple metabolites reporting on CCM were analysed (Figures 3and4).

*Percentage of labelled (13C) and unlabelled (12C) metabolites in H37Rv.Metabolites were extracted from cells fed with13C3-glycerol for 8 days in 50 μM FeCl3(HI: high iron), 0 μM FeCl3(LI: low iron), or 0 μM FeCl3+DFO (deferoxamine [DFO]). The analysed metabolites are shown in black in the schematic pathways. For each metabolite, except fumarate, two plots are shown. The stacked column plot shows the total percentage of labelled and unlabelled molecules per each metabolite pool; the clustered column plot shows the abundance in percentage of each isotopologue. The data represent the average and the standard deviation of four biological replicates from an independent experiment, representative of two independent experiments. The p-values were calculated independently for the two experiments, and the highest p-value was reported. DFO/LI vs HI condition: *=p-value<0.05; *=p-value<0.01; n.s.=non-significant. n.d.=non-detected. M2 vs M1: #=p-value<0.05; ##=p-value<0.01. Ac-CoA: acetyl-CoA; CIT: citrate; FUM: fumarate; α-KG: α-ketoglutarate; ISO: isocitrate; (ISO)CIT: isocitrate and citrate; MAL: malate; OAA: oxaloacetate; PYR: pyruvate; SUCC: succinate; SUCC-CoA: succinyl-CoA.

*Percentage of labelled (13C) and unlabelled (12C) metabolites in the Erdman strain.Metabolites were extracted from cells fed with13C3-glycerol for 8 days in 50 μM FeCl3(HI: high iron), 0 μM FeCl3(LI: low iron), or 0 μM FeCl3+DFO (DFO). The analysed metabolites are shown in black in the schematic pathways. For each metabolite, two plots are shown. The stacked column plots show the total percentage of labelled and unlabelled molecules per each metabolite pool; the clustered column plots show the abundance in percentage of each isotopologue. The data represent the average and the standard deviation of four biological replicates from an independent experiment, representative of four independent experiments (only two for LI condition). The p-values were calculated independently between experiments and the highest value is reported. DFO/LI vs HI: *=p-value<0.01; n.s.=non-significant; n.s.f.=non-significant fold change; the observed trend change was different between independent experiments. n.d.=non-detected. M2 vs M1: ##=p-value<0.01. Ac-CoA: acetyl-CoA; CIT: citrate; FUM: fumarate; α-KG: α-ketoglutarate; ISO: isocitrate; (ISO)CIT: isocitrate and citrate; MAL: malate; OAA: oxaloacetate; PYR: pyruvate; SUCC: succinate; SUCC-CoA: succinyl-CoA.

Malate only partially derives from succinate oxidation

DFO-treated H37Rv shows a 50% reduction in labelled succinate compared to HI condition (Figure 3). This decrease is not observed in isocitrate and α-ketoglutarate, precursors of succinate. These discrepancies suggest that a slower rate conversion of α-ketoglutarate into succinyl-CoA and subsequently into succinate, and of isocitrate to succinate may occur in DFO. The downregulation of KDH complex genes (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)) supports the hypothesis related to α-ketoglutarate; however, the upregulation oficl1andkorAB(see above) contradicts both hypotheses. The percentage of labelling observed in the malate pool (Figure 3) remains constant between the DFO and HI conditions. Notably, the percentage of malate labelling is significantly higher than that of succinate in DFO, indicating that some of the malate may not originate from succinate via fumarate. Unfortunately, the fumarate levels (Figure 3) were insufficient to permit the detection of labelling for this metabolite and verify if it is similar to malate rather than succinate. However, the comparison of the isotopologue distributions of malate and succinate confirms that some of the malate is not a product of succinate oxidation (Figure 3). Indeed, whereas in HI conditions the malate and succinate labelled pools show similar levels of M+1 and M+2 isotopologues (omitted isotopologue word after ‘M+n’ henceforth) and lower levels of M+3, in DFO M+2 succinate is more abundant compared to the M+1 and M+3, and M+3 malate is the most abundant (70–90% of total labelling) over M+1 and M+2 (Figure 3).

The DFO-treated Erdman strain does not show a marked decrease in the total labelling of succinate pool (Figure 4), and the isotopologue distribution shows that M+1 and M+2 are present at similar levels in the DFO and HI conditions. Similar to H37Rv, in Erdman strain the malate labelled pool size was similar in DFO and HI conditions, and contains more M+3 in DFO compared to HI (40–50% of total labelling), with a clear decrease in M+1 and M+2 (Figure 4). In the Erdman strain, higher levels of fumarate were detected and13C incorporation could be evaluated. Surprisingly, the labelled fumarate pool contained only M+1 in all conditions (Figure 4), impeding a more precise tracking of its origin. This specific labelling profile of fumarate is likely linked to the specific medium used in the current work (SI Appendix, Section III).

Altogether, these results strongly suggest that up to 50% of the malate pool is not derived from succinate oxidation under the DFO condition inMtb.

Malate mainly derives from oxaloacetate reduction

The upregulation oficl1andpckAgenes under Fe3+condition ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)) suggests that malate may be produced via isocitrate lyase (ICL) and phosphoenolpyruvate carboxykinase (PCK) activities, despite the conflicting differences in the total labelling (Figure 3) between isocitrate and succinate (see above). The increase of these enzymatic activities was confirmed in the cell-free extracts (Figure 5A and B,Figure 5—figure supplement 1).

*Analysis of iron-independent metabolic routes.Cells were exposed to 50 μM FeCl3(high iron [HI]) or 0 μM FeCl3+DFO (deferoxamine [DFO]). (A) Enzymatic activity of phosphoenolpyruvate carboxykinase (PCK), reaction from phosphoenolpyruvate to oxaloacetate. (B) Enzymatic activity of isocitrate lyase (ICL); the plots show the activity in (mM of NADH×min–1)/mg of total protein detected in cell-free extracts after 3 days of exposure to DFO or HI condition. Data show average and standard deviation from two (H37Rv) or three (Erdman strain) independent experiments and two technical replicates each (A) or from one independent experiment and three technical replicates (B). (C, D) Isotopologue distribution of intracellular malate (C) and isocitrate (D) in Erdman-derived∆pckA::pckAand∆pckAstrains after 8 days of exposure to DFO or HI conditions and fed with13C3-glycerol. The plots show the abundance in percentage of each isotopologue. The histograms represent average and standard deviation from four biological replicates from one experiment, representative of three independent experiments. (E, F) Enzymatic activity of pyruvate carboxylase (PCA), reaction from pyruvate to oxaloacetate; the plots show the activity in (mM of NADH×min–1)/mg of total protein detected in cell-free extracts after 3 days of exposure to DFO or HI condition. Data show average and standard deviation from three independent experiments and two technical replicates each. (G, H) Erdman strain cells were exposed to HI or DFO condition for 8 days with or without 200 μM of inhibitor 3-nitropropionate (3NP) and fed with13C3-glycerol. (G) The plots show the normalised levels of metabolites; the y-axis is shown on Log10scale. Tick labels (0–5) represent the exponents of 10 (100–105); values are reported in arbitrary units. (H) The plots show the isotopologue distribution in percent abundance. The data represent the average and the standard deviation of four biological replicates from an independent experiment, representative of two independent experiments. The p-values were calculated independently between experiments, and the highest value is reported. The p-values were calculated as follows. DFO vs HI for A, B, E, F; HI+3 NP vs HI and DFO+3 NP vs DFO for H; mutant vs complemented for C, D. n.s.f.=non-significant fold change; the observed trend change was different between independent experiments. *=p-value<0.01.

