Cytotoxic necrotizing factor 1 hinders colon tumorigenesis induced by colibactin-producing Escherichia coli in Apc^Min/+ mice

ABSTRACT

Colorectal cancer (CRC) patients are frequently colonized by colibactin-producing Escherichia coli (CoPEC) (>40%), which enhances tumorigenesis in mouse models of CRC. We observed that 50% of CoPEC also contains the cnf1 gene, which encodes cytotoxic necrotizing factor-1 (CNF1), an enhancer of the eukaryotic cell cycle. The impact of its co-occurrence with colibactin (Clb) has not yet been investigated. We evaluated the impact of CNF1 on colorectal tumorigenesis using human colonic epithelial HT-29 cells and CRC-susceptible Apc^Min/+ mice inoculated with the CoPEC 21F8 clinical strain (Clb+Cnf+) or 21F8 isogenic mutants (Clb+Cnf-, Clb-Cnf+ and Clb-Cnf-). Infection with the Clb+Cnf- strain induced higher levels of inflammatory cytokines and senescence markers both in vitro and in vivo compared to those induced by infection with the Clb+Cnf+ strain. In contrast, the Clb+Cnf- and Clb+Cnf+ strains generated similar levels of DNA damage in HT-29 cells and in colonic murine tissues. Furthermore, the Apc^Min/+ mice inoculated with the Clb+Cnf- strain developed significantly more tumors than the mice inoculated with the Clb+Cnf+ strain or the isogenic mutants, and the composition of their microbiota was changed. Finally, rectal administration of the CNF1 protein in Apc^Min/+ mice inoculated with the Clb+Cnf- strain significantly decreased tumorigenesis and inflammation. Overall, this study provides evidence that CNF1 decreases the carcinogenic effects of CoPEC in Apc^Min/+ mice by decreasing CoPEC-induced cellular senescence and inflammation.

KEYWORDS:

• Colorectal cancer; Escherichia coli; colibactin; CNF-1; CoPEC

Introduction

Colorectal cancer (CRC) is the third most common cancer in the world, causing significant morbidity and mortality.¹ CRC is a multifactorial disease involving both genetic and environmental factors. Among the genomic changes associated with CRC, loss-of-function mutations in the Apc (adenomatous polyposis coli) gene are the most prevalent and are considered the initiating event in approximately 80% of CRC cases.² Among the environmental factors linked to CRC, the gut microbiota is increasingly thought to be a key player in CRC pathogenesis.^3,4 Modification of the composition of the gut microbiota, or dysbiosis, has been reported in patients with CRC, with an increase in the abundance of bacteria such as Bacteroides fragilis or Fusobacterium nucleatum and a decrease in the abundance of Faecalibacterium prausnitzii .^5–9

The involvement of the gut microbiota in CRC has been established using murine models of CRC. Germ-free Apc^Min/+ mice display a lower number of intestinal and colorectal tumors than microbiota-bearing Apc^Min/+ mice.¹⁰ A recent study showed that germ-free mice that received fecal samples from patients with CRC exhibited an increase in number of polyps, intestinal dysplasia, and levels of cellular proliferation markers as well as inflammation compared with those of germ-free mice that received fecal samples from healthy individuals.¹¹

At the taxonomic level, analysis of the human CRC microbiome has identified potential microbial candidates implicated in CRC pathology, including Escherichia coli, F. nucleatum, and enterotoxigenic B. fragilis (ETBF).⁸ ETBF induced chronic inflammation and tumorigenesis in Apc^Min/+ mice and led to high levels of interleukin-17 (IL-17) production, which disrupted normal myelopoiesis and resulted in the accumulation of pro-carcinogenic myeloid-derived suppressor cells in the tumor microenvironment.¹² In Apc^Min/+ mice, F. nucleatum increased tumor development without inducing colitis, accompanied by increased infiltration of myeloid cells into tumors.⁷

Recent studies have shown that pathogenic E. coli synthesizes toxins known as cyclomodulins, such as cytolethal distending toxins, cytotoxic necrotizing factor-1 (CNF1), cycle-inhibiting factor, and colibactin, which interfere with the cell cycle.^13,14 Cyclomodulin-encoding genes, especially the colibactin-encoding pks island and CNF1-encoding gene (cnf1), are overrepresented in CRC patients colonized by E. coli strains.^15,16

Colibactin-producing E. coli (CoPEC) strains have been identified in the colonic mucosa of approximately 55–67% of patients with CRC versus 19–21% of control patients.^15,16 CoPEC has been shown to induce DNA double-strand breaks (DSB), chromosomal instability, genomic mutations and cell cycle arrest.^17–21 CoPEC induces senescence of infected cells, accompanied by secretion of inflammatory mediators and growth factors, thus promoting proliferation of adjacent uninfected cells.²² Importantly, CoPEC promotes colon tumorigenesis in multiple murine models of CRC, including Apc^Min/+ mice, AOM-treated Il-10^–/– mice, AOM/DSS-treated mice and Apc^Min/+;Il-10^–/– mice.^16,22–24 Notably, inflammation enhances the development of colon cancer in the Apc^Min/+ model, which was established by specifically deleting the APC gene in epithelial cells,²⁵ as seen with the use of dextran sulfate sodium (DSS)²⁶ and by genetically introducing defective IL-10 signaling.^27,28

