The Enterobacterial Genotoxins: Cytolethal Distending Toxin and Colibactin

Abstract

While the DNA damage induced by ionizing radiation and by many chemical compounds and drugs is well characterized, the genotoxic insults inflicted by bacteria are only scarcely documented. However, accumulating evidence indicates that we are exposed to bacterial genotoxins. The prototypes of such bacterial genotoxins are the Cytolethal Distending Toxins (CDTs) produced by Escherichia coli and Salmonella enterica serovar Typhi. CDTs display the DNase structure fold and activity, and induce DNA strand breaks in the intoxicated host cell nuclei. E. coli and certain other Enterobacteriaceae species synthesize another genotoxin, colibactin. Colibactin is a secondary metabolite, a hybrid polyketide/nonribosomal peptide compound synthesized by a complex biosynthetic machinery. In this review, we summarize the current knowledge on CDT and colibactin produced by E. coli and/or Salmonella Typhi. We describe their prevalence, genetic determinants, modes of action, and impact in infectious diseases or gut colonization, and discuss the possible involvement of these genotoxigenic bacteria in cancer.

INTRODUCTION

Toxins are important virulence factors that contribute to the ability of the producing bacterial strains to achieve their ultimate goal: colonizing tissues and staying there long enough to successfully multiply. In recent years, it has been demonstrated that some of these toxins are genotoxic for the mammalian cells. In addition, the role of the intestinal microbiota in colon cancer has been scrutinized, and genotoxic bacteria are now suspected to contribute to the carcinogenic process (for review see references 1 and 2). The possibility that bacteria, in proximity with intestinal stem cells as soon as a few hours after birth, produce genotoxins, leads to questions regarding the impact of this genotoxicity not only in intestinal carcinogenesis but also, more generally, in long-term digestive health.

KEEPING THE INTEGRITY OF THE GENOME OF THE HOST CELLS: THE DNA DAMAGE RESPONSE

Pathogenic bacteria have evolved strategies to subvert host-signaling pathways for their own benefit. Among eukaryotic processes hijacked by virulence factors such as the actin cytoskeleton dynamics (modulated by RhoGTPase-modifying enzymes) (3), inflammatory pathways (hijacked by immunosuppressive type 3 effectors) (4) or cell cycle regulation (modulated by cyclomodulins) (5, 6), host DNA has now emerged as a target for genomic insults inflicted by bacterial genotoxins. Preservation of genome integrity is a critical process in all living organisms and plays an essential role during growth and cellular renewal. However, this integrity is constantly challenged by intracellular stress generated by cell metabolism and during S phase that generates mistakes during replication. External genotoxic agents such as ionizing radiations, exposure to environmental mutagens, chemotherapeutic molecules, or even biological process such as inflammation or virus infection, also contribute to genomic insults. To deal with this DNA injury, cells have evolved sophisticated mechanisms to repair DNA damage such as DNA breaks, DNA adducts and cross-links, and to resolve replication fork-dependent DNA breaks during replication and transcription (7). Persistent DNA damage is detected at specific cell-cycle checkpoints (during S phase and at the G1/S and G2/M transitions) and triggers the DNA damage response (DDR). This response stops the cell cycle, allowing DNA-repairing machinery to proceed and correct the genomic injury, or induces senescence or apoptosis in case of massive or irreversible damage (8). However, most unrepaired or misrepaired DNA damage results eventually in DNA double-strand breaks (DSBs), the most dangerous type of DNA damage that, when taken over by the unreliable nonhomologous end-joining repairing process, favors DNA mutation and genomic instability leading to cellular transformation and tumorigenesis (7, 9).

CYTOLETHAL DISTENDING TOXINS

CDT Homologs, Diversity, and Distribution

The Cytolethal Distending Toxins (CDTs, formerly CLDT) were first identified in 1988 by Johnson and Lior in culture filtrates of E. coli strains isolated from patients with diarrhea (10, 11). CDT intoxication was characterized by progressive cell distension and cytotoxicity in cultured eukaryotic cells (Fig. 1). Host response to CDT intoxication differs depending on the target cells. Apoptosis was reported in T-cell and B-cell lines, while other cells preferentially arrest in the G1 and/or G2 phases of the host cell cycle (12, 13, 14, 15, 16, 17). Since their initial identification, CDTs and associated activity have been found in numerous Gram-negative and enteropathogenic species including Campylobacter spp. (10, 18), Aggregatibacter actinomycetemcomitans (19), Haemophilus ducreyi (20), Helicobacter spp. (21), Providencia alcalifaciens (22), Shigella dysenteriae (23, 24), Shigella boydii, Escherichia albertii (25), Salmonella enterica serovar Typhi (S. Typhi) (26), and some nontyphoidal Salmonella (NTS) serovars (27). Such functional conservation among distant pathogenic strains is remarkable and supports an evolutionary advantage for CDT-expressing bacteria as well as a role for this toxin in virulence.

Figure 1.

Cell-distending effects of Cytolethal Distending Toxin and colibactin. HeLa cells were treated with purified E. coli CDT (CDTwt) or with CDT mutated on CdtB-conserved histidine (H153A) (CDTm) or infected with E. coli producing or not colibactin (E. coli pks+ and pks–). Cells were stained with methylene blue 3 days after the treatment. Bar represents 100 µm.

