The needle is a superhelical hollow structure comprised of multiple copies of the EscF protein (Wilson et al., 2001), which is essential for the secretion of all T3 substrates and hence for virulence (Deng et al., 2004). EscF associates with the membrane ring proteins EscC, EscD, and EscJ (Ogino et al., 2006), as well as with the filament protein EspA (Wilson et al., 2001), forming a continuous channel that connects the bacterial cytoplasm with the host cell (Figure 2). It also associates in the cytoplasm with two chaperones, EscE and EscG that prevent its premature polymerization and are essential for its assembly (Sal-Man et al., 2013).

In EPEC, the needle is 23 nm in length and 8–9 nm in width (Sekiya et al., 2001; Ogino et al., 2006; Monjarás Feria et al., 2012). Although, the inner diameter of the needle central conduit in A/E pathogens has not been determined, in other T3S systems it ranges from about 1.3 to 2.5 nm (Blocker et al., 2001; Fujii et al., 2012; Loquet et al., 2012). The size of this channel is not wide enough to accommodate folded proteins; therefore, effectors should be unfolded prior to transport through to the T3SS (Feldman et al., 2002; Akeda and Galán, 2005; Fujii et al., 2012). This hypothesis was demonstrated by the fusion of a stable protein, or bulky proteins containing a knotted motif, to the C-terminal ends of T3 substrates. The fusion proteins were trapped inside the needle, blocking further secretion and allowing its co-purification with isolated T3S needle complexes (Dohlich et al., 2014; Radics et al., 2014). Visualization of the trapped recombinant substrates by cryo-EM provided direct evidence for the passage of unfolded proteins through this structure (Radics et al., 2014).

The EPEC needle structure is one of the smallest characterized, compared to those of S. enterica (25–80 nm) (Kubori et al., 1998; Kimbrough and Miller, 2000; Marlovits et al., 2006), S. flexneri (45–50 nm) (Tamano et al., 2000), Y. pestis (41 nm), and Y. enterocolitica (58 nm) (Journet et al., 2003). The needle length is strictly regulated by and correlates with the size of a family of proteins named type III secretion substrate specificity switch (T3S4) proteins (Journet et al., 2003; Büttner, 2012). In A/E pathogens, the EscP protein belongs to the T3S4 protein family, and in EPEC it was demonstrated to directly interact with EscF (Monjarás Feria et al., 2012). This interaction is important for needle length regulation in a process that will be discussed below. Furthermore, in other bacteria it has been proposed that the needle plays an active role in mammalian cell sensing and substrate secretion regulation (Kenjale et al., 2005; Torruellas et al., 2005); however, no direct evidence about this process has been reported in A/E pathogens.

Remarkably, the number of needle complexes in EPEC has been estimated to be 12 per cell, based on the observed EspA filaments (Daniell et al., 2001b; Wilson et al., 2001), which is fewer than the number seen in Salmonella (10–100; Kubori et al., 1998) and Yersinia (30–100; Hoiczyk and Blobel, 2001).

The filament is an extracellular appendage found in the T3S systems of A/E pathogens (also present in Bordetella spp.) that functions as an adaptor between the needle and the translocation pore formed in the host cell membrane (Figure 2; Knutton et al., 1998; Ide et al., 2001; Medhekar et al., 2009). In A/E pathogens, it is assembled by the polymerization of multiple subunits of the EspA protein with a helical symmetry of 5.6 subunits per turn (Knutton et al., 1998; Daniell et al., 2003; Wang et al., 2006), forming a hollow structure of 12 nm in width that allows the passage of substrates through a 2.5 nm central channel (Daniell et al., 2001b, 2003; Crepin et al., 2005b). Unlike the needle, the EspA filament has a variable length that can reach more than 600 nm, which seems to be dependent on the availability of EspA subunits given that it can be enlarged by increasing the amount of EspA in the cytoplasm (Sekiya et al., 2001; Crepin et al., 2005b). The average length of the filament is ca. 90 nm and it elongates by addition of EspA subunits at the tip of the structure (Daniell et al., 2003; Crepin et al., 2005b).

The EspA protein has a coiled-coil domain at its C-terminal region that is required for subunit polymerization and assembly of the filament (Delahay et al., 1999). Since this protein undergoes spontaneous polymerization, it requires the assistance of chaperone proteins in the cytoplasm. CesAB was the first EspA chaperone to be described (Creasey et al., 2003c). The crystal structure of the CesAB-EspA complex showed that the coiled-coil domain of EspA is also the one involved in the interaction with CesAB, thus revealing the mechanism that prevents EspA premature oligomerization (Yip et al., 2005a). Besides CesAB, EspA has a second chaperone named CesA2 (formerly Orf29, renamed EscG in EPEC) that assists EspA stabilization in the cytoplasm (Su et al., 2008). In EHEC it has also been reported that EspA binds to EscL, a component of the ATPase complex, and that this interaction is required to preserve EspA stability, although not via a chaperoning mechanism (Ku et al., 2009). EspA can also directly bind to EscF and EspB, forming a continuous channel from the bacterial cytoplasm to the host cell cytosol (Hartland et al., 2000; Daniell et al., 2001b).