The prevalence of cnf1-harboring E. coli is significantly higher in patients with CRC (37%) than in control patients (13%).¹⁵ Nevertheless, the involvement of cnf1-harboring E. coli in CRC has not been determined. CNF1 is a 115 kDa protein toxin that activates Rho GTPases, leading to cytoskeletal and cell cycle alterations with subsequent macropinocytosis and the formation of megalocytic, multinucleated cells.²⁹ In addition, CNF1-induced activation of Rho GTPases triggers cellular events not directly linked to the actin cytoskeleton, such as the activation of NF-κB³⁰ and the production of cytokines, such as IL-6 and IL-8,^31,32 and provides protection against apoptosis.^33,34 It also promotes quiescent cell entry into the cell cycle.³⁵ Recently, Fabbri et al. showed, in vitro, that CNF1 induces epithelial-to-mesenchymal transition (EMT), a crucial step in malignant tumor conversion and invasiveness, in intestinal epithelial cells.³⁶ A separate study has shown that CNF1 promotes the migration and invasion of prostate cancer cells in vitro.³⁷ Therefore, it appears that many of the cellular activities induced by CNF1 might participate in carcinogenesis. Despite the overrepresentation of the cnf1 gene in E. coli strains isolated from patients with CRC, the effect of the CNF1 toxin has not been studied in CRC.

In this study, we determined the prevalence of E. coli harboring the pks and cnf1 genes in CRC patients and investigated the tumorigenic properties of E. coli strain 21F8 isolated from a human colon cancer biopsy and producing both colibactin and CNF1 in comparison with those of isogenic mutants using human intestinal epithelial HT-29 cells and an Apc^Min/+ mouse model of CRC.

Results

Most E.coli strains harboring the cnf1 gene possess a pks island

The patient data used come from previous studies.^23,38 The prevalence of pks was significantly higher in CRC patients (46%, n = 37/80) than in patients with diverticulosis (21%, n = 6/28; p = 0.037) (Table 1), which is in accordance with previously reported data.^15,16 In contrast, the difference in the prevalence of cnf1 in CRC patients (25%, n = 20/80) and diverticulosis patients (14%, n = 4/28; p = 0.363) was not significant. The majority of E. coli strains harboring the cnf1 gene also carried the pks island: 95% (n = 19/20) and 100% (n = 4/4), in CRC patients and in healthy patients respectively, showing a strong association between the cnf1 gene and the pks genomic island.

Table 1. Distribution of E. coli harboring cnf1 and/or the pks genomic island producing colibactin (Clb) among CRC and control patients (percent in brackets).

Disease	Patients^3	Clb+Cnf+	Clb+Cnf-	Clb-Cnf+	Clb-Cnf-^4
CRC^1	80	19 (23.8)	18 (22.5)	1 (1.3)	42 (52.5)
Control^2	28	4 (12.0)	2 (8.0)	0 (0)	22 (80.0)

¹CRC = patients with colorectal cancer.

²Control = patients with diverticulosis.

³Number of patients carrying a mucosa-associated E. coli that produces cyclomodulins (colibactin or CNF1).

⁴Number of patients carrying a mucosa-associated E. coli that does not produce cyclomodulins (colibactin or CNF1).

Analysis of the virulome of CoPEC strains 11G5 and 21F8

We selected two CoPEC strains, 11G5 and 21F8, from CRC patients. Previous studies have reported that the 11G5 reference strain, which harbors only the pks island, increased the number of tumors in mice.^22,23,39 The 21F8 strain possesses both pks and cnf1. We analyzed the virulome of these two CoPEC strains (11G5 and 21F8). Both of them belong to phylogenetic Group B2 and share 130 genes associated with virulence (Supplemental Figure S1). In silico analysis revealed major virulence factors belonging to the following five categories: adherence (type 1 fimbriae, FC1/S fimbriae, YadA fimbriae, and curli fiber), toxins (colibactin and Vat), iron uptake (enterobactin, ChuA, Sit, and yersiniabactin), protectin (iss), and motility and chemotaxis (che, flg, fliA, and flh). Virulence genes missing in 21F8 but present in the 11G5 genome were those involved in glutathionylspermidine amidase activity (gsp genes), resistance to mercury (mer genes), adhesion (ehaB and espI), invasion (ibeA) and iron uptake (iro genes). Virulence genes missing in 11G5 but present in the 21F8 genome were pap genes, which encode P fimbriae, adhesin genes (iha, upaG) and the pic gene, which encodes a colonization factor. Likewise, pathogenicity island II (PAI II), which harbors the hlyCABD operon and the cnf1 gene, was present in E. coli strain 21F8 but absent in strain 11G5. Thus, the only cyclomodulin present in the 21F8 strain and absent in the 11G5 strain was the CNF1 toxin.

The E.coli 21F8 strain lacking cnf1 promotes colonic tumorigenesis

C57BL/6 Apc^Min/+ mice were gavaged with the 21F8 or 11G5 strain to assess the protumorigenic roles of the CoPEC strains. As expected, the number of colonic tumors increased in mice infected with the 11G5 strain compared to that in the uninfected mice (Figure 1a). Surprisingly, the number of tumors in the mice colonized with the 21F8 strain did not increase relative to that in the uninfected mice (Figure 1a). To evaluate the role of CNF1 in intestinal tumorigenesis, we generated isogenic mutants defective in CNF1 and/or colibactin (Clb) production, which were designated Clb+Cnf- for the mutant defective in CNF-1, Clb-Cnf+ for the mutant defective in colibactin, and Clb-Cnf- for the mutant defective in CNF1 and colibactin. The wild-type CoPEC 21F8 strain (Clb+Cnf+) and the three mutants were administered orally by gavage to Apc^Min/+ mice (Figure 1b, c). Mice inoculated with Clb+Cnf- developed a significantly higher number of colonic tumors than mice inoculated with the isogenic mutants devoid of colibactin (Clb-Cnf+ and Clb-Cnf-). As an increase in the number of tumors may be due to an overabundance of Clb+Cnf- in the gut microbiota, the abundance of 21F8 wild-type or isogenic mutant strains in feces and colon biopsies was determined. We did not observe a significant increase in the bacterial load in the Apc^Min/+ mice fed Clb+Cnf- (Supplemental Figure S2).