Except for Salmonella, CDT is produced as a tripartite holotoxin encoded by the cdt operon that contains the 3 contiguous cdtA, cdtB, and cdtC genes. Among E. coli strains, based on amino acid sequence, CDTs are classified in 5 different subtypes (CDT type I to V). However, no clear correlation of a specific CDT subtype with a particular E. coli pathotype could be established. Initially, CDT types I and II were isolated in 1994 in enteropathogenic E. coli (EPEC) strains (28, 29) but, for example, CDT subtype I has also been found in an extraintestinal pathogenic E. coli (ExPEC) strain isolated from a child with neonatal meningitis (30, 31). The CDT type III operon is found in a pVir plasmid associated with cytotoxic necrotizing factor (CNF)-2 in strains isolated from cattle and sheep with diarrhea or septicemia (23). CDT type IV genes were identified in 2003 in various ExPEC strains isolated from human urinary tract infections, strains isolated from porcine and poultry septicemia, as well as strains from human and porcine diarrheal diseases (32, 33). CDT type V is the most recently identified subtype, isolated from Shiga toxin-producing E. coli (STEC) and enterohemorrhagic E. coli (EHEC) strains, but also in strains of uncertain pathotype (34, 35). The most conserved CDT gene is cdtB with an overall protein sequence identity ranging from 46 to 100% (Fig. 2A). Phylogenetic amino acid sequence analysis indicates that CdtB could be clustered in two main groups (with at least 90% identity in each group), one containing the E. coli CdtB from type I and IV operons and CdtB from S. dysenteriae and the second group containing E. coli types II, III, and V, as well as CdtB from S. boydii, E. albertii, and P. alcalifaciens (Fig. 2B). A 57% identity rate was observed between these two groups. S. Typhi CdtB does not belong to these 2 groups, sharing only 50% global sequence identity with them (Fig. 2A and B). Sequence comparison of CdtA and CdtC subunits displays a similar group distribution to that observed with CdtB, although sequence identity is usually lower, starting from 33% for both subunits (Fig. 2C), than for CdtB (53%). This difference likely reflects a selective pressure higher on CdtB, which represents the active subunit, than on CdtA and C subunits. While phylogenetic divergence between CDT families points to an ancient dissemination of this operon among bacterial strains, sequence similarity in each group suggests that cdt genes have been transmitted more recently by horizontal gene transfer, possibly in E. coli bacterial reservoirs such as healthy cattle or swine farms (36). As with many virulence factor genes or pathogenicity islands, cdt operons are located on, or near, putative mobile genetic elements or on plasmids. The cdt genes of P. alcalifaciens are adjacent to a putative transposase gene (22). In E. coli, CDT-I produced by an enteropathogenic E. coli (EPEC) strain is encoded by an inducible and infectious lambdoid prophage (37). The CDT-IV operon, which is mainly found in ExPEC and avian pathogenic E. coli (APEC) strains (38), is framed by lambdoid prophage in a porcine ExPEC O75 strain, as well as for the CDT-I operon of another human ExPEC O18:K1:H7 strain with similar prophage genetic organization, suggesting a putative common ancestor (33), an assumption supported by sequence similarity of CdtB (Fig. 2A). CDT-V from an EHEC O157 strain was reported to be flanked by bacteriophage P2 type sequences (39, 40), while CDT-III operon is located on the conjugative pVir plasmid from a CNF2-producing E. coli (23). CDT-V-encoding bacteriophages were induced spontaneously from wild-type Shiga toxin-producing E. coli O157:H7 (41) and exhibited a higher persistence than CDT-producing E. coli. Moreover, CDT phages were found in 73% of river samples in Spain and might represent an important genetic vehicle and reservoir for CDT dissemination (42).

Figure 2.

Comparison of CDT subunits. (A) Percent identity matrix of CdtB subunits of E. coli (type I to V), S. dysenteriae, S. boydii, P. alcalifaciens, E. albertii, and Salmonella Typhi; identity score >89% are shown in bold. (B) Phylogenetic tree of CdtB subunits, with real branch length. Brackets indicate an identity score >89%. (C) Percent identity matrix of CdtA and CdtC subunits; identity score >84% are shown in bold.

With the notable exception of certain type V CDT-positive atypical O157 E. coli strains in which CDT is the only toxin identified so far (35), as for many pathogens, CDT-producing E. coli harbor other virulence factors and toxins such as Shiga-like toxin, colibactin, or CNF (36, 43–46). For example, in ExPEC strain IHE3034, the cdt genes are associated with several genomic islands including the pks pathogenicity island producing colibactin, type VI secretion system, enterohemolysin, and siderophore genes (47). This association of virulence genes is common and reflects the notion that pathogens use a repertoire of virulence factors to hijack host functions for the bacterial benefit (48).

Function of CdtA and CdtC

CDT is a two-component type A-B2 toxin, the “A” (active) subunit corresponding to CdtB, while the “B” (binding) moiety, responsible for receptor-mediated transport of the “A” subunit into the host cells, relies on both CdtA and CdtC. The three Cdt subunits contain a signal peptide that is cleaved during their translocation in the periplasm where toxin assembly occurs before further secretion (28, 29, 49). While numerous studies have explored the mode of action of CdtB, the role of A and C CDT subunits is less understood (50). Despite a low level of identity, both units exhibit 40% structural similarity to each other. Crystallographic data of the CDT from H. ducreyi (HducCDT) showed that both A and C subunits contain lectin-type structures (51) that appear to be conserved in all CDT families using structural predictive models (52). E. coli CdtA and CdtC subunits are carbohydrate-binding proteins, homologous to the B-chain of the ricin toxin, a lectin protein that is implicated in receptor recognition and internalization. Preincubation of cells with glycoproteins blocks CdtA and CdtC binding, and removal of N-linked sugars prevents intoxication by E. coli type 2 CDT (53). An aromatic cluster in CdtA subunit, as well as the formation of a deep groove by both CdtA and CdtC subunits (potentially a peptide-binding cleft implicated in membrane receptor recognition), are also shared structures. Another interesting feature common in all CdtC subunits is the presence of an N-terminal extension that could serve to occlude the active site of CdtB, suggesting an autoinhibition conformation (51, 52). Finally, sequence comparison and comparative modeling reveal that the β6-β7 sheets of CdtC are remarkably well conserved among CDT families and might play a crucial role at the interface of both CdtA and CdtB subunits and in complex assembly (52, 54). While both CdtA and CdtC are required for host receptor binding and CdtB entry into target cells (CdtB from S. Typhi being the exception), their respective function is less well defined. In A. actinomycetemcomitans, using fluorescent tagging of Cdt subunits, CdtA was shown to remain on the membrane surface of CHO cells in a cholesterol-independent manner, while CdtC localized mainly in the cytosol, suggesting a possible role in facilitating CdtB internalization and trafficking (55) (Fig. 3). However, cholesterol depletion was shown to reduce the Campylobacter jejuni, E. coli, H. ducreyi, and A. actinomycetemcomitans (Cjej, Ecol, Hduc, and Aact respectively) CDT toxicity in vitro (56, 57, 58), a result consistent with the identification of a putative cholesterol recognition/interaction amino acid consensus (CRAC) in both CdtB and CdtC required for toxin internalization (56, 59). Moreover, a recent study conducted in an in vivo mouse model using cholesterol-rich diet indicates that cholesterol is crucial for CjejCDT-induced inflammation (60). Nonetheless, while the general consensus, supported by experimental data (61, 62, 63, 64) recognizes that both CdtA and CdtC are required for CdtB intoxication, several studies report that CdtAB or CdtBC heterodimers are able to intoxicate host cells, although usually, less efficiently than the tripartite holotoxin (65, 66, 67, 68, 69). These results are supported by evaluation of γH2AX signal-reporting DNA damage with CdtAB in CHO cells (69). However, binding free-energy calculations show that the heterotrimeric structure is more stable than heterodimers (70). These conflicting results are probably a consequence, at least in part, of the source of CDT from different pathogens as well as the recipient cells or doses of toxin used in these assays. However, the fact that all identified cdt gene clusters so far exclusively contain the three subunits does not support a relevant role for CDT dimers in vivo. It should be noted that CDT could be also delivered to the host cells via outer membrane vesicles (OMVs). OMVs result from bacterial membrane budding and were shown to contain periplasmic molecules including CDT in ExPEC strain IHE3034 (31, 71). Vesicle-mediated CDT release was also observed in C. jejuni and in A. actinomycetemcomitans strains (72, 73). In A. actinomycetemcomitans, OMVs were shown to be internalized via fusion with lipid rafts of the plasma membrane of the host cell (73). Together, these results are consistent with the proposed role of OMVs as vehicle for toxin delivery to the host cell (74) (Fig. 3). That OMVs represent the main mode of CDT delivery in vivo is an attractive hypothesis but still requires further support.

