Molecular redundancy defines two or more effectors that modify the same host target using the same molecular mechanism (Figure 1A). In this case, one effector can function in place of the other because it has the same activity and target specificity as its counterpart(s). In some cases, molecular redundancy is likely to be a byproduct of gene duplication however, this is not the sole source with examples of horizontal gene transfer and convergent evolution leading to the presence of molecularly redundant proteins.

Molecular redundancy is exemplified by the L. pneumophila SidE family of IDTS, SidE, SdeA, SdeB, and SdeC (Luo and Isberg, 2004). Individual paralogs consist of a mono-ADP-ribosyltransferase domain, a deubiquitylation domain, and a phosphohydrolase domain that collectively catalyze the ubiquitination of the host proteins Reticulon 4 (Rtn4) (Kotewicz et al., 2016) and Rab33b (Qiu et al., 2016) (Figure 1B). While each member of this family is individually dispensable for intracellular replication, the simultaneous deletion of all four members impairs growth of L. pneumophila in the amoebal hosts Acanthamoebae castellanii (Bardill et al., 2005) and Dictyostelium discoideum (Qiu et al., 2016). The virulence defects can be rescued by SdeA alone (Bardill et al., 2005; Qiu et al., 2016) demonstrating that, at least in these two hosts, a single paralog is sufficient for L. pneumophila intracellular replication.

A second example of molecular redundancy in L. pneumophila is the Lgt family of proteins consisting of Lgt1, Lgt2/LegC8, and Lgt3/LegC5 (de Felipe et al., 2005; Belyi et al., 2006) (Figure 1B). Each paralog is a functional glucosyltransferase that covalently modifies the host protein elongation factor 1A (eEF1A) at serine 53 via mono-O-glycosylation (Belyi et al., 2006, 2008). Modification of eEF1A by any of the three paralogs impairs host protein synthesis (Belyi et al., 2006, 2008). The simultaneous deletion of all three 3 paralogs does not impair L. pneumophila intracellular growth nor does it completely abolish host protein translation in infected cells (Belyi et al., 2006, 2008) suggesting that L. pneumophila encodes additional IDTS that modulate this process.

Target redundancy defines effectors that modulate the same host protein using different molecular mechanisms (Figure 1A). In this case, the activity of one effector cannot replace the other but can have a similar impact on the function of the targeted host protein, and its component pathway. Thus, contrary to molecular redundancy, target redundancy defines redundant strategies rather than redundant activities. Effectors that are redundant at the target level are more difficult to identify because they typically lack sequence, structural and functional similarity.

Target redundancy is exemplified by the IDTS SidM/DrrA (Machner and Isberg, 2006; Murata et al., 2006) and AnkX/LegA8/AnkN (Pan et al., 2008; Mukherjee et al., 2011; Allgood et al., 2017), herein after referred to as AnkX (Figure 1B). Both effectors modulate the activity of the host small GTPase Rab1 to remodel the LCV but do so by different molecular mechanisms. SidM/DrrA AMPylates the GTP-bound form of Rab1 preventing GTP to GDP exchange by its cognate GAP protein (Muller et al., 2010), whereas AnkX phosphocholinates GTP-bound Rab1, locking it in the active state (Mukherjee et al., 2011). Two additional effectors SidD and Lem3 reverse Rab1 constitutive activation by de-AMPylation and de-phosphocholination, respectively (Neunuebel et al., 2011; Tan and Luo, 2011). Thus, L. pneumophila encodes two sets of IDTS, SidM-SidD and AnkX-Lem3 that are both able to regulate Rab1 activity but do so through different mechanisms.