Figure 1. The absence of cnf1 in CoPEC increases tumor development in Apc^Min/+ mice. Apc^Min/+ mice were treated with streptomycin for 3 days and then received water for 24 hours. (a) Mice orally received PBS (Day 0) or 10⁹ colony-forming units (CFU) of 11G5 or 10⁹ CFU of wild-type 21F8 (Clb+Cnf+) bacteria. (b–e) Apc^Min/+ mice were treated with streptomycin for 3 days and then received water for 24 hours. Mice were orally inoculated with wild-type 21F8 (Clb+Cnf+) or 21F8 mutants: Clb+Cnf-, Clb-Cnf+ or Clb-Cnf-. (a-e) Mice were killed at 50 days post-infection. (a, b) the number of colorectal tumors was determined using a dissecting microscope. The data points represent actual values for each individual mouse, and the bars indicate median values. Data were combined from two independent experiments. (c) Representative images of the colons of the inoculated mice. Arrows show macroscopic tumors. (d) Representative images of γH2AX immunohistochemical staining of nontumoral colonic mucosa (scale bars: 50 µm) and (e) quantification of γH2AX-positive cells determined from nontumoral colonic mucosa. Data are presented as means ± SEMs. Statistical analysis: Kruskal–Wallis ANOVA (*P <0 .05, **P < 0.01, ***P < 0.001, ****P <0 .0001).

To test whether Clb+Cnf+ and Clb+Cnf- expressed colibactin and induced DNA damage in vivo, we assessed the occurrence of DSB in colonic epithelial cells by detecting the S139 phosphorylation on the histone H2AX (γH2AX), a well-known DSB marker⁴⁰. Colonic biopsies revealed a significant increase in the number of γH2AX-positive cells in the mice exposed to the CoPEC strains Clb+Cnf+ and Clb+Cnf- compared to that in the mice exposed to the isogenic mutants Clb-Cnf+ and Clb-Cnf- (Figure 1d, e), showing that the pks island was functional and induced DNA damage of colonic epithelial cells. No significant differences in γH2AX staining in the colon of the mice inoculated with Clb+Cnf+ and Clb+Cnf- were observed (Figure 1d, e).

The CoPEC 21F8 strain induces colibactin-dependent cytotoxicity in human colon cancer cells

We next investigated the mechanisms by which CNF1 limits 21F8-induced colorectal tumorigenesis using cell cultures. To test whether E. coli 21F8 expresses the pks and cnf1 genes and induces cytotoxicity, we infected human colon epithelial HT-29 cells with Clb+Cnf+ (21F8 strain), Clb-Cnf+, Clb+Cnf-, Clb-Cnf-, or with the corresponding trans-complemented mutants Clb+Cnf- +pBK-cnf and Clb-Cnf- +pBK-cnf. Colibactin and CNF1 are known to dysregulate cell cycle and induce cytopathic phenotypes. Transient infection with colibactin-producing bacteria (Clb+Cnf+ and Clb+Cnf-) caused cell cycle arrest, apoptosis induction and megalocytosis in HT-29 cells (Figure 2a, as expected.^19,41,42 Clb-Cnf+ and Clb-Cnf- +pBK-cnf were able to induce accumulation of cells in both the S and G2/M phases compared to uninfected cells (Figures 2a and Supplemental Figure S3A). This was also found when cells infected with Clb+Cnf- were incubated with the purified toxin CNF1 (Supplemental Figure S3a). Infection with Clb-Cnf+, Clb-Cnf- +pBK-cnf and Clb+Cnf+ resulted in more flattened cells, elongated cells or cells spreading out compared with infection with Clb-Cnf- (Figure 2a), which are hallmarks of CNF1 cytopathic effects in epithelial cells.^34,41 Only HT-29 cells infected with Clb+Cnf+ and Clb+Cnf- +pBK-cnf exhibited both megalocytosis and elongated cell morphologies. However, the cytopathic phenotype induced by Clb+Cnf- +pBK-cnf appears to be lower than those of Clb+Cnf+ (Figure 2a). Intriguingly, cells infected with this trans-complemented mutant exhibited a cell cycle similar to that of cells infected by Clb-Cnf+ or Clb-Cnf- +pBK-cnf. We quantified the transcription of three key clb genes involved in colibactin production in response to HT-29 cell infection. The mRNA levels of these genes were similar for Clb+Cnf+ and Clb+Cnf- (Supplemental Figure S4a), suggesting that the deletion of cnf1 does not modify the production of colibactin.

Figure 2. The cnf1 gene modulates the genotoxic effect of CoPEC. HT-29 cells were infected for 3.5 h. (a) Cell cycle distribution and cytopathic effects were observed 72 h post-infection. Data are representative of two experiments. The dark and blue arrows show some megalocytes and multinucleated cells respectively as example, and rectangle surrounds the elongated cells. (b) Immunofluorescence of γH2AX indicating double strand breaks was assessed 24 h post-infection. The percentage of γH2AX-positive cells were counted from > 100 cell nuclei per well; the data points represent values for each individual cell wells. Alternatively, the average values of the measured nuclear fluorescence intensities was determined. Data are representative of two or three independent experiments and values are represented in mean ± SEM. Statistical analysis was performed by Kruskal–Wallis tests (NS, not significant, *P <0 .05, **P <0 .01, ***P < 0.001, ****P < 0.0001).