Figure 3—figure supplements 1–5show 31 possible metabolic scenarios for the production of core of CCM intermediates when the glyoxylate cycle and/or the PCK anaplerotic route are active. These scenarios combine labelled and unlabelled pyruvate, phosphoenolpyruvate (PEP), acetyl-CoA, and carbon dioxide molecules. Although they do not consider the re-circulation of metabolites in the global metabolic circuit, these series of reactions offer a useful perspective on identifying operating pathways. We compared these theoretical possibilities with our labelling data (isotopologue distribution) on succinate, malate, and (iso)citrate (Figures 3and4). Oxaloacetate was not included in this analysis due to technical limitations (SI Appendix, Section IV). Of note, the isotopologue distribution of succinate is identical whether derived from isocitrate or from α-ketoglutarate.

Figure 3—figure supplement 1shows scenarios derived from PDH and ICL activities. M+2 (iso)citrate, malate, and succinate are the only isotopologues produced. This matches the experimental results for isocitrate and succinate, for which M+2 is one of the most abundant isotopologues (Figures 3and4). Of note, the medium used in these experiments contains asparagine, which represents an exogenous unlabelled source of oxaloacetate (SI Appendix, Section IV). This indicates thatFigure 3—figure supplement 1scenarios are favoured and explain the dominant abundance of M+2 (iso)citrate and succinate, compared to other isotopologues. However, these combinations do not produce the most abundant malate isotopologue, M+3, detected in our measurements.

Figure 3—figure supplement 2andFigure 3—figure supplement 3show PCK and ICL activities. In these scenarios, the M+3 malate isotopologue is present and derives from the glyoxylate shunt. However, there is no congruence between these theoretical scenarios and our labelling results (more details in SI Appendix, Section V). This raises doubts about an active role of the glyoxylate shunt in the synthesis of malate.

Figure 3—figure supplement 4illustrates alternative metabolic scenarios without the involvement of the glyoxylate shunt that include PCK activity and the oxidative branch of the Krebs cycle stopping at malate. Among these scenarios, the most represented isotopologues are M+3 for (iso)citrate (Figure 3—figure supplement 4B and G), M+2 for malate and for succinate (Figure 3—figure supplement 4B, C, F, and G). These scenarios offer a rationale for the observed increase in M+3 (iso)citrate in our experiments in DFO (Figures 3and4) and imply that M+2 malate and succinate may originate from the oxidative branch of the Krebs cycle, fed via oxaloacetate produced by PCK (Figure 3—figure supplement 4B, C, F, and G). These alternative scenarios align with those shown inFigure 3—figure supplement 1for the synthesis of M+2 malate and succinate, excluding the glyoxylate shunt activity, but do not reveal the origin of M+3 malate.

We then hypothesised that PEP-derived oxaloacetate could proceed to the reductive branch of Krebs cycle (Figure 3—figure supplement 5A–D). The malate isotopologues produced inFigure 3—figure supplement 5A, C, and Dscenarios are all detected in our experiments. The dominance of M+3 over M+1 (Figures 3and4) indicates that the scenario illustrated inFigure 3—figure supplement 5Ais the most likely. The minor abundance of M+4 malate (Figures 3and4) is compatible with the unlikely possibility that an enzyme meets both substrates labelled (Figure 3—figure supplement 5C). A similar consideration is valid for the generation of M+4/M+5/M+6 (iso)citrate and M+4 succinate (Figure 3—figure supplement 4A, E, and F,Figures 3and4). This new analysis, which omits the glyoxylate shunt, is consistent with our experimental results and indicates that M+2 and M+4 malate derive from the oxidative branch of Krebs cycle, while M+1 and M+3 malate from the reductive branch. Further, these results indicate thatpckAhas an important role in the CCM under Fe3+starvation. Surprisingly, an Erdman-derivedpckAmutant, which does not show loss of viability and reducing power under exposure to Fe3+starvation (Figure 1F and I), does not show a significant decrease in M+3 malate and M+3 (iso)citrate in DFO (Figure 5C and D). In fact, there is no consistent between experiments (0–50% of decrease). Oxaloacetate can be produced directly from pyruvate by a pyruvate carboxylase (PCA) ([Basu et al., 2018](#ref-Basu et al., 2018);[Mukhopadhyay and Purwantini, 2000](#ref-Mukhopadhyay and Purwantini, 2000)), generating the same scenarios depicted inFigure 3—figure supplements 1–5, or malate can be directly produced from pyruvate by malic enzyme (MEZ) (Figure 3—figure supplement 5E–H). Measurement of the activity of these two enzymes in cell-free extracts revealed the presence of PCA activity in the DFO condition (Figure 5E and F,Figure 5—figure supplement 1), but not of MEZ activity (Figure 5—figure supplement 1). The absence ofpckAdoes not result in an increase of PCA activity (Figure 5F), suggesting that the two activities work in parallel but serve distinct and non-redundant functions. These results strongly suggest thatMtbutilises anaplerotic reactions mediated by PCK and PCA to supply Krebs cycle intermediates under Fe3+starvation.

The glyoxylate shunt does not supply malate and succinate under iron starvation

To provide independent, genetic evidence supporting or ruling out the role of the glyoxylate shunt under Fe3+-starved conditions in H37Rv, the levels of succinate, malate, and isocitrate were analysed in an ICL-deleted strain. H37Rv has two ICL enzymes, encoded by theicl1 andaceAabgenes; however, the second enzyme has a slower activity ([Höner Zu Bentrup et al., 1999](#ref-Höner Zu Bentrup et al., 1999)), and studies show that the inactivation of justicl1is relevant to alterMtbphysiology ([Serafini et al., 2019](#ref-Serafini et al., 2019);[Gengenbacher et al., 2010](#ref-Gengenbacher et al., 2010)). We used an H37Rv-∆icl1and its complemented strain ([Serafini et al., 2019](#ref-Serafini et al., 2019)), which is not essential for survival in DFO condition (Figure 1E). As the∆icl1strain is not derived from the same H37Rv strain used in this study, metabolite abundance and isotopologue distribution in HI, LI, and DFO conditions were verified and confirmed in the complemented strain (∆icl1::icl1). Except for the lack of increase in intra- and extracellular (iso)citrate levels, all other trends were matched (Figure 2—figure supplement 4). If the glyoxylate shunt is a key supplier of malate and succinate, the absence of a functional ICL should result in a reduction in their abundance or labelling, an increase in isocitrate levels and a decrease in the M+2 in succinate. To our surprise, there are no discernible differences between the∆icl1and the complemented strains, in the intracellular levels of these metabolites (Figure 2—figure supplement 5A) and in the isotopologue distribution of succinate and malate (Figure 2—figure supplement 5B and C) in DFO condition. We reasoned that in∆icl1metabolite secretion could be diminished to maintain the intracellular abundance of malate and succinate or increased in the case of isocitrate. Again, there are no significant differences in the extracellular levels of these metabolites between the∆icl1and the complemented strains (Figure 2—figure supplement 5E).

