While CdiA proteins vary in size (180–630 kDa), they all share the same overall architecture (Figure 1A). CdiA proteins display high conservation at the large N-terminal region which encompasses an N-terminal TPS transport domain, and two large filamentous repeat regions, FHA-1 and FHA-2. Sandwiched between the FHA-1 and FHA-2 regions are two additional domains: the receptor-binding domain (RBD) and the YP domain—a tyrosine and proline enriched region. CdiA FHA regions form extended β-helices resulting in a large rod-like structure that protrudes from the cell-surface and presents the RBD at its tip (Aoki et al., 2010; Hayes et al., 2010). While both the YP and RBD domain are required for target-cell adhesion, the RBD is responsible for contact with target-cell receptor proteins and the YP domain is likely responsible for CdiA secretion arrest (described in Current Understanding of CDI Toxin Delivery) (Ruhe et al., 2018). After the FHA-2 region is the pretoxin (PT or CdiA-PT) domain that terminates with a highly conserved PT motif: a VENN motif in most species, an (E/Q)LYN motif in Burkholderia species (Nikolakakis et al., 2012), one of five distinct motifs in P. aeruginosa (WVHN, TENN, LYVT, DAMV, NEALV) (Mercy et al., 2016; Allen and Hauser, 2019), or an LPEN motif in some classes of Acinetobacter species (De Gregorio et al., 2019). The PT-motif demarcates the beginning of the highly polymorphic CdiA-CT region which is typically 200–450 residues. In general, CdiA-CT encompasses an N-terminal cytoplasmic entry domain and a C-terminal cytotoxic domain.

CdiA is highly modular: individual domains control distinct CDI mechanisms. Single, independently functioning domains are responsible for interactions with the target cell, entry into the target-cell cytosol, and cytotoxicity. For example, in P. aeruginosa distinct RBDs, CdiA-CT entry and cytotoxic domains have been identified; various arrangements of these modules in CdiA proteins result in altered target-cell entry and toxicity mechanisms (Allen and Hauser, 2019). As each of these modules act independently, functional chimeric CdiA proteins can be created to control target-cell recognition, entry mechanism, or toxin activity (Aoki et al., 2011; Nikolakakis et al., 2012; Ruhe et al., 2013). One such example of a CdiA chimera protein is generated by introducing a non-native CdiA-CT region directly following the species-specific PT-motif and results in the exchange of entire toxin activities between diverse CDI⁺ species. These chimeric CdiA proteins retain the ability to deliver non-cognate toxins into the target-cell, resultantly only cells expressing the cognate immunity protein are resistant to growth inhibition by the chimeric CdiA (Aoki et al., 2011; Nikolakakis et al., 2012; Ruhe et al., 2013). The modularity of the cytotoxic CdiA-CT region allows for the exchange and acquisition of new toxins through horizontal gene transfer (HGT - further discussed in Genetics of CDI and Contact-Dependent Signaling) and recombination inside and outside of the cdi loci. As different classes of secreted effectors possess toxins with homology to known CdiA-CT cytotoxic domains, there is substantial evidence for such genetic transfer (Poole et al., 2011; Mercy et al., 2016; Ruhe et al., 2016; Jones et al., 2017; Allen and Hauser, 2019).

Further, the CdiA-CT region is composed of two modular domains; the CdiA-CT entry domain plays a critical role in toxin translocation (Willett et al., 2015a), while the CdiA-CT cytotoxic domain is responsible for cytotoxic activity (Figure 1A). The CdiA-CT entry domains recognize specific inner membrane proteins (IMPs) to mediate toxin transport across the inner membrane (Willett et al., 2015a). CdiA-CT regions with different toxic activity but conserved entry domains rely on the same translocation mechanism to gain entrance into the cytosol. Thus, when fusion proteins are generated that exchange the CdiA-CT entry domains between CdiA-CT regions, the mechanism of target-cell inner membrane translocation is dictated by its entry domain. For example, expression of the Burkholderia pseudomallei 1026b specific IMP was required for the entry of an E. coli CdiA fusion containing the full B. pseudomallei CdiA-CT region into E. coli cells (Willett et al., 2015a), demonstrating the requirement for specific IMPs for CdiA-CT target-cell entry amongst different bacterial species.

When inhibitor-cell CdiA makes contact with a specific target-cell outer membrane receptor, the cytotoxic CdiA-CT region is cleaved and translocated into the cytoplasm of the target bacterium by a multi-step, partially characterized mechanism (Aoki et al., 2005; Aoki et al., 2010), Figure 1B. Different CDI systems recognize distinct target-cell receptors. Class I E. coli CdiA proteins, like the well-characterized CdiA^EC93, bind to OMP BamA (Aoki et al., 2008; Ruhe et al., 2013), Class II E. coli CdiA proteins (e.g., CdiA^EC536) recognize OmpF/OmpC heterotrimers (Beck et al., 2016), and Class III E. coli CdiA proteins (e.g., CdiA^STEC031) recognize Tsx (superscripts indicate the shortened names of the bacterial strain) (Ruhe et al., 2017). While the majority of currently identified target-cell receptors are OMPs, Class VI E. coli CdiA (e.g., CdiA₂ ^STECO31 or CdiA^STEC4) and B. pseudomallei 1026b CdiA_II recognize a glycolipid, lipopolysaccharide (LPS) as their target-cell receptor (Koskiniemi et al., 2015; Halvorsen et al., 2021); notably, the E. coli STEC_O31 CDI operon has an additional cdiC gene that appends a lipid moiety onto CdiA-RBD^STECO31 (Alternate CDI Loci Arrangements). Finally, Class V E. coli CdiA proteins (e.g., CdiA^SWW33) recognize an as yet unidentified cell surface receptor. Interestingly, colicins are also known to utilize OMP receptors for target-cell entry (Cascales et al., 2007).

