The plasmid-borne ε/ζ system was originally discovered on the low-copy number plasmid pSM19035 from a clinical isolate of the gram-positive pathogen S. pyogenes [³²]. This plasmid confers resistances to high levels of erythromycin and lincomycins [³³] and is member of the broad host family of Inc18-like plasmids. Although pSM19035 has a low-copy number (two to five plasmids per cell [³⁴]), it is maintained in absence of any selective antibiotic. This stable inheritance is conferred by plasmid-encoded segregation (seg) loci, whose gene products control the plasmid copy number, improve plasmid partitioning after replication, ensure a better-than-random segregation during cell division, and trigger PSK (see Lioy et al. [⁶]). Among those different maintenance mechanisms, PSK leading to suicide of plasmid-devoid offspring primarily contributes to stable inheritance of pSM19035 [²⁵]. PSK is evoked by the segB1 locus that encodes three proteins: ω, a global transcription repressor of pSM19035, the ε antitoxin, and the ζ toxin [³², ³⁵] (Fig. 3a). The proteins ε and ζ are translated from a bicistronic ε–ζ mRNA, whose transcription is regulated by the ω repressor [³⁶]. Upon plasmid loss, the pool of proteolytically unstable ε antitoxin is eliminated by ATP-dependent proteases and the released ζ prevents further proliferation of plasmid-free cells [²⁴, ²⁵, ³⁷].

More recently, a plethora of homologous ω–ε–ζ loci were identified on multiple resistance plasmids of other gram-positive pathogens such as Enterococcus faecalis and Enterococcus faecium [³⁸–⁴³] or Streptococcus agalactiae [⁴⁴]. Some of these ω–ε–ζ loci evoke stabilization of conjugable Inc18-like plasmids that encode for the vancomycin and teicoplanin resistance determinant vanA in enterococci. Importantly, such Inc18-like plasmids were not exclusively found in hospital strains but also in common human community and animal-derived vancomycin-resistant enterococci (VRE) strains [⁴⁵, ⁴⁶]. This suggests that these glycopeptide-resistant enterococci propagate within the food chain. VRE strains are capable of transferring vanA to other pathogens, such as the methicillin-resistant Staphylococcus aureus (MRSA) [⁴⁷–⁵⁰], and it is conceivable that ε/ζ systems also participate in the stable persistence of vanA in the resulting vancomycin-resistant S. aureus strains, which has evolved to a major public health concern.

A thorough sequence analysis divided ε/ζ systems into two distinct phylogenetic groups that differed in their efficiency to stabilize plasmids in E. faecalis [⁴², ⁴³]. Notably, these two different groups show sequence variation in the ε and ζ protein interface [²⁸, ⁴³] but are conserved in the active site of ζ. This suggests that fine-tuning of the ε–ζ affinity is crucial for the efficiency of plasmid stabilization and is an evolved, host-adaptive feature.

Originally, ε/ζ systems were presumed to be confined to gram-positive bacteria but were recently identified on IncP1 plasmids of Neisseria gonorrhoeae isolates from American and Dutch patients [¹⁴, ⁵¹]. Within the tetracycline determinant tetM of these plasmids, two different ε/ζ loci were identified, but no open reading frame (ORF) for an ω homologue was found. This suggests that the ε/ζ system from IncP1 plasmids is regulated differently when compared with those found in Inc18-like plasmids. ORFs encoding for ε/ζ systems were further identified on resistance plasmids in Bacteroides fragilis [⁵²]. A recent phylogenetic analysis of type II TA systems suggested that homologues of ε/ζ systems are widespread in all bacterial kingdoms [⁵³]. This study also showed that most ζ toxins reside on bacterial chromosomes rather than plasmids.

The human pathogen S. pneumoniae is one of the leading causes for bacterial pneumonia, otitis media, and invasive disease such as bacteremia and meningitis. More than 90 different pneumococcal serotypes have been classified [⁵⁴] which differ strongly in their virulence. However, only a limited number of ∼16 serotypes are found worldwide to cause the bulk of invasive diseases [⁵⁵, ⁵⁶]. Since a strong variation in virulence was reported for strains belonging to identical serotypes, the content of the polysaccharide capsule is not the main and exclusive determinant for pathogenicity [⁵⁷–⁵⁹]. Differences in virulence result rather from genomic heterogeneity caused by variable regions in the genome of S. pneumoniae. Genes in these accessory regions encode important virulence factors such as adhesion factors, transporters, and metabolic enzymes [⁵⁸, ⁶⁰–⁶²]. The chromosomally encoded ζ homologue PezT (Fig. 2) is expressed from the bicistronic pezAT operon (locus tag: SP1050/SP1051) located on the pneumococcal pathogenicity island PPI-1 [²¹, ⁵⁹, ⁶³] and, most importantly, was shown to be required to establish full virulence in mice [²¹]. However, unlike the otherwise stably integrated and conserved PPI region, the pezAT operon resides in the recombination hot-spot PPI-1v that is highly diverse in its gene content [⁵⁹]. Similar to the plasmid-borne ε/ζ systems, the bacteriotoxic UNAG kinase activity of PezT was shown to be inhibited by high-affinity binding of the antitoxin PezA [²⁸, ⁶⁴]. In contrast to plasmid-borne ε/ζ systems whose expression relies on the ω repressor, PezA harbors a transcription repressor domain in addition to its ε homology domain (Fig. 3b).

