Surface expression of antitoxin on engineered bacteria neutralizes genotoxic colibactin in the gut

Abstract

Colibactin, a metabolite produced by gut bacteria carrying the polyketide synthase (pks) island, is associated with host genotoxicity and tumorigenesis. However, no Food and Drug Administration-approved therapeutics directly target colibactin. Here we show that expression of the intracellular colibactin self-resistance protein (ClbS) on the surface of engineered bacteria shields the host from genotoxic effects across multiple pks⁺ isolates. The surface display, due to the fusion of ClbS with outer membrane protein A (ClbS–OmpA) in Escherichia coli, effectively reduced colibactin-induced DNA damage and cell cycle arrest in human cell lines and organoids, outperforming D-serine, a small-molecule inhibitor of colibactin synthesis. The engineered strains mitigated intestinal damage in a mouse model of colitis and suppressed tumorigenesis in mouse models of colon cancer caused by pks⁺ E. coli. Our results show the feasibility of inhibiting bacterial genotoxins in the gut, establishing a starting point for therapeutics targeting other potential cancer-causing bacterial metabolites.

Microbiome research has increasingly highlighted the contributions of individual microbiota members to host health and disease. These links were originally inferred from correlations between the abundance of certain bacteria and disease pathogenesis^1–3. However, such correlative studies often fail to pinpoint causality between individual bacterial species and host processes. Accumulating evidence suggests that microorganisms influence host physiology and pathology in part through microbial genotoxins. Among bacterial genotoxins that have been identified, colibactin has garnered increasing attention owing to its implication in colorectal cancer (CRC) as well as its impact on the composition and functions of the microbiome⁴. Colibactin is a hybrid polyketide-nonribosomal peptide produced by Escherichia coli, Klebsiella pneumoniae and other Enterobacteriaceae harbouring a 54-kb polyketide synthase (PKS) gene cluster, referred to as the pks island. Since its first report in 2006⁵, various studies have shown its linkage to CRC^6–9. Further evidence regarding the genotoxic effects of colibactin has been shown in preclinical mouse CRC models and human intestinal organoids^10,11. However, progress in targeting pks⁺ bacteria has been primarily limited to manipulating and characterizing individual knockout bacterial strains in cell culture or germ-free mice. These reductionist approaches overlook the fact that, in a native environment, multiple different enteric bacteria produce colibactin to impact the host through the acquisition of the conserved pks island. From the translational perspective, the growing recognition of colibactin’s role in CRC and other diseases has spurred the development of a few small molecules aimed at targeting colibactin biosynthesis^12–15. However, no Food and Drug Administration-approved inhibitors directly targeting colibactin are currently available, and alternative strategies to directly neutralize colibactin remain limited. Previously, a colibactin self-resistance protein (ClbS) was identified as an antitoxin mechanism for pks⁺ bacteria to prevent colibactin-induced self-DNA damage^16,17. Here we repurposed this toxin–antitoxin system using a microbial platform to directly neutralize colibactin in complex environments (Extended Data Fig. 1). Inspired by the bacterial surface display of enzymes as whole-cell biocatalysts to enhance enzyme stability and catalytic efficiency in industry applications¹⁸, we showed the ability of surface-displayed ClbS to suppress the genotoxicity of various pks⁺ isolates in multiple cell lines and colon organoids with human and mouse origins. Importantly, this approach suppressed intestinal DNA damage in the colons of wild-type mice and decreased tumorigenesis in a colitis-associated mouse CRC model by inhibiting the colonization and procarcinogenicity of pks⁺ bacteria.

Results

Surface antitoxin neutralizes colibactin genotoxicity

Previous studies showed that colibactin-induced DNA damage requires direct contact between pks⁺ bacteria and host cells, probably owing to the compound’s intrinsic instability^19,20. Inspired by this, we hypothesized that displaying ClbS, the colibactin-inactivating enzyme, on the surface of commensal E. coli could enable localized neutralization of colibactin near pks⁺ bacteria and host cells. To test this, we sought to identify an optimal outer membrane protein (OMP) scaffold to present ClbS on the bacterial surface. We screened three commonly used bacterial OMPs: lipoprotein fused with truncated E. coli outer membrane protein A (Lpp–OmpA, hereinafter OmpA)²¹, C terminal of IgA proteinase (C-IgAP)²² and N terminal of EaeA (Neae)²³ (Fig. 1a and Supplementary Fig. 1). ClbS was efficiently displayed on the surface when fused to the C-terminus of OmpA or Neae (denoted as Ec^OmpA–ClbS or Ec^Neae–ClbS, respectively), but not when fused to the N-terminus of C-IgAP (Ec^{ClbS–C-IgAP}) (Fig. 1b). To isolate the effect of ClbS display, we also constructed control strains expressing OMP scaffolds alone (Ec^OmpA, Ec^Neae and Ec^C-IgAP). We then cocultured these engineered strains with HeLa cells infected with pks⁺ DH10B E. coli harbouring a bacterial artificial chromosome encoding the pks island (DH10B pBAC-pks) or an empty vector (Δpks) that were previously developed by others^5,16,24. Subsequently, bacteria displaying ClbS fusion proteins or the OMP alone were cocultured with HeLa cells infected by DH10B pBAC-pks or the empty vector control (Fig. 1c). To assess DNA damage, we used gamma-H2A histone family member X (γH2AX) staining, a widely adopted method for detecting double-strand DNA breaks caused by pks⁺ bacteria²⁰. As evidenced by γH2AX staining, compared with its control strain Ec^OmpA, Ec^OmpA–ClbS reduced the γH2AX signal by ~60%, indicating effective neutralization of colibactin-mediated genotoxicity. By contrast, Ec^Neae–ClbS did not confer protection (Fig. 1d), despite showing comparable surface display levels to Ec^OmpA–ClbS (Fig. 1b). On the basis of these findings, we concluded that both the display level and protective activity of ClbS depended on the OMP scaffold used. For the remainder of this study, we selected OmpA as the lead scaffold to characterize the surface-displayed ClbS.

Fig. 1 |. Identifying E. coli OmpA as the lead OMP to display ClbS on the bacterial surface for mitigating the genotoxicity of pks⁺ E. coli.

a, A schematic of screening three common OMPs to display ClbS on the E. coli surface. OM, outer membrane; IM, inner membrane. b, Verification of surface-displayed ClbS using three common bacterial OMPs. c, Schematic of coculture experiments to assess DNA damage in host cells by quantifying the levels of phosphorylated-H2AX (γH2AX). d, Quantification of γH2AX in HeLa cells infected with DH10B pBAC-pks (depicted as pks⁺) or pBAC empty vector (Δpks) in the presence of Ec^OmpA–ClbS, Ec^Neae–ClbS, Ec^{ClbS–C-IgAP} or corresponding negative controls (Ec^OmpA, Ec^Neae and Ec^C-IgAP) (n = 3 for each group). DNA damage was quantified as genotoxic indexes, normalizing γH2AX levels to those in control cells. The MOI of DH10B pBAC-pks or pBAC empty vector is 10. The MOI of K-12 strains is 400. Panels b and d show representative data from three independent biological replicates. For d, data are means ± s.d.; two-sided unpaired t-test was used.Panels a and c created with BioRender.com.