Because the effects of colibactin results from DSB, γH2AX staining was monitored 24 hours post-infection (Figure 2b)⁴³. Cells infected with strains not producing colibactin exhibited normal background levels of γH2AX (0.8 to 12%), whereas cells infected with Clb+Cnf+ or Clb+Cnf- showed strong γH2AX staining (51 vs 70%; Figure 2b). The γH2AX levels were not significantly different between those cells infected with Clb+Cnf+ or Clb+Cnf-. However, the incubation of purified CNF1 with HT-29 cells during Clb+Cnf- infection resulted in a ~ 50% reduction in γH2AX levels (Supplemental Figure S3d). Additionally, we unexpectedly found a drastic reduction in the number of γH2AX-positive cells infected by the trans-complemented mutant Clb+Cnf- +pBK-cnf in comparison with those infected with Clb+Cnf- (Figure 2b). The mRNA levels of clbC, clbM and clbP genes were significantly reduced (≥50%) for Clb+Cnf- +pBK-cnf compared to the parent strain (Supplemental Figure S4b). However, deletion of the cnf gene did not modify mRNA levels of the colibactin-synthesis gene (Supplemental Figure S4a). These results therefore suggest that the reduction of γH2AX levels observed with the trans-complemented mutant may be due to the action of CNF1 combined with a modified production of colibactin. To confirm this effect of CNF1 on γH2AX levels, we incubated HT-29 cells with bleomycin, a well-known chemotherapy drug that induces DNA damage, including DSB. We observed that the level of γH2AX induced by bleomycin was reduced when CNF1 was concurrently added. In conclusion, in our experimental conditions, the presence of the cnf1 gene did not appear to affect the genotoxicity of the 21F8 strain. However, we show that the CNF1 toxin is able to reduce the levels of γH2AX induced by DSB.

Deletion of the cnf1 gene increases CoPEC-induced cellular senescence and IL-8 production

CoPEC induces senescence of infected cells, leading to the secretion of inflammatory mediators and growth factors, which promote the proliferation of nearby uninfected cells.^22,44 We infected HT-29 cells with Clb+Cnf+, Clb-Cnf+, Clb+Cnf-, Clb-Cnf-, or with the corresponding trans-complemented mutants Clb+Cnf- +pBK-cnf and Clb-Cnf- +pBK-cnf and detected senescent cells by staining for β-galactosidase at pH 6, a well-accepted senescence marker.⁴⁵ Clb+Cnf+ infection increased the number of β-galactosidase-positive cells compared to that among uninfected cells or cells infected with colibactin-defective mutants (Clb-Cnf+, Clb-Cnf- and Clb-Cnf- +pBK-cnf) (Figure 3a,b). Although senescence-associated β-galactosidase (SA-β-gal) activity observed with Clb-Cnf+ (11%) was much lower than that of Clb+Cnf+ (37%), this activity was significantly increased compared to uninfected cells (0.4%), suggesting that CNF1 induces senescence as has been shown in other models.⁴⁶ If we independently compare cells infected by Clb+Cnf- and Clb+Cnf+, the number of β-galactosidase-positive cells markedly increased when infected with Clb+Cnf- (72% vs 37%; p = 0.0022; Mann-Whitney test). The incubation of CNF1 with HT-29 cells during Clb+Cnf- infection also decreased the number of positive cells (Supplemental Figure S3c). P16 is a cell cycle gene that negatively regulates cell proliferation and is involved in pathways regulating senescence-mediated arrest. The number of p16-positive cells markedly increased when cells were infected with Clb+Cnf- compared to that observed in Clb+Cnf+ infected cells (Supplemental Figure S5). These results showed that the presence of CNF1 limited CoPEC-induced cellular senescence.

Figure 3. The cnf1 gene decreases CoPEC-induced cellular senescence and IL-8 production. HT-29 cells were infected for 3.5 h. (a) on Day 3 post-infection, senescent cells were detected by β-galactosidase staining at pH 6. Representative images are shown (b) the percentage of senescence-associated β-galactosidase-positive cells was determined. The data points represent values the mean of fields with 100 to 200 cells for each individual cell wells (n = 3). Data are representative of two independent experiments and values are represented in mean ± SEM. (c) Heatmaps showing the relative rates of senescence-associated factors secreted by cells infected with Clb+Cnf+ or with the Clb+Cnf- mutant. Control values were set to 0 (negative control) and 100 (positive control). Orange-red indicates the predominant secreted factors. (d) IL-8 amounts secreted in culture supernatant by cells infected with Clb+Cnf+ or with the Clb+Cnf- mutant. The quantification of IL-8 was performed by ELISA. Data are representative of two independent experiments from three different conditioned media. Values represent means ± SEMs. Statistical analysis was performed by Kruskal–Wallis and Mann–Whitney tests (*P <0 .05, **P <0 .01, ****P < 0.0001).

Next, we analyzed the senescence-associated secretory phenotype (SASP), which is known to underlie the pro-proliferative effect of colibactin.²² As expected, conditioned medium derived from cells infected with Clb+Cnf+ or Clb+Cnf- enhanced the proliferation of uninfected cells compared with conditioned medium derived from cells infected with Clb-Cnf+ or Clb-Cnf- (Supplemental Figure S6). However, we observed no significant difference in the pro-proliferative effect of conditioned medium derived from cells infected with Clb+Cnf+ or Clb+Cnf-, showing that cnf1 did not modify the pro-proliferative effect mediated by CoPEC-induced cellular senescence in vitro (Supplemental Figure S6). Conditioned medium derived from HT-29-infected cells was then probed using an antibody array targeting 72 senescence-associated secreted factors (Figure 3c). In agreement with the results regarding the pro-proliferation effect, the production of growth factors was similar in the cells infected with Clb+Cnf+ and Clb+Cnf-. Interestingly, IL-8 production levels were the highest in cells infected with the Clb+Cnf- mutant. The difference in IL-8 production in the Clb+Cnf- infected cells was confirmed by ELISA (Figure 3d).