A major step in connecting PezT with virulence in S. pneumoniae came from the work by Brown et al. who compared clinical serotype 3 isolates with PezT knockout strains in mixed systemic and respiratory infections of mice [²¹]. The authors showed that disruption of the PezT ORF resulted in a variant that was outcompeted by wild-type strains in mouse challenge models. Only a single isolate out of 500 that were recovered from mice either 24 h after systemic infections or 72 h after respiratory infections contained a disrupted PezT ORF. Furthermore, a considerably attenuated infection progress in mice was observed when strains deprived of the PezT ORF were compared with wild-type strains. However, PezT-negative strains had no growth defect in laboratory broth, serum, or blood. This demonstrated that the phenotype of PezT disruption is virulence specific and not a general growth defect. Along this line, a comparison of the gene content of serotype 1 isolates from indigenous Australian patients revealed that the pezAT ORF was exclusively contained in the recombination hot-spot PPI-1v of hypervirulent isolates, whereas it was absent in the PPI-1v of noninvasive and intermediately virulent strains [⁵⁹]. Similar to as found by Brown et al., these hypervirulent strains caused rapid lethal infection in intranasally challenged mice, whereas the infection progress of the pezAT-negative, intermediately virulent isolates was strongly attenuated. Furthermore, when the pezAT-positive PPI-1v element of the virulent laboratory S. pneumoniae D39 strain was replaced by a pezAT-negative PPI-1v element from a noninvasive isolate, this hybrid strain was readily outcompeted by wild-type strains in nasal lavage fluid, lungs, and blood models [⁵⁹]. In contrast, replacing the D39 PPI-1v with one from the hypervirulent serotype 1 isolates resulted in a hybrid strain that was as effective in infection as the wild-type strain. It is noteworthy that the occurrence of pezAT locus is not the only difference within the gene content of these PPI-1v regions [⁵⁹]. Thus, the increased pathogenicity of pezAT-positive PPI-v regions does not unambiguously demonstrate that pezAT is the exclusive virulence factor within PPI-1v. Nonetheless, its exclusive occurrence in hypervirulent strains is striking.

The strong correlation of pezAT loci in PPI-1v and occurrence of hypervirulence suggests that the bacteria have taken advantage of the TA system to increase the speed of infection. It is rather unlikely that PezAT is a simple maintenance system of PPI-1v, since despite the pezAT locus, the other gene content of PPI-1v is highly variable. More likely, the originally as being detrimental perceived UNAG kinase activity of PezT could support pneumococcal virulence directly. PezT might be released from PezA during environmental stresses such as lack of nutrients, presence of antibiotics, or the immune response, as observed for other TA systems [¹⁶, ⁶⁵–⁶⁷]. In this scenario, UNAG-3P production would result in inhibition of cell wall synthesis and lysis of a pneumococcal subpopulation. The ability to undergo lysis under stress conditions is a peculiar feature of pneumococci. It is evoked either by self-induced autolysis [⁶⁸, ⁶⁹] or by fratricide, in which competent cells induce lysis of their noncompetent sister cells [⁷⁰]. Generally, lysis of gram-positive bacteria has great impact on virulence, since it leads to accelerated release of cellular components that are detrimental to the infected host. Components which are well established as virulence factors are teichoic acids, lipoteichoic acids, and bacterial DNA [⁷¹]. In addition, cell fragments from lysed pneumococci inhibit phagocytosis and impair phagocyte-mediated defense against living pneumococci [⁷²]. Most importantly, pneumococcal lysis leads to triggered release of pneumolysin (Ply), which has multiple detrimental effects for the host organism (see [⁷³]). Low Ply concentrations are already sufficient to trigger apoptosis [⁷⁴], to activate host complement [⁷⁵], and to induce proinflammatory reactions [⁷⁶]. At elevated concentrations, Ply causes general cellular damage in the host organism by its cholesterol-dependent ability to form large membrane pores [⁷⁷]. However, unlike other bacterial exotoxins, Ply is not directly secreted into the bacteria environment and its release relies on pneumococcal lysis after peptidoglycan degradation as, for example, caused by the major pneumococcal autolysin LytA [⁷⁸], and similar to PezT, impairment of Ply and LytA function also causes attenuated virulence in infection models [⁷⁹–⁸²]. Thus, it is likely that PezT indeed supports pneumococcal lysis causing cellular fragmentation and by triggered release of virulence factors.