When CdiA is displayed on the inhibitor-cell surface, its RBD is presented at the most distal point to make contact with the target-cell while the FHA-1 domain and the YP-domain provide the scaffold necessary to extend CdiA away from the cell surface (Ruhe et al., 2018), Figure 1B. Though the N-terminal region of CdiA (N-terminus through the PT-motif) is highly conserved (77% identity for CdiA from EC93 and EC536), the RBD has decreased sequence identity (24% between EC93 and EC536) (Ruhe et al., 2017). Like the modularity observed for CdiA-CT, the RBD can be exchanged between CdiA proteins to alter target-cell receptor specificity (Ruhe et al., 2013; Ruhe et al., 2017). Notably, five distinct RBD sequences have been identified for P. aeruginosa (Allen and Hauser, 2019), however, no OMP receptor has been identified to-date.

As mentioned above, BamA is the target-cell receptor for Class I E. coli CdiA proteins. While BamA is highly conserved across enterobacteria with 73–93% sequence identity across E. coli EC93, Salmonella Typhimurium, E. cloacae, Proteus mirabilis, Citrobacter freundii, and Enterobacter aerogenes, the regions with highest sequence variability in enterobacteric BamA are in the extracellular loops (Ruhe et al., 2013). Strikingly, the exchange of E. cloacae extracellular loops six and seven for E. coli loop sequences allowed for chimeric E. cloacae BamA to be recognized by E. coli CdiA with subsequent CdiA-CT translocation (Ruhe et al., 2013). This suggests that BamA loops six and seven appear to be the only requirement for target-cell recognition for Class I E. coli CdiA effectors.

When CdiB presents CdiA onto the surface of the cell, the C-terminus of CdiA is retained in the inhibitor-cell periplasm until the CdiA-RBD binds its cognate target-cell receptor (Ruhe et al., 2017; Lin et al., 2020), Figure 1B. With target-cell receptor recognition, inhibitor-cell CdiA secretion continues, resulting in the translocation of CdiA-PT and CdiA-CT regions into the periplasm of the target-cell. The translocation event relies on contacts between FHA-2 and the target-cell; however, the precise mechanism of target-cell outer membrane translocation is unclear. As there is evidence that FHA-2 is internalized by the target-cell, and FHA-2 is predicted to structurally resemble a β-barrel OMP involved in lipopolysaccharide transport (Shigella LptD), it has been postulated that FHA-2 forms a β-barrel pore in the target-cell membrane for CdiA-CT toxin translocation (Ruhe et al., 2018).

Once the CdiA-PT and CdiA-CT regions enter the target-cell periplasm, cleavage of the CdiA-CT region from the large CdiA N-terminal region has not been well characterized. However, the PT-motif could signal for cleavage. Interestingly, in some Pseudomonas strains, a bacterial intein-like (BIL) domain was identified that terminates just before the CdiA-CT region (Mercy et al., 2016). As BIL domains have been implicated in autoproteolytic cleavage in P. syringae CdiA-CT (Amitai et al., 2003), similar domains could be present in other bacteria. Notably, the Pseudomonas WVHN PT-motif is predicted to be involved in BIL domain protein cleavage, where the terminal Asn residue is the C-terminal residue of the N-terminal domains following cleavage and a putative active site residue in the P. syringae CdiA BIL domain.

The translocation of the CdiA-CT region from the periplasm into the target-cell cytosol is dependent on the CdiA-CT entry domain, which recognizes a specific IMP and dictates the mechanism of cytosol entry (Willett et al., 2015a; Lin et al., 2020). Known CDI IMP receptors are multidrug transport protein component, AcrB (Aoki et al., 2008); ABC transporter membrane permeases MetI, RbsC, and GltJ/GltK; phosphotransferase, PtsG, and the ATP-dependent zinc metallopeptidase, FtsH; and an uncharacterized IMP, YciB (Willett et al., 2015a); and SecY, the channel-forming subunit for Sec pathway IMP internalization (Jones et al., 2021). Lastly, the IMP complex, DppB/DppC, part of an ABC dipeptide import system, was required for the cytosolic entry of a CdiA-CT in P. aeruginosa (Allen et al., 2020), however, as nine different translocation domains have been identified in P. aeruginosa (Allen and Hauser, 2019), other IMPs likely mediate CdiA-CT translocation. Additionally, the energy to cross the target-cell inner membrane requires proton-motive force (pmf) (Ruhe et al., 2014). Interestingly, the translocation of colicin toxins across the outer membrane is similarly dependent on a proton gradient (Cascales et al., 2007). However, CdiA does not require pmf for outer membrane translocation, further delineating the outer and inner membrane translocations steps (Ruhe et al., 2014).