Figure 3.

Cytolethal Distending Toxin pathway from the bacteria to the host nucleus. In E. coli, E. albertii, S. dysenteriae, S. boydii, and P. alcalifaciens, the CDT operon contains 3 genes that synthetize the 3 subunits CdtA (red), CdtB (green), and CdtC (blue) that are translocated and assembled in the periplasm. Following secretion, CDT binds to a nonidentified receptor on the host eukaryotic cell. Following internalization, CdtB is retrotransported through the Golgi apparatus. Through an ERAD-like pathway, CdtB reaches the nucleus and host DNA. Alternatively, CDT is trapped in OMVs that can deliver the toxin into the host cell by fusion with the plasma membrane or through binding to a receptor. S. Typhi is an intracellular pathogen remaining in the cytosol in the Salmonella-containing vacuole (SCV). CdtB is encoded on a CdtB islet (or SPI-11) together with pltA (fuchsia) and pltB (orange). These gene products form the typhoid toxin that contains 2 “active” subunits (CdtB and PltA) and a pentameric “binding” subunit (PltB). The toxin might be entrapped in OMVs and secreted through vacuoles that exit from the SCV and intoxicates cells through autocrine and paracrine pathways. ERAD, endoplasmic reticulum-associated degradation; OMV, outer membrane vesicle; SPI, Salmonella pathogenicity island; plt, pertussis-like toxin).

CdtB Trafficking

Following CDT binding to an, as yet, uncharacterized receptor(s), the CdtB subunit is internalized into the host cells through the endocytic pathway. A haploid genetic screen has revealed a requirement of various membrane proteins for EcolCDT intoxication such as the sphingomyelin synthase 1 (required also for Hduc, Cjej, and Aact CDT intoxication) and more specifically the G-protein-coupled receptor homolog TMEM181, Golgi glycoprotein 1, and the vacuolar ATPase subunit 2 (75). Divergence in the protein requirement, such as TMEM181, Derlin-2, or in the role of endosome acidification as well as variable sensitivity to selective inhibitors between different toxin producers suggest that CDTs use different internalization pathways (76, 77). However, it appears that CdtB is conveyed to the nucleus by a retrograde transport through the Golgi apparatus since brefeldin A, an inhibitor of the transport of proteins from the endoplasmic reticulum (ER) to the Golgi, was found to block CDT-induced cell cycle arrest (76) (Fig. 3). Mutation of the two nuclear localization sequences in EcCdtB-II impairs nuclear localization and CDT-induced cell cycle arrest (78). Conflicting results were published about further exit of CdtB by the endoplasmic reticulum-associated degradation (ERAD) pathway (55, 58, 79, 80). Translocation of the toxin across the ER membrane appears to depend on an unusual ERAD pathway but also on the type of CDT, supporting the fact that CDTs from different pathogens use different intracellular pathways to intoxicate the target cell.

CDT, a Genotoxic Cyclomodulin

CdtB, which is the most conserved subunit of the holotoxin, shows similarity to the deoxyribonuclease I (DNase I) family proteins and, in common with these enzymes, possesses two conserved histidine residues absolutely required for CDT activity. Even though DNase I and E. coli CdtB sequences display only ∼19% identity, determination of the crystal structure of CDT has revealed a near-perfect superposition of the catalytic core of both CdtB and DNase I (51) (Fig. 4). These structural data correlate with evidence of a DNase activity in vitro (51, 62, 81, 82) and CDT-dependent induction of DNA strand breaks in cellulo (58, 61, 83, 84). Although CDT induces an ataxia telangiectasia mutated (ATM)-dependent phosphorylation of histone H2AX, indicative of DNA DSB, a low dose of E. coli CDT (50 pg/ml) was shown to induce DNA single-strand breaks that were further converted to DSB during S phase (85). CDT-dependent DSBs in the host genome activate the DDR causing cell cycle arrest, mobilization of DNA repair complexes, and eventually apoptosis when damage overwhelms the host cell repairing capacity (Fig. 5) (14, 83, 86, 87, 88, 89). Since CDT-dependent DDR activation leads to a cell cycle arrest, CDT can be classified in the family of cyclomodulins, which are bacterial toxin and effectors that deregulate (either accelerate, slow, or stop) the host cell cycle (5, 6). While the DDR elicited by CDT and ionizing radiation (IR) partially overlaps, in contrast to IR that induces a fast and transient phosphorylation of H2AX in fibroblasts, CDT lesions induce delayed and persistent γH2AX foci (88). Studies in the cell-signaling cascade triggered by CDT showed a strong induction of ATM-CHK2 and ATM- and Rad3-related (ATR)-CHK1 axis with H2AX phosphorylation and activation of the canonical DDR implicating 14-3-3 phosphorylation leading to CDC25 phosphatase cytoplasmic sequestration. This cascade results in maintaining the kinase of the M-phase promoting factor, CDK1, in an inactive phosphorylated form leading to cell accumulation in the G2 phase 24 h after genotoxin treatment (Fig. 5) (83, 87, 88, 90). Using H. ducreyi CDT, a myc-dependent activation of ATM (91) and a RhoA-specific Guanine Nucleotide Exchange Factor, Net1- and FEN1-dependent RhoA/MAP kinase pathway activation leading to cell survival was observed (90, 92, 93). Moreover, cells that survive the acute phase of CDT intoxication develop a premature senescent phenotype (94). Following chronic exposure to sublethal doses of CDT, cells fail to activate the DDR and accumulate chromosomal abnormalities leading to a malignant phenotype (95).