Apart from its role as a translocation conduit, and similarly to the needle, the EspA filament has been proposed to participate in sensing the presence of mammalian cells, and there is evidence that involves this structure in adhesion to epithelial cells and biofilm formation (Ebel et al., 1998; Knutton et al., 1998; Cleary et al., 2004; Moreira et al., 2006). However, the EspA filaments are not present over the entire course of infection; once the intimate attachment has been established, they are eliminated and so are absent from the mature A/E lesion. Indeed, it has been reported that expression of the espA gene, and thus the presence of EspA filaments, is downregulated after 6 h of infection (Knutton et al., 1998; Dahan et al., 2004).

The translocation pore is a protein complex formed by the hetero-oligomerization of EspB and EspD subunits. These proteins interact with each other and insert into the host membrane, forming a channel that allows the direct translocation of effectors from the bacteria to the host cell cytoplasm (Figure 2; Ide et al., 2001). The EspB and EspD proteins are predicted to have one and two transmembrane domains, respectively, involved in their membrane anchoring (Delahay and Frankel, 2002; Dasanayake et al., 2011). It has been shown that they can insert into erythrocyte membranes causing red blood cells (RBCs) hemolysis (Warawa et al., 1999). Moreover, both of these hydrophobic translocators interact with the hydrophilic translocator EspA, forming the so-called translocon (Hartland et al., 2000; Luo and Donnenberg, 2011). Mutants in any of the translocon components are still able to secrete proteins to the medium, but fail to translocate them to the host cell, thus impairing adhesion, A/E lesion formation and virulence (Lai et al., 1997; Kresse et al., 1999; Deng et al., 2004).

Analysis with low resolution atomic force microscopy showed that the pores formed by EspB and EspD in diffusely adhering EPEC (DA-EPEC) are composed of six to eight subunits with a minimal pore size of 3–5 nm (Ide et al., 2001). In addition, EspD, which is able to interact with itself through a C-terminal coiled-coil domain (Daniell et al., 2001a), spontaneously incorporates into unilamellar vesicles, forming a pore with an inner diameter of 2.5 nm and a molecular mass of 280–320 kDa, which would consist of six to seven subunits (Chatterjee et al., 2015). In agreement with this, analysis of RBC membranes during EPEC-mediated hemolysis showed that EspD was the only bacterial protein membrane-associated, suggesting it plays a dominant role in pore formation. In contrast, EspB seems to have a secondary role in pore formation since a ΔespB mutant caused only a slight reduction in hemolysis (Shaw et al., 2001).

Although, significant advances have been made in determining the components that form the translocation pore, and to a lesser extent, its stoichiometry, little is known about their topology when inserted into the host membrane. For P. aeruginosa translocator proteins PopB and PopD (EspD and EspB homologs, respectively), it has been shown that PopB is inserted into the host membrane with both the N- and C-terminus facing the outer leaflet of host plasma membrane. Likewise, the C-terminus of PopD is located toward the extracellular milieu, where it docks to the needle tip formed by PcrV (EspA in A/E pathogens), while its N-terminus is facing the host cytoplasm (Discola et al., 2014; Armentrout and Rietsch, 2016). The proposed membrane orientation of PopD differs from that reported for EspB in EPEC where, by introducing sequences recognized by a host kinase, it was demonstrated that the C-terminal domain of EspB is localized in the host cytoplasm while the N-terminal domain remains extracellular (Luo and Donnenberg, 2011).