Enhanced antitoxin display blocks DNA damage and cell cycle arrest

After identifying OmpA as the lead anchor protein, we sought to enhance the surface display levels of ClbS to improve the anti-colibactin function by increasing the plasmid copy number from a medium-copy to a high-copy replication origin. This strategy markedly increased the surface levels of ClbS (Fig. 2a). We found that the improved surface display of ClbS completely suppressed the genotoxicity induced by DH10B pBAC-pks in HeLa cells (Fig. 2b). In addition to using OmpA alone as a negative control, we generated a catalytically inactive mutant of ClbS (Y55F) for surface display based on a previous study by others¹⁷. It was shown that the ClbS mutant failed to protect host cells from colibactin-induced DNA damage in cell culture (Supplementary Fig. 2). These results confirm that the observed protective effects are attributable to the enzymatic activity of ClbS, not merely its presence on the bacterial surface. Consistent results were further validated in other pks⁺ E. coli (MIT A2, MIT A21 and NC101), showing the general anti-colibactin efficacy provided by optimized K-12 Ec^OmpA–ClbS (ref. 25; Fig. 2b and Extended Data Fig. 2a–c).

Fig. 2 |. Enhancing surface display and mining ClbS homologues to optimize protection from genotoxicity and cell cycle arrest by pks⁺ E. coli.

a, Detection of ClbS on the surface of E. coli K-12 carrying a medium-copy or a high-copy plasmid. b, K-12 Ec^OmpA–ClbS from the high-copy plasmid suppressed genotoxicity in HeLa cells induced by pks⁺ NC101 or DH10B pBAC-pks (P < 0.0001). By contrast, Ec^OmpA did not confer any protection (P = 0.2953 and 0.6779). ‘Δpks’ refers to NC101 with the pks knocked out or DH10B carrying an empty vector, pBAC. c, A schematic of cell cycle analysis after coculture. d, Flow cytometry analysis of cell cycle arrest in HeLa cells infected with NC101 or NC101Δpks in the presence of Ec^OmpA–ClbS or Ec^OmpA (P = 0.0012, NC101/Ec^OmpA–ClbS versus NC101/Ec^OmpA). A lower percentage of cells characterized by ‘2N (diploid) DNA content’ in the G1 phase corresponds to higher cell cycle arrest. e, Surface display of different ClbS homologue proteins reduced colibactin-mediated host DNA damage. The heat map and clustering of the ClbS homologue proteins are based on amino acid identity relative to pks⁺ E. coli ClbS. An MOI of 50 (NC101) or 20 (DH10B) was used, and an MOI of 400 was used for surface-displayed ClbS homologues or OmpA only (P < 0.0001, OmpA versus OmpA–E. coli–ClbS). Panels a, b, d and e show representative data from three independent biological replicates. For b, d and e, data are means ± s.d.; two-sided unpaired t-test was used.

To further highlight the benefits of our optimized E. coli strain displaying ClbS, we compared it with D-serine, a small molecular inhibitor against colibactin biosynthesis with documented safety in other preclinical diseases^12,26,27. Notably, D-serine failed to reduce the genotoxicity by DH10B pBAC-pks in HeLa cells while Ec^OmpA–ClbS completely suppressed the genotoxicity (Extended Data Fig. 2b). This observation agrees with the finding of a previous study that the inhibitory activity of D-serine against colibactin requires factors unique to natural colibactin-producing isolates¹². Supporting this hypothesis, our results showed that D-serine did not completely prevent the genotoxic effects induced by the three natural pks⁺ isolates (MIT A2, MIT A21 and NC101) (Extended Data Fig. 2c). Collectively, we have shown that the surface-displayed ClbS in Ec^OmpA–ClbS can efficiently protect host cells from colibactin-mediated genotoxicity by various pks⁺ E. coli isolates, outperforming the small-molecule inhibitor D-serine.

As pks⁺ NC101 is known for its procarcinogenicity and adherent-invasive nature in mice^28–30, we focused on NC101 for further studies. In addition, while knocking out the entire pks may have other consequences compared with deleting or mutating an essential clb gene of pks in NC101 (refs. 29,31,32), we chose NC101Δpks (the full knockout of pks) as the negative control strain for evaluating the genotoxic effects from the pks island. In addition to γH2AX staining to examine DNA damage, cells exposed to colibactin also experience cell cycle arrest, which can be evaluated by staining nuclear DNA contents and conducting flow cytometry (Fig. 2c). The optimized ClbS display in Ec^OmpA–ClbS prevented the cell cycle arrest in HeLa cells. Consistent with previous findings^5,13, infection of HeLa cells with NC101 alone markedly reduced the percentage of cells characterized by 2N (diploid, G1 phase) DNA content compared with infection with the negative control NC101Δpks. However, the cell cycle arrest did not occur when HeLa cells were treated with NC101 and Ec^OmpA–ClbS, while Ec^OmpA failed to prevent the cell cycle arrest in HeLa cells infected with NC101 (Fig. 2d). These findings suggest that the surface-displayed ClbS mitigated genotoxicity and prevented cell cycle arrest in host cells infected with pks⁺ bacteria.

Surface localization of antitoxin is required for protection

To probe the interactions between bacteria and host cells, we fluorescently labelled Ec^OmpA–ClbS and pks⁺ E. coli with distinct fluorescent dyes and cocultured them with epithelial cells. Confocal microscopy revealed that Ec^OmpA–ClbS localized with pks⁺ E. coli and host cells at the microorganism–host interface (Supplementary Fig. 3a). To reliably distinguish intracellular and surface-adherent bacteria, we used a membrane-impermeable antibiotic, gentamicin, to validate the localization of bacteria and understand how the ClbS-displaying K-12 may affect the adhesion and invasion of the adherent-invasive pks⁺ E. coli, NC101. To this end, we used a chloramphenicol-resistant pks⁺ NC101, NC101^CmR. After coculturing bacteria with HeLa cells for 4 h, the majority of NC101^CmR remained surface bound with minimal invasion into epithelial cells (Supplementary Fig. 3b,c), which agreed with a similar study by others²⁰. Moreover, the adhesion and invasion of NC101^CmR were not affected by Ec^OmpA–ClbS or Ec^OmpA (Supplementary Fig. 3b,c). To further assess the importance of ClbS localization, we compared surface-displayed ClbS with cytoplasmic ClbS expression in E. coli K-12 and found that only the surface-displayed form provided protection against pks⁺ E. coli-induced DNA damage, as measured by γH2AX staining (Supplementary Fig. 4a,b). This highlights the critical role of surface localization in conferring protection from colibactin.