Figure 4.

Structural conservation of the catalytic cores of E. coli and S. Typhi CdtB and DNase I. E. coli CdtB (Protein Data Bank [PDB] 2F1N, light blue), S. Typhi CdtB (PDB 4K6L, purple) aligned with DNase I (light grey) complexed with target DNA (PDB 1DNK, green). The main catalytic residues D211, H136, and H243 are highlighted in red.

Figure 5.

Genotoxin-induced DSBs induce the DNA damage response. Genotoxin-induced DNA double-strand breaks activate ATM/ATR, resulting in phosphorylation of H2AX (γH2AX in magenta), checkpoint protein kinase (CHK) recruitment, activation of DNA repair machineries, and ultimately apoptosis and senescence. Cell cycle arrest results from nuclear exclusion of the phosphatase CDC25 and accumulation of CDK inhibitor p21. Cytoplasmic CDC25 cannot relieve the cyclin-dependent kinase (CDK) from its inhibitory phosphorylation, and CDK-cyclin complexes are further inhibited by p21. Since activation of CDK-cyclin is required for cell cycle progression, activation of the DDR results in cell cycle arrest. In the case of incomplete DNA repair following a chronic or low level of genotoxin exposure, cells accumulate mutations and develop genomic instability.

Although DNase is the canonical enzymatic activity of CdtB, Shenker and colleagues observed that A. actinomycetemcomitans CdtB subunit exhibits phosphatidylinositol-3,4,5-triphosphate (PIP3) phosphatase activity and generates PIP2 production correlating with G2 cell cycle arrest, lymphocyte toxicity, and induction of proinflammatory cytokine secretion via a PI-3K/GSK3β-signaling pathway in macrophages (96, 97). A mutant that retains phosphatase activity but lacks DNase activity is still able to induce G2 arrest in the absence of γH2AX signaling (98). Whether PIP3 phosphatase activity is relevant or not in CDT-producing Enterobacteriaceae requires further studies.

Salmonella Typhoid Toxin

Salmonella enterica serovar Typhi is responsible for typhoid fever, a human-restricted life-threatening disease resulting in 222,000 annual deaths worldwide according to the most recent estimates published in 2014 (WHO). Genome comparison of various Salmonella serovars has revealed that a unique combination of virulence factors, operons, and salmonella pathogenicity islands (SPI) specific to the Typhi serovar may contribute to its unusual virulence (27). Among these genes, cdtB, which is located in the CdtB islet framed by two insertion sequences (also called SPI-11) together with pertussis-like toxin (plt)A and pltB genes (99), encodes a homolog of the active subunit of CDT (26). CdtB associates with PltA and PltB to form the typhoid toxin (26, 100). Functional and 100% identical CdtB were also identified in Salmonella enterica serovar Javiana responsible for food-borne infections (101) and in a subpopulation of S. enterica subsp. enterica (27, 102). This strict conservation suggests a common recent ancestor for this islet in the Salmonella population. Even if Salmonella CdtB shares only 50% identity with the other active subunits, at the structural level, its catalytic residues superimpose almost perfectly with those of CdtB from E. coli and DNase I (Fig. 4). While both CDT and typhoid toxin contain the CdtB subunit, these toxins differs dramatically in both the composition and mode of delivery into the host cell since S. Typhi does not encode CdtA and CdtC subunits (26) (Fig. 3). In this invasive strain, the typhoid toxin shows an unprecedented organization of an A2-B5 type toxin with two different A (active) subunits: PltA that possesses an ADP ribosyltransferase activity and CdtB that exhibits DNase activity. The B (binding) subunit PltB exhibits a typical pentameric organization and interacts with specific carbohydrate moieties on cell surface glycoproteins, allowing the toxin to intoxicate a broad range of epithelial and immune cells. The crystal structure of this chimeric A2-B5 toxin reveals a pyramidal-shaped complex with the CdtB subunit culminating at the summit of this toxin, linked to PltA by a disulfide bond essential for the integrity and activity of the holotoxin (103). This organization is a remarkable example of coevolution in that both PltA and CdtB possess a unique Cys residue lacking in their respective homologs, allowing the covalent linkage of the two enzymatic activities. Upon internalization of S. Typhi, the toxin is secreted in the lumen of the Salmonella-containing vacuole and then packaged in small vacuoles that exit the infected cells to release the typhoid toxin in the extracellular medium and intoxicate the cells in autocrine and paracrine manners (Fig. 3) (100, 104).

Interestingly, Guidi et al showed that S. Typhimurium-infected epithelial cells could release, via an anterograde transport, extracellular OMV-like vesicles containing typhoid toxin that could intoxicate bystander cells via a dynamin-dependent endocytosis pathway and retrograde transport through the Golgi complex (105).

CDT, a Virulence Factor

The fact that CDT is expressed primarily by pathogens argues for a role of this toxin in virulence. Indeed, when administered intravenously in mice, the typhoid toxin recapitulates most of the acute symptoms of typhoid fever. Disruption of the DNase activity carried by the CdtB subunit of the holotoxin abolishes the symptoms as well as animal death and prevents the sharp decrease in white blood cell count, thus demonstrating the role of the CdtB and its associated DNase activity in bacterial virulence of S. Typhi (103). CDT was also shown to be required for persistence of Helicobacter hepaticus infected mice (106, 107) and was also demonstrated to promote a NF-κB proinflammatory response and hepatocarcinogenesis in a model of susceptible A/JCr mice (108). Similarly, S. Typhimurium expressing the active typhoid toxin caused an enhanced inflammation in liver and spleen of an infected mouse whereas the isogenic CdtB mutant did not (105). Administration of S. dysenteriae CDT from culture supernatant or partially purified CDT preparation were associated with watery diarrhea and induced epithelial damage in the colon of treated mice more often than CDT-negative preparation (24). In C. jejuni, wild-type CDT was shown to favor colonization in C57BL/6 mice and to induce more severe gastritis compared with CDT mutant (109). CDT is required for efficient colonization of H. hepaticus in Swiss mice (106). However, the scarcity of in vivo experiments studying CDT-producing bacteria in the intestinal context renders it difficult to evaluate the actual role of this toxin in gut colonization and pathogenesis. Moreover, whether the bacteria could benefit from inducing DNA damage in the host cell remains unclear. CDT has been proposed to exhibit immunosuppressive function since it exerts cytotoxic activity toward B and T cells (12, 14). Such inhibitory effect on immune cells of the gastrointestinal tract is an attractive hypothesis but lacks experimental support.