Next, we verified the operation of glyoxylate shunt in the Erdman strain by analysing the levels and labelling profiles of intracellular metabolites in DFO-treated cells exposed to a potent inhibitor of mycobacterial ICL activity, 3-nitropropionate (3NP) ([Höner Zu Bentrup et al., 1999](#ref-Höner Zu Bentrup et al., 1999);[Moynihan and Murkin, 2014](#ref-Moynihan and Murkin, 2014)), which does not affect survival under such condition (Figure 2—figure supplement 3B). The presence of 3NP did not increase the abundance of (iso)citrate, but rather that of α-ketoglutarate, and no relevant changes were detected in malate and succinate levels (Figure 5G). The isotopologue distribution analysis revealed no change in the abundance of the M+2 succinate in the DFO condition, but rather a diminution of the M+2 malate (Figure 5H). 3NP is also an inhibitor of Sdh ([Alston et al., 1977](#ref-Alston et al., 1977)), and the observed result appears to report on this, instead of on inhibition of ICL. The results demonstrate that the glyoxylate shunt is not playing an important role under Fe3+starvation inMtb.

However, the presence of ICL activity in lysates indicates that the enzyme is active and could generate a pool of succinate, yet not detectable under our experimental conditions, likely due to compensatory metabolic routes (see below). The labelling profile analysis (Figures 3and4) excludes malate as a derivative of glyoxylate. We therefore investigated an alternative metabolic fate of this molecule. Glyoxylate can be converted to glycine via reductive amination catalysed by alanine dehydrogenase (Ald), whose transcript is upregulated under iron limitation ([Giffin et al., 2012](#ref-Giffin et al., 2012);[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)), leading us to hypothesise that this metabolic route might be active under iron starvation. Due to technical limitations, glyoxylate could not be directly detected, but we assessed glycine abundance and its isotopologue distribution (Figure 4—figure supplement 1A–D), which should mirror that of glyoxylate if glycine is derived from it. While total glycine levels were unchanged between HI and DFO conditions, isotopologue analysis revealed an ~3-fold increase in M+2 relative to M+1 under DFO in both H37Rv and Erdman strains. The comparison (Figure 4—figure supplement 1E) of these results with the most probable labelling scenarios of glyoxylate illustrated inFigure 3—figure supplements 1–5suggests that glycine does not originate from glyoxylate. Additionally, no differences in labelling were observed between theicl1mutant and the complemented strain (Figure 4—figure supplement 1F and G). Taken together, these results indicate that glyoxylate is not a precursor of glycine under iron starvation, further supporting the non-operational state of the glyoxylate shunt under these conditions.

The lack of evidence for an operational glyoxylate shunt under Fe3+starvation led us to conclude that succinate must derive from α-ketoglutarate oxidation. The upregulation of thekorABgenes ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)) suggests that KorAB actively participates in succinyl-CoA synthesis. However, its iron-dependent nature together with the downregulation of succinyl-CoA synthetase genes (sucCD) suggests that an additional pathway, likely iron-independent, might be active under Fe3+starvation conditions to maintain the succinate pool. This pathway might be the γ-aminobutyric acid (GABA) shunt, an active route of CCM inMtb([Serafini et al., 2019](#ref-Serafini et al., 2019)), which converts α-ketoglutarate into succinate bypassing succinyl-CoA synthesis (Figure 5—figure supplement 2A). Genes encoding GABA shunt enzymes are not differentially expressed under Fe3+starvation conditions (Source data 1and[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)). In H37Rv, the levels and labelling of glutamate do not change to a major extent under HI and DFO conditions (Figure 5—figure supplement 2B). In contrast, we see a significant decrease in the GABA pool size in DFO (0.001–0.428 FC in four independent experiments), with a concomitant decrease in total labelling (reduction >50%) (Figure 5—figure supplement 2C). Changes in GABA abundance resemble that of succinate (Figure 5—figure supplement 2D), with a greater magnitude. This suggests that GABA might be used to produce succinate. We wondered if this reduction of the GABA pool could be due to its secretion; however, we were not able to detect extracellular GABA (data not shown). The isotopologue distribution of the GABA and succinate pools (Figure 5—figure supplement 2C and D) is comparable to those of α-ketoglutarate (Figure 3) with a notable reduction in the M+1 compared to the M+2. Similar results were obtained in the Erdman strain (Figure 5—figure supplement 2E–G) in which the isotopologue distribution of succinate resembles that of GABA (Figure 5—figure supplement 2F and G) and glutamate; yet, it differs from that of α-ketoglutarate (Figure 4,Figure 5—figure supplement 2E). In particular, in succinate, GABA, and glutamate, the levels of M+1 and M+2 are similar in DFO (Figure 5—figure supplement 2E–G), whereas M+2 α-ketoglutarate consistently exceeds M+1 (Figure 4). This different isotopologue distribution in α-ketoglutarate suggests that the GABA shunt is significantly sustained by other metabolic circuits beyond α-ketoglutarate in Erdman.

The absence of a differential expression of GABA shunt genes under Fe3+starvation and the similarity between the labelling profiles of succinate and GABA are consistent with the hypothesis that the GABA shunt may contribute to maintaining the succinate pool under Fe3+starvation conditions inMtb, although direct genetic evidence is still required to substantiate this conclusion.

