Tuberculosis (TB) remains among the top 10 causes of death worldwide ¹ . In 2017, an estimated 10 million people developed new TB disease, of which 90% were aged 15 or older ¹ . This lacking control of TB in a large number of people emphasizes the complex disease TB represents, involving a highly persistent bacterium, Mycobacterium tuberculosis, and the variable abilities of individual human immune systems to cope with this infection. As TB disease may also occur in certain individuals who have been vaccinated during early childhood with the widely used attenuated Mycobacterium bovis Bacillus Calmette–Guérin (BCG) vaccine, this situation also points to the shortcomings of BCG, the only licensed vaccine against TB today . BCG provides protection against severe, disseminated TB in children, but protection often fades in adolescence and fails to protect against pulmonary TB. The reported efficacy of BCG against adult pulmonary TB varies hugely (0–80%) and seems to depend on many factors, including social conditions, co-infection with helminths, and other variables ^{2– 5} . BCG can generally be regarded as safe, although individuals with immune deficiencies can develop disseminated BCG disease ² . Thus, despite the millions of childhood TB cases prevented by large-scale BCG vaccination strategies, there is a need for a more effective vaccine that can interrupt the vicious infection cycle of M. tuberculosis, which is driven by adolescent and adult patients with pulmonary TB disease. In this review, we will thus focus on selected mycobacterial virulence factors that play a role as potent immunogens as well as potential targets for the attenuation of novel vaccine strains.

An accurate definition of mycobacterial virulence is quite difficult to find, as many factors may influence the survival of M. tuberculosis in specific environmental conditions. Apart from essential genes that are needed by M. tuberculosis for its survival in commonly used mycobacterial growth media ^{6, 7} , its survival in the hostile environment inside host phagocytes requires additional skills from the bacterium. An overview of such in vivo-required genes was obtained by using libraries of transposon mutants of M. tuberculosis strains in murine infection models, unraveling between 200 and 500 genes essential for the growth of M. tuberculosis in vivo ^{8, 9} . Among these genes, several were also identified as important virulence determinants of M. tuberculosis by comparative analyses of attenuated and virulent tubercle bacilli, including genes encoding proteins involved in secretion systems, persistence mechanisms, or lipid metabolism ^{10– 17} , as further discussed below.

Most successful vaccines act against infectious agents which can be defeated by humoral immune responses, while it has been proven difficult to design successful vaccines that rely on cellular immune responses, such as in TB ⁴ . It is undisputed that in TB the CD4 ⁺ Th1 response is the major factor conferring protection ^{1, 83– 85} . Only fairly recently has evidence emerged that other immune cell subsets such as CD8 ⁺ T cells, NK cells, Th17 cells, or B cells may provide an important contribution to protection ^{4, 86– 90} . Other factors are well known to play a destructive role, such as Th2 cells and T-regulatory cells/secreted IL-10. Moreover, active TB disease appears to be associated with strong recruitment of neutrophils and type I interferon (IFN) signaling, though the role of type I IFNs in TB remains disputed ^{4, 91– 94} . The innate immune response has a crucial role in protection, especially in early infection control ^{95, 96} .

ESAT-6 causes phagosomal rupture in the infected host cells, permitting cytosolic contact of the mycobacteria or their secreted products ^{18, 60, 97} . The results of this process are varied and may also have consequences for infection control.

