Microorganisms are in constant competition for growth niches and environmental resources. In Gram-negative bacteria, contact-dependent growth inhibition (CDI) systems link the fate of one cell with its immediate neighbor through touch-dependent, receptor-mediated toxin delivery. Though discovered for their ability to confer a competitive growth advantage, CDI systems also play significant roles in intersibling cooperation, promoting both auto-aggregation and biofilm formation. In this review, we detail the mechanisms of CDI toxin delivery and consider how toxin exchange between isogenic sibling cells could regulate gene expression.
Within the polymicrobial communities that predominate in nature, bacteria adapt their physiology in response to the myriad of metabolites and signaling molecules in the environment. This dynamic interplay reflects the continuous adjustment to fluctuating nutrient/chemical landscapes in the environment and within animal and plant hosts. For example, some enteric pathogens establish transient syntrophic relationships with members of the microbiota, using the sugars released by saccharolytic bacteria to fuel growth and colonization . Invaders also compete with resident bacteria whose metabolic requirements overlap with their own. For example, beneficial commensal bacteria appear to suppress pathogen colonization by competing for preferred carbohydrate sources [2,3]. As metabolic demands evolve, so does the demeanor of established microbial interactions. Such changes can be abrupt when antibiotics or host immune responses impinge on the community. Previously antagonistic associations may turn collegial or vice versa, sometimes leading to co-infections [4–6].
Bacterial contact-dependent growth inhibition (CDI) is a common mechanism used by some Gram-negative bacteria to either initiate hostilities or forge cooperative interactions. CDI was discovered in Escherichia coli isolate EC93, which rapidly kills laboratory strains of E. coli K-12 during co-culture [7,8]. E. coli EC93 uses its CdiB and CdiA two-partner secretion proteins to deliver an ionophoric toxin into other E. coli cells . CdiA is an extended filamentous protein that is exported to the cell surface by the outer-membrane-localized CdiB transporter. Upon binding its receptor on a neighboring cell, CdiA delivers its C-terminal toxin domain (CdiA-CT) into the target (Figure 1). E. coli EC93 cells protect themselves from toxins delivered from neighboring siblings with CdiI immunity proteins. Thus, CDI imparts a competitive growth advantage over strains that lack immunity. However, CDI systems also promote cooperation between sibling cells, facilitating social behaviors that underlie biofilm formation and pathogenesis. Here, we review the molecular mechanism of CDI-mediated competition and cooperation.
CDI-mediated toxin delivery.
Distribution and organization of cdi loci
Since its discovery in E. coli EC93, cdi loci have been identified in a wide variety of Gram-negative proteobacteria [10,11]. The systems are invariably encoded within genomic/pathogenicity islands or plasmids ; therefore, not all strains of a given species necessarily carry cdi genes. Furthermore, CdiA-CT and CdiI sequences are extraordinarily polymorphic between bacteria, such that individual strains often deploy different toxin–immunity pairs [10,11,13]. CDI systems are typically organized as cdiBAI gene clusters, though most loci in the Burkholderiales are arranged in an alternative cdiAIB order (Figure 2). In all instances, the toxic 3′-coding sequence of cdiA is closely linked to the downstream immunity gene, with the cdiI initiation codon often overlapping the cdiA termination codon. Consistent with their presence on mobile genetic elements, cdiA-CT/cdiI sequences appear to be shared between bacteria through horizontal gene transfer. Moreover, related toxin–immunity modules are often associated with bacteriocins and type VI and type VII secretion systems [10,11,14–19], indicating that the toxins are deployed by several competition systems. Further evidence of horizontal exchange is manifest by the tandem arrays of ‘orphaned’ cdiA-CT/cdiI gene pairs often found downstream from cdiBAI gene clusters (Figure 2) [15,17,20]. Insertion sequence elements, integrases and transposases are usually interspersed among the orphan toxin–immunity gene pairs. The function of orphan gene clusters remains largely unexplored. Because many orphan toxin–immunity pairs retain growth inhibition and immunity functions, they may expand the cell's toxic repertoire through recombination with the upstream cdiA gene [17,20]. Alternatively, orphan pairs may represent ancestral toxins that were displaced during the integration of new incoming cdiA-CT/cdiI sequences.
Variation in cdi gene clusters.