Recently, the structure of a MetI-dependent CdiA-CT entry domain from E. coli STEC_MHI813 was characterized by NMR (Bartelli et al., 2019). The structure revealed a predominantly helical protein with a dynamic N-terminal region structured by disulfide bonds that are required for CdiA-CT translocation into the target-cell cytosol. Strikingly, these disulfide-forming cysteine residues are conserved across CdiA-CT entry domains while most CdiA proteins are otherwise devoid of cysteine residues. The prevalence of entry domain disulfide bonds suggests widespread importance for disulfide bonds in transport to the cytosol. Further, the CdiA-CT entry domain adopts a molten helical structure (Bartelli et al., 2019) akin to the increased flexibility of some colicin proteins in the presence of anionic lipids (Cascales et al., 2007). This physical property is also likely important in the transport of some CdiA-CT toxins across target-cell membranes aiding in translocation upon contact with membrane anionic lipids.

Substantial ground has been made in characterizing the mechanism of CdiA-CT delivery into the target-cell, however significant questions remain. In particular, the precise determinants of how CdiA-CT is translocated across the target-cell outer and inner membrane, and the molecular details behind the cleavage event that frees CdiA-CT from CdiA are still unknown. Much work is required to understand the mechanism of toxin cleavage and translocation.

CdiA is presented onto the surface of CDI⁺ bacteria by the TpsB protein CdiB, which is an Omp85 family member. The structures of CdiB from Acinetobacter baumannii and E. coli (Guerin et al., 2020) have been solved and each comprise of a 16-stranded β-barrel with an N-terminal α-helix H1 that transverses the center of the β-barrel from extracellular to periplasm side followed by two N-terminal periplasmic POTRA domains and a canonical extracellular loop L6 (Figure 1C). Based on similarity to other TpsB proteins (Clantin et al., 2007; Gruss et al., 2013; Noinaj et al., 2013; Ni et al., 2014; Maier et al., 2015; Guerin et al., 2020), L6 is proposed to act as a ‘lid-lock’ inside the β-barrel and is probably essential for CdiA secretion to the inhibitor-cell surface. Further, POTRA domain P2 has been implicated in CdiA recognition and secretion (Hodak et al., 2006). Even though CdiA proteins are highly homologous, CdiA secretion appears to be species specific, wherein CdiA from A. baumannii cannot be secreted by E. coli CdiB, and similarly, E. coli CdiA is not displayed on the cell-surface in the presence of A. baumannii CdiB (Guerin et al., 2020). Notably, CdiB structures have high similarity to that of FhaC, an OMP responsible for the secretion of the FHA toxin (Clantin et al., 2007; Maier et al., 2015). Thus, research on FhaC is likely relevant to CdiB. The structures of CdiB show a “closed” conformation with α-helix H1 blocking the lumen of the β-barrel (Guerin et al., 2020). This internal helix may be important in ensuring unidirectional CdiA secretion. Questions remain as to how α-helix H1 is displaced to allow for CdiA secretion and what processes would be required to make this movement energetically favorable.

Many cdi loci are complicated by accessory genes, transposable elements, and the presence of orphan toxin and immunity protein modules (Poole et al., 2011). Indeed, in Burkholderia ten distinct cdi locus types have been described (Anderson et al., 2012; Nikolakakis et al., 2012). Most B. pseudomallei strains encode a bcpO gene in their cdi cluster, resulting in a bcpAIOB loci. The accessory protein BcpO localizes to the inner leaflet of the outer membrane, and bcpO mutants in B. thailandensis are associated with reduced CDI activity and a defect in autoaggregation (Anderson et al., 2012). BcpO is a predicted lipoprotein but has not been well characterized and its specific activity is unknown.

Class IV E. coli CdiA proteins are encoded by the cdiBCAI type loci, encoding a CdiC accessory protein (Willett et al., 2015a; Ogier et al., 2016). As CdiC bears sequence similarity to the hemolysin activator (HlyC), CdiC is a predicted lysyl acyltransferase that potentially activates CdiA through lysine acetylation (Ogier et al., 2016). Recently, it has been shown that E. coli STEC_O31 CdiC appends 3-hydroxydecanoate to a specific lysine residue within CdiA-RBD^STECO31 (Halvorsen et al., 2021). As the loss of CdiA-RBD^STECO31 acylation resulted in reduced toxicity to the target-cell, and Class IV CdiA-RBD recognize LPS on the target-cell, it seems likely that CdiA-RBD acylation could be involved in outer membrane entry, increasing the affinity and stabilizing the interaction of CdiA^STECO31 with the target-cell LPS-rich leaflet (Halvorsen et al., 2021). The precise mechanisms of CdiA lipidation by CdiC and CdiA fatty acid insertion into the target-cell outer membrane remain unknown.

Many organisms, like E. coli, Y. pestis, and P. aeruginosa, encode a series of orphan toxin (CdiA-CT) and immunity (CdiI) modules downstream of the cdi loci (Poole et al., 2011; Allen and Hauser, 2019). Burkholderia species sometimes encode these modules between cdiI and cdiB (Anderson et al., 2012; Nikolakakis et al., 2012). While the orphan immunity genes appear to be expressed, the orphan toxin genes lack a translation initiation codon, typically starting at the PT-motif (VENN (E/Q)LYN, or other) (Poole et al., 2011). These orphan toxin modules have toxic activity when expressed (Poole et al., 2011), thus, reorganization of the cdi loci could result in novel CdiA proteins encoding orphan toxin modules. These orphan toxin/immunity modules may represent ancestral toxin/immunity pairs that were displaced through the introduction of new toxin/immunity pairs (Jones et al., 2017). Some modules have homology to toxin and immunity proteins from other bacterial strains and are perhaps evidence of horizontal gene transfer (HGT). Furthermore, orphan immunity proteins likely provide protection against alternative CDI systems. Because immunity to additional CDI systems would confer evolutionary advantages, an organism would likely face selective pressures to maintain additional cdiI genes. Such evolutionary pressure would explain the observation that cdi loci in P. aeruginosa strains have as many as 15 orphan immunity genes (Allen and Hauser, 2019). Similarly, analysis of a S. Typhi genomic islet revealed an encoded orphan CdiI protein homologous to an E. coli O157:H7 strain EC869 CdiI (Barretto and Fowler, 2020), while the rest of the S. Typhi is devoid of cdi genes. This finding suggests that S. Typhi encodes the cdiI gene to protect itself against competing bacteria that encode EC869-like toxins.