Overall, the presence of cnf1 did not significantly modify the pro-proliferative effect of the colibactin-induced SASP in uninfected cells. However, it affected the induction of senescence mediated by colibactin and induced a subtle modification of the SASP, notably a decrease in the secretion of the proinflammatory cytokine IL-8.

The E.coli 21F8 strain lacking cnf1 induces an increase in colonic inflammation and senescence in Apc^Min/+ mice

Given the differences in senescence and SASP observed in our in vitro assays, we investigated inflammatory responses in infected Apc^Min/+ mice. We analyzed the expression of several pro-inflammatory factors by qRT – PCR in the colon of Apc^Min/+ mice. Pro-inflammatory gene mRNA levels, including those of KC, the murine homolog of human IL-8, were significantly higher in mice inoculated with Clb+Cnf- than in those inoculated with Clb+Cnf+, Clb-Cnf+, or Clb-Cnf- (Figure 4a). These results were corroborated by histological analyses of colonic biopsies from Apc^Min/+ mice. Colon sections from the Clb+Cnf- inoculated Apc^Min/+ mice showed submucosal edema and cellular infiltration (neutrophils and mononuclear cells), whereas colon sections from the Clb+Cnf+-inoculated Apc^Min/+ mice showed only few inflamed areas with weak inflammatory cellular infiltrate (Figures 4b and Supplemental Figure S7). Accordingly, the colonic inflammation score was significantly increased in Apc^Min/+ mice inoculated with Clb+Cnf- compared to that in those inoculated with Clb+Cnf+ or mutants defective in colibactin production (Clb-Cnf+ and Clb-Cnf-) (Figure 4c). The degree of inflammation induced by Clb+Cnf- remained low, with no ulcers or extensive crypt damage. The increase in inflammation was in accordance with the increase in the tumor number observed in the mice inoculated with the Clb+Cnf- strain compared to that of the mice inoculated with the Clb+Cnf+ and Clb-Cnf+ mutants (Figure 1b). We observed a significant positive correlation between the inflammation score and the number of colonic tumors in the Apc^Min/+ mice (Figure 4d), suggesting that CNF1 decreased colibactin-mediated colon tumorigenesis by inhibiting inflammation.

Figure 4. The cnf1 gene limits CoPEC-induced colonic inflammation in Apc^Min/+ mice. Apc^Min/+ mice were orally administered 10⁹ colony-forming units (CFU) of wild-type E. coli 21F8 (Clb+cnf+) or 10⁹ (CFU) of its isogenic mutants: Clb+Cnf-, Clb-Cnf+ and Clb-Cnf-. Mice were killed on Day 50 after administration. (a) Kc, Il-6, Tnf-α, and Il-1β mRNA levels in the colonic mucosa were quantified by Qrt – PCR. The data points represent values for each individual mouse. Data are presented as means ± SEMs. (b) Representative images of H&E-stained colonic sections showing submucosal edema and inflammatory cell infiltration. (c) Inflammation score for each individual mouse, with the bars indicating median values. (d) Correlation between the inflammation score and the number of tumors [Clb+cnf+ (++, n = 10); Clb+Cnf- (+ -, n = 10); Clb-Cnf+ (- +, n = 7) and 7 Clb-Cnf- (- -, n = 10)]. The given r values indicate Spearman’s rank correlation, and the P value represents the significance of the test result. Statistical analysis was performed by the Kruskal–Wallis test (*P < 0.05, **P < 0.01, ***P <0 .001, ****P < 0.0001).

Next, we investigated whether the presence of the cnf1 gene limited CoPEC-induced senescence in vivo. A PCR array designed to analyze a panel of 84 genes associated with senescence was performed with RNA extracted from colonic biopsy samples of mice colonized by Clb+Cnf+ or by its mutant Clb+Cnf-. Twelve genes were ≥ 2-fold upregulated in the Clb+Cnf- inoculated group compared to the group inoculated with Clb+Cnf+ (Figure 5a). qRT – PCR tests confirmed the upregulation of the most deregulated genes (Figure 5b), including Creg-1, an enhancer of the p16^INK4a-dependent senescence pathway.⁴⁷ Accordingly, there was a significant decrease in p16^INK4a-positive cells in the murine colonic tissues colonized with Clb+Cnf+ compared to those in the murine colonic tissues colonized with Clb+Cnf- (Figure 5c), with only a few p16^INK4a-positive cells detected in the colonic epithelium of the mice inoculated with the mutants defective in colibactin production (Figure 5d). Overall, these results suggested that CoPEC induces senescence in the colon of Apc^Min/+ mice and that CNF1 limits this process.

Figure 5. The cnf1 gene limited CoPEC-induced senescence in the colon of Apc^Min/+ mice. Apc^Min/+ mice were orally administered 10⁹ colony-forming units (CFU) of wild-type E. coli 21F8 (Clb+Cnf+) or 10⁹ (CFU) of its isogenic mutants: Clb+Cnf-, Clb-Cnf+ and Clb-Cnf-. Mice were killed on Day 50 after administration. (a) Scatter plot of differential gene expression in the colonic mucosa of mice inoculated with Clb+Cnf- and Clb+Cnf+ determined using a cellular senescence RT2 Profiler PCR Array (the conditioned medium used was a mix of 3 replicates of mice infected by Clb+Cnf+ or Clb+Cnf- strains). The yellow circles show ≥ 2-fold upregulated genes in the mice inoculated with Clb+Cnf- compared to those in mice inoculated with Clb+Cnf+. (b) Relative Rbl1, Map2k6, Creg1 and Terf2 mRNA levels quantified by Qrt – PCR (RT2 Profiler PCR) in the colonic mucosa of mice inoculated with Clb+Cnf- or Clb+Cnf+. The data points represent values for each individual mouse (c) Quantification of p16^INK4a-positive cell number/10 villi determined from 100 villi/mouse and 5 mice/group. Data are presented as means ± SEMs. Statistical analysis was performed using the Kruskal–Wallis test (*P < 0.05, **P <0 .01, ****P <0 .0001). (d) Representative images of immunohistochemical expression of p16^IN4a in Apc^Min/+ colonic mucosa of infected mice. The arrowheads show positive cells (scale bars: 20 µm).