Though cdiB, cdiA and cdiI constitute the minimal core of a CDI system, some loci contain additional genes that are important for function. A subset of Burkholderia systems encodes a small lipoprotein (BcpO) between cdiI and cdiB (Figure 2). BcpO plays a significant role in CdiA secretion and toxin delivery for these systems . Other loci encode a putative lysyl acyltransferase related to the hemolysin activator HlyC (Figure 2) [22,23]. This latter association suggests that some CdiA effectors may be subject to post-translational modification with fatty acyl chains. We also note that Citrobacter rodentium ICC168 encodes an unusual CdiA annotated as a fimbrial adhesin (Figure 2). This protein contains the characteristic filamentous hemagglutinin (FHA) peptide repeats and pretoxin–VENN domain of CdiA, but lacks the N-terminal two-partner secretion (TPS) transport domain required for export by CdiB (Figure 3). Instead, the Citrobacter effector contains an N-terminal domain that is closely related to the StbD fimbrial protein from Salmonella species (Figure 3). Collectively, these observations suggest that the fimbrial chaperone and usher proteins encoded immediately upstream mediate secretion and presentation of this unique CdiA effector (Figure 2).
CdiA protein domain architecture.
Toxin delivery and activation
CdiA binds to specific cell surface receptors on target bacteria (Figure 1). The molecular basis of CdiA–receptor interactions is best understood for the CDI systems of E. coli strains EC93 and 536. The CdiAEC93 effector binds to BamA — a key subunit of the essential outer-membrane β-barrel assembly machine (BAM) complex [24,25]. BamA is found in all Gram-negative bacteria, and its sequence is highly conserved among the Enterobacteriaceae. However, E. coli EC93 cells are unable to inhibit the growth of closely related enterobacteria because the surface exposed loops of BamA vary considerably between species . Expression of E. coli bamA in Salmonella Typhimurium, Citrobacter freundii and Enterobacter aerogenes sensitizes these bacteria to CdiAEC93-mediated growth inhibition. Similarly, exchange of E. coli bamA with alleles from Enterobacter cloacae or S. Typhimurium protects cells from growth inhibition and abrogates CdiAEC93-dependent cell–cell adhesion [8,25]. CdiAEC536 recognizes heterotrimeric complexes of OmpF and OmpC . OmpC and OmpF are highly expressed outer-membrane osmoporins that are subject to intense selective pressure from bacteriophages and adaptive immune systems [27–29]. Consequently, the surface exposed residues of OmpC vary between E. coli isolates, and this antigenic variation protects many potential target strains from CdiAEC536 activity . Intriguingly, group A colicins appear to use OmpF as a conduit to translocate nuclease domains across the outer membrane [30,31]. It is presently unclear whether CdiAEC536 also uses OmpC/OmpF for toxin translocation. However, because CdiA proteins exploit a variety of cell-surface receptors, including lipopolysaccharide ( and unpublished data), it appears that porins are not an obligate part of the CDI toxin translocation pathway.
Recent work has localized the receptor-binding domain within the CdiA filament. Alignment of the closely related CdiAEC93 and CdiAEC536 proteins shows divergence in a central region between the FHA-1 and FHA-2 peptide repeats (Figure 3), suggesting that this region could be responsible for differential receptor-binding activity . This conclusion is somewhat unexpected, because prevailing models predict that the C-terminal toxin domain forms the distal end of the CdiA filament. However, truncated CdiAEC93 lacking ∼1000 C-terminal residues retains BamA-binding activity , demonstrating that the C-terminus is not required to bind target cells. Furthermore, CdiAEC93 fragments containing the central region bind directly to purified BamA, and receptor specificity is altered when this region is exchanged between CdiA proteins . These findings underscore the modular nature of CdiA, but also raise important questions about effector structure and topology on the inhibitor cell surface.