Additionally, it has been reported that cholesterol membrane content plays an important role in EPEC and EHEC adherence, protein translocation and pedestal formation (Hayward et al., 2005; Riff et al., 2005). In fact, it was shown that the EspD homologs in Salmonella (SipB) and Shigella (IpaB) bind with high affinity to cholesterol, which is required for efficient delivery of effectors into host cells (Hayward et al., 2005). However, even though it was initially reported that EspD inserts into Triton X-100 resistant membrane domains (Wachter et al., 1999), and binds to vesicles with a lipid composition that resembles that of the eukaryotic outer leaflet of the membrane, it was later demonstrated that cholesterol is not required for EspD binding or pore formation; instead, addition of anionic lipids such as phosphatidylserine or phosphaditylglycerol induces pore formation of EspD in unilamellar vesicles (Chatterjee et al., 2015). It is possible that cholesterol is necessary for the function of effectors associated with the host membrane rather than for the translocon assembly per se (Allen-Vercoe et al., 2006; Chatterjee et al., 2015); however, this issue remains a matter of debate. Recently, the translocon pore-forming activity was demonstrated to be regulated by the serine protease autotransporters EspC and EspP in EPEC and EHEC, respectively (Guignot et al., 2015). EspC degrades EspA and EspD after host cell contact, regulating the translocation pore forming activity, which in turn downregulates the cytotoxicity induced by EPEC (Guignot et al., 2015). This is in agreement with previous findings showing that EspC interacts with EspA (Vidal and Navarro-García, 2008) and that a mutant defective in the EspC ortholog in C. rodentium displayed increased virulence in an in vivo infection model (Vijayakumar et al., 2014).

Besides its role as a pore-forming component, EspB is as well translocated into host cells, where it binds to myosin, inhibiting its interaction with actin and contributing to microvilli effacement and EPEC phagocytosis inhibition (Taylor et al., 1998; Wolff et al., 1998; Iizumi et al., 2007). It has also been demonstrated to bind and recruit α-catenin at the EHEC adherence site, helping to the development of the A/E lesion (Kodama et al., 2002). EspD and EspB are assisted in the bacterial cytoplasm by three chaperones CesD, CesD2, and CesAB that maintain them in an unfolded state and prevent their premature oligomerization (Wainwright and Kaper, 1998; Creasey et al., 2003c; Neves et al., 2003).

The ATPase complex is formed by the EscN, EscL, and EscO proteins (Figure 2), all of which are important for T3SS function (Deng et al., 2004; Andrade et al., 2007; Ku et al., 2009; Biemans-Oldehinkel et al., 2011; Romo-Castillo et al., 2014). This virulence-associated complex is homologous to the flagellar FliI/FliH/FliJ ATPase complex (Minamino and MacNab, 2000b; Ibuki et al., 2011). The crystal structure of the ATPase EscN has been solved, and like FliI, it shows structural similarity to the F1-ATPase α and β subunits (Imada et al., 2007; Zarivach et al., 2007). EscL, like FliH, has been evolutionarily linked to the b and δ subunits of the F1 ATPase (Pallen et al., 2005, 2006). Even more, despite the lack of sequence conservation, in silico and functional data suggest that EscO is the evolutionary counterpart of the flagellar FliJ protein, which in turn has been related to the γ subunit of the F1-ATPase (Ibuki et al., 2011; Romo-Castillo et al., 2014). The functional relatedness of EscO and FliJ was confirmed by heterologous complementation of the motility of a ΔfliJ mutant with a plasmid encoding EscO (Romo-Castillo et al., 2014). Taken together, these similarities would imply that an energy-producing enzyme, a motility machinery, and an interkingdom protein transport device, share a common evolutionary ancestor.

EscN is a peripheral membrane protein located at the base of the needle complex that energizes the secretion process, probably by releasing the substrate from its cognate chaperone and unfolding it for further secretion, as has been reported for Salmonella InvC, and functioning as a docking site for chaperone-substrate complexes (Gauthier and Finlay, 2003; Akeda and Galán, 2005; Thomas et al., 2005; Chen et al., 2013). This enzyme oligomerizes into a homohexameric ring structure and the oligomeric state of the protein affects its specific activity (Andrade et al., 2007; Zarivach et al., 2007). EscN interacts with EscL, a negative regulator that inhibits its ATPase activity, and with EscO, which conversely, stimulates EscN ATPase activity (Biemans-Oldehinkel et al., 2011; Romo-Castillo et al., 2014).

A model proposed for the ATPase complex function that includes previous findings in the flagellar and virulence T3S systems, suggests that the formation of the EscN-EscL complex in the cytoplasm ensures that the ATPase activity is inhibited until ATP hydrolysis can be coupled to protein secretion, preventing futile energy expenditure (Minamino and MacNab, 2000b; Blaylock et al., 2006; Stone et al., 2011). Once the EscN/EscL/EscO complex is formed near the vicinity of the export apparatus through protein-protein interactions with the C-ring, both EscL and EscO, in a similar way to their T3SS counterparts, interact with the major export gate component EscV (Zhu et al., 2002; González-Pedrajo et al., 2006; Morita-Ishihara et al., 2006; Hara et al., 2012; Cherradi et al., 2014; Lee et al., 2014), promoting a conformational change that allows EscO to stimulate EscN oligomerization and, subsequently, its ATPase activity (Claret et al., 2003; Evans et al., 2006; Ibuki et al., 2011; Romo-Castillo et al., 2014).