A second example of target redundancy is observed between the glucosyltransferases Lgt1, Lgt2, Lgt3 (Belyi et al., 2008), and SidI (Shen et al., 2009) (Figure 1B). While the Lgt proteins and SidI both impair host protein synthesis by targeting eEFA1, SidI appears to do so by an alternative mechanism. Similar to Lgt1, Lgt2, and Lgt3, SidI directly interacts with eEFA1 to impair its function however, direct binding is not solely responsible for this effect (Shen et al., 2009). If SidI inactivates eEF1A through modification, the lack of a glycosyltransferase domain suggests it is likely to differ from glycosylation. In addition to Lgt1, Lgt2, Lgt3, and SidI, a fifth effector, SidL has been implicated in impairing host protein synthesis (Fontana et al., 2011), although the mechanism has yet to be elucidated including whether this occurs through eEF1A or another component of the translation machinery. Moreover, the deletion of all five of these IDTS only partially restores host protein synthesis (Fontana et al., 2011), suggesting that additional IDTS regulate this process. Thus, this example encompasses multiple types of redundancy from molecular and target redundancy to pathway and possibly, cellular process redundancy (see below). We predict that many virulence strategies will similarly consist of more than one type of redundancy.

Pathway redundancy defines effectors that modulate a single host pathway but target different components of that pathway (Figure 1A). Sets of effectors that belong to this category can manipulate different proteins in a single complex, different components at various steps along the pathway or regulators of the pathway. However, while the mechanisms and the host proteins used to modulate the pathway differ, the outcome of that modulation is the same and these effectors collectively serve to achieve a common goal.

Pathway redundancy is illustrated by the IDTS VipD and SidK that both modulate the endocytic pathway but do so by targeting different components at different stages of LCV maturation on the way to the lysosome (Figure 1B). On early endosomes, VipD dephosphorylates PI3P, which functions as an anchor for the tethering protein EEA1 (Gaspar and Machner, 2014). The lack of EEA1 at endosomal surfaces prevents endosome fusion with the LCV (Gaspar and Machner, 2014). Vacuole acidification occurs downstream of early endosome fusion events and is mediated by vATPases, multi-component proton pumps (Forgac, 2007). SidK directly binds VatA, a component of the vATPase to inhibit its function (Xu et al., 2010). While VipD can impair early endosome fusion with the LCV, it is not sufficient to avoid endosomal fusion completely as 40% of LCVs containing wild type bacteria stain positive for the early endosomal marker Rab5 (Gaspar and Machner, 2014). SidK acts as part of a contingency plan when endocytic maturation of the LCV is not completely thwarted. Moreover, while vacuoles containing the ΔvipD mutant are more likely to accumulate Rab5 than those containing wild type bacteria, the frequency is significantly lower than that observed for a dot- mutant (Gaspar and Machner, 2014): this suggests that other effectors function to modulate the endocytic pathway. Several effectors including VipA, VipF, SetA, and Ceg19 are likely candidates based on their ability to disrupt trafficking along the vacuole sorting pathway in yeast (Shohdy et al., 2005; Franco et al., 2012).

A second example of pathway redundancy is observed between the effectors RidL, LpnE, and AnkX which target separate mediators of retrograde trafficking between endosomes and the trans-Golgi network to alter the fate of the LCV (Figure 1B). RidL directly interacts with the retromer complex subunit Vps29 to compete with endosome sorting nexins for retromer and PI3P binding (Finsel et al., 2013). LpnE directly interacts with OCRL1 (Weber et al., 2009), a phosphoinositol 5-phosphatase that regulates retrograde trafficking by altering phosphoinositide phosphate pools. AnkX phosphocholinates Rab35 (Mukherjee et al., 2011), a regulator of cargo sorting and recycling from recycling endosomes. Modification of Rab35 prevents microtubule-dependent endosomal vesicle transport to the LCV (Pan et al., 2008). Loss of RidL, LpnE, or AnkX moderately increases the frequency of LAMP1 staining of LCVs demonstrating that the all three effectors independently contribute to disrupting maturation of the LCV along the endocytic pathway (Newton et al., 2007; Pan et al., 2008; Finsel et al., 2013).

Cellular process redundancy occurs when sets of effectors compensate for one another by targeting redundant or complementary host pathways that collectively govern a single cellular process (Figure 1A). An example of a cellular process that is mediated by multiple pathways is the unfolded protein response (UPR). The UPR is activated through three separate sensory pathways: inositol requiring enzyme-1 (IRE1), protein kinase RNA-like ER kinase (PERK), and activating transcription factor 6 (ATF6) (Walter and Ron, 2011). Distinct pathways allow the cell to respond to multiple signs of ER stress enhancing the sensitivity and breadth of the sensory system but all pathways lead to a common response that includes global translation inhibition, upregulation of ER stress proteins, ER membrane expansion and under extreme conditions, activation of pro-apoptotic pathways (Walter and Ron, 2011). While multiple pathways provide robustness to the host, it affords pathogens multiple ways to hijack a cellular process and when necessary, the ability to do so without completely abolishing the cellular process, which can have negative, even detrimental effects on the pathogen itself.