Toxin translocation into the cytoplasm
Most characterized CDI toxins are nucleases, which must be translocated into the target cell cytoplasm to degrade substrates. Crystallography and comparative sequence analyses indicate that the variable CdiA-CT region is often composed of two domains with distinct functions during CDI [22,34,35]. The extreme C-terminal domain is responsible for toxicity, whereas the N-terminal domain governs transport from the periplasm into the target cell cytoplasm (Figure 3). Genetic evidence suggests that the N-terminal domain interacts directly with membrane protein receptors to mediate this translocation (Figure 1) . Integral membrane proteins implicated in CDI include several ABC transporters (GltJ/GltK, RbsC and MetI), the d-glucose transporter PtsG, the ATP-dependent zinc metallopeptidase FtsH and a predicted inner-membrane protein of unknown function (YciB) . How these disparate membrane proteins mediate toxin translocation is not clear. One possibility is that the N-terminal domains use membrane proteins as receptors to bring the tethered nucleases into close proximity with the lipid bilayer. Because toxin transport requires the proton-motive force , this electrochemical gradient could energize translocation. Additionally, the proton gradient lowers the periplasmic pH, which could induce toxin unfolding for insertion directly into the lipid bilayer. This mechanism is similar to that proposed for the import of colicin nuclease domains, which adopt a molten globule-like state upon interaction with membrane lipids [38–40]. This destabilization of tertiary structure suggests that the toxins must unfold during transport across the cytoplasmic membrane.
In some instances, CDI toxins are inherently inactive, with activation occurring upon entry into target cells. The CdiA-CTEC536 toxin deployed by uropathogenic E. coli 536 is a latent anticodon nuclease that is only active when bound to the cysteine biosynthetic enzyme CysK . Purified CdiA-CTEC536 does not cleave tRNA in vitro unless the reaction is supplemented with CysK, and ΔcysK mutants are completely resistant to CdiA-CTEC536 intoxication [41,42]. Intriguingly, CdiA-CTEC536 inserts its C-terminal GYGI peptide into the active-site cleft of CysK, mimicking the highly conserved ‘cysteine synthase’ complex, in which CysE uses its C-terminal GDGI motif to bind the CysK active site [43,44]. CysK significantly increases toxin thermostability and also promotes binding to tRNA substrates [44,45]. Because CDI toxins must presumably unfold to enter target cells, the interaction with CysK may have evolved to compensate for the toxin's intrinsic instability. In this model, the binding partner ensures that the toxin refolds efficiently after delivery into the cytoplasm. More recently, we have discovered that diverse CdiA-CT toxins from E. coli strains EC869, NC101 and 96.154 interact functionally with translation elongation factors EF-Tu and EF-Ts . CdiA-CTEC869 binds to EF-Tu with high affinity and specifically cleaves within the aminoacyl acceptor stem of tRNAGln and tRNAAsn molecules. Remarkably, the toxin only cleaves substrate in the context of tRNA·EF-Tu·GTP ternary complexes , suggesting that the interaction with EF-Tu is required for nuclease activity. Moreover, EF-Ts is required for toxin activity in vivo, though guanine nucleotide exchange activity per se is not required for the nuclease reaction. Instead, it appear that EF-Ts promotes the formation of tRNA·EF-Tu·GTP complexes for rapid cleavage by CdiA-CTEC869. Finally, the trimeric integral membrane protein AcrB may play a role in the activation of membrane-pore toxins. E. coli ΔacrB mutants are resistant to the CdiA-CTEC93 pore-forming toxin during CDI , but also grow normally when the toxin is produced intracellularly (unpublished data). Thus, AcrB is required for intoxication regardless of whether the toxin is delivered from a neighboring cell or expressed internally. These observations raise the possibility that AcrB anchors CdiA-CTEC93 in the membrane or perhaps activates the toxin to allow ion flow.
The yin and yang of bacterial competition and cooperation
Cell–cell adhesion and cooperative behavior
The fitness advantage of E. coli EC93 over laboratory strains led to the competition model of CDI function. However, the adhesive properties of CdiA also promote collaboration between sibling cells (Figure 4). Expression of cdi is associated with auto-aggregation and biofilm formation in several species including Erwinia chysanthemii , Xylella fastidiosa , Neisseria meningitidis , Xanthomonas axonopodis , B. thailandensis , E. coli  and Pseudomonas aeruginosa . Biofilm formation is a prime example of cooperative behavior, in which individual bacteria collaborate to build multicellular structures. Such communities are beneficial because they protect inhabitants from predation and antimicrobial compounds. Moreover, this collective behavior often contributes to virulence, perhaps accounting for the large number of bacterial pathogens that carry CDI systems. For example, HecA/CdiA of Erwinia (Dickeya) chrysanthemii promotes adherence to leaf epidermal cells and allows planktonic bacteria to join pre-established cell aggregates (Figure 4) . Similarly, CdiA homologs from N. meningitidis (HrpA) , Moraxella catarrhalis (MhaB)  and X. fastidiosa (HxfA) [54,55] are proposed to mediate initial adhesion to host cells. For X. fastidiosa, HxfA/CdiA contributes to the colonization of insect vectors, thereby promoting subsequent transmission into plant hosts. In some instances, CDI may play a more active role in pathogenesis. N. meningitidis hrpA/cdiA mutants are less fit inside human cells and are also defective for endosomal escape into the cytoplasm .
