Our analysis suggests that during dormancy M. tuberculosis performs limited protein synthesis. The pathogen spends its available resources on maintaining cell wall, membrane potential and genome integrity as well as resisting host defenses.
The main objective of our analysis was to develop a prioritized list of genes that might be suitable targets for drugs effective against non-replicating bacilli (Fig 1, Step 4). The foremost assumption is that genes that are highly expressed in a number of experimental situations are important for persistence of non-replicating M. tuberculosis. This was not an absolute criterion but pathways or multimeric complexes in which only a single gene was up-regulated were judged to be less important or anomalous results and, therefore were considered to be lower priority targets. The knockout experiments address a fundamentally different question than do the dormancy experiments (e.g. replication versus non-replicative survival). The focus of this analysis is on the non-replicative state. Targets that show both up-regulation in the dormant state and inhibit growth are certainly more interesting, but it may well be the case that genes required for survival in the non-replicating state have no effect on growth. As an example, Rengarajan et al. found that genes expressed under hypoxic or low nutrient conditions simulating the granuloma environment, are not necessarily essential for intracellular growth in macrophages . In bacteria, growth essential genes are often constitutively expressed while virulence and environmental response genes tend be highly regulated yet essential during those periods of high induction. Thus one would expect that different groups of genes play distinctive roles, and have transient importance for viability, in the non-replicative versus replicative phases of the bacterium's life cycle.
Here we discuss the relative tractability of proposed dormancy phase modulated genes and pathways as drug targets for the eradication of dormancy phase M. tuberculosis. The individual genes and scores discussed below are listed in Additional file 4. Since not every gene in a particular pathway scored highly, we included additional low scoring pathway proteins in order to provide context for targeting particular pathways.
devR regulon and other regulatory systems
An effective immune response that halts disease progression typically results in formation of granulomas. It is in these granulomas that the non-replicative bacilli are believed to reside [32,33]. Based on the nature of the granuloma, it is very likely that the non-replicating bacilli are surviving in an hypoxic environment. Granulomas are necrotic, avascular lesions that are walled off from the immune system. They are made up of caseous (chalky) centers and contain dead macrophages, T-cells and the tubercular bacilli [34,35]. This environment also has ramifications for the nutrient supply.
The results here and elsewhere [18,19,36,37] clearly show that the devR regulon (also known as the dosR regulon) is a significant regulator of genes in response to the experimental stressors of hypoxia and nitric oxide. This, combined with the production of nitric oxide by host immune cells and the likely hypoxic nature of the granuloma, makes a strong, albeit circumstantial, case for this regulon to be a primary trigger in the metabolic shiftdown to achieve the non-replicative state.
Additional evidence comes from Malhotra et al.  where guinea pigs were infected with wild type H37Rv M. tuberculosis and a devR::kan mutant. After 47 days the lungs of the animals infected with the mutant strain showed small granulomas (Mycobacterium bovis BCG, Boon and Dick showed that a devR::kan mutant grew at the same rate as wild type during the aerated phase of the Wayne model, but had a 1500 fold decrease in viability during the 5 week hypoxic phase . Mice lacking iNOS are rapidly killed by M. tuberculosis, whereas mice with functional iNOS yield chronic infections. One interpretation is that nitric oxide activates devS and devR, resulting in metabolic downshifting and latent infection. Without the nitric oxide signal, the regulon is not upregulated and cell division continues until the host is killed [40,41].