A critical event in Legionella pathogenesis is remodeling and maintenance of the LCV to support bacterial replication: this is accomplished through the recruitment of ER-derived membrane material. Three parallel mechanisms by which Legionella achieves this have been described (Figure 1B). The SdeA, SdeB, and SdeC family of IDTS drives rearrangement of tubular ER and its association with the LCV through ubiquitination of Rtn4 (Kotewicz et al., 2016), a regulator of tubular ER dynamics (Zurek et al., 2011) (Figure 1B). SidM/DrrA and RalF target components of the early secretory pathway to redirect vesicles trafficking between the ER and the Golgi to the LCV. SidM/DrrA does so by recruiting and activating Rab1 at the LCV and promoting non-canonical functional pairing between the plasma membrane tSNARE syntaxins at the LCV and the ER-derived vesicle vSNARE Sec22b (Arasaki et al., 2012). RalF does so by recruiting and activating the host protein ARF1 at the LCV (Nagai et al., 2002) (Figure 1B). The loss of either SidM/DrrA or RalF alters the timing and efficiency of ER protein accumulation at the LCV (Nagai et al., 2002; Ingmundson et al., 2007) demonstrating redundant roles for these proteins in LCV remodeling.

A second example of redundancy at the level of cellular processes is SidF and SidP (Figure 1B). Each effector contributes to modulation of host lipid metabolism to modulate the relative abundance of phosphoinositides (PIs) at the LCV, specifically conversion from a PI(3)P rich environment to a PI(4)P rich environment. SidF is a phosphoinositide 3-dephosphatase with specificity for PI(3,4)P₂ and PI(3,4,5)P₃ preventing PI(3)P accumulation at the LCV (Banga et al., 2007) while SidP is a phosphoinositide 3-phosphatase that hydrolyzes PI(3)P and PI(3,5)P₂ removing PI(3)P from the LCV (Toulabi et al., 2013). The host protein ORCL1, a PI(4,5)P₂ 5-phosphatase also localizes to the LCV and thus may also promote PI(4)P accumulation at the surface (Weber et al., 2009) (Figure 1B). OCRL1 targeting to the LCV is Icm/Dot-dependent but the specific IDTS required for this has yet to be determined (Weber et al., 2009). Phosphoinositides distinguish individual organelle membranes in the host cell and serve as anchors for organelle-specific host proteins. Several IDTS exploit PIs decorating the LCV to anchor themselves to the surface. Many of these IDTS have been implicated in LCV remodeling including SidM/DrrA, SidC, LidA, and RidL (Machner and Isberg, 2006; Murata et al., 2006; Finsel et al., 2013; Hsu et al., 2014). Loss of SidF impairs Rab1 recruitment to the LCV (Toulabi et al., 2013), likely as a consequence of the inability of SidM to attach itself to the LCV surface. Thus, PI dynamics play a central role determining the repertoires of IDTS at the vacuole surface and thus the fate of the Legionella vacuole.

System redundancy defines effectors that modulate more than one host cellular process to accomplish a single task (Figure 1A). An example in biology of a single event that is governed by multiple host cellular processes is cell death. Cell death can be achieved through apoptosis, necrosis, pyroptosis, or autophagy. Each process may be triggered by different cues and the mechanisms by which the cell is terminated may vary but the result is the same—death. In some cases, components mediating these pathways are completely distinct; in other cases they may overlap. For a pathogen, the more options at its disposal for manipulating the host cell to accomplish a specific goal, the greater the likelihood of its success. System redundancy provides yet another layer of insurance by allowing a pathogen to tap into multiple cellular processes to ensure completion of a critical event.