Recently, mobile genetic elements were characterized in B. thailandensis that carry bcpAIOB on extrachromosomal DNA “megacircles” (Ocasio and Cotter, 2019). In these megacircles, the cdi locus is framed by IS2-like elements, allowing for insertion of CDI genes into chromosomes. While these megacircles have not been observed in other bacteria, this system shows a mechanism by which CDI genes can be replicated and inserted directly into DNA. Further these transposable megacircles have been implicated in contact-dependent signaling (CDS) (Ocasio and Cotter, 2019).

While CDI describes the competition and discrimination bacteria impose against “nonself” cells, CDS describes a phenomenon where CDI proteins mediate cooperative behaviors amongst “self’ or sibling cells (Danka et al., 2017; Ocasio and Cotter, 2019). CDS behaviors include the formation of DNA megacircles, along with biofilm formation, pigmentation, and polysaccharide production in Burkholderia species (Garcia et al., 2013; Garcia et al., 2016; Ocasio and Cotter, 2019). CDS is an emerging field of investigation, and other organisms need to be investigated for CDS phenotypes.

The BECR superfamily is incredibly diverse and includes barnase, a ribonuclease toxin from Bacillus amyloliquefaciens; the EndoU (Endoribonuclease specific for uridylate) superfamily of nucleases; colicins ColE5, ColD and ColE3 (Figure 2A), discussed earlier, with tRNase or rRNase activity; and the RelE/ParE superfamily, which includes the RelE family of ribosome dependent mRNA endonucleases and the ParE family of plasmid partition proteins that inhibit DNA gyrase and DNA replication. Characterized CdiA-CT toxin structures have homology to proteins from each of these BECR subfamilies apart from barnase (Table 1).

While several CDI toxins have sequence similarity to BECR nucleases, most structurally characterized BECR CdiA-CT proteins were only recognized as BECR family members based on their structure and not primary sequence alone. To date, CdiA-CT^EC3006, CdiA-CT^Kp342, CdiA-CT^NC101, CdiA-CT^STECO31, CdiA-CT^ECL, CdiA-CT^PABL017, and CdiA-CT^Ykris all have structural homology to BECR family members (Table 1; Figure 2). BECR-fold CdiA-CT proteins highlight the diversity of the BECR-fold family and the difficulty in identifying BECR family members and/or their specific nuclease activity by sequence alone.

Despite reasonable structural similarities, where CdiA-CT proteins typically align to BECR core domains with 2.4–3.5 Å root mean square deviation (rmsd), the sequence identity between these proteins is low with 7–17% sequence identity (Table 1). The BECR CdiA-CTs all have nuclease activity and comprise the BECR α/β-core, however, like the canonical BECR family members, the BECR α/β-core is marked by modifications, insertions and substitutions of secondary structure elements that are distinct for each CdiA-CT. This phenomenon has also been observed in non-CDI PTS toxins (Zhang et al., 2014). Interestingly, CdiA-CT^Kp342 and CdiA-CT^NC101 structurally resemble ColD, ColE5, and RelE toxin family members (BrnT, RelE, ParE) (Table 1; Figures 2A–C). Like CdiA-CT^Kp342 and CdiA-CT^NC101, other BECR CdiA-CTs have structural homology to several distinct proteins of the BECR superfamily. However, some toxins more closely resemble a single BECR family, where CdiA-CT^STECO31 only has structural homology to EndoU type toxins, and CdiA-CT^ECL only resembles ColE3 (Figures 2D,F). Despite differences between BECR CdiA-CTs, structural similarity to canonical BECR toxins has guided the elucidation of substrate selectivity and the mechanism of action for each toxin.

Currently there are four CdiA-CT toxins with structural homology to the PD-(D/E)XK superfamily nucleases: CdiA-CT^Bp1026b, CdiA-CT^E479, and CdiA-CT^TA271 and its closely related family member CdiA-CT^YPIII (Morse et al., 2012; Morse et al., 2015; Johnson et al., 2016b). The PD-(D/E)XK nuclease superfamily are a diverse set of proteins with extreme sequence variability (Knizewski et al., 2007; Steczkiewicz et al., 2012). For this reason, identifying PD-(D/E)XK family members by sequence alone is difficult. PD-(D/E)XK proteins are characterized by a αβββαβ core, well-conserved catalytic lysine and aspartate residues, and one or more metal-binding sites coordinated by carboxylate groups (i.e., Asp or Glu residues). The proteins have diverse nuclease related functions, including amongst others: DNA restriction, tRNA splicing, transposon excision, DNA recombination and Holliday junction resolution (Knizewski et al., 2007; Steczkiewicz et al., 2012).