Discussion

This study was conducted on twoMtbstrains that exhibit similar overall behaviour, with minor differences in the carbon flux – likely reflecting variations in the intensity of Krebs cycle activity (SI Appendix, Section II) – and in viability readouts (Figure 1,Figure 1—figure supplement 1). The results of this investigation are summarised inFigure 6. The consistent accumulation of pyruvate and α-ketoglutarate highlights two key checkpoints controlling carbon flux through the CCM. Because DlaT and LpdC are shared components of both the pyruvate and the KDH complexes ([Tian et al., 2005](#ref-Tian et al., 2005);[Shi and Ehrt, 2006](#ref-Shi and Ehrt, 2006);[Maksymiuk et al., 2015](#ref-Maksymiuk et al., 2015)), they likely play a critical role in this buildup. As central metabolites linking energy and carbon/nitrogen metabolism, pyruvate and α-ketoglutarate accumulation indicate that Fe3+starvation slows down the overall cellular metabolism. Under these conditions,Mtbcells survive and maintain a physiological energy balance (Figure 1G and HandFigure 1—figure supplement 1A and B) but display an altered redox state marked by increased NADH levels (Figure 1—figure supplement 1C and D). It has been observed that under Fe3+starvation,Mtbdownregulates genes encoding the proton pumping enzymatic complexes type I NADH dehydrogenase (nuoABCDEFGHIJKLMN) and cytochrome oxidase bc1-aa3(qcrCAB/ctaBCED), whereas it upregulates the gene for non-proton pumping type II NADH dehydrogenase (ndh) and the gene for assembly of the less efficient and non-proton pumping cytochrome oxidasebd(cydABDC) ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013);[Cruz-Ramos et al., 2004](#ref-Cruz-Ramos et al., 2004)). The iron independent nature of Ndh and its lack of proton pumping activity may explain the opposite regulation of type I and II NADH dehydrogenases under Fe3+starvation. In contrast, the reason behind the divergent regulation of the two cytochrome oxidases remains unclear. Together with the increased NADH/NAD ratio, these changes in the electron transport chain (ETC) suggest that the alternative ETC is less efficient at NADH reoxidation. Notably, the replacement of proton pumping components with the non-proton pumping alternatives, along with the reported downregulation ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)) of the ATP synthase operon (atpBEFHAGDC), contrasts with the stable ATP levels observed in Fe3+-starvedMtbcells.

Remodelling of central carbon metabolism (CCM) under iron starvation inM. tuberculosis(Mtb).The picture depicts a schematic representation of the CCM pathways active inMtbexposed to Fe3+deprivation in the presence of D-glucose and glycerol as carbon sources and asparagine as sole nitrogen source. The thickening arrows indicate the increased levels of the metabolite; thicker arrows indicate a preferred route. Bold and larger font indicates accumulated metabolites. Under iron starvation, the pool of iron-dependent enzymes (denoted by Fe in parentheses in the figure) contains a reduced number of fully active molecules, which then slows the carbon flux through the Krebs cycle. The reduction in the transcript levels of iron-independent enzymes of the CCM pathways is likely a consequence of this. The disparity in efficiency between iron-dependent and iron-independent enzyme pools gives rise to the accumulation of (iso)citrate, pyruvate, and α-ketoglutarate.Mtbresponds to these accumulations by expelling these metabolites from the cell and splitting the carbon flux from PEP/pyruvate (via phosphoenolpyruvate carboxykinase [PCK], pyruvate carboxylase [PCA], and pyruvate dehydrogenase [PDH] activities) into both oxidative and reductive branches of the Krebs cycle. Both fluxes terminate in malate synthesis, which is then secreted. To maintain malate synthesis by succinate oxidation,Mtblimits its secretion. Malate secretion relieves the slowdown of carbon flux through the oxidative branch of Krebs cycle. PCK and PCA anaplerotic reactions control pyruvate levels and recycle carbon dioxide stoichiometric to the α-ketoglutarate accumulation. GDH: glutamate dehydrogenase; GADB: glutamate decarboxylase; GABT: 4-aminobutyrate aminotransferase; GABD: succinate-semialdehyde dehydrogenase. For the other enzyme acronyms, see main text.

Intriguingly, the differential expression ofatpBEFHAGDC, ndh,andnuoABCDEFGHIJKLMNgenes during Fe3+starvation resembles that seen during the efficientMtbgrowth in L-lactate as a sole carbon source ([Serafini et al., 2019](#ref-Serafini et al., 2019)), suggesting thatMtbassembles an alternative ETC tailored to distinct physiological demands. Even more intriguing is the very recent work onMycobacterium smegmatis, which shows that a minimal ETC composed of cytochrome oxidaseb,Ndh, and F1F0-ATPase can generate ATP orpmfusing atmospheric hydrogen ([Soom et al., 2025](#ref-Soom et al., 2025)). This raises the possibility that additional, yet unidentified mechanisms support ATP synthesis under the conditions used in our study.

The downregulation of transcripts for many CCM enzymes in Fe3+starvation ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)) is presumably a consequence of the slowed carbon flux, and the induction ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)) of a few of them (korAB,pckA, andicl1) is likely a strategy to sustain slowed carbon flow. The induction of the iron-dependent KorAB route represents a mechanism to bypass or parallel the KDH complex and GABA shunt, in order to sustain succinate synthesis and dispose of the accumulated α-ketoglutarate. This strategy combines multiple routes to maintain succinate synthesis under Fe3+-starved conditions. A combination of multiple routes is also deployed to sustain carbon flux from PEP/pyruvate to the Krebs cycle. The anaplerotic reaction of PCK converts PEP into oxaloacetate, and it is flanked by PCA, which converts pyruvate into oxaloacetate. These two activities, together with PDH, represent three routes to control the level of pyruvate and sustain the carbon flux.

Detailed13C tracking through Krebs cycle intermediates revealed an unexpected split of carbon flux from PEP- and pyruvate-derived oxaloacetate to both the oxidative and reductive branches of the Krebs cycle. Oxaloacetate is partitioned between oxidation to citrate and reduction to malate, with the latter route being the most favourable (Figures 3and4,Figure 3—figure supplements 1–5). The two fluxes converge to produce malate, whose extracellular levels increase (Figure 6). The production of oxaloacetate via PCK and PCA under Fe3+starvation occurs despite the presence of an exogenous source of oxaloacetate (the asparagine in the medium, SI Appendix, Section IV) and clearly demonstrates the importance of these anaplerotic reactions under Fe3+starvation beyond oxaloacetate synthesis. These activities likely control pyruvate levels but also may contribute to assimilating the excess of carbon dioxide stoichiometrically produced with α-ketoglutarate by isocitrate dehydrogenase (Icd,Figure 6). ThepckAgene is also induced in Fe3+starvation when glutamate and ammonium, but not asparagine, are present as nitrogen sources in the medium ([Serafini et al., 2013](#ref-Serafini et al., 2013)), suggesting that PCK and likely PCA are the main producers of oxaloacetate under these conditions.

The significant quantities of extracellular malate under Fe3+starvation, despite its intracellular levels being similar to those in Fe3+replete condition, suggest that secretion of this metabolite has a functional role. Also, the reduced secretion of succinate under Fe3+starvation seems to indicate that succinate (Figure 2andFigure 2—figure supplement 1) is required to produce malate. The secretion of malate may promote carbon flux through the anaplerotic node to relieve the accumulation of metabolic intermediates from the oxidative branch of the Krebs cycle. NADH reoxidation mediated by malate dehydrogenase, together with Ndh, may help to maintain a redox state compatible with cell survival.