devS/devT devR two component sensor
Two component response regulator systems are found in bacteria, fungi, and plants [42,43] and are already being exploited as new antibiotic targets [44,45]. These systems consist of a sensor kinase (component 1) which autophosphorylates histidine, then transphosphorylates a conserved aspartate of its cognate transcriptional regulator (component 2). M. tuberculosis has eleven complete systems and seven orphans belonging to sensor kinase and response regulator families . It has been shown that recombinant sensor kinases devS and devT autophosphorylate to stable phosphoproteins, but rapidly transphosphorylate devR [46-48], confirming these as members of the two component family. Additionally several papers have identified devS/devT devR as a strong candidate for therapeutic intervention [15,47,49,6,52]. The results collected here are consistent with these analyses. It interesting to speculate that this approach may work not by killing M. tuberculosis directly, but by forcing them out of the non-replicative state,  leaving them susceptible to the usual anti-microbial agents. If the regulatory system having significant control over non-replicative metabolism were disrupted, this could lead to exit from the non-replicative state or a metabolic catastrophe. Ideally the cell would suffer unregulated expression with insufficient resources from which it could not recover. Paving the way for further work are recent papers which describe a validated high-throughput screening assay . and x-ray crystal structures of the C-terminus of devR, with and without DNA bound. . A fluorescence polarization assay has been described for binding of component 2 to its cognate DNA sequence .
A bifunctional inhibitor or a combination of compounds may be necessary to attack the sensor kinase since it took knockouts of both devS and devT to prevent phosphorylation of devR after 2 h of hypoxia . Blocking devR binding to its cognate DNA-binding sites might require a less complex therapeutic strategy. The ATP site here and in the universal stress proteins and mprA, mprB regulators (below) offers the opportunity to take advantage of the growing expertise in development of kinase inhibitors .
Universal stress proteins
Universal stress proteins (USPs) are a family of proteins that are produced by bacteria in response to a range of environmental stressors and cause a significant change in metabolism . In our analysis, five of the eight Usp genes in the M. tuberculosis genome score in the upper 5% of expression results. The exact function of USPs in TB pathogenicity is not fully understood although studies in other species suggest they likely play a key cellular role. The function of UspA from E. coli has been characterized by Nyström et al. . Increased expression of UspA to physiological levels caused a global change in protein expression and decreased cell growth on minimal media, and decreased expression (via insertional inactivation) results in a disruption of glucose metabolism and excretion of acetate. The authors proposed that UspA may have a direct effect on alteration of the flux of carbon through the primary metabolic pathways during arrested growth . UspA has autophosphoryating activity  but is also phosphorylated by other proteins, including TypA . If the parallels to other species hold, USPs in M. tuberculosis might affect the regulation of primary carbon metabolism, cause changes in the expression of a significant number of proteins, and generally modulate the metabolic shift to stasis. All eight of the Usp members found in M. tuberculosis show the ATP binding signature found in MJ0577 , for which there is a crystal structure with ATP bound . Since ATP-competitive inhibitors are well-known, the USPs are very attractive targets. There is also a crystal structure of Rv1636 (unfortunately the lowest expressing member) opening up rational design approaches.
mprA, mprB two component sensor
A second two component sensor kinase/transcriptional regulator, mprA/mprB, also appears to be involved in persistent infections. Biochemical characterization of the phosphorylated intermediates has established this as a bona fide two-component system . Knockout experiments show a survival disadvantage in long term infection in mouse models, though the effect is tissue specific. Mice co-infected with wild type H37Rv M. tuberculosis and an mprA::Kmr mutant were analyzed after 147 days. In the spleen, 2% of the culturable M. tuberculosis were mutants, in the lung, 25% were mutants, and in the liver, 75% were mutants .
Bacterial sigma factors play a role in transcriptional initiation. A recent review summarizes the current state of knowledge of the Mycobacterium family . Several sigma factors are up-regulated (sigB, sigE, sigH, and sigF) and there is a tantalizing functional connection between the Bacillus subtilus sigma factor sigF involved in sporulation and the M. tuberculosis homolog [65,66]. Cappelli et al. show sigG upregulation after 7 days in human macrophages, suggesting a role in adjusting transcription to survival inside the macrophage . However, any novel anti-TB strategy based on the targeting of transcriptional components such as sigma-factors, would need to demonstrate some advantage over current rifampin therapy which also disrupts transcription through the inhibition of RNA polymerase activity. The mprA/mprB two-component system has been shown to directly regulate sigB and sigE . The effect of the mprA::Kmr mutant was not as dramatic as could be hoped for, but if the link with the sigma factors and the metabolic shutdown in sporulation is substantiated, then the result of inhibition of mprA or mprB may be to reactivate M. tuberculosis.