COLIBACTIN

Colibactin, a Potent Genotoxin

Almost 20 years after the discovery of CDT, a second genotoxin named colibactin was discovered. Colibactin is a polyketide nonribosomal peptide hybrid compound produced by several species of Enterobacteriaceae. It was first identified in 2006 in an extraintestinal pathogenic E. coli (ExPEC) strain isolated from neonatal meningitis (110). This secondary metabolite is produced by the clbA-S genes present in the 54-kb pathogenicity pks island. This gene cluster codes for a biosynthetic machinery constituted with nonribosomal peptide and polyketide synthases (NRPS and PKS), together with accessory and tailoring enzymes (Fig. 6). Originally identified as inducing a megalocytosis phenotype 3 days after a transient infection assay of eukaryotic cells (Fig. 1), colibactin was shown to provoke DSBs in the host cell nucleus as demonstrated by comet assay (revealing DNA fragmentation) and phosphorylation of histone H2AX (a marker of DSBs). The genotoxic activity of colibactin is dependent on a direct host cell-bacteria interaction and cannot be recovered from culture supernatant, killed bacteria, or bacterial lysate (110). Together, these observations suggest that colibactin is a highly reactive and unstable product precluding its purification and/or that contact with the host cell is required to trigger “inoculation” of the genotoxin into the recipient cell. Similarly to CDT, colibactin-induced DNA damage triggers the activation of the DDR involving the ATM/ATR, CHK1/2, 14-3-3, CDC25 axis leading to CDK inactivation, cell cycle arrest, and eventually apoptosis and senescence (megalocytosis, cell enlargement) (Fig. 5). However, in contrast to CDT intoxication that results in a G2 arrest without affecting the S-phase length in HeLa cells (17, 86, 87, 111), infection with colibactin-producing bacteria induces a delay in the accomplishment of the S phase (Fig. 5 and Table 1) (110). This disparity suggests that cells activate a different DDR, indicative of a distinct type of genotoxic insult or type of DNA damage (see below). Following low multiplicity of infection, after a delay in the cell cycle concomitant with H2AX phosphorylation and DNA repair, the cells resume their cell cycle. However, a subpopulation could proliferate with residual and persisting DNA damage leading to chronic anaphase bridging, chromosome instability (with lagging chromosome, chromosome break, micronuclei, ring chromosome), and aneuploidy. This chromosomal instability increases gene mutation frequency leading to an oncogenic transformation of in vitro infected mammalian cells (112). In contrast, following acute infection, the colibactin-induced massive DNA damage results in premature programmed cellular senescence, characterized by a cell enlargement, irreversible cell cycle arrest associated with accumulation of CDK inhibitors (p16 and p21), prolonged DSB, and chromatin remodeling. These senescent cells also secrete proinflammatory cytokines, chemokines, growth factors, and proteases in the medium. This senescence-associated secretory phenotype (SASP) induces a bystander genotoxic effect in naive recipient cells (113, 114). In addition, the colibactin-induced SASP can promote the growth of tumor cells, through secretion of the hepatocyte growth factor (113, 114, 115). Importantly, oral administration of pks-positive E. coli inflicted DNA damage in vivo in rodent enterocytes, whereas an isogenic mutant not producing colibactin did not, revealing the mutagenic potential of these bacteria that could participate in the development of colorectal cancer (see below) (112, 114, 116, 117, 118).

Figure 6.

The pks island. The pks genomic island contains 19 genes (clbA to clbS) encoding polyketide synthases (PKS, orange), nonribosomal peptide synthases (NRPS red), hybrid PKS-NRPS (orange/red) and genes with unknown function (black). Other “accessory” enzymes and proteins are encoded by clbA (phosphopantetheinyl transferase), clbM (efflux pump), clbP (peptidase) and clbS (resistance protein).

Table 1.

Comparative key characteristics of CDT, typhoid toxin, and colibactin

Features	CDT and Typhoid Toxin	Colibactin	References
Type of DNA damage	Single-strand break, double-strand break, chromosomal instability (chronic exposure)	DNA cross-link?, double-strand break chromosomal instability	51, 81, 84, 85, 94, 110, 112, 122
Cell cycle arrest	G2 phase (G1 phase)	G1 phase, G2 phase, S phase	23, 90, 110, 111
Secretion	Yes (present in supernatant)	No (contact dependent)	11, 49, 110
Distribution	Pathogenic E. coli, Shigella E. albertii P. alcalifaciens Salmonella Typhi and iNTS	E. coli of phylogroup B2 (ExPEC, commensal) Citrobacter, Klebsiella, Enterobacter	17, 27, 89, 110, 137, 138, 140
Senescence	Yes, SASP	Yes, SASP	88, 94, 113
Nature	Heterotrimeric complex type AB2 (CdtA, CdtB, CdtC) or heptameric complex type A2B5 (CdtB, PltA, PltB5)	Nonribosomal peptide-polyketide hybrid compound(s)	29, 110, 130
Genes	3 genes (cdtA, cdtB, and cdtC) clustered in operon or in Salmonella, cdtB localized in CdtB islet with pltA and pltB	19 genes hosted on the pks genomic island	26, 27, 29, 110
Enzymatic activity	DNase (PIP3 phosphatase?)	DNA intercalation and alkylation?	51, 97, 122
Trafficking	Endocytosis, anterograde transport	?	58, 76, 80
Receptor	Lipid raft, carbohydrates?	?	53, 54, 76
Bacterial DNA protection	Secretion, autoinhibition with N terminus of CdtC occluding CdtB active site?	Defense protein ClbS, maturation of colibactin in the periplasm	51, 52, 129, 131
Prodrug	No	Yes, precolibactin is cleaved by ClbP	127, 128, 129
Homology	DNase I	Duocarmycin, acylfulvenes?	51, 133
Procarcinogen	No evidence in Enterobacteriaceae	Yes (in inflammatory context)	114, 116