EscO has also been localized in the EPEC periplasm and as previously mentioned, it was found to interact with the OM ring protein EscC (Sal-Man et al., 2012a). In addition, similarly to what has been observed for FliJ, EscO interacts with the chaperones CesA2 (EscG) and CesL (Mpc) (Lin et al., 2014), suggesting that this protein performs additional roles during the T3S process. Likewise, EscL has been recently implicated as a component of the sorting platform as will be discussed next (Lara-Tejero et al., 2011; Hu et al., 2015).

In A/E pathogens the C-ring is proposed to be a cytoplasmic annular structure located at the base of the basal body, formed mainly by the EscQ protein (formerly known as SepQ), which belongs to the YscQ/FliN (SctQ) protein family (Figure 2; Pallen et al., 2005; Biemans-Oldehinkel et al., 2011). Other members of this family are YscQ (from Yersinia spp.), SpaO (from S. enterica SPI-1), Spa33 (from S. flexneri), SsaQ (from S. enterica SPI-2), and HrcQ (from Xanthomonas spp.) (Morita-Ishihara et al., 2006; Lara-Tejero et al., 2011; Yu et al., 2011; Bzymek et al., 2012; Lorenz et al., 2012; Notti et al., 2015). Three different proteins, FliM, FliN, and FliG, form the corresponding flagellar C-ring, and it has been shown that members of the SctQ family have an evolutionary relationship with FliM and FliN (Zhao et al., 1996; Hueck, 1998; Pallen et al., 2005; Thomas et al., 2006). The mRNA from most of the sctQ homologs contains an internal alternative translation site, giving rise to a full-length protein (SctQ-Full of approximately 300 amino acids) resembling the flagellar FliM protein, and to a shorter C-terminal version (SctQ-C of ca. 100 amino acids) similar to FliN. Both translation products, SctQ-Full and SctQ-C, are required for T3S assembly and function (Bzymek et al., 2012; Notti et al., 2015; McDowell et al., 2016) with the exception of SsaQ-C from Salmonella SPI-2 that serves as a chaperone for SsaQ-Full but is not essential for T3S (Yu et al., 2011). To date it remains unclear whether EscQ has an internal translation start site.

The SctQ-Full and SctQ-C proteins have been shown to interact in a 1:2 ratio, and it is proposed that this complex assembles into higher order oligomers that form a ring shaped structure (Bzymek et al., 2012; Diepold et al., 2015; McDowell et al., 2016). The SctQ-Full:SctQ-C protein ratio seems to be important for proper injectisome assembly as it was shown that there is a linear correlation between the SctQ-Full structure assembly and the expression levels of SctQ-C protein (Diepold et al., 2015). The injectisome SctQ-Full/SctQ-C complex is believed to mirror the FliM/FliN basic building block of the flagellar C-ring (Bzymek et al., 2012; McDowell et al., 2016). The solved structures of several SctQ proteins revealed the presence of SpoA domains that mediate both SctQ-C-SctQ-C homotypic and SctQ-Full-SctQ-C heterotypic associations (Bzymek et al., 2012; Notti et al., 2015; McDowell et al., 2016). It was recently reported that the Y. enterocolitica C-ring is formed by 22 ± 8 YscQ-Full subunits per injectisome and has an estimated diameter of 30.2 nm, making it similar to that of the flagellum (Diepold et al., 2015). Furthermore, Hu and colleagues determined the structure of the S. flexneri C-ring by cryoelectron tomography and showed that it is formed by six pod-like structures made out of multiple copies of Spa33, arranged in a hexagonal array of 32 nm in diameter and 24 nm in height. The top portion of the pods, suggested to be composed of Spa33-Full and MxiK, links the entire C-ring to the membrane-associated components of the basal body. The bottom of each pod, proposed to be formed by a homotetramer of Spa33-C, is connected to the ATPase Spa47 (SctN) through six spoke-like densities made out of MxiN (SctL) (Hu et al., 2015). The ultrastructural details of the C-ring in A/E pathogens have not yet been addressed. In Y. enterocolitica it has been shown that the C-ring is a highly dynamic substructure that exchanges subunits between a YscQ cytosolic pool and the assembled structure at the base of the injectisome, and that the rate of this subunit exchange correlates with the effector secretion status (Diepold et al., 2015).