Antitoxin homologues mitigate genotoxicity

ClbS homologues have been identified in various distantly related bacteria²⁴, and the sequence divergence of these ClbS homologues may provide an opportunity to identify more efficient ClbS enzymes for surface display to neutralize colibactin. Therefore, we asked whether E. coli-derived ClbS remains the most efficient homologue. To answer this question, we codon optimized and encoded seven additional ClbS homologues on a high-copy plasmid, with these homologues sharing 30–80% amino acid identity with E. coli ClbS (Fig. 2e and Supplementary Fig. 5a,b). Interestingly, the surface display of different ClbS homologue proteins showed different efficiencies in reducing colibactin-mediated host DNA damage, with E. coli ClbS being the most effective one (Fig. 2e). Although it is possible to identify a more efficient ClbS enzyme by screening a larger library of ClbS homologues, our results, combining OMP screening, plasmid copy number optimization and ClbS homologue mining, support the continued use of E. coli ClbS for subsequent studies.

Antitoxin display lowers intrinsic and bystander genotoxicity

We next explored whether the surface display strategy can be extended to other E. coli chassis used for the delivery of biologics in preclinical models. Notably, probiotics such as E. coli Nissle 1917 (EcN), a phylogenetic lineage B2 strain, harbour the pks island encoding for colibactin biosynthesis^33,34. Although EcN is a popular chassis for live therapeutics, concerns have been raised about its potential to promote DNA mutations such as SBS88 in mammalian cells³⁵. Some researchers have circumvented this issue by knocking out genes in the pks island of EcN to eliminate colibactin production as the alternative chassis³⁶. However, the pks island is also associated with the probiotic properties of EcN³³. To determine whether surface-displayed ClbS in EcN could mitigate its genotoxicity without disrupting the pks island, we first confirmed the genotoxicity of EcN and another pks⁺ mouse E. coli isolate, NGF-1, by comparing the wild-type strains to their isogenic Δpks mutants (Extended Data Fig. 3a). Next, we introduced the optimized OmpA–ClbS construct into EcN (EcN^OmpA–ClbS) and confirmed the high-level expression of ClbS on the bacterial surface (Extended Data Fig. 3b). Subsequently, we performed a coculture consisting of HeLa cells, EcN^OmpA–ClbS and NC101Δpks (Extended Data Fig. 3c). Of note, NC101Δpks mimics a control E. coli strain lacking the pks island involved in the interplay with pks⁺ bacteria and the host cells. Interestingly, compared with EcN displaying OmpA alone (EcN^OmpA), which induced DNA damage in host cells, EcN^OmpA–ClbS had abolished genotoxicity despite the presence of the pks island in both EcN^OmpA–ClbS and EcN^OmpA. As EcN^OmpA–ClbS or EcN^OmpA was the only source of colibactin in the coculture with HeLa and NC101Δpks, surface-displayed ClbS abrogated the intrinsic genotoxicity of EcN (Extended Data Fig. 3d). In addition, when NC101Δpks was replaced with NC101 to generate a coculture containing HeLa cells and two different pks⁺ bacteria with one strain displaying ClbS (that is, EcN^OmpA–ClbS), we showed that EcN^OmpA–ClbS further protected host cells from NC101-induced genotoxicity, suggesting bystander protection against other pks⁺ bacteria (Extended Data Fig. 3d). Nevertheless, the surface-displayed ClbS was less efficient in EcN than K-12 at preventing NC101-induced genotoxicity at higher multiplicities of infection (MOIs; the ratio of NC101 to HeLa). While both EcN and K-12 strains displaying ClbS suppressed NC101-induced genotoxicity in HeLa cells at an MOI of 25, only K-12 maintained complete protection at higher MOIs (that is, increased concentrations of NC101) (Supplementary Fig. 6a,b). In addition to EcN, the dual protective effects were also observed in ClbS-displaying NGF-1, a pks⁺ strain isolated from a healthy BALB/c mouse^37,38 (Extended Data Fig. 3d). In summary, while certain E. coli strains (EcN and NGF-1) show in vitro genotoxicity owing to the pks island, the surface display of ClbS in these strains can block their intrinsic genotoxicity and decrease DNA damage in host cells infected with other pks⁺ bacteria. However, the lower efficacy of ClbS in pks⁺ EcN probably stems from the need to neutralize its own colibactin and colibactin produced by NC101. By contrast, ClbS in K-12, which lacks the pks island, is required only to neutralize colibactin from NC101 (Supplementary Fig. 6a,b). Therefore, K-12 was the preferred chassis for subsequent studies.

Antitoxin-displaying E. coli protect CRC cell lines and colonoids

To evaluate our approach in intestinal epithelial cells, we established a coculture model consisting of K-12 Ec^OmpA–ClbS, NC101 and various CRC cell lines (human: HT-29, Caco-2 and HCT-116; murine: MC38), using K-12 Ec^OmpA and NC101Δpks as negative controls.

Across all CRC cell lines, Ec^OmpA–ClbS markedly reduced genotoxicity and restored normal cell cycle profiles compared with Ec^OmpA controls, as shown by both γH2AX staining and cell cycle analysis (Fig. 3a and Supplementary Fig. 7a,b). These results show that ClbS display alleviates colibactin-induced DNA damage and its downstream consequences on cell cycle arrest.

Fig. 3 |. E. coli displaying ClbS mitigate the genotoxicity of pks⁺ E. coli in CRC cell lines and human colonoids.

a, Ec^OmpA–ClbS protect colon cancer cell lines (HC116, Caco-2 and MC38) from DNA damage. An MOI of 50 (for Caco-2 and MC38) or 100 (for HCT116) was used for pks⁺ or Δpks NC101, and the MOI for Ec^OmpA–ClbS or Ec^OmpA is 400. b, A schematic describing the workflow of coinfection of human colonoids and immunostaining. c, A representative image of human colonoids after recovery from bacterial coinfection. Scale bar, 100 μm. d,e,Representative confocal fluorescence images (d) and quantification of γH2AX staining (e) in human colonoids infected with NC101 or NC101Δpks (negative control) in the presence of Ec^OmpA–ClbS or Ec^OmpA (P = 0.0002, NC101/Ec^OmpA–ClbS versus NC101) for 3 h. Cisplatin (50 μM) was included as a positive control for the induction of DNA damage in colonoids. The MOI for NC101 is 10. The MOI for Ec^OmpA–ClbS or Ec^OmpA is 400. Scale bars, 20 μm (d). Panels a, d and e show representative data from three independent biological replicates. Two-sided unpaired t-test was used for a, and one-way ANOVA test followed by one-sided Tukey’s multiple-comparison test was used for e. For a and e, data are means ± s.d. MFI, median fluorescence intensity; ctl, positive control. Panel b created with BioRender.com.