Structural similarities between PD-(D/E)XK CdiA-CT toxins is readily apparent (Figure 3). CdiA-CT^TA271 and CdiA-CT^YPIII belong to the same toxin/immunity family with a rmsd of 1.7 Å and high sequence identity (64%); the Burkholderia CdiA-CT^Bp1026b and CdiA-CT^E479 are also similar with 2.9 Å rmsd and medium sequence similarity (19%) (Table 1). While these two pairs are disparate in sequence (5–12% identity), the two groups bear striking structural homology to each other with 3.5–4.4 Å rmsd. All four proteins have the conserved αβββαβ core (Table 1; Figure 3). The crystal structure of CdiA-CT^TA271 is unique as preceding the C-terminal nuclease domain is an N-terminal helical domain, which is likely a portion of the N-terminal cytoplasmic entry domain (CdiA-CT entry domain) (Figure 3A). PD-(D/E)XK family members frequently have core structure variability in β-strand length, angling of the α-helices, and insertions of additional secondary structure elements into the core (Knizewski et al., 2007; Steczkiewicz et al., 2012). Though the α-helices in all the characterized PD-(D/E)XK CdiA-CTs are angled similarly in relation to the central β-sheet, the final two β-strands for CdiA-CT^TA271 (and CdiA-CT^YPIII) are extended when compared to those in the Burkholderia toxin structures (Figure 3). Other differences include an extended, dramatically curved C-terminal core α-helix in CdiA-CT^E479, and an insertion in CdiA-CT^TA271 (and CdiA-CT^YPIII) where the final α-helix is preceded by a short α-helix that falls along the bent CdiA-CT^E479 α-helix trajectory. Despite reasonable structural homology between all the PD-(D/E)XK CdiA-CTs (Table 1), the CdiI^E479 and CdiI^1026b immunity proteins share no significant sequence or structural homology with each other or with CdiI^TA271 (and CdiI^YPIII). CdiI^TA271 and CdiI^YPIII, which again belong to the same toxin/immunity family, share ∼50% sequence identity and bear significant structural homology (1.0 Å rmsd) (Morse et al., 2015).

The active site for closely related CdiA-CT^TA271 and CdiA-CT^YPIII is fully conserved, containing Glu177, Asp198, Ser209 and Lys211 (TA271 numbering) (Morse et al., 2012; Morse et al., 2015). CdiA-CT^Bp1026b has a similar active site whereby Ser209 is an aspartate residue (Asp223) (Morse et al., 2012). The active site of CdiA-CT^E479 is unusual in that the catalytic lysine residue has been replaced with a histidine residue (His275) (Table 1) (Johnson et al., 2016b). Despite similarity between the PD-(D/E)XK CdiA-CT active sites, these toxins show two distinct catalytic activities. CdiA-CT^TA271 is a Zn²⁺-dependent DNase, completely degrading plasmid DNA in the presence of Zn²⁺ and shows very reduced DNase activity if the divalent metal is Mg²⁺ (Morse et al., 2012). CdiA-CT^YPIII is also a Zn²⁺-dependent DNase and is not activated by Mg²⁺ (Morse et al., 2015). In contrast, both CdiA-CT^Bp1026b and CdiA-CT^E479 have tRNase activity. CdiA-CT^Bp1026b specifically cleaves tRNA^Ala at the aminoacyl-acceptor stem (Nikolakakis et al., 2012), resulting in an accumulation of uncharged tRNA and disrupting translation (Morse et al., 2012). Conversely, CdiA-CT^E479 cleaves tRNA₂ ^Arg at the T-loop between a conserved thymidine (T55) and pseudouridine (ψ56) (Nikolakakis et al., 2012). CdiA-CT^1026b and CdiA-CT^E479 active sites overlay extremely well. Examining the size and shape of each active site revealed that CdiA-CT^E479 has a wider active site pocket compared to CdiA-CT^1026b (12.6 Å vs. 10.4 Å, respectively), which may play a significant role in substrate specificity (Johnson et al., 2016b).

The toxin/immunity protein interface for all PD-(D/E)XK CdiA-CT/CdiI complexes are extensive. The four toxin/immunity complexes have interfaces with ∼960–1,170 Å² (10–15%) toxin buried surface area and are marked by multiple salt bridges, 14–19 hydrogen bonds and hundreds of van der Waals interactions (Table 1). Notably, CdiA-CT^TA271 (and CdiA-CT^YpIII) has an additional β-hairpin inserted into the nuclease core (Figure 3A). For CdiA-CT/CdiI^TA271 (and CdiA-CT/CdiI^YpIII), the toxin is inactivated through a highly unusual mechanism of β-augmentation, wherein the toxin inserts this β-hairpin into the pocket of cognate immunity, producing a highly stable six-stranded antiparallel β-sheet where the toxin β-hairpin is sandwiched between two immunity β-hairpins (Morse et al., 2012; Morse et al., 2015). The formation of this β-sheet between the two proteins yields a highly specific and high affinity complex (Morse et al., 2012). Despite significant sequence identity and structural homology, CdiI^TA271 and CdiI^YpIII do not confer protection to the non-cognate toxin (Morse et al., 2015). Notably, both CdiA-CT^1026b and CdiA-CT^E479 lack this extended β-hairpin element and have a larger toxin/immunity interface (14–15% of the buried toxin surface area) than CdiA-CT/CdiI^TA271 and CdiA-CT/CdiI^YpIII (10–12% buried toxin surface area) (Table 1). For Bp1026b and BpE479, immunity protein specificity appears to stem from unique distributions of electrostatic charges at the protein-protein interfaces, preventing the interaction between non-cognate toxin/immunity pairs.