Intracellular growth of L. pneumophila requires the viability of the host cell but cell death is induced by host cells when bacteria cannot be eradicated through lysosomal targeting. L. pneumophila regulates host cell death by targeting host signal transduction, translation, and apoptosis (Figure 1B). The IDTS LnaB and LegK1 activate the host transcription factor, nuclear factor κB (NFκB) causing upregulation of anti-apoptotic pathway-associated genes (Losick and Isberg, 2006; Ge et al., 2009). While the mechanism of action of LnaB is unknown, in vitro studies suggest that LegK1 promotes the degradation of the NFκB inhibitor IκB through direct phosphorylation (Ge et al., 2009). Lgt1, Lgt2, Lgt3, SidI, and SidL promote prolonged NFκB signaling by blocking host protein synthesis and thus cellular levels of IκB from being replenished (Fontana et al., 2011). SidF promotes host cell survival by inhibiting the activity of the pro-apoptotic proteins BNIP3 and Bcl-Rambo through direct binding (Banga et al., 2007). Thus, L. pneumophila orchestrates the induction of host cell survival mechanisms while simultaneously obstructing host cell death pathways by targeting distinct cellular processes.

Lysosomal avoidance by L. pneumophila is orchestrated through four separate cellular processes: the endocytic pathway using VipD (Gaspar and Machner, 2014) and SidK (Xu et al., 2010); retrograde transport via RidL (Finsel et al., 2013), LpnE (Weber et al., 2009), and AnkX (Mukherjee et al., 2011); actin cytoskeleton dynamics through LegK2 (Michard et al., 2015) and VipA (Franco et al., 2012); and autophagy by RavZ (Choy et al., 2012) (Figure 1B). The mechanisms of action of VipD, SidK, RidL, LpnE, and AnkX have been discussed previously (see section Pathway Redundancy). Altering endosome transport to the LCV is also achieved by manipulating the actin cytoskeleton. LegK2 phosphorylates the Arp2/3 complex subunits ARPC1B and ARP3 (Michard et al., 2015): this prevents actin nucleation at the site of the LCV thus perturbing endosome trafficking to the LCV (Michard et al., 2015). In contrast, VipA localizes to endosomes and promotes actin polymerization by directly binding to actin. In yeast, VipA impairs vacuole sorting and thus is predicted to similarly alter organelle trafficking during infection (Shohdy et al., 2005; Franco et al., 2012). Host cells are not without their own forms of redundancy. When bacteria fail to be delivered to the lysosome, host cells can also target pathogens to the lysosome via autophagy (Xie and Klionsky, 2007). The IDTS RavZ localizes to the LCV where it irreversibly deconjugates the autophagic protein Atg8 thereby preventing autophagasome membrane nucleation at the site of the LCV (Choy et al., 2012).

System redundancy is also exemplified by SidM/DrrA, SdeA, SdeB, SdeC, RalF, and a functional complex formed by LegC2/YlfB, LegC3, and LegC7 (Figure 1B). SidM/DrrA, SdeABC, and RalF modulate LCV remodeling by hijacking tubular ER dynamics and vesicle trafficking along the early secretory pathway (see section Cellular Process Redundancy). LegC2/YlfB, LegC3, and LegC7 collectively mimic Q-SNARE proteins and directly bind the R-SNARE protein VAMP4 (Shi et al., 2016). LegC2/YlfB-LegC3-LegC7/YlfA complex pairing with VAMP4 diverts VAMP4-containing vesicle trafficking along the retrograde transport pathway between endosomes and the trans-Golgi network to the LCV (Shi et al., 2016). L. pneumophila mutants lacking LegC2/YlfB and LegC7/YlfA show reduced accumulation of the ER marker calnexin at the LCV but do not exhibit an increase in LAMP1 staining (Campodonico et al., 2016). Thus, recruitment of VAMP4-containing vesicles serves to remodel and maintain the LCV but does not impact LCV trafficking to the lysosome (Campodonico et al., 2016; Shi et al., 2016). Differential targeting of endosomes to the LCV suggests the existence of distinct populations of endosomal vesicles, some of which are actively recruited to the LCV to enable L. pneumophila replication while others are actively excluded because they promote L. pneumophila trafficking to the lysosome.