Co-immunoprecipitation experiments in EPEC revealed that EscQ interacts with the ATPase complex components EscN and EscL. Moreover, EscQ was found to be located in the cytoplasm as well as associated with the membrane and this subcellular localization was independent of its binding partners (Biemans-Oldehinkel et al., 2011). Likewise, the EscQ homologs Spa33, YscQ, and SpaO have been shown to interact with members of the SctK and SctL families (MxiK/MxiN, YscK/YscL, OrgA/OrgB; Jackson and Plano, 2000; Jouihri et al., 2003; Lara-Tejero et al., 2011). In S. enterica, SpaO, OrgA, and OrgB form a large molecular weight complex that interacts with translocators and effectors in a sequential order, leading to the suggestion that the C-ring may act as a sorting platform, queuing the different substrate categories to establish the correct secretion hierarchy. Both chaperones and molecular switch proteins have an active role in this selection process, acting in concert with the sorting platform to ensure an orderly secretion (Lara-Tejero et al., 2011). In Y. enterocolitica and S. enterica the formation of the C-ring/sorting platform substructure depends on the presence of both SctK (YscK and OrgA) and SctL (YscL and OrgB) proteins (Lara-Tejero et al., 2011; Diepold et al., 2015). In contrast, the sorting platform visualized from Shigella ΔmxiN minicells showed that structural pods are assembled even in absence of the MxiN (SctL) protein (Hu et al., 2015). Based on the information for the homologous proteins, in A/E pathogens the sorting platform might be formed by EscQ, EscL, and EscK (formerly Orf4), a component that has been associated with the SctK protein family (Abby and Rocha, 2012; Barison et al., 2013). Like other members of the SctK protein family such as OrgA from Salmonella and MxiK from Shigella, EscK is critical for T3 secretion (Sukhan et al., 2001; Jouihri et al., 2003; Deng et al., 2004). The precise composition and function of the sorting platform in A/E pathogens remains to be investigated.

The ultimate goal of T3SS assembly is to modulate host cell functions for the benefit of the bacterium, and this is accomplished by the interplay of the biochemical activities displayed by injected effectors. A/E pathogens share seven effectors encoded in the LEE-PAI, and additionally each pathogen has its own suite of effectors encoded outside the LEE, termed non-LEE-encoded (Nle) effectors, which are encoded on prophages or other integrative elements (McDaniel et al., 1995; Deng et al., 2003, 2004, 2010; Tobe et al., 2006; Iguchi et al., 2009).

The LEE-encoded effectors are Tir, Map, EspF, EspG, EspH, EspZ, and EspB (Garmendia et al., 2005). It has been shown in EPEC that LEE-encoded effectors are hierarchically secreted/translocated into host cells, being Tir the first effector to be injected followed by EspZ, EspF, EspH, EspG, and Map (Thomas et al., 2007; Mills et al., 2008). As previously mentioned, although EspB is a structural translocon component, it is also translocated into host cells but in much lower levels compared to the other effectors, thus precluding a meaningful analysis of its translocation hierarchy (Wolff et al., 1998; Mills et al., 2008). The total number of effectors varies between pathogens, C. rodentium possesses a reported repertoire of at least 29 effectors, while EPEC E2348/69 encodes ca. 22 and EHEC O157:H7 encodes 39 (Tobe et al., 2006; Dean and Kenny, 2009; Petty et al., 2010). Some of the Nle effectors, like NleA/EspI, NleB, NleE, NleF, NleG, NleH, EspJ, and EspL, are common to all LEE-encoding species sequenced to date (Petty et al., 2010). A more recent hierarchical translocation order was reported including both LEE and Nle effectors, in which Tir is again the first one to be injected followed by EspZ, NleA, NleH1, EspF, EspH, NleH2, EspJ, Map, EspG, NleD, NleF, NleB1, NleE1, NleB2, NleC, NleG, NleE2, EspG2, and EspL2 (Mills et al., 2013). For a detailed description of the specific function of effectors in A/E pathogens, we refer to reviews on the subject (Dean and Kenny, 2009; Jayamani and Mylonakis, 2014; Santos and Finlay, 2015).

In general, effectors contain one or more C-terminal catalytic domains that manipulate host cell functions, while their N-termini mediates chaperone recognition and transport via the T3SS (Ghosh, 2004). The first ca. 20 amino acids contain the secretion signal sequence necessary and sufficient for T3 secretion (Crawford and Kaper, 2002; Munera et al., 2010; Deng et al., 2015). The chaperone-binding domain is located downstream of the N-terminal secretion signal, usually between amino acids 50 and 100 (Blocker et al., 2003). This domain adopts a linear conformation and wraps around the chaperone dimer through a β-motif (Lilic et al., 2006).