Under conditions where amino acids or carbon are limiting, bacteria shift their metabolism via a series of well-studied pathways known as the stringent response. Disruption of the stringent response is a potential mechanism to force M. tuberculosis to exit from dormancy phase. In M. tuberculosis, the stringent response is mediated by relA (aka RelMtb, Rv2583c) which synthesizes and degrades hyperphosphorylated guanine nucleotides that in turn affect RNA polymerase and gene expression [69-72]. Dahl et al. compared relA knockouts with wild type under nutrient sufficient and starvation conditions and analyzed the expression patterns . They concluded that relA globally down-regulates the cellular translation apparatus, including 54 of 58 ribosomal proteins. Fig. 4B shows the comparison of the expression profile of the down-regulated genes for the whole genome and the genes in Dahl et al. shown to be differentially regulated during starvation in M. tuberculosis wildtype but not in relA deletion mutants. relA showed a modest change in up-regulated expression, scoring 3.8, not far from the genome median of 3.1, and the positive scores do in fact come from the starvation conditions (Additional file 4). Hasan et al.  similarly scored relA low as a persistence target. However, despite the low score we feel that relA should be a good therapeutic target, especially if it disrupts the regulation of protein synthesis. It seems probable that devR and relA are working in concert, each responding to different environmental cues, to achieve the metabolic downshift to the non-replicative state.
The dual stresses of an unbalanced nutrient pool and hypoxia place significant demands on the regulation of metabolites through the two-carbon pathways. Transcriptional analyses [29,73] suggest M. tuberculosis survives in the host on a nutrient pool rich in fatty acids and poor in carbohydrates [17,74]. Fatty acids are broken down to acetyl-CoA by the five-step beta-oxidation pathway. M. tuberculosis has 34 acyl-CoA synthases, 34 acyl-CoA dehydrogenases, 21 enoyl-CoA hydratases, six 3-hydroxyacyl-CoA dehydrogenases, and six acetyl-CoA C-acetyltransferases. There are over 250 lipid metabolizing enzymes in M. tuberculosis compared to 50 in E. coli, which underlines the importance of fatty acids as a nutrient source for this bacillus . The postulate has been put forward that the main storage molecule for fatty acids are triacyl glycerols and 15 M. tuberculosis genes showed triacylglycerol synthase activity . A knockout of the most active and highly upregulated synthase, Rv3130c, named tgs1, completely eliminates accumulation of triacyl glycerols in hypoxic M. tuberculosis. As the authors suggest, inhibition of tgs1 could lead to starvation of non-replicative cells, but may be compromised by the redundant activity of the other synthases.
Acetyl-coenzyme A (Ac-CoA) is a central intermediate in primary metabolism with roles in the TCA cycle as well as fatty acid and amino acid biosynthesis. The flux of carbon through Ac-CoA is particularly critical to non-replicating M. tuberculosis. Fatty acids, via breakdown to Ac-CoA and use of the glyoxylate shunt, provide the carbon for carbohydrate synthesis and acetyl-coenzyme A appears to be the gate through which much of the utilized carbon pool must pass. The M. tuberculosis membrane, being rich in unusual carbohydrates [76,77], is likely critically dependent upon this carbon flux through Ac-CoA for the maintenance of its integrity.