Colibactin Biosynthesis

Colibactin is produced by a complex biosynthetic machinery involving the sequential action of proteins ClbA to ClbS (Fig. 6 and 7). The core machinery consists of three polyketide synthases (PKS: ClbC, ClbI, ClbO), three nonribosomal peptide synthases (NRPS: ClbH, ClbJ, ClbN), and two hybrids (PKS-NRPS: ClbB, ClbK). The machinery also uses additional maturation proteins (ClbA, D-G, L, P) and an efflux pump (ClbM). ClbA is a phosphopantetheinyl (PPant) transferase, essential for the activation of the NRPS and PKS enzymes. ClbA adds a PPant arm onto the NRPS and PKS carrier protein domains that allows the tethering of the nascent synthesis intermediates and substrates (119). ClbA is also implicated in the biosynthesis of other NRP/PK compounds; it was shown that it participates in the synthesis of the siderophores yersiniabactin and enterobactin (and the enterobactin-derived salmochelins) (120). Thus a clbA mutant not only does not express colibactin, but could also exhibit altered production of siderophores; care should be taken when using a clbA or full pks island deletion mutant because they do not allow conclusion to be made solely on the activity of the colibactin pathway. Following activation by ClbA, the initiating NRPS ClbN uses Asn as a substrate to generate an N-myristoyl-d-Asn prodrug motif (and in low abundance, other analogs with various carbon chain lengths). The NRPS-PKS assembly line (ClbB-C-H-I-J-K) continues the synthesis of precolibactin compound(s) using as substrates malonyl-CoA and amino acids (including Ala, Ser, Gly, Cys, and the unusual nonproteinogenic aminocyclopropane-carboxic acid) (Fig. 7) (121, 122, 123). It was recently shown that ClbD-E-F-G and the adenylation domain of the NRPS ClbH generate from l-Ser an unusual PKS extender unit, aminomalonyl-ACP. ClbG is a free-standing, trans-acting acyltransferase, which transfers the aminomalonyl-ACP-ClbE to the multiple PKS modules of the assembly line (124, 125). Other enzymes encoded on the pks island are required for genotoxin synthesis but do not yet have an assigned role in biosynthesis (ClbL, O, Q). The metabolite(s) resulting from the assembly line is an inactive “precolibactin,” a polyketide-peptide hybrid that is eventually exported into the periplasm of the bacteria by ClbM. ClbM is 12-transmenbrane efflux pump inserted in the bacterial inner membrane, whose mutation leads to a reduced genotoxic activity (126). The current hypothesis for this attenuated phenotype assumes that other pumps partially complement the absence of ClbM (110, 126). The X-ray structure of ClbM indicates that it is a cation-coupled antiporter containing a 40 Å wide binding pocket that could confer its specificity for the 700 to 900 Da precolibactin (126). Once in the periplasm, the prodrug motif of precolibactin is hydrolyzed to release the active genotoxin colibactin and the cleavage product (Fig. 7). This prodrug activation is achieved by the periplasmic membrane-bound ClbP protein with a d-amino peptidase activity (127, 128, 129). Such prodrug strategy is also observed with other polyketide biosynthetic pathways such as those of the antibiotics pre/zwittermicin and pre/xenocoumacin (129, 130). It is believed that this periplasmic prodrug activation serves as a self-resistance mechanism. The pks island encodes an additional self-resistance system, the ClbS protein (131). ClbS is not required for colibactin synthesis but plays a protective role for the bacterial producer from the genotoxin by modifying or sequestering residual cytoplasmic active colibactin (Fig. 7). Interestingly, ectopic expression of ClbS in HeLa cells confers a partial resistance phenotype to colibactin-induced genotoxicity, suggesting that the genotoxin is actually injected into the host cell, possibly conveyed to the nucleus, to induce DNA damage (131).

Figure 7.

Model of colibactin biosynthesis. The NRPS-PKS assembly line is posttranslationally matured with a 4′-phosphopantetheine (PP) arm on the carrier proteins by the ClbA PP transferase. The active assembly line synthesizes precolibactin that is translocated in the periplasm by the ClbM efflux pump. The prodrug is cleaved by the ClbP peptidase to generate the cleavage product (black) and active mature colibactin (red). Self-damage to the producing bacterium DNA by active colibactin is prevented by the periplasmic maturation and by the cytoplasmic ClbS resistance protein.

Structure and Activity of Colibactin

Although the high instability of the activated colibactin has precluded its purification, the relatively stable precolibactin intermediate biosynthesis derailment products were recently characterized in clbP mutants, shedding light on the colibactin “warhead” (122, 123, 132). Resulting from the cleavage of the prodrug, the colibactin warhead is an electrophilic spiro cyclopropane ring system similar to that found in the DNA-alkylating secondary metabolites such as illudins, duocarmycins, and yatakemycin (Fig. 8) (133). Mass spectrometry analyses and in silico predictions indicate that the warhead is connected to a thiazolinyl-thiazole or bithiazole unit (respectively, compounds 2 and 3 in Fig. 8) (122, 132). A bithiazole-containing precolibactin was recently confirmed in a clbP clbG double mutant (125). Such thiazole tails have been shown to be involved in DNA intercalation in the DNA-damaging natural products bleomycins and phleomycins, thus supporting the hypothesis that colibactin directly targets DNA. Interestingly, Vizcaino and Crawford detected a weak cross-linking activity in plasmid DNA that was incubated 24 h with millimolar concentrations of a candidate precolibactin compound, suggesting that colibactin could generate DNA interstrand cross-links by alkylation (122). Consistent with this hypothesis, bacteria producing colibactin that were mutated for both the self-resistance protein ClbS and the nucleotide excision repair system (that repairs DNA adducts and cross-links) exhibit an exquisite sensitivity to colibactin (131). However, to date, no colibactin-DNA adduct has been directly identified, and if indeed DNA cross-links are produced, at least a part of the DSB, revealed by the occurrence of γH2AX foci in the host cell nuclei, may be due to secondary events linked to their presence. In addition, it appears that the pks island machinery does not produce one molecule but rather a set of secondary metabolites whose composition and quantity differ according to the bacterial strain (134). The function of several pks-encoded genes (especially the PKS ClbO and amidase ClbL, which were shown to be essential for genotoxicity) are not yet attributed to the biosynthesis model, raising the possibility that (pre)colibactin(s) structure could be larger than currently predicted (130, 135). Much work is needed to elucidate the structures of the full-length precolibactin(s) and how they contribute to the bioactivity, or even to different bioactivities.