Cellular interactions mediated by CDI.
The role of CDI in cooperative biofilm formation has been examined most extensively in B. thailandensis E264. The B. thailandensis cdi locus is critical for wild-type biofilm formation, and the number and size of pillar structures are altered in cdiA mutants (Figure 4) . Furthermore, B. thailandensis cells that constitutively express cdi genes produce abnormally dense and flat biofilms that lack discrete pillar structures. This latter finding demonstrates that cdi expression must be precisely regulated for cooperation. Transcription of the B. thailandensis cdi locus is induced by quorum-sensing homoserine lactones , and there is evidence of complex spatiotemporal regulation . Thus, cdi genes are preferentially expressed at high cell densities when cell–cell contacts are maximized. Of course, CDI-mediated cooperation is only possible between cells that express the appropriate cognate CdiI immunity protein. Non-isogenic bacteria are excluded from the biofilm community through toxin delivery . However, when non-isogenic cells are provided with the appropriate immunity gene, the unrelated populations are able to congregate and form mixed pillars. These studies demonstrate that self/non-self-discrimination in bacteria can be controlled by specific toxin–immunity protein-binding interactions.
Contact-dependent cell–cell signaling
Remarkably, CdiA-mediated cell–cell adhesion is not sufficient to promote full biofilm formation in B. thailandensis, and the exchange of catalytically active toxin domains is also required . This observation suggests that delivered toxin domains have an unanticipated signaling function when delivered into immune sibling cells. Recent work from Cotter and colleagues has revealed that toxin delivery is associated with changes in gene expression, including the up-regulation of a putative transcriptional regulator involved in biofilm formation . The molecular pathway linking toxin delivery to transcriptional regulation is not yet known, but there are several possible mechanisms. CdiA-CTE264 has Mn2+-dependent DNase activity [21,61], and therefore its intrinsic DNA-binding activity could contribute to signaling. Consistent with this model, we have found that the neutralized CdiA-CT·CdiIE264 complex binds to DNA (unpublished data), suggesting that the toxin–immunity protein complex could act as a transcription factor to regulate gene expression (Figure 4). Alternatively, the complex may exhibit residual (and sub-lethal) ‘nickase’ activity, thereby modulating DNA supercoiling and indirectly influencing transcription. Whether contact-dependent cell signaling extends to other types of toxins remains an open question. We note that many CDI toxins cleave tRNA [34,61,62], and that tRNA fragments regulate protein synthesis and contribute to stress responses in vertebrates . Angiogenin represents an intriguing example of RNase-mediated response regulation. Angiogenin is a potent angiogenesis factor and a paralog of RNase A/RNase 1. Circulating angiogenin is endocytosed and trafficked to the nucleolus, where it cleaves a promoter-associated RNA to de-repress transcription of ribosomal RNA [64–66]. The resulting increase in ribosome production is critical to support cell growth and proliferation. In addition, cytoplasmic angiogenin is activated to cleave the anticodons of mature tRNA during stress [67,68]. The resulting tRNA fragments inhibit protein synthesis, promote stress granule assembly and enhance cell survival [69,70]. Though these regulatory mechanisms have only been described in vertebrates, tRNA fragments accumulate in Streptomyces coelicolor during aerial hyphae formation and diminish later in development as spores are produced . These latter findings suggest that tRNA fragments have regulatory activities in prokaryotes and raise the possibility that CDI delivered RNase toxins could also participate in signaling. Determining how CDI toxin exchange regulates gene expression represents an outstanding and exciting challenge for the field.
CDI systems contribute to interstrain competition and self/non-self-recognition.
CDI promotes cell–cell adhesion, which facilitates cooperation between isogenic sibling cells.
CDI-mediated toxin exchange appears to induce changes in sibling cell gene expression.
Research in the Low and Hayes laboratories is supported by grants from the National Science Foundation [MCB1545720] and the National Institutes of Health [GM117373 and GM117930]. A.M.J. is supported in part by the Ralph M. Parsons Fellowship.
The Authors declare that there are no competing interests associated with the manuscript.