To evaluate this approach in a more physiologically relevant system, we used human colonoids, which offer a robust reductionist model for studying host–microorganism interactions^39–41. Previous studies have used short-term infection models to investigate host DNA damage induced by pks⁺ bacteria as well as chronic infection to characterize colibactin-specific mutations, such as indels and single base substitutions^35,39,40. Building on these studies, we adopted a short-term infection model to assess whether surface-displayed ClbS could mitigate colibactin-induced damage³⁹ (Fig. 3b), while later evaluating the in vivo efficacy of this approach using mouse models to capture the complexity of the gut environment. After acute infection and recovery (Fig. 3c), the organoids were stained for γH2AX, F-actin and DAPI. Compared with Ec^OmpA, Ec^OmpA–ClbS significantly reduced colibactin-induced host DNA damage (Fig. 3d,e and Supplementary Fig. 8). Meanwhile, similar results were observed in C57BL/6J-derived murine colonoids (Supplementary Fig. 7c).

Antitoxin-displaying E. coli effectively and safely mitigate intestinal DNA damage

We next evaluated the efficacy of Ec^OmpA–ClbS in suppressing NC101-mediated host DNA damage in mice maintained under specific-pathogen-free conditions. Mice were pretreated with 2 g l⁻¹ streptomycin in drinking water for 3 days, followed by 2% dextran sodium sulfate (DSS) in water for 1 week. The DSS treatment was included to induce intestinal inflammation and exacerbate the mucosal damage by colibactin⁴². Subsequently, we gavaged the mice with NC101 or the isogenic mutant NC101Δpks (10⁸ CFU per mouse) weekly for 3 weeks to establish a chronic infection. Concurrently, we administered 10⁹ CFU per mouse of Ec^OmpA–ClbS or the empty scaffold control Ec^OmpA via oral gavage twice weekly for five doses owing to its rapid clearance (within 2–3 days), while NC101 was gavaged weekly owing to its murine origin and ability to colonize more efficiently than human E. coli K-12 in the gut⁴³ (Fig. 4b and Supplementary Fig. 9a,b).

Fig. 4 |. Ec^OmpA–ClbS can effectively and safely reduce host DNA damage in mice infected with pks⁺ NC101.

a, Dosing schedule for C57BL/6J mice. b, The surface-displayed ClbS reduced DNA damage induced by NC101 compared with the control strain displaying OmpA alone in mouse colons. Representative IHC images from the colons of mice orally receiving bacteria indicated in the dosing schedule in a. Scale bars, 50 μm (zoomed in) and 1 mm (Swiss roll). c, Percentage of nuclei positive for γH2AX foci shown in b was calculated by dividing γH2AX-positive cells by nuclei-counterstained cells. IHC quantification (n = 10 per treatment group, P < 0.0001 compared with NC101/Ec^OmpA) was performed by QuPath 0.5.1 software. d, Ec^OmpA–ClbS reduced colibactin-mediated mucosal injury upon DSS treatment. Representative image of H&E-stained colon Swiss roll sections. Scale bars, 50 μm. e, Histological scoring of crypt and mucosal integrity performed by a gastrointestinal pathologist in a blinded manner (n = 5 per treatment group). f, Lack of liver toxicity in mice following oral gavage. C57BL/6J mice were gavaged with 10⁹ CFU Ec^OmpA–ClbS, Ec^OmpA or PBS as a negative control. Sera were collected from the mice (n = 5 except n = 4 for PBS) before gavage (day 0, baseline) and on days 1, 3 and 7 after gavage for measuring circulating AST and ALT levels. Panels b–f show representative data from two independent biological replicates. One-way ANOVA test followed by one-sided Tukey’s multiple-comparison test was used for c and e. For c, e and f, data are means ± s.d. Panel a created with BioRender.com.

One day after the last oral administration of Ec^OmpA–ClbS or the controls, colons were collected for immunohistochemistry (IHC) for γH2AX. While the treatment group of Ec^OmpA and NC101 induced a large percentage of γH2AX-positive nuclei in the colon of mice, Ec^OmpA–ClbS and NC101 displayed very low levels of γH2AX staining comparable to the mice treated with NC101Δpks and Ec^OmpA–ClbS or Ec^OmpA (Fig. 4b,c). Of note, NC101Δpks in combination with Ec^OmpA–ClbS or Ec^OmpA helped account for non-pks-associated background DNA damage. To further characterize the cell-type-specific protective effects of ClbS-displaying E. coli, we performed immunofluorescence staining on colon tissues from the NC101/Ec^OmpA group, which showed the highest DNA damage. Results illustrated that γH2AX-positive nuclei were predominantly detected in differentiated surface colonic epithelial cells (E-cadherin⁺), with a smaller proportion detected in goblet cells (MUC2⁺), and minimal to undetectable DNA damage in Paneth (lysozyme⁺) and enteroendocrine cells (chromogranin A⁺) (Extended Data Fig. 4). These results are consistent with a recent study²⁰ showing that attachment of pks⁺ E. coli is enriched on colonic surface epithelium and goblet cells within colonic crypts.

Colibactin has been shown to exacerbate DSS-induced mucosal injury. As expected, the control groups of mice infected with NC101Δpks and Ec^OmpA–ClbS or Ec^OmpA displayed much lower inflammatory responses than those infected with NC101 and Ec^OmpA. Furthermore, the treatment group of NC101 and Ec^OmpA–ClbS had less destruction of mucosal architecture, lower infiltration of inflammatory cells and more goblet cells than mice receiving NC101 and Ec^OmpA (Fig. 4d,e and Supplementary Fig. 10). These results confirm that oral administration of Ec^OmpA–ClbS protected the intestinal epithelium from NC101-induced DNA damage and mucosal injury during mucosal inflammation. In addition to the efficacy, we also assessed whether ClbS-displaying E. coli are associated with any adverse effects in mice. Mice received oral gavage of Ec^OmpA–ClbS, Ec^OmpA or PBS with the same dose, and blood was collected on days 0 (baseline), 1, 3 and 7. Two common biomarkers for liver damage (aspartate transaminase (AST) and alanine transaminase (ALT)) were evaluated considering the gut–liver axis. The results showed no detectable toxicity compared with the PBS control within 1 week post-dosing (Fig. 4f). To evaluate potential neutralizing systemic immune responses by repeated dose of Ec^OmpA–ClbS, we measured ClbS-specific IgG levels in serum using enzyme-linked immunosorbent assay. Mice receiving multiple oral doses of Ec^OmpA–ClbS did not show elevated ClbS-specific IgG titres compared with the untreated control group, indicating minimal systemic immune activation. This is consistent with previous findings that oral antigen exposure typically promotes mucosal tolerance rather than systemic immunity⁴⁴. As a positive control, intravenous administration of E. coli K-12 led to modest but detectable IgG responses (Supplementary Fig. 11). These results support the safety and durability of this therapeutic approach.