The T3SS-associated chaperones are small and acidic cytoplasmic proteins important for efficient secretion of their cognate substrates (Feldman and Cornelis, 2003). They promote T3 substrate secretion by contributing to protein stability, aiding in substrate recognition and targeting, preventing premature oligomerization and unspecific interactions, and participating in the establishment of a secretion hierarchy (Wainwright and Kaper, 1998; Abe et al., 1999; Elliott et al., 1999a; Gauthier and Finlay, 2003; Neves et al., 2003; Parsot et al., 2003; Creasey et al., 2003a,c; Thomas et al., 2005, 2007; Su et al., 2008; Chen et al., 2013; Sal-Man et al., 2013; Allison et al., 2014). T3 chaperones function in an ATP-independent manner, although chaperone dissociation from the substrate-chaperone complex is an ATP-dependent process (Akeda and Galán, 2005). These proteins have been grouped into three classes depending on their binding substrates: class I chaperones that bind effectors (class IA, one effector and class IB, multiple effectors), class II chaperones that bind translocators, and class III chaperones that bind the needle component (Cornelis and Van Gijsegem, 2000; Page and Parsot, 2002; Parsot et al., 2003; Cornelis, 2006).

The LEE encodes eight proteins that have been characterized as chaperones: CesF, CesL, CesT, CesAB, CesD, CesD2, EscE, and EscG/CesA2 (Wainwright and Kaper, 1998; Elliott et al., 1999a, 2002; Luo et al., 2001; Creasey et al., 2003a,c; Neves et al., 2003; Thomas et al., 2005; Younis et al., 2010; Ramu et al., 2013; Sal-Man et al., 2013). These proteins, with the exception of CesF and CesD2, are critical for C. rodentium virulence in mice (Deng et al., 2004). CesT was originally classified as a class IA chaperone for the Tir effector, but subsequent reports showed that it interacts with 10 additional effectors: Map, NleA, NleF, NleH, NleH2, EspH, EspZ, EspF, NleG, and EspG, being re-classified as a multicargo class IB chaperone (Abe et al., 1999; Elliott et al., 1999a; Creasey et al., 2003a,b; Thomas et al., 2005, 2007; Mills et al., 2008). CesF is a class IA chaperone that binds EspF (Elliott et al., 2002) and is required for its translocation (Mills et al., 2008). Additionally, based on sequence similarity searches, CesL was recently classified as a class I chaperone for SepL although it is a special case, since its substrate seems to be an aberrant effector that is not secreted (Younis et al., 2010).

Four class II chaperones have been described in A/E pathogens: CesD, CesD2, CesAB, and CesA2. CesD assists EspD and EspB secretion, although it has only been shown to directly interact with EspD (Wainwright and Kaper, 1998), which has an additional chaperone, CesD2 (Neves et al., 2003). The CesAB chaperone differs from the others in that it has a basic pI. It interacts specifically with EspA and EspB, preventing their polymerization, maintaining them partially unfolded and promoting their stabilization (Creasey et al., 2003b; Yip et al., 2005b). Finally, CesA2 was identified in EHEC as a second chaperone for EspA (Su et al., 2008). However, it was later demonstrated in EPEC that CesA2 is a class III chaperone that binds to the needle subunit EscF and therefore, it was named EscG, which together with EscE avoid premature polymerization of the needle (Sal-Man et al., 2013).

The observation that the size of YscP correlates with the length of the needle, led Journet and colleagues to propose that this protein functions as a molecular ruler that directly measures needle length (Journet et al., 2003). In this static ruler model, YscP is positioned within the inner channel of the growing needle, with its N- and C-terminal domains probably contacting the needle tip and the T3SS base, respectively, allowing the secretion of needle subunits. When the molecular ruler is completely stretched, it interacts with YscU at the T3SS base to switch substrate secretion, and gets secreted, finishing the export of needle subunits and permitting middle and late substrate secretion (Journet et al., 2003). Afterwards it was demonstrated that only one YscP molecule was required for length control and that, while measuring needle length, it preserved its helical structure (Wagner et al., 2009, 2010b). However, since the diameter of an α-helix is approximately 10 Å and that of the secretion channel is 13 Å, it is impossible to fit a ruler protein and a secreted needle subunit in it (Fujii et al., 2012). Under these same physical constraints, the simultaneous secretion of FliK and hook subunits in the flagellar system was also suggested to be improbable, and therefore an alternative model was proposed for hook length regulation (Shaikh et al., 2005; Moriya et al., 2006).