The immunity protein for CdiA-CT^TA271 (and CdiA-CT^YpIII) binds at an alternate location to the immunity proteins of CdiA-CT^1026b and CdiA-CT^E479 (Figure 3B). While CdiI^1026b and CdiI^E479 directly coordinate toxin active site residues, effectively rendering the protein inactive, no similar interaction is observed in the TA271 and YpIII toxin/immunity complexes. Instead, for this family of toxin/immunity proteins the toxin active site residues remain solvent exposed and perhaps available to bind substrate. It is currently unknown how CdiI^TA271 neutralizes its toxin. However, perhaps formation of the toxin/immunity complex could prevent or restrict toxin mobility upon binding DNA substrate, inactivating the toxin.

While all of the PD-(D/E)XK CdiA-CT/CdiI complex structures reveal heterodimeric oligomerization, experiments with CdiA-CT^E479 indicate the toxin adopts higher order oligomerization in the presence of tRNA. Size exclusion chromatography with CdiA-CT^1026b and CdiA-CT^E479 indicate 1:1 binding of CdiA-CT to tRNA, but while CdiA-CT^1026b forms a monomeric protein:tRNA complex, CdiA-CT^E479 forms tetrameric protein:tRNA complex (Johnson et al., 2016b). These results were supported experimentally by fitting tRNA-docked CdiA-CT^1026b and CdiA-CT^E479 models into small-angle X-ray scattering (SAXS) derived electron density envelopes of CdiA-CT^1026b and CdiA-CT^E479 in complex with tRNA substrate (Johnson et al., 2016b). Thus, while CdiA-CT^1026b interacts with a single tRNA molecule as a monomer, CdiA-CT^E479 tetramerizes to form a complex with four tRNA molecules (Johnson et al., 2016b).

As PD-(D/E)XK proteins are characterized by highly variable sequences and active-site plasticity, CDI systems are able to generate a diverse array of cytotoxic proteins with unique substrate specificity, nuclease activity and oligomerization. These characterized PD-(D/E)XK CdiA-CT toxins are likely representative of a number of uncharacterized CdiA-CT proteins that belong to this superfamily, however, alternative toxin activities and structures would not be unexpected.

The structure of the toxin/immunity complex from Yersinia kristensenii ATCC 33638 revealed that CdiA-CT is a member of the RNase A superfamily, which previously had only been observed in eukaryotes (Batot et al., 2017; Cuthbert et al., 2017). Though CdiA-CT^Ykris shares little sequence similarity with RNase A family members, CdiA-CT^Ykris adopts a kidney shape that is formed by two curved β-sheet domains, and strongly resembles several RNase A family members (Holloway et al., 2011; Thiyagarajan and Acharya, 2013; Lomax et al., 2014), Figure 4A. Like RNase A proteins, CdiA-CT^Ykris has metal-independent RNase activity. However, while RNase A proteins have a conserved His-Lys-His triad, the active site of CdiA-CT^Ykris comprises His175, Thr276, Tyr278, and Arg186, suggesting an alternate mechanism of ribonuclease action (Table 1). In CdiA-CT^Ykris, these residues are required for full RNase activity and cyclic cytidine monophosphate (cCMP) hydrolysis, another RNase A family activity (Raines, 1998). However, the structural homology of CdiA-CT^Ykris to the RNase A family, and its RNase and cCMP hydrolytic activities, support its classification as a novel bacterial member of the RNase A superfamily.

Unlike eukaryotic RNase A proteins, CdiA-CT^Ykris has no disulfide bonds (Batot et al., 2017). Strikingly, RNase A undergoes cooperative unfolding at ∼60°C while CdiA-CT^Ykris gradually unfolds between 40 and 80°C (Batot et al., 2017). Thus, its thermal stability and lack of disulfide bonds enables CdiA-CT^Ykris to be more tolerant to unfolding/refolding than eukaryotic RNase A proteins. As CdiA-CT is transported from the periplasm of the CDI⁺ bacteria and through the target-cell outer and inner membranes to reach the target-cell cytosol, these biophysical characteristics of CdiA-CT^Ykris may allow for successful unfolding and refolding of the toxin in its passage through these environments.

Like other CdiA-CT/CdiI complexes, CdiI^Ykris forms an extensive interface with CdiA-CT^Ykris (Table 1; Figure 4Aiii). Like BECR-type CdiA-CT/CdiI complexes, CdiI^Ykris inactivates its cognate toxin through direct interaction with active-site residues: His175, Thr276, and Tyr278. Eukaryotes express an RNase A inhibitor (RI) to prevent RNase cytotoxicity (Dickson et al., 2005). Notably, CdiI^Ykris has no sequence or structural homology to RI.

Until the characterization of CdiA-CT^Ykris (Batot et al., 2017), the only known RNase A family members were in vertebrates (Dyer and Rosenberg, 2006). Extending the RNase A superfamily to include CdiA-CT^Ykris, creates an RNase A subfamily with bacterial proteins/domains from both Gram-negative and Gram-positive bacteria, many of which play a role in bacterial virulence or competition and all of which have predicted immunity proteins and associated secretion systems (Batot et al., 2017).