Targeting coenzyme A biosynthesis should disrupt the flux through several pathways. Supporting this notion are inhibitors of pantothenate kinase that suppress the growth of the low GC Gram-positive bacterium, Staphylococcus aureus, in a typical MIC assay . However pantothenate biosynthesis knockouts in M. tuberculosis in mouse infection experiments suggest the operation of some salvage mechanism . PanC and panD are two of the five genes needed for de novo biosynthesis. A double deletion mutant was constructed and tested in immunocompromised SCID and normal BALB/C mice. In both strains the panCD knockouts show greatly attenuated virulence. The SCID mice injected with the double mutant live seven times longer than those injected with the wild type H37Rv M. tuberculosis. The double mutant does not kill any of the BALB/c mice, even after 1 year. Yet in both cases there is significant bacterial load. The lungs of the SCID mice eventually reach >108 cfu, likely the cause of death. The lungs of the BALB/C mice reach almost 106 cfu after 2 weeks and show severe, diffuse granulamatous pneumonia; the infection is slowly cleared over the next months. The spleen and liver show similar pathologies. The mechanism that allows this persistence is not clear. The panCD knockouts are able to grow in SCID and BALB/C mice yet have greatly attenuated virulence. The authors suggest an unidentified permease is salvaging enough pantothenate to survive but not enough to cause disease. In order for a pantothenate biosynthesis inhibitor to be effective, the supply of exogenous pantothenate must be exhausted. This is possible since, as mentioned above, the granulomas formed in the mouse do not show the central caseating necrosis found in the human lung. Testing the panCD knockout in a model that closely replicates the human granuloma is critical to the validation of this pathway as a potential drug target.
Under hypoxic conditions there is a lack of terminal electron acceptors and M. tuberculosis must reoxidize reduced metabolites to maintain viability. Maintaining redox balance, especially in the face of heavy use of fatty acids, is critically dependent upon the fate of the acetyl unit as it is either oxidized for energy, utilized for biosynthesis, or shunted off to intermediates to hold reducing equivalents. Eventually some of these reduced compounds must make their way to terminal electron acceptors.
Knockouts of the genes encoding two isocitrate lyase isoforms (icl1 and icl2) have shown that disruption of reactions close to or within these primary metabolic steps are effective and it has long been noted as a promising therapeutic target [74,80]. Its effectiveness in knockout experiments and with a prototype inhibitor (3-nitropropionate) is likely due to its role in enabling the utilization of the Ac-CoA pool for carbohydrate synthesis [81,74,80,82]. The 3-dimensional structure of the enzyme will also aid the discovery of inhibitors of ICL .
Genes for PEP (phosphoenolpyruvate) carboxykinase and a citrate synthase, both involved in carbohydrate metabolism, are also up-regulated. Three genes are annotated as citrate synthases, converting oxaloacetate and Ac-CoA to citrate and CoA. Cit3 shows significant up-regulation and could be an attractive target if the other two enzymes do not provide a functional alternative. Malate dehydrogenase (mez), another enzyme intimately involved in primary metabolism, shows modest up-regulation.
Alanine dehydrogenase has been suggested to be important both for the production of alanine (utilized in peptidoglycan) and for maintenance of the NAD pool [84,85]. The kinetic behavior of the enzyme strongly suggests that its primary function is to catalyze the formation of alanine. The M. smegmatis alanine dehydrogenase is also suggested to have glycine dehydrogenase activity . If this holds for M. tuberculosis, inhibition of this enzyme would also disrupt the glyoxylate pool. The potential effect on a peptidoglycan precursor, redox balance, and carbohydrate synthesis coupled with its upregulation make this enzyme a viable target. The protein was also identified in a proteomics analysis of the Wayne model .
Another strategy to disrupt block redox balance is to attack the key carrier of reducing equivalents: NAD(P)+. Taking into account both avoidance of enzymes with close human orthologues and inclusion of those genes with high expression scores, the best target appears to be nadR. NadR is a bifunctional enzyme with nicotinamide-nucleotide adenylytransferase and ribosylnicotinamide kinase (RNK) activities.
Another drug targeting opportunity created by hypoxia is the need for M. tuberculosis to utilize alternate terminal electron acceptors as the final step of respiration. Wayne and Sohasky  have proposed that nitrate is reduced by a nitroreductase (narGHJ) and is then excreted by a nitrite extrusion protein (narK1, narK2, narK3). This scheme is supported by the transcript levels of proteins in the respiratory chain.  Knockouts of narG in a SCID mouse model with M. tuberculosis BCG show a decreased virulence of the mutant compared to wild type , leaving open the possibility that blocking nitrate reduction could be effective against a truly hypoxic adapted M. tuberculosis. Nitrate reduction is controlled by the strongly up-regulated narK2, also a member of the devR regulon . Sohaskey has shown that the presence of narK2 is responsible for the 75 fold difference in nitrite synthesis between actively growing, oxygenated cultures and hypoxic, non-replicative bacilli. This represents another strategy to take advantage of hypoxia.