In agreement with the previous model, FliK, Spa32, and InvJ have also been demonstrated to act as molecular rulers (Shibata et al., 2007; Botteaux et al., 2008; Erhardt et al., 2011; Wee and Hughes, 2015). However, the proposed needle/hook measurement mechanisms are different. The alternative dynamic-ruler model was originally proposed in the flagellar system by Moriya et al. (2006). This model suggests that during hook assembly FliK is occasionally exported, and that when found in the central channel secretion is briefly paused, allowing the measurement of hook length through the temporary interaction of the N- and C-terminal domains of FliK with the hook cap FlgD and FlhB, respectively. Short hooks do not allow productive interactions with FliK, inducing its rapid secretion, however, as the hook assembles, more frequent interactions between FliK and the growing structure occur, slowing the secretion rate. Finally, when the hook length reaches approximately 55 nm, the slow secretion rate enables the completely stretched FliK to successfully interact with FlhB, which flips the switch in substrate secretion (Erhardt et al., 2010a, 2011). A similar mechanism was proposed for the needle length control in EPEC. EscP, which directly interacts with EscF, measures needle length upon intermittent secretion during needle assembly (Figure 4A). When this structure reaches its final length, all subdomains of EscP can make contact with the needle, promoting a pause in secretion that allows EscP to interact with EscU, inducing a conformational change that modifies its specificity from early to middle and late substrates (Figure 4B; Monjarás Feria et al., 2012).

An alternative model was proposed for the SPI-1 injectisome in S. enterica, which states that the assembly of the inner rod determines the size of the needle. Therefore, the substrate switching is the result of conformational changes produced at the base of the T3SS in response to inner rod assembly completion (Marlovits et al., 2006). Many pieces of evidence support this model. An invJ null mutant, which assembles long needles (Kubori et al., 2000), secretes high levels of the inner rod subunit PrgJ, but fails to assemble this structure, thus, InvJ was proposed to promote inner rod assembly. In addition, the overexpression of PrgJ results in the assembly of shorter needles, apparently because of faster inner rod assembly (Sukhan et al., 2003; Marlovits et al., 2006). Moreover, mutations in PrgJ that slowed the rate of rod assembly, presumably due to impaired subunit-subunit interactions, lead to the assembly of longer needles (Lefebre and Galán, 2014). Likewise, a role of the inner rod component in substrate specificity switching has been proposed in Y. pseudotuberculosis. It was shown that the inner rod protein YscI is hypersecreted in a yscP mutant background and that a suppressor mutation in YscU reestablishes normal YscI secretion, suggesting that the inner rod assembly is important for the substrate specificity switching mechanism (Wood et al., 2008). In EPEC, EscP was shown to directly interact with the inner rod subunit EscI, and it is proposed to regulate its secretion, since an escP null mutant hypersecretes EscI (Monjarás Feria et al., 2012). Besides, EscI has also been shown to interact with the export gate protein EscU (Creasey et al., 2003b; Sal-Man et al., 2012b).

It is possible that a combination of these models will contribute to the molecular mechanism of needle length control and substrate specificity switching. Finally, the outcome of the first molecular switch is the assembly of a proper size needle and the trigger of middle and late substrate secretion.

The mechanism of translocator secretion regulation is not completely understood; however, in Salmonella it was shown that InvE directly interacts with the chaperone/translocator complex, though not with its individual components (Kubori and Galán, 2002). Additionally, the Chlamydial gatekeeper CopN interacts with the translocator-specific chaperone Scc3, and this interaction is important not only for translocator secretion but also for timely secretion prior to effectors (Archuleta and Spiller, 2014). Likewise, Shigella MxiC was shown to interact with the translocator chaperone IpgC (Cherradi et al., 2013). These results highlight the importance of the interaction of the gatekeeper with the translocator/chaperone complex or its individual components in the establishment of the secretion hierarchy. Furthermore, as aforementioned, in a seminal work Lara-Tejero et al. demonstrated the existence of a sorting platform that selects substrates for secretion. Translocators are loaded into the sorting platform in an InvE dependent manner, and elimination of translocators or the gatekeeper results instead in the loading of effectors (Lara-Tejero et al., 2011).

Recently, the crystal structure of SepL from EPEC was determined (Burkinshaw et al., 2015b). It showed that despite the low sequence similarity between gatekeeper proteins, its overall structure is well conserved, with a general architecture consisting of three helical X-bundle domains. When compared to the solved structures of CopN, YopN/TyeA and MxiC (Schubot et al., 2005; Deane et al., 2008b; Archuleta and Spiller, 2014), the SepL structure showed that among the regions conserved between these proteins one corresponds to the binding site of the translocator chaperone Scc3 in CopN. Protein sequence analysis highlighted the presence of three highly conserved residues: Tyr188, Phe327 and Arg333 (Burkinshaw et al., 2015b). In CopN, Arg365 (equivalent to Arg333 in SepL) is essential for the CopN/Scc3 interaction (Archuleta and Spiller, 2014). In contrast, alanine substitution of Arg333 in SepL did not affect the secretion phenotype (Burkinshaw et al., 2015b), though an interaction between SepL and translocators or their chaperones in A/E pathogens has yet to be demonstrated. In conclusion, the precise molecular mechanism through which gatekeepers promote translocator secretion remains an open question.