We investigated the fecal microbiota composition by 16S rRNA gene sequencing. In addition to increases CoPEC-induced cellular senescence and inflammation, the deletion of cnf1 gene was associated with a change in intestinal microbiota composition (these results are detailed in the supplementary data).

The CNF1 toxin impairs the development of colon tumors in Apc^Min/+ mice

To confirm the importance of CNF1 in preventing the tumorigenic activity of colibactin, we investigated the impact of weekly rectal administration of the CNF1 protein on the development of colonic tumors in Apc^Min/+ mice colonized by pro-tumorigenic Clb+Cnf-. We observed no difference in intestinal colonization by Clb+Cnf- between the mice treated with CNF1 and the mice treated with a saline solution (PBS) (Supplemental Figure S8). The mice treated with CNF1 had significantly fewer tumors than the PBS-treated mice (Figure 6a). The decrease in tumor number was associated with a reduction in both inflammation and senescence marker levels (Figure 6b–e). Inflammatory cell infiltration and submucosal edema were significantly less pronounced in mice treated with CNF1 than in PBS-treated mice (Figure 6d). Accordingly, there was a significant decrease in the histological colonic inflammation score of the Apc^Min/+ mice colonized with E. coli Clb+Cnf- and treated with CNF1 in comparison to that of PBS-treated mice (Figure 6c). As shown in (Figure 6f, g) the histological inflammation score and KC rate were significantly and positively correlated with the number of tumors in the Apc^Min/+ mice. The mice with the highest number of tumors and the highest level of colonic inflammation were those that did not receive CNF1. This observation was found with Apc^Min/+ mice colonized by pro-tumorigenic 11G5 (Supplemental Figure S9). Overall, these results demonstrated that CNF1 limited the development of colonic tumors in CoPEC-infected Apc^Min/+ mice, like by decreasing senescence and/or chronic low-grade inflammation induced by CoPEC.

Figure 6. In CoPEC-infected Apc^Min/+ mice, intrarectal administration of CNF1 impedes the development of colon tumors and decreases inflammation. Apc^Min/+ mice were orally administered 10⁹ colony-forming units of the E. coli 21F8 mutant producing colibactin but defective in cnf1. Two days post-infection, the mice received an intrarectal injection of 10 µg of CNF1 protein or PBS every 7 days for 7 weeks. (a) the number of colorectal tumors by mouse was determined using a dissecting microscope. The data points represent actual values for each individual mouse, and the bars indicate median values. (b) the levels of secreted cytokines (KC, IL-6 and TNF-α) in colonic tissue were quantified by ELISA. (c) Inflammation score is presented as the mean ± SEM. (d) Representative images of H&E-stained colonic mouse sections of PBS- or CNF1-treated mice. The red arrow shows submucosal edema. e) Rbl1 and Map2k6 mRNA relative levels in colonic mucosa were quantified by Qrt – PCR (RT2 Profiler PCR). (f) Correlation between the inflammation score and the tumor number. (g) Correlation between the KC levels in colonic tissue and the tumor number. Blue dots represent mice treated with PBS (n = 5) and red dots represent those treated with CNF1 (n = 5). Statistical comparisons were carried out by unpaired t test (*P <0 .05) after normality testing. Spearman correlation analysis was performed between the inflammation score or KC concentration and the tumor number. The given r values indicate Spearman’s rank correlation, and the P value represents the significance of the test result.

Discussion

Colibactin and CNF1, two common toxins of E. coli that affect the eukaryotic cell cycle, are putative pro-tumorigenic factors.^{20,22,29,36,37,48} Our epidemiological data demonstrated that CNF1 is almost always associated with colibactin in E. coli strains isolated from CRC patients. We thus hypothesized that coproduction of these two cyclomodulins might enhance colorectal carcinogenesis. However, our results show that a human CoPEC strain coproducing colibactin and CNF1 did not exhibit pro-tumorigenic activity in Apc^Min/+ mice, unlike human E. coli strain 11G5, which produces only colibactin. We constructed isogenic mutants defective for CNF1 and/or colibactin production from the clinical strain 21F8. We observed independent impacts of CNF1 and colibactin on eukaryotic cells in terms of cytopathic effects that were in agreement with their production from independent genetic structures.⁴⁹ Interestingly, isogenic deletion of cnf1 in 21F8 revealed the pro-tumorigenic activity of the strain, and rectal administration of CNF1 in Apc^Min/+ mice colonized by the 21F8 mutant defective in cnf1 decreased colonic tumor development. Counterintuitively, these results revealed that CNF1 hinders tumorigenesis induced by colibactin-producing E. coli in the Apc^Min/+ CRC model. Further analyses highlighted the underlying mechanisms.

The carcinogenic effect of CNF1 has previously been linked to its ability to promote Rho GTPase-dependent cellular effects, proinflammatory NF-kB pathway activation, cell growth and apoptosis suppression, tumor invasiveness, epithelial-to-mesenchymal transition (EMT) and metastasis.^{30,32,36,37,48,50,51} However, in vitro and in vivo studies have demonstrated the anti-proliferative and cytotoxic effects of CNF1 in cancer cell lines,^52–55 suggesting a two-sided paradigm of CNF-1 impact.⁴⁸ In the present study, isogenic E. coli 21F8 producing only CNF1 did not induce metastasis or enhance colonic tumorigenesis in the Apc^Min/+ murine model of colon cancer. CNF1 behaved therefore less as an anti-tumorigenic factor than as a protective factor against colibactin.