Knockouts of the 'fused reductase' narX have no effect so this gene appears to be inactive . Additionally, three genes which are significantly up-regulated contain the nitroreductase ortholgous group, COG0778 (Rv2032 (acg), Rv3131 and Rv3127). However the lack of nitrate reductase activity after a narG knockout argues against these as respiratory inhibition targets .
Moving up the reaction path leads to the proteins in the respiratory chain. There are several redox proteins that have high to moderate up-regulation scores and are likely involved as alternate electron carriers in the hypoxic environment: fumarate reductase (frdA), probable NAD(P)H dehydrogenases (Rv3054c, Rv0082, Rv1812c, and Rv1854c), and ferridoxin (fdxA). Ferridoxin is the most significantly up-regulated of the low-redox potential electron carriers . Recent studies show that phenothiazines inhibit NADH:menaquinone oxidoreductase (Rv1854c, Rv0392c) and stop growth in acute models [91,92]. This class of inhibitor may force the respiratory chain into an oxidized state and M. tuberculosis to behave as if it were in an aerobic environment. The potential of this mechanism to signal the bacilli to switch its metabolism back to the replication, hence more vulnerable to existing antibiotics, is an intriguing therapeutic approach.
The liability with the respiratory chain targets is the potential for redundancy. It is unknown which or how many enzymes to inhibit for a lethal blockade. However, as mentioned above, inhibition may not be the best approach. Uncoupling electron transfer with generation of the proton motive gradient or phosphorylation of ATP should be rapidly lethal. Rv1812, Rv1854c and the fumarate reductase complex are believed to be components of the oxidative phosphorylation chain and could take advantage of the uncoupling strategy. The human orthologue to fumarate reductase, NP_004159, shows 35% identity and 51% similarity to the M. tuberculosis enzyme at the amino acid level, so there is the potential for toxicity that would have to be addressed.
Downregulation of ATP synthesis
Since M. tuberculosis appears to have limited available ATP in dormancy phase, targeting ATP synthesis also looks to be a promising strategy. The newly discovered diarylyquinoline inhibitor of the ATP synthase proton pump  could be a valuable tool for establishing whether this approach would be effective against non-replicating cells. Compounds with similar modes of action as 2,4-dinirophenol that uncouple ATP synthesis from proton movement may also work provided they could be made selective for M. tuberculosis [94,95]. It is unclear whether targeting processes that utilize ATP would result in cell death. It is possible that they are occurring so slowly already (due to low availability of ATP) that inhibition would have a minor effect. An alternative approach would be to deplete available ATP stores. A compound that caused a futile ATP cycle could be very effective in killing dormant cells. For example if an ATP dependent transporter could be reduced so as to be only 1% efficient, this could drain the ATP pool and leave the bacterium unable to recover.
Sulfate transport and metabolism
Wooff et al.  have studied sulfate transport knockouts in M. bovis in detail. The sulfate transporter is made up of the proteins from cysA1, cysT, cysW, and subI. Transposon insertions into cysA1 or subI yield methionine auxotrophs completely unable to import sulfate which indicates that cysA1 is the sole sulfate transporter. They conclude that cysA2, and cysA3 are in fact thiosulfate sulfurtransferases. M. tuberculosis has two additional genes that are up-regulated and also annotated as possible sulfate transporters (Rv1707 and Rv1739c); it is not known whether these are active.
Wooff et al. also inoculated BALB/c mice with the cysA1 auxotroph and found no difference in growth compared to wild type, suggesting that the mutant was able to survive well using exogenous sources of methionine. Recently, it has shown that M. tuberculosis can convert methionine into cysteine . The expression results analyzed here imply that sulfur is deficient within M. tuberculosis. Whether this situation can be remedied by scavenging methionine is likely dependent upon the nutrient pool available in the local environment of the granuloma.