While most CDI toxins possess active cytotoxic domains in the absence of a protein partner, some CDI toxins require one or more target-cell protein partners to activate the toxin. This phenomenon was first identified in the CDI system of uropathogenic E. coli 536 (UPEC536), which requires a target-cell metabolic protein to bind and activate its toxin. More recently, the discovery of several CDI systems that require both EF-Tu and EF-Ts for toxin activity suggests that toxin activation by endogenous target-cell proteins may be utilized by other CDI systems.

The requirement of a target-cell protein to activate delivered toxins might be more widespread than our current knowledge implies, but under sampled due to the difficulty of toxin protein partner detection and verification by in vitro methods. Further, interactions with endogenous cytosolic proteins may be another avenue through which CDI proteins have a significant role in contact-dependent signaling (CDS) (Danka et al., 2017; Ocasio and Cotter, 2019), where interactions between CdiA-CT/CdiI and specific cytosolic proteins in CDI⁺ cells could result in altered cell processes or gene regulation (Garcia, 2018).

The CdiA-CT^UPEC536 toxin is a potent tRNase that is activated upon binding to target-cell cysteine synthase A (CysK, O-acetylserine sulfhydrylase A), an enzyme involved in cysteine biosynthesis (Diner et al., 2012). CysK is extremely well conserved in Gram-negative bacteria, with ∼99% sequence identity between bacterial species (Burkhard et al., 1998; Rabeh and Cook, 2004; Claus et al., 2005). However, CysK is a non-essential enzyme as CysK-expressing cells also express an isoenzyme, CysM (Claus et al., 2005). In a target cell expressing cytosolic CysK, CdiA-CT^UPEC536 binds to CysK with high affinity to form an active tRNase complex. Interestingly, CysM does not bind to CdiA-CT^UPEC536, granting CysK-deficient cells immunity to CdiA-CT^UPEC536. The CysK/CdiA-CT^UPEC536 complex is inactivated upon forming a ternary complex with the cognate UPEC immunity protein (CysK/CdiA-CT^UPEC536/CdiI^UPEC536).

The structure of the CysK/CdiA-CT^UPEC536/CdiI^UPEC536 complex reveals an interesting mechanism of interaction between CysK and CdiA-CT^UPEC536. The UPEC toxin interacts with CysK by inserting its C-terminal tail into the CysK active site cleft (Figure 4B) (Johnson et al., 2016a). This interaction mimics the native interaction between CysK and a serine acetyltransferase (CysE), another cysteine biosynthetic enzyme. CdiA-CT^UPEC536 uses a conserved C-terminal Gly-X-Gly-Ile CysE motif to make similar contacts with CysK to those observed in the CysK/CysE complex (Francois et al., 2006; Zhao et al., 2006; Salsi et al., 2010; Diner et al., 2012; Wang and Leyh, 2012). The CysK-CdiA-CT^UPEC536 complex cleaves cytosolic tRNA at the anticodon loop, rendering it unusable and halting protein translation in the cell (Diner et al., 2012). Notably, CysK is a homodimer and thus the ternary complex with CdiA-CT/CdiI^UPEC536 forms a dimer of heterotrimers (Figure 4B).

The CdiA-CT^UPEC536 structure is completely α-helical with no structural homology to any known tRNase to-date. Mutational and biochemical analysis revealed an active site consisting of Asp155, Trp176, His178, and Glu181 (Figure 4B) (Johnson et al., 2016a). Notably, Trp176 is involved in interactions with CdiI^UPEC536, forming hydrophobic interactions with a hydrophobic patch on the immunity protein surface. The rest of the toxin/immunity interface is formed by a series of hydrogen bonds, including direct coordination of the toxin active site residues, His178 and Glu181.

To elucidate the mechanism of toxin activation by CysK, an Ntox28 homolog of CdiA-CT^UPEC536 was identified from the Gram-positive Ruminococcus lactaris, Tox28^Rlac. Tox28^Rlac is functionally homologous to CdiA-CT^UPEC536 but does not require CysK for tRNase activity. Extensive experimental evidence shows that Tox28^Rlac is significantly more thermostable than CdiA-CT^UPEC536 (Johnson et al., 2016a). Notably, upon CysK binding CdiA-CT^UPEC536 becomes significantly more stable with an improved ability to bind tRNA substrate. This data suggests that CysK may act as a chaperone to CdiA-CT^UPEC536. CdiA-CT toxins may be intrinsically unstable to allow the passage of partially unfolded toxin across bacterial membranes into the target-cell cytosol. Thus, cytosolic chaperones like CysK may be a necessary stabilizing scaffold for some CdiA-CT proteins to ensure a fully active toxin. The lack of disulfide bonds and increased thermal stability of the bacterial RNase A CdiA-CT^Ykris (CdiA-CT ^Ykris , a Bacterial RNase a Family Member) also supports this stability/folding hypothesis.