Mycobacterial dsDNA that is likely released because of phagosomal rupture into the cytosol is sensed by absent in melanoma 2 (AIM2), which induces the NLRP3 inflammasome to activate caspase-1, which in turn induces the cleavage of pro-IL-1β and pro-IL-18 to active, pro-inflammatory IL-1β and IL-18 ^{18, 98} . Moreover, the link between cytosolic access of mycobacteria and type I IFN signaling has clearly been established. dsDNA in the cytosol activates cyclic GMP-AMP synthase (cGAS), leading to the signaling of the second messenger cGAMP, which activates the STING pathway (stimulator of IFN genes). The latter is located at the endoplasmic reticulum and triggers Tank-binding kinase 1 (TBK-1), which activates IFN regulatory factor 3 (IRF3), leading to the production of IFN-β ( Figure 3). Importantly, cGAMP also activates neighboring cells for the production of type I IFN, augmenting the production of IL-10 and IL-1Ra in these bystander host cells, leading to increased host cell necrosis and tissue damage ^{18, 94, 99– 101} . On the other hand, Banks et al. recently reported a protective bactericidal effect of IFN-β, activating nitric oxide (NO) production, which has been shown to act in a bactericidal and anti-inflammatory manner, leading to killing of the bacilli and preventing tissue damage ⁹⁴ . Banks et al. suggest a scenario in which M. tuberculosis inhibits possibly beneficial autocrine IFN-β signaling while allowing possibly detrimental paracrine IFN-β signaling, while non-pathogenic mycobacteria such as M. smegmatis cannot inhibit autocrine IFN-β signaling, which is why the beneficial properties of autocrine type I IFN signaling predominate ^{94, 102} . It is probable that the role for type I IFN will not be black and white in TB but ambivalent, for example, depending on the timing of signaling during infection (early, late) or the mode of signaling (autocrine, paracrine).

It has been shown that virulent M. tuberculosis H37Rv induces necrosis of infected host cells, leading to damage to the mitochondrial inner and outer membrane, and it has been proposed that this could be mediated by ESAT-6, which also leads to the induction of type I IFN by sensing of mitochondrial DNA, while non-virulent H37Ra causes damage only to the outer membrane of host mitochondria, not leading to necrosis ¹⁰³ . Necrosis is an uncontrolled process of cell death and a destructive process in mycobacterial infection. Meanwhile, apoptosis represents a highly controlled process of cell death. Parts of the apoptotic cell as well as mycobacteria disassemble, safely packaged in apoptotic bodies, and are eventually taken up via efferocytosis by phagocytes. This process prevents tissue damage and spread of the bacilli and may enhance the activation of naïve professional antigen-presenting cells (APCs), especially dendritic cells (DCs), taking up mycobacterial components. However, this may also lead to infection. The “fate” of the phagocyte is largely dependent on the cocktail of cytokines present, Th1 cytokines being correlates of protection. A Th1-dominated cytokine cocktail produces classically activated macrophages (CAM, M1) and the production of NO, promoting killing of the bacilli and augmenting the Th1 response, while a Th2-dominated cocktail produces alternatively activated macrophages (AAM, M2) that express arginase-1, are impaired in bacterial killing, and augment Th2 responses ¹⁰⁴ . Moreover, Th1 cytokines stimulate autophagy while Th2 cytokines inhibit autophagy ^{103, 105– 107} . Autophagy has been described as an essential process in the control and clearance of mycobacterial infection, as mice lacking the essential autophagy factor Atg5 are highly susceptible to infection ^{108, 109} . However, this process may also be entirely due to excess neutrophilic recruitment and resulting tissue damage in ΔAtg5 mice, not to a lack in autophagy, as another study with mice deficient in several players of the autophagy pathway in macrophages, other than Atg5, did not show any higher susceptibility to M. tuberculosis ^{18, 110} . In the latter study, autophagy in macrophages appeared to be insignificant for the outcome of infection with M. tuberculosis. However, this finding might have been influenced by the fact that macrophages are not the most important effectors of autophagy compared to DCs, as DCs are needed for the priming of naïve Th1 cells, which eventually produce IFN-γ to activate phagocytes for the clearance of the bacilli ¹¹⁰ . Evidence has pointed towards a scenario in which ESAT-6 inhibits autophagy at the step of phagosome–lysosome fusion, resulting in reduced secretion of IL-12 in DCs needed for the induction of Th1, which could be restored by rapamycin therapy ^{111, 112} . However, cytosolic access of mycobacteria may also stimulate autophagy via marking of mycobacterial compounds in the cytosol with ubiquitin for selective autophagy or via the AIM2/inflammasome pathway, as seen with the vaccine candidate VPM-1002 ¹¹⁵ . Moreover, cytosolic contact of mycobacteria via the AIM2/inflammasome/IL-18 pathway has been reported as an essential mechanism for the production of IFN-γ by M. tuberculosis-independent memory CD8 ⁺ T cells and NK cells, thus emerging as an important, immediate, and unexpected source of Th1 cytokines ⁸⁶ .