A remaining conundrum is how activation of PezT is accomplished during infection, since PezA and PezT establish an extremely tight complex and house-keeping proteases of E. coli and B. subtilis were unable to disassemble this complex [⁶⁴]. This implies that a specific pneumococcal protease or protease adaptor protein is required to remove and degrade PezA from PezT. More experimental data are required to decipher the molecular mechanisms of PezAT's contribution to pneumococcal virulence. Likewise, the role of PezAT homologues in other streptococcal pathogens such as S. agalactiae and Streptococcus suis [⁶³] remains completely unexplored. Moreover, the presence of ζ homologues in the chromosomes of pathogenic E. coli and Capnocytophaga strains (Fig. 3) raises interesting questions about a general implication of the UNAG kinases family in virulence of gram-negative pathogens.

Drugs that disrupt normal operation of TA systems (Fig. 4a) are considered an attractive target for development of new therapeutic agents [⁸³–⁸⁵]. Pathogenic species could be specifically affected, without eliminating the bulk of the mutualistic bacterial flora. On the one hand, compounds could be developed that directly interfere with toxin inhibition by the antitoxin such as for instance disrupting the TA interaction [⁸⁵] (Fig. 4b). On the other hand, specific compounds that impair expression of a TA operon could trigger toxin release indirectly (Fig. 4c). In fact, interference with antitoxin inhibition seems to be explored by bacteria themselves. The quorum-sensing pentapeptide extracellular death factor (EDF) has been shown to stimulate the MazF endoribonuclease of the chromosomal MazEF system [⁸⁶]. Although the antitoxin MazE is present, the endoribunclease activity is enhanced by EDF and this is thought to trigger stress-induced cell death in E. coli [⁸⁶].

Similarly, an artificial disruption of the toxin–antitoxin interaction of the ε/ζ TA system from S. pyogenes has recently been described [⁸⁷]. In a high-throughput approach, a mixture of oligopeptide libraries was screened for their ability to impair the ε–ζ interaction. Thereby, sub-libraries containing 17 amino acids long peptides were shown to interfere with the ε–ζ interaction. However, further subfractionation of this mixed oligopeptide libraries lead to loss of this capability and the authors suggested that avidity effects from several weak binders could have caused disruption of the protein interaction [⁸⁷]. It remains to be shown whether those peptide inhibitors can also restore UNAG kinase activity of ζ, as they could block the ATP binding similarly as the antitoxin ε does. Whether dismantling of the ε–ζ interaction is always feasible remains questionable, since this relies heavily on the dissociation kinetics of the complex. For instance, dissociation of the ultra-tight PezAT complex is extremely slow, rendering it difficult to disrupt the complex by any peptide inhibitor [⁶⁴].

Another very promising approach to achieve ζ activation and toxins of other TA systems directly is interference with de novo synthesis by use of antisense RNA mechanisms (Fig. 4c). This would provoke specific degradation of the TA mRNA and thereby impair de novo synthesis of the TA proteins. For instance, such successful knockdowns of genes essential for bacterial viability were achieved by liposome-delivered antisense RNA [⁸⁸] or peptide nucleic acids [⁸⁹–⁹²].

It remains to be shown whether ζ or PezT activation is indeed always a beneficial approach for microbial defense. In cases where the TA system is associated with virulence, activation of the toxin could increase infectiousness by enhanced cell lysis and boosted virulence factor release from the pathogenic bacteria. In this case, an as-yet-unknown but specific inhibitor of ζ toxin itself would be expected to eliminate the toxic effects of the TA system (Fig. 4d).