Following the discovery of the critical role of CysK in CdiA-CT^EC536 activity, there was speculation on whether other CDI systems employ target-cell proteins in CdiA-CT activity. Recently, it was observed that the translation elongation factors EF-Ts and EF-Tu are involved in the tRNase activity of several E. coli CDI toxins: CdiA-CT^EC869, CdiC-CT^96.154, CdiA-CT^NC101, and CdiA-CT^Kp342 (Jones et al., 2017; Michalska et al., 2017; Gucinski et al., 2019). CdiA-CT^NC101 and CdiA-CT^Kp342 are both BECR CdiA-CT proteins and are discussed above in BECR Family CdiA-CT Toxins and Their Complexes. EF-Tu delivers aminoacyl-tRNAs (aa-tRNAs) to the free site of the ribosome during protein synthesis, while EF-Ts acts as a guanine nucleotide exchange factor for EF-Tu, catalyzing the release of guanosine-5′-diphosphate (GDP) from EF-Tu. CDI toxins that rely on these elongation factors are GTP-dependent tRNases and perhaps only recognize tRNA in an EF-Tu bound context or rely on EF-Tu to position tRNA correctly for toxin cleavage.

The CdiA-CT toxin of enterohemorrhagic E. coli 869 (CdiA-CT^EC869) was the first CDI protein identified that requires EF-Ts and EF-Tu for tRNase activity in vivo (Jones et al., 2017). CdiA-CT^EC869 requires EF-Tu and GTP to specifically cleave the 3′-end of the acceptor stem of tRNA^Gln and tRNA^Asn, while EF-Ts further stimulates the tRNase activity of this complex (Jones et al., 2017). EF-Tu and EF-Ts are both necessary for CdiA-CT^NC101 activity, which cleaves the single-stranded 3′-tail of the tRNA acceptor stem. Interestingly, CdiA-CT^NC101 cleaves several tRNAs, where substrate specificity stems from a guanine discriminator nucleotide adjacent to the cleavage site. CdiA-CT^Kp342, as discussed earlier, cleaves uncharged tRNA_GAU ^Ile; this tRNase activity is greatly stimulated in the presence of both EF-Tu and EF-Ts (Gucinski et al., 2019). Notably, unlike CdiA-CT^EC869 and CdiA-CT^NC101, neither EF-Ts nor EF-Tu are required for CdiA-CT^Kp342 activity as the same reaction can be carried out in their absence, albeit at a considerably slower rate.

While other BECR fold CdiA-CT toxin/immunity complexes are heterodimeric, both CdiA-CT/CdiI^Kp342 and CdiA-CT/CdiI^NC101 possess a homodimeric CdiI where each subunit binds one toxin, resulting in the formation of a heterotetramer or heterohexamer, depending on the inclusion of EF-Tu in the complex (Figure 4C). CdiI^Kp342 is almost entirely composed of β-strands (Gucinski et al., 2019), and CdiI^NC101 is completely helical (Jones et al., 2017) (Figures 2B,C); thus, CdiI homodimerization is quite distinct between these complexes. The CdiA-CT/CdiI^NC101/EF-Tu complex is a dimer of heterotrimers, where CdiI forms a central dimer that interacts directly with EF-Tu and CdiA-CT (Jones et al., 2017), Figure 4C. The interaction between CdiI^NC101 and its cognate toxin is much like other BECR CdiA-CT/CdiI complexes described above (Table 1; BECR Family CdiA-CT Toxins and Their Complexes). EF-Tu interacts with the N-terminal face of CdiA-CT^NC101 through an extensive interface involving the formation of an anti-parallel 3-stranded β-sheet between the toxin and an EF-Tu β-hairpin alongside a vast network of interactions. Strikingly, a CdiA-CT^NC101 Tyr192Arg mutation prevents the inclusion of EF-Tu into a complex with CdiA-CT/CdiI, and results in an inactive toxin. Docking CdiA-CT^Kp342 onto EF-Tu at the CdiA-CT^NC101/EF-Tu interface results in extensive clashes between CdiA-CT^Kp342 and EF-Tu structural elements, suggesting that interactions between CdiA-CT^Kp342 and EF-Tu are unique compared to the CdiA-CT^NC101/EF-Tu complex (Gucinski et al., 2019).

Despite the crystal structure and biochemical analysis, there are still unanswered questions about how CdiA-CT^NC101 cleaves tRNA in complex with EF-Tu. For example, when the CdiA-CT^NC101/EF-Tu complex is superimposed onto a structure of the 5′-guanylyl imidodiphosphate (GDPNP, a non-hydrolysable GTP analog)/EF-Tu/aa-tRNA complex, the toxin active site is located more than 10 Å away from the tRNA cleavage site. Furthermore, the complexes do not superimpose perfectly as the superimposed toxin has intermolecular clashes with domain I of EF-Tu and bound aa-tRNA from the GDPNP/EF-Tu/aa-tRNA complex. Thus, CdiA-CT^NC101 binding must impose structural changes to the GTP/EF-Tu/tRNA complex, likely resulting in the correct tRNA placement and exposure for cleavage by CdiA-CT^NC101.

Both CdiA-CT/CdiI ^EC869 and CdiA-CT/CdiI^NC101 complexes copurify with EF-Tu to form high-affinity ternary complexes (Jones et al., 2017; Michalska et al., 2017). Notably, while CdiA-CT^96.154 copurifies with EF-Tu, the complex is extremely unstable, indicating interactions between EF-Tu and CdiA-CT^96.154 are likely weaker than those between EF-Tu and CdiA-CT^EC869 or CdiA-CT^NC101. The differences in affinity between CdiA-CTs and EF-Tu highlight the difficulty in identifying novel CdiA-CT effectors as many heretofore undetected interactions will not be observed under typical experimental conditions.