A recombinant BCG expressing the ESX-1 from M. tuberculosis provided improved protection over BCG but, as a downside, was also more virulent in murine models and therefore not an easily usable vaccine candidate ^{10, 114} . Following this observation, the idea to recombine BCG with the ESX-1 of less-virulent mycobacteria such as M. marinum was proposed, which resulted in the construction of a recombinant BCG expressing ESX-1 ^Mmar , which confers improved protection compared to BCG in different murine models. This candidate vaccine induced higher levels of CD4 ⁺/CD8 ⁺ T-cell responses, likely through the establishment of cytosolic contact via the heterogeneous ESX-1 ^Mmar system, which induces innate and adaptive signaling events that are not induced by standard BCG strains ¹¹³ ( Figure 3), while remaining at a similar safety level as parental BCG.

The picture of how exactly ESAT-6 influences intracellular signaling remains an unsolved puzzle. In the context of M. tuberculosis infection, ESAT-6 represents a harmful virulence factor, but, in the context of vaccination with recombinant BCG, ESAT-6 acts as a potent immunogen and initiator of important, additional signaling pathways ^{114, 116} .

For the ESX-5-associated PE/PPE proteins, strong immunogenicity has been demonstrated in murine models, inducing CD4 ⁺ and CD8 ⁺ T-cell responses. Importantly, this could also be induced via cross-reactivity by non-ESX-5-associated PE/PPE proteins in a knock-out mutant not expressing the ESX-5-associated PE/PPE, thanks to the redundancy of PE/PPE proteins in the mycobacterial genome and the conservation of the essential ESX-5 secretion system needed for the secretion of a multiplicity of PE/PPE proteins ^{43, 46} . The exact function of PE/PPE proteins remains largely unknown, but they have been proposed to be involved in mycobacterial capsule/cell-wall integrity, nutrient uptake, antigenic variation, immunodominance, mycobacterial growth in macrophages, and inhibition of phagosome maturation through the induction of phagosomal rupture ^{14, 43, 117, 118} . Thus, certain PE/PPE proteins might have an impact on intracellular signaling that is similar to that of ESAT-6/CFP-10 discussed above, while other features such as nutrient uptake may not be such an important factor for immunogenicity but an interesting target to be considered for the attenuation of possible vaccine candidates.

Lipid surface factors such as pathogen-associated molecular patterns play a crucial role during the early phase of infection, aiding the establishment and persistence of infection. They interact with various types of pathogen recognition receptors (PRRs) on phagocytes, which can lead to the priming of phagocytes, the production of chemokines/cytokines for the recruitment of other players of the immune system, and the production of antimicrobial products or can reversely hamper activation of the phagocyte and mediate the silent entry of the bacilli ^{119, 120} . One of the most thoroughly studied lipid surface factors is cord factor/TDM, which was first described in the 1950s ¹²¹ . TDM stimulates PRRs, inducing the production of cytokines and NO, influencing granuloma formation, and playing an important role as an adjuvant; however, it may also be involved in the delay of phagosome maturation ^{57, 122, 123} . Cord factor is abundantly present in all mycobacteria, although subtle differences in mycolate structure seem to play a role in the virulence and pathogenicity of varying mycobacterial species ⁵⁷ . Other lipid surface factors such as DAT and PAT are associated only with mycobacteria belonging to the MTBC and SL is present only in M. tuberculosis strains ¹²⁴ . Synthesis of DAT, PAT, and SL needs specific polyketide synthase systems (PKSs). DAT and PAT are associated with the delay of phagosome maturation, and SL has been described as a TLR-2 antagonist, which makes PKSs suitable targets for possible attenuation and drugs ^{57, 125, 126} . Other surface lipids have been lost in the MTBC, such as the above-mentioned LOS, because of a recombination event in the pks5 locus ^{59, 127} . Interestingly, the lineage 4 strains of the Euro-American lineage of M. tuberculosis strains harbor a frameshift mutation in the pks15/1 gene, which is not present in lineage 2 Beijing strains and in turn leads to the synthesis of phenolic glycolipids (PGLs) in these latter strains ^{57, 128, 129} . This phenomenon has been postulated to be linked with increased virulence of Beijing strains ¹³⁰ , although other lineage-specific factors also seem to interfere ¹³¹ . Lipid surface factors thus play a dual role in mycobacteria as immune response boosters, as virulence factors, or even as attenuation factors and could simultaneously be employed as vaccine adjuvants, as drug targets, or for attenuation ¹¹⁹ . Moreover, lipid surface factors can directly activate the adaptive immune responses via recognition by CD1; the exact role of this alternative activation of T-cells will need to be further elucidated ^{132, 133} .