The enzymatic product of this kinase family, UNAG-3P, is a promising lead compound for the development of a new broad-spectrum antibiotic. UNAG-3P inhibits MurA [³⁰] and most likely also its paralogue MurZ, which are conserved enzymes in peptidoglycan synthesis in all gram-positive and gram-negative bacteria. This family of enzymes is an attractive drug target for microbial defense, since no eukaryotic homolog exists [⁹³]. However, apart from the naturally occurring, broad-spectrum antibiotic fosfomycin, no specific MurA inhibitors have been applied so far [⁹⁴]. Because an increasing number of bacteria have developed resistance, fosfomycin has turned into a relatively inefficient drug [⁹⁴]. Thus, the discovery of UNAG-3P discloses a novel and attractive compound that has the potential to evolve into a drug inhibiting MurA.

Moreover, UNAG-3P most likely interferes with bacterial biosynthetic pathways of other glycoconjugates as well. For instance, formation of UNAG-3P might interfere with lipopolysaccharide (LP) production in gram-negative bacteria. LP biosynthesis requires the unique and distinctive building block lipid A. This phosphoglycolipid is highly conserved [⁹⁵], and its formation is initiated by condensation of the 3′-OH group of the N-acetylglucosamine moiety of UNAG with a lipid anchor [⁹⁶]. It remains to be shown for this pathway whether UNAG-3P is an inhibitor similar as it is for peptidoglycan synthesis or is just a dead-end nucleotide sugar. LPs are part of the outer membrane of gram-negative bacteria and protect bacterial cells from cell-damaging agents. Moreover, LPs act as endotoxins and can cause septic shocks due to stimulation of excessive levels of proinflammatory cytokines [⁹⁷]. Thus, UNAG-3P could not only impair bacterial growth but could also reduce LP-induced toxic shock responses during infection.

Moreover, teichoic acids, which are diverse multifunctional glycopolymers that reside in the cell wall of gram-positive bacteria, require UNAG as precursor. Similar to inhibition of teichoic acid synthesis leading to bacteriostasis [⁹⁸] and impaired host colonization and tissue invasion in MRSA [⁹⁹], UNAG-3P might impair these bacterial mechanisms as well.

The discovery of UNAG-3P as a new natural product has uncovered the basic principles for the development of a new broad-spectrum antibiotic, and new challenges need to be managed to turn it into an effective drug. An obvious and required improvement is to render this compound permeable for cell membranes. UNAG-3P is a highly charged anionic molecule and thus most likely will not penetrate through membranes. Further synthetic modifications of UNAG-3P should aim at creating membrane-permeable derivates of UNAG-3P and probe their activity against gram-positive and gram-negative bacteria. For instance, reversible masking of the phosphate groups should help to reduce the permeability barrier [¹⁰⁰, ¹⁰¹].

Furthermore, it remains unclear whether UNAG-3P has any detrimental effect on eukaryotes, since UNAG is also an essential metabolite for these organisms. Previous studies have shown that ectopic expression of ζ from S. pyogenes in Saccharomyces cerevisiae caused growth retardation, albeit the general toxicity of ζ was much weaker when compared with the bacteria [¹⁰²]. This growth retardation in yeast might be explained by a toxic effect of UNAG-3P. However, a more likely explanation is an artificial depletion of cells from UNAG by ζ's enzymatic activity. Cytosolic concentrations of UNAG are approximately 10 times lower in S. cerevisiae (∼30 μM) than in E. coli [¹⁰³], and a rapid depletion of UNAG by ζ will most probably impair chitin synthesis in S. cerevisiae [¹⁰⁴]. Although UNAG-3P is a very promising lead compound for drug development, it might also have undesired adverse reactions similar to those of other lytic antibiotics when detrimental effects upon endotoxin release have to be expected [¹⁰⁵].

Both triggered induction of cell death upon treatment with UNAG-3P-derived drugs or upon activation of PezT toxins will boost the release of virulence factors from pathogenic bacteria into their environment. The involvement of PezT toxins in virulence makes it conceivable that an attenuation of fulminant pneumococcal infections could be achieved by the inhibition of PezT's lytic activity. If the increase of the survival time during fulminant infections in humans is similar to what was observed after PezT inactivation in mice, such an inhibitor could help to increase the time window for therapeutic intervention and allow a retarded kill-off of pathogens.

Another interesting, although still speculative, aspect of PezT's contribution to virulence is a possible involvement as endotoxin during infection. Similar to the cytolysin-mediated translocation of the toxic S. pyogenes NAD glycohydrolase into host cells [¹⁰⁶], PezT could enter the cytosol of the infected host cells. In this scenario, PezT would aggravate eukaryotic cells by its fatal enzymatic mechanism similarly as observed during ectopic expression in S. cerevisiae [¹⁰²]. A PezT inhibitor that enters eukaryotic cells as well would have dual function and could not only counteract pneumococcal autolysis but additionally abolish the toxin's activity in the infected host.