Figure 8.

Candidate precolibactin molecules synthetized by the pks biosynthetic pathway. Cleavage product released by the ClbP peptidase (1) and precolibactins molecules (2 and 3) that were elucidated in clbP mutants, based on mass spectrometry and nuclear magnetic resonance data. The electrophilic warhead, putative DNA intercalating motif, and the peptidase cleavage site are shown.

Colibactin, a Threat for Human Health?

Among E. coli strains classified in 5 main phylogroups (A, B1, B2, D, and E), the pks genomic island is mainly restricted to the phylogroup B2 (136). This phylogroup comprises ExPEC and commensal strains. Epidemiological surveys show that approximately 40% of ExPEC strains are pks island positive, a rate reaching 60% in uropathogenic E. coli isolates and up to 72% in E. coli isolates of men with acute prostatitis (110, 130, 137, 138, 139, 140). The pks island is genetically associated with the High Pathogenicity Island (HPI) that hosts the cluster of genes required for the biosynthesis of the siderophore yersiniabactin (120). The pks island is also found in other Enterobacteriaceae opportunists implicated in nosocomial infections, such as Klebsiella pneumoniae, Enterobacter aerogenes, and Citrobacter koseri (138, 140, 141). In Klebsiella, the pks gene cluster, which is 100% identical to the E. coli pks island, was associated with phosphorylation of H2AX in the hepatocytes of orally infected mice (141). The role of the pks island in virulence was demonstrated in a sepsis model with the ExPEC strain of serotype O18:K1:H7 (isolated from newborn meningitis) where colibactin production was associated with a more severe lymphopenia and a lower survival rate of infected mice compared with a mutant impaired for colibactin synthesis (142). In a rat neonatal systemic infection model, mutation of the peptidase ClbP is also associated with a decreased virulence, but to a lesser extent, compared with the mutation of the PPTase ClbA (117). This interesting result underlines the ubiquitous role of ClbA in colibactin and siderophore biosyntheses that are correlated with the virulence of ExPEC (120, 143). This interplay between colibactin and siderophore synthesis reveals a complex regulatory network of various virulence features with different functions.

Besides the pathogenic potential of E. coli, this bacterium belongs to the pioneer microbiota colonizing the mammalian gut immediately after birth and remaining resident throughout the life of the host as the predominant facultative anaerobic species (144). Recent epidemiological studies indicate a striking increase of the prevalence of E. coli strains of the phylogenetic group B2 in the neonatal gut microbiota in high-income countries. During the past decades, studies on the E. coli strains that colonize humans (including infants) have shown that the E. coli population shifts from the phylogenetic group A to the phylogenetic group B2, with a correlation on long-term colonization with commensal strains carrying the pks island (145, 146). The prevalence of the pks island was reported from 38 to 58% in the B2 phylogroup (137, 138). One likely reason for the high prevalence of E. coli strains harboring the pks island is the increased capacity of these strains to persist in the microbiota, including in infants (145). Several studies have investigated the impact of these genotoxic strains in the gut microbiota. First, the pks island is found in the E. coli probiotic strain Nissle 1917, marketed as Mutaflor in Germany. Nissle 1917 was demonstrated to be as efficient as mesalazine in maintaining remission in ulcerative colitis (147, 148). Surprisingly, in an in vivo rodent model, colibactin expression was shown to be linked with the probiotic effect of Nissle 1917, as mutation of clbA abolishes its anti-inflammatory property (149). In contrast, primo-colonization (bacterial colonization that occurs at birth) with a commensal pks+ E. coli strain (that was isolated from a healthy adult) induced transient DNA damage in rodent enterocytes during the preweaning period. No DNA damage was detected at adulthood, but the frequency of abnormal mitosis in the crypt-proliferating compartment, indicative of a persistent chromosome instability, was increased in animals colonized with the wild-type pks+ E. coli compared with the clbA mutant not producing colibactin, suggesting that the early DNA damage inflicted by colibactin imprinted a chronic chromosomal instability (118). Moreover, the adult animals colonized by the wild-type strain exhibited an alteration of the gut barrier, with an increased paracellular permeability and an increased transepithelial passage through the Peyer’s patches (118, 150). This alteration of the gut barrier increased the immune response against luminal antigens, as observed in an ovalbumin-driven oral tolerance model (150). Collectively, these data indicate that gut colonization with colibactin-producing E. coli strains could contribute to dysregulated immune-mediated diseases at adulthood.

Colibactin and Cancer

Colibactin-producing strains of E. coli inflict DNA double-strand breaks and exert a strong mutagenic power, at least in cultured cells. The DNA damage has been observed in vivo as well as in rodent models (112, 114, 115, 116, 118). DNA damage has been long identified as a mutagenic factor, possibly leading to gene mutations, genomic instability, and cellular transformation, resulting in tumor initiation and progression (151). Several publications have revealed the higher prevalence of pks-positive E. coli strain in biopsy from colorectal cancer (CRC) patients (116, 152, 153). Strikingly, a recent long-term retrospective cohort study showed a significant increase in the incidence of CRC diagnosed in young adults, especially in the 20- to 34-year-old age group (154). This higher incidence of young adult’s CRC in developed countries coincides with the increasing prevalence of pks-positive B2 commensal strains that colonizes the gut at birth. Other epidemiological data suggest that pks-positive K. pneumoniae, which is the predominant pathogen in liver abscess, might contribute to carcinogenesis since these lesions correlate with increased risk of colorectal cancer (141). The first experiment demonstrating a role for the pks island in the development of cancer was published in 2012 using an axenic IL10^–/– mouse model. These axenic mice developed an inflammatory response upon gut colonization with the B2 colibactin-producing E. coli NC101 strain. Following azoxymethane (AOM) treatment, increased tumor multiplicity and promotion of carcinoma invasiveness were shown to be pks-dependent (116). Using a mouse AOM/DSS proinflammatory model, oral gavage with a clinical pks-positive strain isolated from a human colorectal tumor significantly increased the number of tumors compared with the isogenic pks-deleted strain (114). Moreover, in nude mice, increased xenograft tumor size was observed with epithelial cells transiently exposed to pks+ E. coli. The mechanism sustaining this effect was associated with a colibactin-induced senescence microenvironment secreted by cells exposed to pks strains and higher expression of hepatocyte growth factor. This SASP pathway was dependent on myc induction of the microRNA miR-20a-5p synthesis. Supernatant of clinical pks+ E. coli-infected cells was shown to stimulate proliferation of naive uninfected cells. Interestingly, human colorectal adenocarcinomas colonized by pks+ strains were shown to display increased level of γH2AX, miR-20a-5p, and senescence markers (SA-β-gal, p21^cip) compared with non-pks-colonized human carcinoma (114, 115). These experimental studies have demonstrated that, in an inflammatory context, expression of colibactin is procarcinogenic. However, whether the presence of such bacteria in a “normal” environment represents a significant hazard in colorectal cancer is still far from being established. For a more comprehensive evaluation of the risk associated with the higher prevalence of genotoxic B2 “commensal” E. coli in the microbiota of the human population in industrialized countries, a large long-term follow-up survey with analysis of microbial population in stool samples from controls and patients with intestinal diseases is necessary.