Once sulfate is transported in the cell, it is activated by conversion into adenosine 5'-phosphosulfate (APS) and 3'-phosphoadenosine 5'-phosphosulfate (PAPS). The genes for these enzymes (cysD and cysN) are significantly up-regulated. M. tuberculosis CysN, also known as cysNC, contains two fused activities, sulfate adenylyltransferase (cysN) and adenylylsulfate kinase (cysC) . These genes have been cloned and the recombinant enzymes are functional and produce PAPS from sulfate, ATP and GTP [98,99]. Recent work has shown that in M. tuberculosis APS, not PAPS, is utilized for cysteine and methionine biosynthesis . Several genes in this pathway are up-regulated: cysD, cysN (transferase), metC, and cysK2. The same issue arises here as with the transporters: will methionine salvage outflank the inhibitory blockade? As above this remains to be tested with a necrotic granuloma.
There is an additional possibility, however. Another role of sulfate is in the synthesis of sulfated lipids. (See reference  for a review of sulfate metabolism in M. tuberculosis.) Sulfated lipids are synthesized from PAPS and known targets to date are cysC (kinase domain) and Rv1373 . Inhibition of sulfolipid synthesis may be effective in the early phases of infection ; it is not yet known what the role of sulfated lipids is in persistent infection, but the hope is that the up-regulation of the genes for APS, PAPS and sulfolipid synthesis are indicative of a critical function.
CtpF and ctpG are annotated as probable ATP dependent cation transporters. If the annotations are in fact correct then these are promising targets. They have the advantage of avoiding the salvage problem above and opening up the opportunity for futile hydrolysis of ATP. However, the high similarity of these two proteins with human orthologs could provide challenges in the development of M. tuberculosis selective inhibitors (Additional file 4).
Downregulation of protein synthesis
Rifampin, a transcriptional inhibitor, is one of the few drugs effective against M. tuberculosis in the NRP-2 phase of the Wayne model . Rifampin thus serves as an important benchmark: it has efficacy against persistent M. tuberculosis, so any strategy that does not have a rationale for a shorter therapy should not make the list of new targets. We put proteins synthesis inhibitors into this category. This is borne out in clinical trials: streptomycin, an inhibitor of protein synthesis the binds the 30S ribosome, is not as effective as rifampin throughout a long treatment period [104,35].
The effectiveness of inhibiting ribonucleotide reductase in actively growing cells is shown by Dawes et al. . The large and small subunits of the type Ib reductase is encoded by nrdE and nrdF2, respectively. Knockout of nrdF2 is lethal, both validating this gene as a target, and showing that neither nrdF1 nor nrdZ could substitute. In contrast to our analysis (which show up-regulation of all the nrd genes), quantitative RT-PCR studies of nrdEF2 in the Wayne model shows a 10 fold decrease in expression from log phase to NRP-1, followed by a 2–3 fold increase from there as the cells progress into NRP-2. Dawes et al. also found surprising results with knockouts of nrdZ which encodes a putative class II reductase that is also up-regulated by devR. The expected scenario where nrdEF2 is down-regulated and nrdZ takes over during hypoxia was not observed. NrdZ knockouts had no effect on survival in the Wayne model or in B6D2/F1 mice, even after 14 months. So it appears that nrdEF2 is still supplying the needed pool of deoxyribonucleotides.
With regard to the main focus of our analysis, the real question is one of rate: is the rate of DNA damage to non-replicating M. tuberculosis high enough so that blocking repair will shorten therapy? The main impetus for developing ribonucleotide reductase inhibitors will certainly be to kill actively dividing cells. They may have the additional benefit of preventing cells from reactivating or leaving cells with a damaged genome to die.