We observed both in vitro and in vivo that CNF1 reduced colibactin-induced (i) inflammatory cell infiltration, (ii) senescence, and (iii) senescence-associated secretion of the key pro-inflammatory cytokine, KC/IL-8. Senescence and the resulting SASP have been identified as cellular processes sustaining tumor development in a xenograft model.^22,56 Chronic inflammation is known to be an important risk factor for numerous forms of cancer, including CRC,⁵⁷,and inflammation has also been identified as a key player in colibactin tumorigenic activity in vivo.^16,22,58 In addition, several studies have implicated IL-8 in the progression of various types of cancer,^59–63 including CRC.^64,65 CNF1 has pro-inflammatory activity in acute infections, such as urinary tract infections.⁶⁶ In contrast, it may also counteract the overexpression of proinflammatory cytokines such as IL-8 during chronic colonization, as reported by Loizzo et al. in the context of the chronic inflammation associated with Alzheimer’s disease.⁶⁷ Therefore, the impact of CNF1 on colibactin-induced tumorigenesis can be explained in the context of the tissue organization field theory. According to this theory, alterations in tissue organization by cellular processes such as inflammation or senescence lead to carcinogenesis.⁶⁸ Additionally, the bacterium-engulfing activity of CNF1, linked to the activation of Rho GTPases by deamidation,^50,69 could contribute to the observed phenotype. Upon expressing CNF1, the bacteria may acquire invasive capacities and thereby shelter from the host immune system, generating less inflammation.^29,69 Thus, CNF1 appears to function as a protective factor against colibactin, impeding the emergence of a microenvironment and inflammatory cell infiltration promoting tumorigenesis in the Apc^Min/+ mouse model.

On the other hand, double-strand DNA breaks are the primary effect of colibactin and induce oncogenic mutations in human CRC.^18,20 Our results showed that the presence of cnf1 gene under normal conditions, i.e when not overexpressed, does not modify the intensity of colibactin-induced DNA damage. However, the capacity to repair injury may be different in presence of CNF1. In this regard, the level of γH2AX was considerably reduced when the cells were infected with the Clb+Cnf- strain trans-complemented with cnf1. Rho GTPases such as Rho and Rac proteins, that are involved the regulation of DNA repair systems,⁷⁰ are the target of CNF1, which induces their constitutive overactivation through the deamination of a specific glutamine residue in the infected cells. A high Rho GTPases activity have been directly correlated with a high level of DNA repair, and inhibition of Rho GTPases dramatically reduces γH2AX and the formation of DNA damage foci.⁷⁰ Furthermore, CNF1-induced Rac1 activation positively regulates RhoB expression cultures of epithelial cell lines, including HT-29 cells.⁷¹ CNF1 may therefore enhance DNA repair and then weaken the carcinogenic effect of colibactin.

In addition, there is emerging evidence for a cancer suppressive role for RhoB through inhibitory effects on cell proliferation, survival, invasion and metastasis.⁷² These in vitro observations were supported by in vivo findings. RhoB-depleted cells form tumors more efficiently than cells expressing RhoB when injected intraperitoneally into mice.⁷³ Cells transfected with RhoB and subcutaneously implanted into nude mice suppress tumor growth.⁷⁴ CNF1-induced activation of Rho GTPases may thereby hinder colon tumorigenesis induced by colibactin-producing E. coli. However, Rho GTPases interact with a wide range of effectors and cellular signaling cascades. Their role in cancer is dependent of cellular context and they can also contribute to tumor formation.⁷² Further studies are required to decipher the features that determine the impact of Rho GTPases in carcinogenesis.

The findings of this study underline the strong epidemiological link between CNF1 and colibactin in E. coli strains associated with human CRC. In this study, more than 90% of E. coli strains harboring the cnf1 gene carried the genomic island pks producing colibactin. This finding is in accordance with previous reports that demonstrated this association in uropathogenic E. coli (83%) and E. coli isolated from fecal samples of different animals (62–100%).^41,75,76 Thus, the association of CNF1 and colibactin is common in E. coli, which is the predominant aerobic organism observed in the gastrointestinal tract and common in the human intestinal microbiota (i.e., >90%).⁷⁷ The prevalence of patients colonized with strains of E. coli bearing both the pks genomic island and cnf1 gene (Table 1) was not significantly higher in CRC patients than in controls (24%, n = 19/80 versus 14%, n = 4/28; p = 0.432). Thus, the frequent association of CNF1 with colibactin in E. coli, an extremely common bacterium in the gut, does not result in higher incidence of CRC.

In conclusion, we found that the pks island and cnf1 gene are frequently co-harbored in E. coli. Our work showed for the first time that CNF1 hinders CoPEC-induced colorectal carcinogenesis by decreasing CoPEC-induced cellular senescence and inflammation. The presence of E. coli strains producing only colibactin might represent a higher risk of CRC than the presence of strains producing both CNF1 and colibactin. Another finding that emerges from this study is the need to consider the genetic diversity of bacteria colonizing CRC patients and especially their virulome to understand microbiota-induced carcinogenesis and to determine whether the bacterium is deleterious to the host.

Materials and methods

Information on the bacterial strains, cell culture, infection assays, CoPEC colonization, quantitative reverse transcription polymerase chain reaction (qRT – PCR), antibody array chips, enzyme-linked immunosorbent assays (ELISAs), senescence-associated β-galactosidase staining, immunofluorescence microscopy, histological observations, and immunohistochemical staining appears in the Supplementary materials.