The upregulation oficl1([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017);[Serafini et al., 2013](#ref-Serafini et al., 2013)) and the resulting increase of ICL activity under Fe3+starvation appear to be dispensable for the synthesis of malate and succinate (Figure 2—figure supplement 5andFigure 5), which is surprising considering that the glyoxylate shunt has been demonstrated to be important for the survival ofMtbunder various growth arrest-causing conditions ([Eoh and Rhee, 2013](#ref-Eoh and Rhee, 2013);[Gengenbacher et al., 2010](#ref-Gengenbacher et al., 2010);[Nandakumar et al., 2014](#ref-Nandakumar et al., 2014)). On the other hand, the key function of the glyoxylate shunt is to bypass the two Krebs cycle decarboxylation reactions, and thereby ‘save up’ carbon. Under our experimental conditions, carbon is replete, and therefore the glyoxylate shunt is not needed. ICL might participate in a previously unrecognised metabolic pathway or other process. The induction of ICL under Fe3+starvation is not unique to mycobacteria. An increase of its activity has been observed inPseudomonas aeruginosaunder Fe3+starvation ([Ha et al., 2018](#ref-Ha et al., 2018)) and inPseudomonas fluorescensunder Al3+stress, which mimics iron deficiency ([Middaugh et al., 2005](#ref-Middaugh et al., 2005)). InP. aeruginosa, the absence of ICL activity causes an increase in succinate dehydrogenase activity, suggesting its involvement in succinate synthesis ([Ha et al., 2018](#ref-Ha et al., 2018)). It is noteworthy thatP. aeruginosa iclmutant accumulates intracellular iron, suggesting a role beyond CCM. Unfortunately, metabolomic studies have not been performed in this mutant. Interestingly, it has been demonstrated thatP. fluorescensexposed to Al3+stress induces an acylating glyoxylate dehydrogenase (AGODH) that transforms glyoxylate to oxalyl-CoA and then oxalate, a molecule necessary to seize Al3+ions ([Singh et al., 2009](#ref-Singh et al., 2009)). Coenzyme A is recycled to produce succinyl-CoA and then succinate, releasing one ATP ([Singh et al., 2009](#ref-Singh et al., 2009)). The finding that glyoxylate is used in a metabolic pathway functional for Al3+stress, rather than for malate synthesis, supports the hypothesis thatMtbmay use ICL for an alternative metabolic function under Fe3+starvation.

Collectively, these findings suggest that, under iron depletion,Mtbrelies on anaplerotic routes to divert carbon flux from glycolysis towards the reductive branch of the Krebs cycle. This results in a truncated cycle in which both carbon fluxes converge towards malate synthesis. The secretion of malate appears to be functionally relevant to sustaining these fluxes, ensuring the production of metabolic intermediates and maintaining redox balance through NADH reoxidation via malate dehydrogenase. Although the alternative ETC comprising Ndh and cytochrome oxidasebdis less efficient in both ATP synthesis and NADH reoxidation, it may represent an adaptive mechanism consistent with the reduced energetic demands of non-replicating bacteria. Under these conditions, the alternative ETC may be insufficient to fully reoxidise NADH, thereby inducing malate dehydrogenase activity as a compensatory mechanism. This mechanistic hypothesis would explain the persistence of Mtb under iron-depleted conditions.

Further studies are needed to identify which enzymes of CCM are essential for this mechanism. In particular, genetic and biochemical approaches targeting MDH would provide direct evidence for the role of the PCK/PCA-mediated reductive Krebs cycle in malate biosynthesis and secretion under iron-limiting conditions. Additionally, further investigation of the GABA shunt would help clarify its contribution to succinate biosynthesis.

While this study has certain limitations related to the specific experimental conditions used, these observations provide a basis for exploring whether this adaptive mechanism also occurs during infection. An important question is whether metabolites secreted byMtbcould substantially modify the surrounding microenvironment, thereby influencing host responses and bacterial pathogenicity.

Mycobacteria are known to exhibit a dynamic interplay between CCM and cell envelope biosynthesis under stress conditions such as hypoxia and antibiotic exposure ([Eoh et al., 2022](#ref-Eoh et al., 2022)). In this context, the metabolic shift induced by iron starvation may contribute to remodelling of the cell envelope. Consistent with this, iron-starved mycobacterial cells show altered cell envelope thickness and differential expression of genes involved in mycolic acid and cell wall biosynthesis ([Vijay et al., 2017](#ref-Vijay et al., 2017);[Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)). Notably, under iron starvation,Mtbdisplays differential sensitivity to antibiotics with distinct structural properties: resistance is observed for smaller molecules such as kanamycin,d-cycloserine, ethionamide, ciprofloxacin, and isoniazid, whereas sensitivity to rifampicin, which has a larger and more complex structure, is maintained ([Kurthkoti et al., 2017](#ref-Kurthkoti et al., 2017)). This pattern suggests that cell envelope remodelling under iron starvation may selectively affect antibiotic permeability. Whether such changes also influence direct interactions with host cells, particularly at the bacteria surface-host cell interface, remains an important question for future investigation.

Materials and methods

Bacterial strains and growth conditions

In this study, H37Rv and ErdmanMtbstrains were used.ΔpckAandΔpckA::pckA([Marrero et al., 2010](#ref-Marrero et al., 2010)) (Erdman genomic background) were kindly provided by Prof. Sabine Ehrt (Weill Cornell Medicine). TheΔicl1 ([Lee et al., 2013](#ref-Lee et al., 2013)) strain was kindly provided by Prof. Brian VanderVen (Cornell University). TheΔicl1::icl1strain was previously generated ([Serafini et al., 2019](#ref-Serafini et al., 2019)). For the metabolomic experiments in the Erdman genomic background, the strainΔpckA::pckAwas used (named Erdman strain in the text and figures and namedΔpckA::pckAin the comparison withΔpckA).

Bacteria were routinely cultivated at 37°C in 7H9/7H10 medium supplemented with ADC (0.5% bovine serum albumin, 0.2% glucose, 0.085% NaCl, 0.003% catalase), 0.05% Tween 80, and 0.2% glycerol. Mutant and complemented strains were grown in the presence of selective antibiotics in the 7H9/7H10 pre-cultures. Antibiotics were added at the following concentrations: 20 μg/mL kanamycin (Km); 10 μg/mL gentamicin (Gm); 100 μg/mL hygromycin (Hyg).

For all the experiments, bacteria were cultivated in minimal medium (MM; 0.5% wt/vol KH2PO4; 0.5 % wt/vol asparagine; 0.2% glycerol; 0.05% tyloxapol; 0.5% bovine serum albumin, 0.2% D-glucose, 0.085% NaCl, pH 6.8) treated with Chelex 100 resin (Sigma) at RT for 24 hr, replacing Chelex 100 resin after 12 hr. The resin was removed by filtration, and the medium was supplemented with 40 mg/L MgSO4, 0.1 mg/L MnSO4, and 0.5 mg/L ZnSO4. For the labelling experiments, 0.2% glycerol was replaced with a mixture of 0.1%13C3-glycerol and 0.1%13C3-glycerol. The choice of glycerol as the labelled carbon source was based on the superior ability of H37Rv to metabolise this substrate compared to glucose ([Serafini et al., 2019](#ref-Serafini et al., 2019)).