Genotoxic Bacteria, Not All the Same...

Accumulating evidence indicates that there is not a unique colibactin compound but a mixture of several molecules, possibly with different genotoxic strength and chemical stability. In addition, the mixture of compounds synthesized by a particular strain could depend on its genetic background and thus results in different impact on the host (134). Notably, the probiotic E. coli strain Nissle 1917 shows a pks-dependent ability to reduce inflammation while, in contrast, other pks-positive commensal strains impair intestinal permeability, leading to an increased immune response and oral tolerance failure (118, 149, 150). Animal models of gut colonization with pks+ strains revealed that colibactin production could participate in CRC in inflammatory conditions (114, 116). In ExPEC, colibactin participates in the virulence of this pathogen (142), while epidemiological studies suggest that it could act as a fitness factor favoring colonization by commensal pks-positive bacteria (145). This fitness advantage or increased virulence could be linked to the presence of clbA that increases the capacity of the bacteria to produce siderophores (120). In other context, the NC101 strain favors inflammation and, in a pks-dependent manner, induces carcinogenesis in IL10^–/– mice (116). Together, these studies showed a variety of colibactin impact depending on the bacterial species, the model, and the aim of the experiment. Nonetheless, these studies clearly illustrate the fact that the genetic diversity of the bacterial producer is critical and determines the distinctive consequence of the carriage of pks island in the E. coli population on human health. Whether these differences between strains are related to the level of production of colibactin or to the composition and stability of the secondary metabolites produced by the pks machinery need to be addressed to evaluate the health threat represented by exposure to genotoxic bacteria.

OTHER GENOTOXINS

Since CDT and colibactin genotoxins were uncovered, recent studies have revealed novel bacterial factors in Enterobacteriaceae that exhibit mutagenic properties on host cells.

Some reports have shown that most of E. coli from the B2 group (approximately 90%) possess a genomic island containing the gene encoding for the uropathogenic specific protein (USP) and 3 associated open reading frames corresponding to so-called immunity genes versus 24% of usp-positive strains in fecal isolates. The C-terminal domain of USP displays homology with the DNase-like domain of the S-pyocin and to the HNH endonuclease motif of the colicin E7 of E. coli. USP exhibits genotoxicity toward cultured eukaryotic endothelial cells as revealed by induction of host DSBs detected in a comet assay (155). The associated immunity genes are required for self-protection of USP producers from their own DNase activity since USP is lethal when expressed alone. Interestingly, USP immunity protein was shown to exhibit affinity to nucleic acids that protects genomic DNA not only from USP genotoxicity, but also from other DNase activities such as colicin (156).

The invasive Shigella flexneri strain was also reported to induce an early genotoxic stress detectable as soon as 1 h after infection in a virulence plasmid-dependent manner. Moreover, Shigella was shown to inhibit the p53 response via VirA-mediated calpain degradation of p53, thus impairing cell apoptosis and disrupting the p53-dependent DNA repair activation (157). Such bacterial impact on DNA repair pathways has been also reported in EPEC. Consistent with epidemiological data supporting a role of adhesive E. coli in colon cancer (158), recent studies suggest that the type 3 effector EspF has the ability to deplete DNA mismatch repair enzymes in infected cells resulting in accumulation of unrepaired DNA mutations induced by EPEC infection-dependent intracellular reactive oxygen species elevation (159, 160). However, in vivo data supporting the existence and relevance of such effects on enterocytes is still required.

CONCLUDING REMARKS

More and more enterobacterial strains present in the microbiota are discovered to express genotoxins. This genotoxicity likely represents a way to modulate the host cell cycle and host immune system and eventually favor the efficient colonization of the tissues. However, the long-term consequences for the host—the colonic mucosa and possibly the urinary tract as well—could be cellular transformation and ultimately tumorigenesis. However, while mutagenesis is recognized as a procarcinogenic event, both the nature of the damage (DNA adducts, strand breaks, chromosomal translocation, illegitimate recombination…) and its level modulate the final risk of developing a tumor (161). Hence, examining the nature and the level of the DNA damage that genotoxic enterobacteria can induce in the gut cells is a major question. In addition, the nature of the cells exposed to such genotoxic stresses (such as stem cells) as well as environmental conditions (such as inflammation) are also partly dependent on the bacteria producing the genotoxin (i.e., whether they are able to reach the crypt cells and/or induce inflammation) and thus remain a fundamental issue to be addressed in order to identify the danger of bacterial strains in the tumorigenesis multistep process. Those studies should, in addition, be convincingly grounded to epidemiology.

Bacterial genotoxins represent an increasingly important field of research. While Helicobacter pylori has been the first WHO-recognized carcinogenic bacterium, one could speculate that genotoxin-expressing strains may be future candidates. On the one hand, those toxins are powerful tools used by E. coli and other enterobacterial species to dominate over others in their territory, impair host immune response, and favor tissue invasion. On the other hand, the strains producing genotoxins clearly represent a tumorigenic risk, the knowledge of which needs to be deepened at the molecular as well as epidemiological levels.