Antitoxin-displaying E. coli reduce tumorigenesis in vivo

After validating the prevention from host DNA damage, we next examined whether this strategy could lower the tumorigenesis induced by pks⁺ E. coli in a colitis-associated mouse CRC model using DSS-treated Apc^Min/+ mice^28,31,45. We first orally gavaged DSS/Apc^Min/+ mice with 10⁸ CFU of NC101 or NC101Δpks once a week for three doses under conventional housing conditions (Extended Data Fig. 5a). Two weeks after the third dose, higher numbers of colonic tumours were detected in the NC101 treatment group than in the NC101Δpks treatment group, consistent with the procarcinogenicity of NC101 (Extended Data Fig. 5b,c). We then used this model to assess the efficacy of surface-displayed ClbS in targeting NC101-induced colonic tumour formation. A separate cohort of mice were gavaged with NC101 or the isogenic mutant NC101Δpks (10⁸ CFU per mouse) weekly for three doses. Concurrently, we orally administered 10⁹ CFU per mouse of Ec^OmpA–ClbS or the empty scaffold control Ec^OmpA every 3 days for six doses (Fig. 5a). Two weeks after the third gavage of NC101 (day 28), mice were euthanized for examination of colonic tumour counts and other adverse effects. There was a significant (P = 0.0074) reduction in macroscopic tumour counts in the colons of DSS/Apc^Min/+ mice treated with NC101 and Ec^OmpA–ClbS, compared with the treatment group of NC101 and Ec^OmpA (Fig. 5b,c and Supplementary Fig. 12). By contrast, there were much lower colonic tumour counts in DSS/Apc^Min/+ mice treated with NC101Δpks in combination with Ec^OmpA–ClbS or Ec^OmpA (Fig. 5b,c and Supplementary Fig. 12). These findings suggested that K12 Ec^OmpA–ClbS decreased the tumour formation in the colons of DSS/Apc^Min/+ mice in a ClbS- and pks-specific manner. In addition, there were no overt body weight changes over the course of the treatment compared with the initial baseline before oral administration of different bacterial combinations (Supplementary Fig. 13a,b).

Fig. 5 |. Ec^OmpA–ClbS decreased NC101-mediated tumorigenesis in a colitis-associated mouse CRC model (DSS/Apc^Min/+).

a, Timeline for the DSS/Apc^Min/+ mice treated with pks⁺ or NC101Δpks in combination with Ec^OmpA–ClbS or Ec^OmpA. b, Representative tumours in the colon on day 28 and quantification of colonic tumour counts. Results are pooled from two independent biological replicates (n = 10 for NC101 + Ec^OmpA–ClbS, n = 9 for NC101 + Ec^OmpA, n = 8 for both NC101Δpks + Ec^OmpA–ClbS and NC101Δpks + Ec^OmpA). c, The relative fold changes in the bacterial abundance of NC101 over time measured by qPCR. d, CFU counts from the caecum, colon and faeces of NC101 or NC101Δpks at the end-point (day 28) in wild-type C57BL/6J mice receiving oral gavage of Ec^OmpA–ClbS or Ec^OmpA. The figures show representative data from two independent biological replicates (c, d). One-way ANOVA test followed by one-sided Tukey’s multiple-comparison test was used for b and d; two-way ANOVA test was used for c. For b–d, data are means ± s.d. Panel a created with BioRender.com.

The above study involved co-administration of NC101 and Ec^OmpA–ClbS or Ec^OmpA, which may aid colocalization and initial competition between two strains. From the translational perspective, we next implemented a staggered administration protocol in DSS/Apc^Min/+ mice, in which NC101 was administered 4 h before the dosing of Ec^OmpA–ClbS or Ec^OmpA. It was shown that Ec^OmpA–ClbS remained able to decrease tumorigenesis in the colons compared with the Ec^OmpA treatment group (Supplementary Fig. 14). To further explore the generalizability of this approach beyond pks⁺ E. coli, we showed the in vitro inhibition of genotoxicity by Ec^OmpA–ClbS in Citrobacter koseri ATCC BAA-895 (ref. 46), a pks⁺ strain from a different genus within the Enterobacteriaceae family (Extended Data Fig. 6a). Oral administration of Ec^OmpA–ClbS significantly (P = 0.0233) reduced tumour burden induced by C. koseri compared with that of the Ec^OmpA control in DSS/Apc^Min/+ mice (Extended Data Fig. 6b–d). In summary, our in vivo studies show the translational potential of the antitoxin display strategy against diverse colibactin-producing pathogens.

Antitoxin display suppresses pks⁺ bacterial colonization

The increased abundance of procarcinogenic NC101 has been correlated with a higher colonic tumour burden in DSS/Apc^Min/+ mice²⁸. In addition to showing the prevention of intestinal DNA damage and tumorigenic potential of pks⁺ E. coli (Figs. 4b,c and 5b), we further asked whether the surface-displayed ClbS affected the abundance of NC101 in mice. To this end, we analysed the abundance of NC101 longitudinally in three different treatment groups receiving (1) NC101 alone, (2) NC101 and Ec^OmpA–ClbS and (3) NC101 and Ec^OmpA in DSS/Apc^Min/+ mice following the same treatment scheduling in Fig. 5a. We quantified the copy number of the clbP gene unique to the pks island as a proxy for NC101 abundance in mouse faeces by quantitative polymerase chain reaction (qPCR) and normalized the clbP copy number to the total bacterial abundance measured by a pair of universal primers. The relative abundance of NC101 in the treatment group receiving NC101 and Ec^OmpA–ClbS decreased by approximately 30 times compared with the treatment group receiving NC101 and Ec^OmpA on day 28, 10 days after the last administration of Ec^OmpA–ClbS or Ec^OmpA when mice were euthanized (Fig. 5c and Supplementary Fig. 15). In addition, E. coli displaying the empty scaffold OmpA (Ec^OmpA) did not affect the colonization of NC101 compared with the control group receiving NC101 alone (Fig. 5c and Supplementary Fig. 15), indicating that the decrease of NC101 resulted from engineered E. coli displaying ClbS as opposed to non-specific competition between different bacterial strains in vivo. As an important negative control, the clbP-specific primer pair did not detect any signal from the faecal DNA of the mice that were orally gavaged with the isogenic mutant NC101Δpks (Supplementary Fig. 16). In summary, the qPCR data from faecal samples showed that Ec^OmpA–ClbS could decrease the colonization of NC101, offering another mechanism to reduce the genotoxicity and procarcinogenicity of NC101 in vivo.