A source of Fe3+(50 μM of FeCl3) was added only in the control condition (high iron [HI]). All the tests of this study were performed in bacterial cells sub-cultured three times (4–7 days of growth) in MM supplemented with (HI) or without (0 μM of FeCl3, low iron [LI]) an Fe3+source. To get growth arrest, twice sub-cultured bacterial cells grown without a source of Fe3+were treated with the siderophore DFO mesylate salt at a final concentration of 50 ng/mL15. For each experiment, parallel HI and LI cultures were also treated. For metabolomic samples, MM was prepared omitting tyloxapol, which interferes with LC-MS analysis (MM No Tyloxapol, MMNT).

All the growth experiments in liquid cultures were performed in standing in T75 (25 mL of culture)/T25 (10 mL of culture) flasks laid down on the incubator floor to optimise the oxygenation, with daily gentle manual shaking.

The input of cells was approximately 1×107cells/mL for CFU/growth assay and 2×107cells/mL for all the other analysis (metabolite content, ATP, and NADH assays).

Cell viability was assessed by counting CFU/mL or by spot density on 7H10/ADC plates. CFU/mL: bacterial culture was serially diluted 10-fold, and 50 μL of each dilution was plated in duplicate. Spot’s density: bacterial culture was diluted to 107cells/mL (OD600~0.1) and subjected to four 1:10 serial dilutions (106–102cells/mL); 5 μL of each dilution was spotted on the plate.

Protein extraction and enzymatic activity

Approximately 10–15 OD600of a bacterial culture were harvested and washed three times with cold PBS 1× and then the pellet resuspended in 600 µL of extraction buffer (DPBS 1× and Complete EDTA-free protease inhibitor cocktail). The cells were disrupted by using acid-washed glass beads (2:1=buffer volume:beads volume) and a refrigerated bead-beater (6.0 set, 35’’ pulses/1’ pause twice). After centrifugation at 13,000 rpm for 15 min at 4°C, the supernatant was filtered with a PVDF filter tube before removal from the BLS3 laboratory. Within 6 hr after extraction, enzymatic activity assays were performed using 10 µL of extract in a 96-well polystyrene microplate, with a final reaction volume of 200 µL. Protein content was determined using the Bradford reagent.

The anaplerotic activity of phosphoenolpyruvate carboxylase was determined as previously described ([Machová et al., 2014](#ref-Machová et al., 2014)) coupling the phosphoenolpyruvate carboxylation to oxaloacetate and malate synthesis by using malate dehydrogenase (pig heart, Sigma). The reaction was performed at 37°C in a buffer containing HEPES/NaOH pH 7.2 100 mM, KHCO3100 mM, DTT 37 mM, sodium phosphoenolpyruvate 2 mM, GDP 1 mM, MgCl22 mM, MnCl20.1 mM, malate dehydrogenase 3.5 U/ml, NADH 0.25 mM. The reaction was initiated by adding the MgCl2and MnCl2. The decrease of absorbance at 340 nm was measured to monitor NADH oxidation.

PCA activity was determined as previously described ([Mukhopadhyay and Purwantini, 2000](#ref-Mukhopadhyay and Purwantini, 2000)) coupling the phosphoenolpyruvate carboxylation to oxaloacetate and malate synthesis by using malate dehydrogenase. The reaction was performed at 37°C in a buffer containing Tris/HCl pH 8.0 50 mM, KHCO320 mM, MgCl28 mM, di-sodium ATP 8 mM, acetyl-CoA 50 µM, di-sodium NADH 0.2 mM, malate dehydrogenase 3.5 U/mL and sodium pyruvate 20 mM. The reaction was initiated by adding pyruvate. The decrease of absorbance at 340 nm was measured to monitor NADH oxidation.

MEZ activity was determined as previously described ([Basu et al., 2018](#ref-Basu et al., 2018)). The reaction was performed at 30°C in a buffer containing Tris/HCl pH 6.0 100 mM, KHCO3200 mM, MnCl20.5 mM, NAD(P)H 0.6 mM, and sodium pyruvate 50 mM. The reaction was initiated by adding pyruvate. The decrease of absorbance at 340 nm was measured to monitor NAD(P)H oxidation.

ICL activity was determined as previously described ([Muñoz-Elías et al., 2006](#ref-Muñoz-Elías et al., 2006)). The reaction was performed at 25°C in a buffer containing MOPS-HCl pH 6.8 50 mM, MgCl25 mM, NADH 0.1 mM, 7 U L-lactate dehydrogenase (Rabbit). The reaction was pre-incubated 5 min at 25°C and then 1 mM of DL-threo-isocitrate was added. The decrease of absorbance at 340 nm was measured to monitor NADH oxidation.

Metabolite extraction and LC-MS analysis

Cells were sub-cultured twice in MM supplemented with 50 or 0 µM of FeCl3, then washed once to remove tyloxapol and resuspended to a final absorbance of 0.2 in 10 mL of MMNT supplemented with 50, 0 µM of FeCl3and 0 µM of FeCl3+DFO. MMNT contained 0.2% of D-glucose, 0.1% of glycerol, and 0.1% of (U)-13C-labelled glycerol. Cultures were incubated standing at 37°C. Intracellular polar metabolites were extracted after 8 days by mechanical rupture in an acetonitrile:methanol:water (2:2:1, vol/vol/vol) solution as previously described ([Serafini et al., 2019](#ref-Serafini et al., 2019)). Extracellular metabolites were extracted after 1, 3, and 8 days by diluting the culture filtrate 1:5 using cold acetonitrile/methanol (1:1) supplemented with 0.1% formic acid; the samples were vigorously vortexed and stored at –20°C for 2 hr; they were spun down for 10 min at 14,000 rpm and 4°C, then filtered through 0.22 µm filter tubes.

2 µL of pellet extract and 10 μL of culture-filtrate extract were injected in a 1200 Liquid Chromatography System (Agilent) coupled to an Accurate Mass 6220 TOF (Agilent). Polar elution was performed as previously described using a gradient of two solvents, A (mQ water and 0.1% of formic acid) and B (acetonitrile and 0.1% of formic acid) ([Serafini et al., 2019](#ref-Serafini et al., 2019)).

The data were analysed by Profinder B.08.00 software and Masshunter Qualitative Analysis B07.00. Intracellular metabolites were normalised by the residual peptide content detected using a BCA assay. Extracellular metabolites were normalised by total ion counts detected by Progenesis software. To plot the data, each normalised value was multiplied or divided for the same number, then the y-axis plots report an arbitrary unit.

Due to technical limitations, the phosphorylated intermediates of glycolysis and oxaloacetate could not be analysed.

ATP and NADH/NAD assays

The ATP assay was performed using the Promega BacTiter-Glo Microbial Cell Viability assay (G8232). Three technical replicates (three aliquots of cells) were performed for each experiment using approximately 107cells/well (0.1 OD600). ATP levels were assessed using an ATP standard curve (μM), and values normalised to OD600. The NADH/NAD ratio was determined using the Promega NAD/NADH-Glo assay (G9071). Cells were collected, cooled on ice, and resuspended in a buffer containing PBS, NaOH, and dodecyltrimethylammonium bromide according to the kit. The cells were then disrupted with a bead-beater (30’’ 6.5 power at 4°C), and the extracts were filtered through 0.22 µm filter tubes. The levels of NADH and NAD were then determined according to the kit instructions using NAD and NADH calibration curves. Two technical replicates were performed for each experiment using approximately 3×107cells. The experiments shown inFigure 1,Figure 1—figure supplement 1were performed in independent experimental sessions, using different batches of reagents and by different operators. These factors account for the differences in the y-axis scale.