Bacterial strains and construction of isogenic mutants

The clinical E. coli 11G5 and 21F8 strains were isolated from tumors of patients with CRC. The following isogenic mutants of the 21F8 strain were generated using the method described by Datsenko et al . ⁷⁸ and modified by Chaveroche et al. ⁷⁹ 21F8Δcnf (Clb+Cnf-) with deletion of the cnf1 gene, 21F8ΔclbQ (Clb-Cnf+) with deletion of the clbQ gene of the pks island and 21F8ΔcnfΔclbQ (Clb-Cnf-) with deletion of the cnf1 and clbQ genes. The ClbQ thioesterase regulates colibactin synthesis and consequently its genotoxic activity. CoPEC strains deficient in clbQ are unable to produce functional colibactin.²² In brief, the method consisted of the replacement of the gene of interest by a selective antibiotic cassette (kanamycin) generated by PCR using primers reported in the Table S1 from supplementary materials. The hemolysin A (hlyA) was deleted from the 21F8 strain and its isogenic mutants with the same method to avoid lysis of the HT-29 cells. The presence of deletions and the absence of additional genetic modifications were verified by sequencing the clinical 21F8 strain and its isogenic mutants. With the In-Fusion^Ⓡ HD Cloning (Takara), cnf1 gene was cloned into the pBK-CMV plasmid (Table S1). The 21F8ΔhlyAΔcnf and 21F8ΔhlyAΔcnfΔclbQ was electroporated with sequenced pBK-CMV-cnf1 plasmid. For experiments, strains were growth in Luria-Bertani (LB) broth overnight at 37°C with 110 rpm agitation. All the strains for this study were summarized in Table S2.

Colonization of the Apc^Min/+ murine model

C57BL/6 Apc^Min/+ females (6–7 weeks of age) were used. The mice were inoculated as previously described.²³ All the mice were sacrificed 50 days post-infection. Colonic tumor number and tumor volume ([width² × length]/2) were determined using a dissecting microscope. Colonic tissue adjacent to tumors were fixed in buffered 4% formalin and embedded in paraffin. Non-tumoral colonic mucosa was frozen at −80°C for protein and RNA extraction. For the experiment employing rectal administration of the CNF1 protein, Apc^Min/+ mice were inoculated as previously described²³ and then received an intrarectal injection of 10 µg of CNF1 protein or PBS under anesthesia with isoflurane. The injections were administered 2 days post-infection and then once per week for 7 weeks. CNF1 was purified as described previously.⁵¹

Ethical statement

Animal protocols were in accordance with French and European Economic Community guidelines (86–60, EEC) for the care of laboratory animals. The study was approved by the French Ministry of Higher Education Research and Innovation (Apafis no. 22798).

Biological samples were collected from CRC patients (ethical approval for human study no. DC-2017–2972). All patients underwent surgery for resectable CRC in the Digestive and Hepatobiliary Surgery Department of the University Hospital of Clermont-Ferrand.³⁸ All patients were adult volunteers and signed an informed consent form before inclusion in the study. The exclusion criteria included clinically suspected hereditary CRC based on the revised Bethesda criteria, neoadjuvant chemotherapy receipt, a history of previous colonic resection, emergency surgery, and use of antibiotics within 4 weeks before the surgery.

RT² Profiler PCR Array

Eighty-four genes or biological pathways involved in cellular senescence were analyzed using the RT² Profiler PCR Array Mouse Cellular Senescence system (PAMM-050Z; Qiagen, Maryland, USA). According to the manufacturer’s protocol, real-time PCR was performed using RT² Profiler PCR Arrays in combination with RT² SYBR Green PCR Master Mix (Qiagen, Maryland, USA) using a mixture of cDNA obtained from three colonic biopsy samples of mice colonized by 21F8 (Clb+Cnf+) or its mutant Clb+Cnf-. The expression levels of the 84 genes were quantified relative to the values obtained for housekeeping genes (ACTB, B2M and GAPDH). Data analyses were performed using web-based analysis software (http://pcrdataanalysis.sabiosciences.com/pcr/arrayanalysis.php). In addition, we further performed a similar RT² Profiler PCR assay using custom plates including four genes (RBL1, MAP2K6, CREG1 and TERF2) in addition to the housekeeping genes (ACTB, B2M and GAPDH). The analysis included five or six mice in each group (mice with intrarectal injection of CNF1 protein or PBS or mice colonized by Clb+Cnf+, Clb+Cnf-, Clb-Cnf+ or Clb-Cnf-).

Statistical analysis

GraphPad Prism software was used for all statistical calculations. Data comparisons with multiple groups were analyzed by one-way Kruskal – Wallis test. Data comparisons between 2 groups were performed with unpaired t test or a Mann-Whitney U-test depending on the normality test. Spearman’s correlation analysis was performed for correlation testing. A value of P <0 .05 was considered to indicate a statistically significant difference.

Acknowledgments

We thank Laurent Guillouard for providing technical assistance. We thank the CLIC (Clermont-Ferrand Imagerie Confocale, Université Clermont Auvergne) with help from Caroline Vachias and Anipath histology technical platforms (GReD, Université Clermont Auvergne) for assistance with tissue preparation and immunohistochemical staining. We also thank Christelle Blavignac from the CICS for cell cycle analysis.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available in Mendeley data at http://doi.org/10.17632/zwck97zw4k.1.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2023.2229569.

Additional information

Funding

This study was supported by the Ministère de la Recherche et de la Technologie; Inserm (UMR 1071); INRAe (USC-1382), Région Auvergne Rhône Alpes; the French government’s IDEX-ISITE initiative 16-IDEX-0001 (CAP 20-25 project of the University of Clermont Auvergne); and the National Program “Microbiote” Inserm. ”