After showing that Ec^OmpA–ClbS could decrease the abundance of NC101 in mice, we aimed to determine how the surface-displayed ClbS might differentially affect NC101 and NC101Δpks, respectively. Similar to evaluating the protection from intestinal DNA damage in Fig. 4a, we gavaged C57BL/6J mice with NC101 or NC101Δpks weekly for 3 weeks. Concurrently, we administered 10⁹ CFU per mouse of Ec^OmpA–ClbS or Ec^OmpA every 3 days for five doses. NC101 and NC101Δpks carry a functional lacZ gene while K-12 Ec^OmpA–ClbS and Ec^OmpA strains have a mutated lacZ, a non-essential gene for lactose metabolism. This genotypic difference allowed us to distinguish NC101 derivatives (pink colonies) from K-12 derivatives (white colonies) on MacConkey agar plates. At the end-point (day 28), Ec^OmpA–ClbS, but not Ec^OmpA, notably lowered the abundance of NC101 in mouse faeces. By contrast, Ec^OmpA–ClbS failed to affect the colonization of NC101Δpks in mice compared with Ec^OmpA (Fig. 5d). The observed reduction of NC101 in faeces could potentially be attributed to prolonged retention of the strains in the intestines, resulting in decreased shedding of NC101 into faeces. To rule out this possibility, we homogenized the caecum and colon tissues and enumerated the live bacterial counts (normalized to tissue weight) of NC101 or NC101Δpks on the MacConkey agar plates from treated mice. Consistent with faecal CFU findings, Ec^OmpA–ClbS, but not Ec^OmpA, resulted in approximately a 10⁵-fold reduction in the caecum and a 10³-fold reduction in the colon, when comparing CFUs of NC101 with those of NC101Δpks (Fig. 5d).

To further evaluate the impact of Ec^OmpA–ClbS on gut microbiota composition, we performed 16S rRNA gene sequencing on faecal samples from NC101-colonized mice in the midpoint and end-point of the in vivo work, in which mice received NC101 alone, NC101 + Ec^OmpA or NC101 + Ec^OmpA–ClbS. Alpha diversity (Shannon and Chao1 indices) remained unchanged across different groups, suggesting comparable microbial richness and evenness. Beta diversity analysis revealed a relative separation of mice receiving NC101 + Ec^OmpA–ClbS compared with those colonized with NC101 alone or NC101 + Ec^OmpA (Extended Data Fig. 7a,b). In addition, we observed a progressive decrease in the relative abundance of Enterobacteriaceae, a family often expanded during intestinal inflammation and linked to adverse clinical outcomes⁴⁷. Specifically, Enterobacteriaceae abundance was highest in mice colonized with NC101 alone, moderately reduced in mice co-treated with Ec^OmpA and further suppressed in mice co-treated with Ec^OmpA–ClbS (Extended Data Fig. 7c,d). The 16S sequencing, along with the CFU quantitation, suggests that Ec^OmpA–ClbS may attenuate the expansion of inflammation-associated Enterobacteriaceae. Future studies are warranted to further explore the observed ClbS-dependent inhibition of pks⁺ NC101 in colonizing mice gut. Nevertheless, our in vivo findings suggest that engineered E. coli displaying ClbS not only suppresses colibactin-mediated host DNA damage and tumorigenesis, but also reduces the abundance of procarcinogenic pks⁺ NC101 in vivo.

Discussion

The rising incidence of pathogenic gut bacteria producing procarcinogenic molecules has garnered increasing attention, with colibactin representing one of the most extensively studied microbial metabolites linked to CRC. Of note, colibactin-producing pks⁺ bacteria are major contributors to early-onset colorectal cancer (EOCRC) by inducing the earliest driver mutations in adenomatous polyposis coli (APC) during the first decade of life—well before the clinical onset of disease⁴⁸. These findings highlight a critical window in early life when reducing the colonization of pks⁺ bacteria may offer preventive benefit. Despite increasing evidence of colibactin’s role in DNA damage and tumorigenesis, few therapeutic strategies directly target this genotoxin. Inspired by the natural ClbS-mediated self-resistance mechanism in pks⁺ bacteria, we engineered non-pathogenic E. coli to display ClbS on its surface, enabling in situ neutralization of colibactin at the host–bacteria interface. We showed that surface-displayed ClbS can (1) prevent DNA damage in human and mouse colon-derived cell lines and organoids, (2) reduce tumorigenesis in the colitis-associated DSS/Apc^Min/+ CRC model and (3) selectively suppress pks⁺ E. coli without affecting isogenic knockout strains in conventional mice. One potential limitation of our study is that our evaluation of colibactin-induced genotoxicity has primarily relied on γH2AX staining and cell cycle analyses. While these assays are widely used and informative for assessing DNA damage and its cellular consequences, they do not capture the full spectrum of mutational events associated with colibactin exposure. As a future direction, we aim to perform whole-genome sequencing of tumour-bearing tissues or organoids to investigate whether our intervention reduces the prevalence of colibactin-associated mutational signatures, thereby providing deeper insights into the long-term genomic impact of this therapeutic strategy.

From the translational perspective, our study provides early proof of concept for a preventive strategy targeting early-life exposure to colibactin, a known contributor to mutagenesis decades before CRC becomes clinically detectable. A recent 2025 Nature study provided compelling evidence that colibactin-producing bacteria present in the gut microbiota during the first decade of life are associated with the initiation of APC mutations in patients with EOCRC⁴⁸. These findings highlight the need for microbiome-targeted preventive strategies early in life. Our data support this concept by showing that ClbS-displaying E. coli can reduce pks⁺ bacterial abundance and genotoxicity in vivo. Moreover, to broaden the relevance beyond E. coli, we tested our system in DSS/Apc^Min/+ mice colonized with C. koseri, a pks⁺ strain from a different genus. Ec^OmpA–ClbS substantially reduced tumour burden induced by C. koseri, confirming that the therapeutic efficacy of ClbS surface display extends to other colibactin-producing Enterobacteriaceae. It is worth noting that in all mouse experiments, we did not administer antibiotics to maintain the OmpA–ClbS plasmid. While plasmid-based expression was sufficient to produce the observed phenotypes, plasmid systems can incur fitness costs and be diluted during longer or competitive colonization. Future work can involve integrating the OmpA–ClbS expression cassette into a neutral chromosomal locus to ensure durable expression. In addition, our 16S microbiome analyses were performed in the context of pks⁺ NC101 colonization. However, community shifts observed with OmpA–ClbS may partially reflect indirect effects mediated through suppression of pks⁺ NC101 rather than a direct impact of the engineered chassis on resident gut bacteria. Future work can compare OmpA–ClbS versus OmpA in mice without colonization of pks⁺ NC101, to disentangle direct effects from NC101-dependent effects.