Statistical analysis

Data were expressed as average and standard deviations. Statistical correlations of data were checked for significance using the paired Student’s t test (p-value<0.05).

References

1. AllisonKRBrynildsenMPCollinsJJ2011Metabolite-enabled eradication of bacterial persisters by aminoglycosidesNature473216220 doi.org/10.1038/nature10069
2. AlstonTAMelaLBrightHJ19773-Nitropropionate, the toxic substance of Indigofera, is a suicide inactivator of succinate dehydrogenasePNAS7437673771 doi.org/10.1073/pnas.74.9.3767
3. BasuPSandhuNBhattASinghABalhanaRGobeICrowhurstNAMendumTAGaoLWardJLBealeMHMcFaddenJBesteDJV2018The anaplerotic node is essential for the intracellular survival ofMycobacterium tuberculosisJournal of Biological Chemistry29356955704 doi.org/10.1074/jbc.RA118.001839
4. BaughnADGarforthSJVilchèzeCJacobsWR2009An anaerobic-type alpha-ketoglutarate ferredoxin oxidoreductase completes the oxidative tricarboxylic acid cycle ofMycobacterium tuberculosisPLOS Pathogens5e1000662 doi.org/10.1371/journal.ppat.1000662
5. Cruz-RamosHCookGMWuGCleeterMWPooleRK2004Membrane topology and mutational analysis ofEscherichia coliCydDC, an ABC-type cysteine exporter required for cytochrome assemblyMicrobiology15034153427 doi.org/10.1099/mic.0.27191-0
6. DartoisVDickT2024Therapeutic developments for tuberculosis and nontuberculous mycobacterial lung diseaseNature Reviews. Drug Discovery23381403 doi.org/10.1038/s41573-024-00897-5
7. EohHRheeKY2013Multifunctional essentiality of succinate metabolism in adaptation to hypoxia inMycobacterium tuberculosisPNAS11065546559 doi.org/10.1073/pnas.1219375110
8. EohHLiuRLimJLeeJJSellP2022Central carbon metabolism remodeling as a mechanism to develop drug tolerance and drug resistance inMycobacterium tuberculosisFrontiers in Cellular and Infection Microbiology12958240 doi.org/10.3389/fcimb.2022.958240
9. FangZSampsonSLWarrenRMGey van PittiusNCNewton-FootM2015Iron acquisition strategies in mycobacteriaTuberculosis95123130 doi.org/10.1016/j.tube.2015.01.004
10. GengenbacherMRaoSPSPetheKDickT2010Nutrient-starved, non-replicatingMycobacterium tuberculosisrequires respiration, ATP synthase and isocitrate lyase for maintenance of ATP homeostasis and viabilityMicrobiology1568187 doi.org/10.1099/mic.0.033084-0
11. GiffinMMModestiLRaabRWWayneLGSohaskeyCD2012ald ofMycobacterium tuberculosisencodes both the alanine dehydrogenase and the putative glycine dehydrogenaseJournal of Bacteriology19410451054 doi.org/10.1128/JB.05914-11
12. HaSShinBParkW2018Lack of glyoxylate shunt dysregulates iron homeostasis inPseudomonas aeruginosaMicrobiology164587599 doi.org/10.1099/mic.0.000623
13. HicksNDYangJZhangXZhaoBGradYHLiuLOuXChangZXiaHZhouYWangSDongJSunLZhuYZhaoYJinQFortuneSM2018Clinically prevalent mutations inMycobacterium tuberculosisalter propionate metabolism and mediate multidrug toleranceNature Microbiology310321042 doi.org/10.1038/s41564-018-0218-3
14. Höner Zu BentrupKMiczakASwensonDLRussellDG1999Characterization of activity and expression of isocitrate lyase inMycobacterium aviumandMycobacterium tuberculosisJournal of Bacteriology18171617167 doi.org/10.1128/JB.181.23.7161-7167.1999
15. KawaiYMercierRMickiewiczKSerafiniASório de CarvalhoLPErringtonJ2019Crucial role for central carbon metabolism in the bacterial L-form switch and killing by β-lactam antibioticsNature Microbiology417161726 doi.org/10.1038/s41564-019-0497-3
16. KurthkotiKAminHMarakalalaMJGhannySSubbianSSakatosALivnyJFortuneSMBerneyMRodriguezGM2017The capacity ofMycobacterium tuberculosisto survive iron starvation might enable it to persist in iron-deprived microenvironments of human granulomasmBio8e01092-17 doi.org/10.1128/mBio.01092-17
17. LeeWVanderVenBCFaheyRJRussellDG2013IntracellularMycobacterium tuberculosisexploits host-derived fatty acids to limit metabolic stressThe Journal of Biological Chemistry28867886800 doi.org/10.1074/jbc.M112.445056
18. LimJLeeJJLeeSKKimSEumSYEohH2021Phosphoenolpyruvate depletion mediates both growth arrest and drug tolerance ofMycobacterium tuberculosisin hypoxiaPNAS118e2105800118 doi.org/10.1073/pnas.2105800118
19. LopatkinAJStokesJMZhengEJYangJHTakahashiMKYouLCollinsJJ2019Bacterial metabolic state more accurately predicts antibiotic lethality than growth rateNature Microbiology421092117 doi.org/10.1038/s41564-019-0536-0
20. LopatkinAJBeningSCMansonALStokesJMKohanskiMABadranAHEarlAMCheneyNJYangJHCollinsJJ2021Clinically relevant mutations in core metabolic genes confer antibiotic resistanceScience371eaba0862 doi.org/10.1126/science.aba0862
21. MachováISnašelJZimmermannMLaubitzDPlocinskiPOehlmannWSinghMDostálJSauerUPichováI2014Mycobacterium tuberculosisphosphoenolpyruvate carboxykinase is regulated by redox mechanisms and interaction with thioredoxinThe Journal of Biological Chemistry2891306613078 doi.org/10.1074/jbc.M113.536748
22. MadiganCAMartinotAJWeiJRMadduriAChengTYYoungDCLayreEMurryJPRubinEJMoodyDB2015Lipidomic analysis links mycobactin synthase K to iron uptake and virulence inM. tuberculosisPLOS Pathogens11e1004792 doi.org/10.1371/journal.ppat.1004792
23. MaksymiukCBalakrishnanABrykRRheeKYNathanCF2015E1 of α-ketoglutarate dehydrogenase defendsMycobacterium tuberculosisagainst glutamate anaplerosis and nitroxidative stressPNAS112E5834E5843 doi.org/10.1073/pnas.1510932112
24. MarreroJRheeKYSchnappingerDPetheKEhrtS2010Gluconeogenic carbon flow of tricarboxylic acid cycle intermediates is critical forMycobacterium tuberculosisto establish and maintain infectionPNAS10798199824 doi.org/10.1073/pnas.1000715107