As shown in Fig. 5, M. tuberculosis solely utilizes the pyruvate/glyceraldehyde phosphate pathway for isopentenyl-pyrophosphate biosynthesis and lacks the mevalonate pathway found in eukaryotes and some low GC Gram positive organisms . Isopentenyl-pyrophosphate is a precursor in the synthesis of decaprenyl phosphate. Decaprenyl phosphate is a 50-carbon lipid that has at least 2 functions in the productions of the mycolic acid-arabinogalactan-peptidoglycan complex of the cell wall . Nearly all of the enzymes of the cell wall biosynthetic pathway show low up-regulation scores (typically 0 – 4; as an example, murA, murB, murC, murD, murE, murF, and murG score 1.5, 0, 0, 2.2, 3.9, 0.3, and 3.8 respectively). Cell wall biosynthesis is already targeted by isoniazid (which is most effective in the first few days of TB treatment) , and the expression scores are low, so the argument can be made that targeting the cell wall is not interesting in the context of attacking non-replicating M. tuberculosis. However a recent review argues that maintenance of the membrane in persistent M. tuberculosis is important for long term viability, but the biochemical mechanisms are very difficult to determine . The function of decaprenyl phosphate as a carrier molecule may be more disruptive to downstream processes. We believe that this coupled with the knockout and up-regulation seen in the long term experiments makes this pathway a viable target.
Lysine is biosynthesized starting with aspartate and pyruvate, the so-called diaminopimilate pathway. The data here shows that this pathway is significantly downregulated. An altenative pathway exists in fungi, which starts with Ac-CoA and 2-keto-glutarate. The final enzyme in the fungal pathway, lysine-ε-aminotransferase, has been identified in M. tuberculosis while the other nine enzymes are either absent or yet to be annotated. Lysine-ε-aminotransferase is also found in several bacterial and fungal species. Its role in non-replicative metabolism is not clear. The high up-regulation score makes this an intriguing potential target. A crystal structure of this enzyme has recently been reported which could facilitate rational drug design .
Membrane protease regulating envelope composition
Rv2869c has been identified as an intramembrane cleaving protease (iCLIP) that regulates the composition of mycolic acids and some phosphatidylinositol mannosides in the M. tuberculosis cell envelope, as well as having transcriptional effects on other lipid metabolizing enzymes . A deletion of the protease had no effect on cell counts in infected mice after 24 h indicating that it is not critical for the acute phase of infection, however after 22 weeks the bacterial load in the lung was down 10,000 fold compared to wild type. The up-regulation score for this enzyme is not even at the genomic median, but the knockout results, as well as the lack of a closely related human homolog, make a compelling case for this as a therapeutic target.
Wayne and Sohaskey have suggested that the decreased effectiveness of rifampin in the non-replicative state in the Wayne model could be due in part to the stabilizing effect of the chaperonins . Thus a combination therapy of rifampin and a chaperonin inhibitor has the potential to shorten the therapeutic regimen. Alpha-crystallin 1 (Rv031c, hspX, acr1) has a large molecular chaperone domain  and is among the genes most highly induced in the devR regulon. The biology of alpha-crystallin 1 is complex as evidenced by the effect on growth when the genes are knocked out. Recent work has shown more rapid initial growth and increased colony counts in mouse lung and spleen in acr1 knockouts [109,110]. These findings suggest that acr1 has a role in the growth rate both during the initial stage of infection and long term persistence. The role of acr1 as a growth modulator raises the attractive possibility that interfering with its function could cause a beneficial switch to active replication.
The second of two alpha-crystallins in M. tuberculosis, encoded by the gene acr2, has one of the highest expression scores, consistent with Stewart et al.  where it was up-regulated at every stage of infection. A knockout of the acr2 gene had no significant effect on bacterial load in the mouse lung or spleen, though the mice did not experience as significant weight loss as those infected with wild type. An interesting experiment would be the effect of dual knockouts of both acr1 and acr2 genes in the presence of rifampin.
M. tuberculosis has other chaperonins; groEL1 and groEL2 show moderate up-regulation scores and should also work in concert with rifampin. A concern with these two proteins is that there are reasonably close mammalian orthologues. A human chaperonin (NP_955472) has 41% identity and 65% similarity with groEL1, and groEL2 (for which there is a crystal structure ) is 46% identical and 68% similar to another human chaperonin (NP_002147). Again there is the problem of redudancy, with at least four proteins capable of stabilizing proteins during persistence.