The prevention of effector secretion before host cell contact is a common feature of all gatekeeper proteins. For some of them, this regulatory function relies in their ability to be secreted, e.g., YopN, MxiC, CopN, and PopN (Day and Plano, 1998; Fields and Hackstadt, 2000; Yang et al., 2007; Botteaux et al., 2009). In the case of Yersinia, YopN functions like a plug that physically blocks the secretion channel while being attached to the T3SS base through its interaction with TyeA, however, in response to low calcium concentration, this interaction is disrupted, YopN is secreted and the blockade is relieved allowing the secretion of Yops (Cheng et al., 2001). In Chlamydia, the secretion of CopN is also essential for secretion of effectors (Archuleta and Spiller, 2014; Shen et al., 2015). Additionally, a recent model of effector secretion control in Pseudomonas suggests that the PopN complex (PopN, Pcr1, PscB, and Pcr2) is tethered to the T3S apparatus via the Pcr1/PcrD (SepL C-terminal domain/EscV) interaction, blocking the access of effectors to the secretion apparatus. However, when the appropriate signal for effector secretion is received, PopN is secreted, permitting the loading of effectors onto the sorting platform and its subsequent docking to the export gate for secretion (Lee et al., 2014). A similar mechanism was proposed for MxiC from Shigella, in which its interaction with the inner rod protein MxiI, blocks access of effectors to the entry gate. Nevertheless, since this mechanism is also conserved for InvE in Salmonella, which is not secreted, its regulation might rely instead in the ability of MxiC to engage with the rod and not in its secretion capacity (Botteaux et al., 2009; Cherradi et al., 2013).

In A/E pathogens, neither SepL nor SepD have been observed to be secreted. Therefore, a different secretion regulation mechanism was proposed in EHEC, in which SepL was shown to bind to the effector Tir. It was demonstrated that while translocators are secreted, the interaction of Tir with SepL delays its secretion, avoiding the premature export of all other effectors, which suggests that this interaction controls the proper timing of secretion (Figure 4C; Wang et al., 2008). This mechanism is supported by the demonstration that Tir secretion is important for the secretion of the rest of the effectors (Thomas et al., 2007). In addition, as aforementioned, there is an established hierarchy in the secretion of effectors in which Tir is the first effector to be secreted (Figure 4D; Mills et al., 2008, 2013). In this regard, it has also been shown that although Tir translocation is not essential for the translocation of other effectors, its absence negatively influences their translocation efficiency (Mills et al., 2013). Finally, it was suggested that different interacting partners of SepL are required for its secretion regulatory function. This is sustained by the fact that the same region of SepL that is required for Tir binding is also essential for its interaction with EscD (Wang et al., 2008).

Even though SepL is not secreted, it was proposed to be an aberrant effector because when fused to a reporter protein, its N-terminal domain is able to mediate protein secretion. Moreover, the SepL-SepD complex was shown to interact with a third protein named CesL, which like SepD, is suggested to function as a SepL chaperone (Younis et al., 2010). The SepL-SepD-CesL complex resembles the YopN/TyeA-YscB-SycN complex in Yersinia (Day and Plano, 1998), PopN/Pcr1-PscB-Pcr2 complex in Pseudomonas (Yang et al., 2007), CopN-Scc4-Scc1complex in Chlamydia (Silva-Herzog et al., 2011) and SsaL-SpiC-SsaM complex in Salmonella SPI-2 T3SS (Yu et al., 2010). An outstanding work by Yu and colleagues demonstrated that the SsaL-SpiC-SsaM complex regulates the secretion hierarchy in response to pH changes. This protein complex associates with the T3SS base under the acidic conditions of the Salmonella-containing vacuole, allowing the secretion of translocators while blocking the secretion of effectors. When the translocation pore assembles in the vacuolar membrane, an increase in pH is detected and transmitted to the T3SS base, which in turn causes dissociation of the SsaL-SpiC-SsaM complex and the degradation of its individual components, permitting the secretion of effectors (Yu et al., 2010). In A/E pathogens it has been shown that, upon removal of calcium from the growth medium, translocator secretion is reduced while effector secretion is increased, a phenotype that partially mimics that of the sepL or sepD mutants (Kenny et al., 1997a; Ide et al., 2003; Deng et al., 2005). This suggests that the absence of calcium might be the environmental cue that regulates the secretion hierarchy, probably through the SepL/SepD complex (Deng et al., 2005).