Combined biochemical, molecular, and cell biological characterization of effectors is the most comprehensive way to identify redundant proteins. It provides detailed information about their mechanism of action, their host cell targets and their direct impact on host cellular processes. Moreover, deciphering the intricate details of an effector's function can define subtle distinctions between effectors with overlapping functions and thus circumvent their improper classification as redundant. However, this method is not without its drawbacks. The amount of time required to exhaustively characterize protein function can be lengthy, especially if the techniques to do so are not available or the function of the host target protein or its component pathway have yet to be characterized. For a pathogen like L. pneumophila that employs at least 270 IDTS, such an endeavor would be an arduous one. In addition, for many effectors, sequence homology and structure prediction tools are not always informative. For L. pneumophila, as many as one third of all IDTS lack domain homology to any other protein characterized to date and often very little is learned from structural predictions. Finally, while there are several methods to define host targets, including more recently adapted high throughput methods (Mousnier et al., 2014; Yu et al., 2015), this can be challenging as host targets can range from proteins to lipids to small molecules (Machner and Isberg, 2006; Toulabi et al., 2013; Isaac et al., 2015). Thus, while characterizing individual effector functions can be highly informative, the road to defining redundant virulence mechanisms can be bumpy and painstakingly slow.

iMAD is a genetic screening strategy developed to resolve redundancy amongst effectors by defining sets of bacterial proteins that target common host pathways and parallel pathways exploited by a pathogen to accomplish a single task (O'Connor et al., 2012; O'Connor and Isberg, 2014). To do so, iMAD integrates bacterial mutagenesis and host RNA interference to systematically identify genetic interactions between a pathogen gene and a host gene based on impaired replication of the pathogen (O'Connor et al., 2012; O'Connor and Isberg, 2014). In the case of L. pneumophila, a library of transposon mutants were assessed for their ability to replicate within host cells depleted of one of five early secretory proteins that promote L. pneumophila intracellular growth (O'Connor et al., 2012). Hierarchical clustering of bacterial gene mutations in IDTS with similar behavioral patterns across all host conditions examined revealed several important functional relationships: (1) Common phenotypic signatures identified sets of bacterial proteins that target common host pathways: these sets of proteins defined distinct functional groups; (2) Deleting pairs of bacterial genes from separate functional groups impaired intracellular growth of L. pneumophila: these functional groups defined separate but redundant host pathways targeted by L. pneumophila to generate a replication vacuole; (3) Specific defects in host cell biology resulting from loss of bacterial proteins could be predicted for genes based on the characterization of other members of its group: this identified three sets of proteins that independently contribute to the maintenance of replication vacuole membrane integrity; (4) Different combinations of bacterial genes were required for optimal growth in different hosts defining sources of adaptation to host variation. By grouping individual effectors that commonly manipulate a single host pathway and redundant pathways that contribute to a single process, iMAD defines functional relationships between effectors at the target, pathway, cellular process, and system levels. With more efficient methods for generating arrayed bacterial mutant libraries, commercially available RNAi libraries and the development of CRISPR technology to facilitate host cell protein depletion, and the replacement of DNA microarrays with massively parallel sequencing techniques to monitor bacterial mutant populations, more comprehensive, high-throughput iMAD screens are now possible.

Genome reduction followed by effector repertoire reconstitution is another strategy used to identify redundant effectors and the host pathways they target (Cunnac et al., 2011). Progressive removal of all 28 effector genes from the plant pathogen Pseudomonas syringae Pto DC3000 determined 15 of the effector genes are collectively dispensable for growth in the plant host Nicotiana benthamiana (Kvitko et al., 2009). The subsequent reintroduction of different combinations of effectors defined several redundant-effector groups (REGs) that promote P. syringae growth in N. benthamiana, two of which were determined to mediate resistance to independent arms of plant innate immunity (Block and Alfano, 2011; Cunnac et al., 2011). By analyzing correlates between effector combinations and rescued P. syringae growth during infection, a minimal set of 8 effectors was defined that was sufficient to promote growth of P. syringae to near wild type levels. The effector repertoire reconstitution linked individual REGs with the host pathways they target, identified specific combinations of effectors that are sufficient to cause disease and demonstrated the ability to swap different members of individual REGs and still achieve robust P. syringae growth. For genetically amenable pathogens with manageable sizes of effector repertoires, preferably clustered in a minimal set of genetic loci to facilitate combinatorial effector reintroduction into the genome, effector repertoire reduction, and reconstitution strategies provides a comprehensive method to define redundant effectors, the cellular processes they modulate and the minimum set of effector functions required for pathogenesis.