From the fundamental perspective, our findings on the targeted inhibition of pks⁺ NC101 colonization by antitoxin display may reflect the complexity of the pks island⁴⁹. Previous studies showed that, compared with the wild-type NC101, knocking out the entire pks or an essential clb gene of pks in NC101 did not alter the colonization ability of different isogenic mutant NC101 strains^29,31,32. However, it has been speculated that there is a selective pressure for acquiring the pks island. For example, colibactin produced by pks⁺ bacteria could lower the abundance of other gut microbiota members in neonatal mice⁴, suggesting that colibactin may confer competitive ecological benefits to pks⁺ bacteria. While colibactin may confer a competitive advantage to pks⁺ bacteria, our study and others suggested that the competitive advantage conferred by colibactin is likely context dependent, being affected and shaped by contextual factors. In addition, the potential benefits from colibactin might be counterbalanced by other factors within the pks island. For example, the pks island can interact with other pathways that influence bacterial fitness and interbacterial competition⁴⁹. Moreover, colibactin biosynthesis is energy intensive. This hypothesis, if correct, could explain two outcomes we observed: (1) direct colibactin neutralization by ClbS could put NC101 at a disadvantage during the in vivo colonization as the other opposing factors were still present in the pks island such as the possible metabolic burden from the energy-intensive colibactin biosynthesis, and (2) the removal of the pks island rendered NC101Δpks insensitive to colibactin neutralization by ClbS. As a result, the in vivo abundance of NC101Δpks may not be affected by ClbS-displaying E. coli or the loss of the pks island. Further validation of these hypotheses would warrant future studies using germ-free mice and in vitro coculture containing a mixed synthetic gut community to understand the competition and colonization resistance among ClbS-displaying E. coli, pks⁺ bacteria and other gut commensals. Nonetheless, our study may pave the way for a targeted approach to ‘knock down’ colibactin in situ to interrogate the causal relationships between colibactin and other host processes. In a broader context, it offers a promising strategy to engineer microbial therapeutics targeting other cancer-promoting bacterial metabolites.

Methods

This study complies with all relevant ethical regulations. All animal experiments were approved by the Institutional Animal Care and Use Committees at Northeastern University and the University of Michigan. De-identified human adult colon tissues were collected with approval from the University of Michigan Institutional Review Board (HUM00105750) and were obtained from Gift of Life Michigan from cadaveric organ donors.

Bacterial and cell culture

Unless specified elsewhere for cocultures, all bacterial plasmids and strains listed in Supplementary Tables 1 and 2 were routinely maintained in lysogeny broth (LB) media and incubated at 37 °C under aerobic conditions. Kanamycin (50 μg ml⁻¹) and chloramphenicol (35 μg ml⁻¹) were included in LB media for strains with corresponding resistance genes.

HeLa, HCT116, Caco-2 and HT-29 were purchased from American Type Culture Collection. MC38 was requested from D. Irvine’s laboratory at the Koch Institute originally purchased from Kerafast. L-WRN (derived from a mouse fibroblast cell line, L cells, and transfected with vectors to stably express Wnt-3A, R-spondin 3 and Noggin) was obtained from AddexBio. HeLa, HCT116, MC38 and L-WRN were maintained in complete Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (P/S). Caco-2 was cultured in complete DMEM supplemented with 20% FBS, 100 U ml⁻¹ P/S and 1× non-essential amino acid. HT29 was cultured in Roswell Park Memorial Institute Medium-1640 (RPMI-1640) supplemented with 10% FBS, 1% P/S and 1× non-essential amino acid. All the cell lines mentioned or coculture assay listed below were maintained at 37 °C in a humidified incubator with 5% carbon dioxide (CO₂). Cells at passages 2–10 were used for the experiments.

Surface display

Plasmids pLyGo-Esc-7, pLyGo-Ec-8 and pDSG323 were used. The DNA sequence for encoding ClbS with the DYKDDDDK tag (FLAG tag) in the N terminal and Myc-tag in the C terminal was inserted between two SapI sites for pLyGo-Ec-7 and pLyGo-Ec-8 and between SpeI and PstI sites for pDSG323 by Gibson Assembly. For display verification, E. coli K-12 NEB 5-alpha with the correct plasmid were inoculated in fresh LB medium with 50 μg ml⁻¹ kanamycin. After overnight culture in a shaker (37 °C, 250 rpm unless otherwise indicated), bacterial suspensions were diluted by 10-fold in the fresh LB with 50 μg ml⁻¹ kanamycin and 10 mM L-rhamnose (pLyGo-Ec-7 and pLyGo-Ec-8 derived plasmids) or 100 ng ml⁻¹ anhydrotetracycline (aTc; pDSG323 derived) for a 3-h induction in a shaker.

To increase the display levels, a high-copy plasmid, pDS861-dsRed, was digested with XbaI and NotI. Regulator proteins (RhaR and RhaS) and a rhamnose-responsive promoter (PrhaBAD), along with Lpp–OmpA, were amplified from pLyGo-Ec-8. The DNA fragments for ClbS or variants with an N-terminal Myc tag and a C-terminal FLAG tag were synthesized by Twist Bioscience, assembled with digested pDS861-dsRed backbone by Gibson Assembly. For high-level display induction, overnight bacterial cultures were diluted ten times in the fresh LB containing 50 μg ml⁻¹ kanamycin and 10 mM L-rhamnose monohydrate, and induced at 18 °C, 250 rpm for 72 h or 25 °C, 250 rpm for 24–48 h.

All plasmids were validated by Sanger sequencing.

In vitro infection assay

NC101 and NC101Δpks were inoculated in LB and grown overnight. In parallel, epithelial cell lines were seeded in 24-well plates or 96-well plates (40,000–200,000 cells per well). Next day, diluted overnight bacteria were expanded in fresh media until the exponential phase. The optical density of bacteria at 600 nm (OD₆₀₀) ranged from 0.6 to 0.8, and bacterial density was calculated by multiplying OD₆₀₀ by 8 × 10⁸ ml⁻¹. Bacteria were resuspended by cell culture media without antibiotics, and the concentrations were adjusted to the desired MOI. In the meantime, media in the cell plates were removed and cells were washed with Dulbecco’s phosphate-buffered saline (DPBS). Afterwards, the media containing ClbS-displaying E. coli or other controls were added to the wells followed by media containing NC101 or other controls (MOI is indicated in the figure legends of relevant data). The coculture wells were incubated for 4 h, washed once with DPBS and replenished with fresh media containing 10% FBS and 200 μg ml⁻¹ gentamicin to remove bacteria. After 2–4 h, the cells were detached and subjected to immunocytochemistry staining for γH2AX phosphorylation (Ser139). For the D-serine assay, 1-mM or 10-mM D-serine was added in diluted overnight culture until the exponentially growing phase. D-serine was further added in the cell culture medium during bacteria–HeLa coculture, followed by DMEM, 10% FBS and gentamicin for another 2–4 h before staining with anti-γH2AX. For cell cycle arrest detection, cells were trypsinized after 24 h of incubation with gentamycin-containing media.

Flow cytometry

For γH2AX (Ser139) staining, cells were washed with PBS, fixed by 4% paraformaldehyde (PFA) in 37 °C for 15 min and permeabilized by 0.4% Triton X-100 in 4 °C for 15 min after washes. The antibody-staining buffer (1% BSA, 0.05% Tween-20 in PBS) containing 1 μg ml⁻¹ Alexa Fluor 647 anti-H2AX (Ser139) antibody was added to cell washes for overnight incubation. Next, the antibody was removed and cells were washed with PBST (0.05% Tween-20 in PBS), resuspended in PBS and analysed by flow cytometry. For cell cycle arrest, detached cells were fixed by 90% ethanol for 20 min in 4 °C and incubated with 500 ng ml⁻¹ 7-AAD and 50 μg ml⁻¹ RNase after PBS washes before flow cytometry. For bacterial surface display, 20 μl of bacterial suspension was stained with 1 μg ml⁻¹ Alexa Fluor 647 anti-DYKDDDDK Tag Antibody in PBS for 20 min at room temperature (RT) before analysis.