25. MiddaughJHamelRJean-BaptisteGBeriaultRChenierDAppannaVD2005Aluminum triggers decreased aconitase activity via Fe-S cluster disruption and the overexpression of isocitrate dehydrogenase and isocitrate lyase: a metabolic network mediating cellular survivalThe Journal of Biological Chemistry28031593165 doi.org/10.1074/jbc.M411979200
26. MoynihanMMMurkinAS2014Cysteine is the general base that serves in catalysis by isocitrate lyase and in mechanism-based inhibition by 3-nitropropionateBiochemistry53178187 doi.org/10.1021/bi401432t
27. MukhopadhyayBPurwantiniE2000Pyruvate carboxylase fromMycobacterium smegmatis: stabilization, rapid purification, molecular and biochemical characterization and regulation of the cellular levelBiochimica et Biophysica Acta1475191206 doi.org/10.1016/s0304-4165(00)00064-7
28. Muñoz-ElíasEJUptonAMCherianJMcKinneyJD2006Role of the methylcitrate cycle inMycobacterium tuberculosismetabolism, intracellular growth, and virulenceMolecular Microbiology6011091122 doi.org/10.1111/j.1365-2958.2006.05155.x
29. MurdochCCSkaarEP2022Nutritional immunity: the battle for nutrient metals at the host-pathogen interfaceNature Reviews. Microbiology20657670 doi.org/10.1038/s41579-022-00745-6
30. MurrayJFSchraufnagelDEHopewellPC2015Treatment of tuberculosis. A historical perspectiveAnnals of the American Thoracic Society1217491759 doi.org/10.1513/AnnalsATS.201509-632PS
31. NandakumarMNathanCRheeKY2014Isocitrate lyase mediates broad antibiotic tolerance inMycobacterium tuberculosisNature Communications54306 doi.org/10.1038/ncomms5306
32. PaulAMishraASuroliaAVijayanM2013Cloning, expression, purification, crystallization and preliminary X-ray studies of argininosuccinate lyase (Rv1659) fromMycobacterium tuberculosisActa Crystallographica. Section F, Structural Biology and Crystallization Communications6914221424 doi.org/10.1107/S1744309113031138
33. PisuDHuangLGrenierJKRussellDG2020Dual RNA-Seq of Mtb-infected macrophages in vivo reveals ontologically distinct host-pathogen interactionsCell Reports30335350 doi.org/10.1016/j.celrep.2019.12.033
34. RaoSPSAlonsoSRandLDickTPetheK2008The protonmotive force is required for maintaining ATP homeostasis and viability of hypoxic, nonreplicatingMycobacterium tuberculosisPNAS1051194511950 doi.org/10.1073/pnas.0711697105
35. SchraderSMBotellaHJansenREhrtSRheeKNathanCVaubourgeixJ2021Multiform antimicrobial resistance from a metabolic mutationScience Advances7eabh2037 doi.org/10.1126/sciadv.abh2037
36. SerafiniAPisuDPalùGRodriguezGMManganelliR2013The ESX-3 secretion system is necessary for iron and zinc homeostasis inMycobacterium tuberculosisPLOS ONE8e78351 doi.org/10.1371/journal.pone.0078351
37. SerafiniATanLHorswellSHowellSGreenwoodDJHuntDMPhanMDSchembriMMonteleoneMMontagueCRBrittonWGarza-GarciaASnijdersAPVanderVenBGutierrezMGWestNPde CarvalhoLPS2019Mycobacterium tuberculosisrequires glyoxylate shunt and reverse methylcitrate cycle for lactate and pyruvate metabolismMolecular Microbiology11212841307 doi.org/10.1111/mmi.14362
38. SerafiniA2021Interplay between central carbon metabolism and metal homeostasis in mycobacteria and other human pathogensMicrobiology16734080971 doi.org/10.1099/mic.0.001060
39. ShiSEhrtS2006Dihydrolipoamide acyltransferase is critical forMycobacterium tuberculosispathogenesisInfection and Immunity745663 doi.org/10.1128/IAI.74.1.56-63.2006
40. SinghRLemireJMaillouxRJChénierDHamelRAppannaVD2009An ATP and oxalate generating variant tricarboxylic acid cycle counters aluminum toxicity in Pseudomonas fluorescensPLOS ONE4e7344 doi.org/10.1371/journal.pone.0007344
41. SoomSMoningSUCookGMLingfordJPKroppATranSGrinterRGreeningCvon BallmoosC2025ATP synthesis driven by atmospheric hydrogen concentrationsPNAS122e2506353122 doi.org/10.1073/pnas.2506353122
42. TianJBrykRShiSErdjument-BromageHTempstPNathanC2005Mycobacterium tuberculosisappears to lack alpha-ketoglutarate dehydrogenase and encodes pyruvate dehydrogenase in widely separated genesMolecular Microbiology57859868 doi.org/10.1111/j.1365-2958.2005.04741.x
43. VijaySHaiHTThuDDAJohnsonEPielachAPhuNHThwaitesGEThuongNTT2017Ultrastructural analysis of cell envelope and accumulation of lipid inclusions in clinicalMycobacterium tuberculosisisolates from sputum, oxidative stress, and iron deficiencyFrontiers in Microbiology82681 doi.org/10.3389/fmicb.2017.02681
44. WangZSoniVMarrinerGKanekoTBoshoffHIMBarryCERheeKY2019Mode-of-action profiling reveals glutamine synthetase as a collateral metabolic vulnerability ofM. tuberculosisto bedaquilinePNAS1161964619651 doi.org/10.1073/pnas.1907946116
45. WellsRMJonesCMXiZSpeerADanilchankaODoornbosKSSunPWuFTianCNiederweisM2013Discovery of a siderophore export system essential for virulence ofMycobacterium tuberculosisPLOS Pathogens9e1003120 doi.org/10.1371/journal.ppat.1003120
46. WHO2024WHO bacterial priority pathogens list 2024: bacterial pathogens of public health importance, to guide research, development, and strategies to prevent and control antimicrobial resistance licence: CC BY-NC-SA 3. 0 IGOWorld Health Organization
47. WongDKLeeBYHorwitzMAGibsonBW1999Identification of fur, aconitase, and other proteins expressed byMycobacterium tuberculosisunder conditions of low and high concentrations of iron by combined two-dimensional gel electrophoresis and mass spectrometryInfection and Immunity67327336 doi.org/10.1128/IAI.67.1.327-336.1999
48. ZhengXAv-GayY2016New era of TB drug discovery and its impact on disease managementCurrent Treatment Options in Infectious Diseases8299310 doi.org/10.1007/s40506-016-0098-0

---

Republished from the open web under CC-BY. Authors: Serafini A, Garza-Garcia A, Sorze D, de Carvalho LPS, Manganelli R. Read the original.