Proteome-based analyses that use high throughput mass spectrometry allow the interactomes of entire effector subfamilies to be mapped. Affinity purified-mass spectrometry has been used to define the host interacting partners of 58 secreted virulence factors in Chlamydia trachomatis called inclusion membrane proteins (Incs) (Mirrashidi et al., 2015). The results not only allowed groups of effectors to be assigned to specific cellular processes but identified sets of Inc proteins that target the same host proteins or different members of the same multiprotein complex. Mapping the interactome network of all 58 Inc proteins revealed sets of Inc proteins that converge on common targets, pathways and cellular processes: this defined focal points of host modification by C. trachomatis and thus, potential sources of redundancy. Moreover, the C. trachomatis Inc-human interactome had significant overlap with that of other pathogens. Comparisons with three viral-human interactomes (Jager et al., 2011; Davis et al., 2015; Ramage et al., 2015) identified 98 shared host targets between C. trachomatis and at least one of the three viruses. Similarly, a number of the Inc host protein targets are also common targets of L. pneumophila IDTS including Rtn4 (Kotewicz et al., 2016), vATPases (Xu et al., 2010), and the retromer complex (Finsel et al., 2013). The lack of similarly between the respective C. trachomatis Inc proteins and L. pneumophila IDTS and differences in the host protein complex subunits targeted demonstrates that each bacterial pathogen has acquired or evolved different mechanisms to modulate the same host proteins, pathways and/or cellular processes. Comparing effector functions between pathogens not only allows additional redundant virulence mechanisms to be defined but establishes a critical set of events central to microbial pathogenesis.

Comparative genomics combined with phenotypic analyses provide a means to define correlates between effector conservation and redundant virulence mechanisms (Baltrus et al., 2011). The P. syringae pan-genome effector repertoire consists of 57 effectors but different subsets of effectors are sufficient for growth in the same plant host. Computational analyses that correlate specific combinations of effectors and host tropism (Baltrus et al., 2011) allow redundant virulence mechanisms to be elucidated on a global scale. Alternatively, comparative genomes can be used as a more targeted approach. For example, in L. pneumophila SdhA is a critical virulence determinant in macrophages (Laguna et al., 2006). While the precise function of SdhA is still unclear, the severe growth defect of the ΔsdhA mutant is due to loss of vacuole integrity that leads to a robust host innate immune response and consequently either bacterial or host cell death (Laguna et al., 2006; Creasey and Isberg, 2012). L. pneumophila encodes two paralogs of SdhA, SidH, and SdhB, but their deletion only moderately enhances the already severe intracellular growth defect of the ΔsdhA mutant (Laguna et al., 2006). Legionella feeleii lacks a sdhA paralog but grows almost as well as the wild type strain of L. pneumophila in macrophages (Figure 2). While the presence of sidH in L. feeleii may compensate for the absence of sdhA, deletion of sidH does not impair L. feeleii growth in macrophages (Figure 2). The dispensability of SdhA (and SidH) in L. feeleii is not due to the lack of plaA and/or traI, which suppresses the ΔsdhA mutant phenotype in L. pneumophila (Creasey and Isberg, 2012). Thus, while SdhA plays a critical role in L. pneumophila pathogenesis, the entire family of SidH paralogs is dispensable in L. feeleii. While there are a number of explanations for this discrepancy, L. feeleii encodes 27 additional putative IDTS that are not conserved in L. pneumophila (Burstein et al., 2016), one or more of which may compensate for the absence of SdhA despite their lack of homology. As more genomes of pathogen isolates are sequenced, correlates between effector conservation and phenotypes will allow alternate virulence mechanisms employed by pathogens to be defined.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.