There is some skepticism over whether it will be possible to develop a truly effective vaccine against TB, but there is also some evidence that gives hope for a positive outcome. First of all, it is the existence of BCG, which does provide excellent protection against miliary and meningeal TB in children. Second, it is the fact that previous M. tuberculosis infection or controlled LTBI does seem to provide some protection against the development of active TB disease after reinfection, as 90–95% of people infected with M. tuberculosis exhibit a protective immune response and as a result do not develop active TB in their lifetime ¹ . In the 1930s, a study from Norway of nurses working in a TB hospital, a high-risk population, showed that nurses with a positive TST prior to their employment at the hospital had a 96% risk reduction of developing active TB compared to nurses who had a negative TST ¹³⁴ . A review of 18 prospective clinical trials from 2012 confirmed this with a 79% reduced risk of progression to active TB after reinfection in individuals with LTBI compared to previously non-infected people ¹³⁵ . However, LTBI in itself poses a risk to the development of active TB when control of M. tuberculosis is lost, for example because of age or immunosuppression. Moreover, it has been described that the rate of reinfection TB, defined as a recurrent TB episode with a different strain, in individuals who had previously already developed active TB disease and had successfully been treated, is higher than the rate of new TB cases in the population, suggesting that people who had TB once are at a strongly increased risk of developing TB when reinfected ¹³⁶ . Indeed, this underlines the variability and importance of individual human immune systems to cope with the infection. One interesting attempt to improve vaccine research is thus to identify correlates of protection by studying the protective immune response in people with LTBI ⁵¹ . Simultaneously, new vaccine candidates are often designed following two simple principles to either make BCG more immunogenic by introducing strong immunogens such as ESAT-6 or render M. tuberculosis attenuated by the deletion of important virulence factors ⁴ . Given the major advances that have been made recently in the genomic and cellular characterization of virulence factors and virulence strategies of the members of the MTBC and closely related tubercle bacilli, such as M. canettii, the challenge is now to transpose this knowledge into practice concerning vaccine research. There are some promising leads, which include a number of recombinant BCG strains that show improved protection in preclinical models ^{86, 113, 137, 138} . In addition, some recombinant BCG strains have been or are presently in clinical trials ^{72, 139} , and very recent results have also suggested that revaccination with BCG has a positive impact against infection with M. tuberculosis ¹⁴⁰ . Alternatively, there is hope that rationally attenuated M. tuberculosis strains may induce better protection than standard BCG vaccination, as suggested by different preclinical studies ^{46, 141} . The fact that the attenuated M. tuberculosis strain MTBVAC is in clinical evaluation is an encouraging sign that attenuated M. tuberculosis strains may be part of the future vaccine strategies against M. tuberculosis infection and TB disease.

From a more upstream research prospective, the identification of novel virulence factors of pathogenic mycobacteria will certainly continue and might provide new perspectives for vaccine research. Several mycobacterial species that have been considered as most closely related to M. tuberculosis, such as M. marinum or M. kansasii, have been used to identify novel mycobacterial virulence factors ^{142– 145} . However, according to the most recent genome analyses, several new, recently described mycobacterial species are much more closely related to M. tuberculosis than M. marinum and M. kansasii ^{146, 147} . Indeed, several virulence factors, such as the fumarate reductase locus, the SL locus, or numerous toxin–antitoxin systems that have been considered as being present only in M. tuberculosis and/or M. canettii, can also be found in this group of closely related mycobacteria, which constitute a common clade with MTB that was named MTB-associated phylotype (MTBAP) ¹⁴⁷ . Future research will show if some of these virulence factors might have importance in vaccine research.