CFU-based adhesion and invasion assays

The adhesion and invasion assays were adapted from a previous study²⁰. First, NC101^CmR was generated by inserting a chloramphenicol (Cm)-resistant gene through Tn7. For the adhesion assay, HeLa cells were infected by NC101^Cm (MOI 50) for 4 h along with Ec^OmpA–ClbS or Ec^OmpA. Cells were washed to remove non-adherent bacteria, lysed with 1% Triton X-100, serially diluted and plated on LB plus Cm agar plates. For the invasion assay, following the same infection, wells were washed and incubated for 1 h in medium with gentamicin (100 μg ml⁻¹) to remove extracellular bacteria. Cells were then washed, lysed and plated as above to enumerate intracellular bacteria.

Organoid culture media

The process of making conditioned media for organoid culture is adapted from the Translational Tissue Modeling Laboratory at the University of Michigan. Briefly, L-WRN cells were cultured in a 15-cm Petri dish in DMEM supplemented with 10% FBS and 100 U ml⁻¹ P/S. The next day, the medium was refreshed with the addition of 500 μg ml⁻¹ G418 and 500 μg ml⁻¹ hygromycin B, and L-WRN cells were cultured for 3–4 days until the cells became fluent, detached by trypsin and expanded in five 15-cm Petri dishes containing DMEM supplemented with 10% FBS and 100 U ml⁻¹ P/S for another 3 days or 4 days until full confluency. Then, the medium was changed to 25 ml Advanced DMEM/F12 supplemented with 20% FBS and 1% Glutamax. After 24 h, the entire supernatant was aspirated, centrifuged to remove debris, collected in 50-ml conical tubes and further supplemented with 25 ml fresh Advanced DMEM/F12 (20% FBS and 1% Glutamax) as murine-conditioned medium. The supernatant was collected daily and pooled from aliquots over a period of 4 days. In addition, the human colonoid medium was prepared with the following reagents: 15% (v/v) Intesticult, 42.5% (v/v) L-WRN-conditioned medium, 21.25% basal medium (12.5 ml; 8 mM GlutamAx, 40 mM HEPES, 4× N-2 media supplement, 4× B-27 supplement minus vitamin A, 4 mM N-acetyl-L-cysteine in Advanced DMEM/F-12), 21.25% Advanced DMEM/F12, 85 ng ml⁻¹ human recombinant EGF, 85 μg ml⁻¹ Primocin, 500 nM A8301, 10 μM SB202190, 10 μM Y27632 and 2.5 μM CHIR99021. Aliquots of the media were stored at −20 °C for long-term use. Note that any antibiotics were excluded in the conditioned media to allow coculture with bacteria later.

Colon organoid culture and passaging

For murine organoids, the colons were isolated after euthanizing C57BL/6J mice (8–10 weeks, both genders, approved by the Institutional Animal Care and Use Committee at Northeastern University). The colons were cut longitudinally, cleaned with PBS and minced into small pieces. The tissue fragments were incubated in advanced DMEM/F12 containing 2 mg ml⁻¹ collagenase type IV, 50 μg ml⁻¹ gentamicin, 1% Glutamax, 100 U ml⁻¹ P/S, 10 mM HEPES and 10% FBS for 30–60 min at 37 °C and vigorously pipetted to isolate the colon epithelial units. The mixture was filtered through after washes with media, the pellet was resuspended in Advanced DMEM/F-12 and 10% FBS, and crypts were counted. A total of 500 crypts were resuspended and seeded in a 15-μl Matrigel drop in a 24-well plate. The Matrigel was polymerized at 37 °C for 15 min and then supplemented with 500 μl murine-conditioned medium supplemented with 10 mM HEPES, 100 U ml⁻¹ P/S and 10 μM Y-27632 dihydrochloride. The organoids were cultured with the change of media three times weekly. The organoids were weekly passed 1:2–1:5 by mechanical dissociation. The required number of cells were resuspended and seeded per 15-μl Matrigel drop in a 24-well plate for polymerization with subsequent fresh media. For human colonoids, de-identified human adult colon tissues were collected with approval from the University of Michigan Institutional Review Board (HUM00105750) and were obtained from Gift of Life Michigan from cadaveric organ donors. The human colonoids were passed in a similar manner with different media mentioned in ‘Organoid culture media’.

Colon organoid bacterial coculture

The colon organoids embedded in the Matrigel were scraped mechanically, resuspended with infection medium (Advanced DMEM/F12, 1% Glutamax, 20% FBS, 10 mM HEPES) after washes and disrupted mechanically. Bacteria in 0.9% NaCl, cisplatin dissolved in 0.9% NaCl or equal volume 0.9% NaCl were added to each treatment group of organoids. After a 3-h incubation, the organoids were washed and re-embedded in Matrigel. Then, 500 μl of infection medium was added to each well and the organoids were incubated for another 4 h, after which the infection medium was replaced with regular culture medium. The organoids were cultured for 3 days before the next steps.

Animal experiment

Mice were housed in a specific-pathogen-free facility and fed normal chow and water ad libitum under standard animal facility conditions (12-h light–dark cycle, temperature of 22 °C, relative humidity of 40–70%). C57BL/6J (strain 000664) and C57BL/6J-Apc^Min/J (strain 002020, Apc^Min/+) mice were initially purchased from the Jackson Laboratory, and C57BL/6J-Apc^Min/J (Apc^min) were further bred in a specific-pathogen-free (SPF) facility. The offspring were genotyped using polymerase chain reaction (PCR) analysis on tail biopsy. Mice were housed under a 12-h light–12-h dark cycle at a controlled temperature of 20–25 °C and relative humidity of 30–70%. Female C57BL/6J mice aged 8–10 weeks old were maintained under specific-pathogen-free conditions. Mice were pretreated with 2 g l⁻¹ streptomycin in drinking water for 3 days, followed by 2% DSS in water for 1 week for inducing intestinal inflammation and exacerbating the mucosal damage by colibactin. Subsequently, the mice were gavaged with NC101 or NC101Δpks (10⁸ CFU per mouse) weekly for 3 weeks. Concurrently, 10⁹ CFU per mouse of Ec^OmpA–ClbS or Ec^OmpA were gavaged every 3–4 days for 5 doses. One day after the last oral administration, colons Swiss rolled in cassettes, fixed in 4% PFA overnight at 4 °C and transferred to 70% ethanol were subjected to IHC for γH2AX. Paraffin embedding, sectioning, IHC or haematoxylin and eosin (H&E) staining were processed in iHisto. Images were scanned by a bright-field digital scanner. Histological scoring of colons was performed blindly by one investigator. IHC results were quantified by QuPath 0.5.1.

For the colon adenoma-forming experiment, Apc^Min/+ mice (12–18 weeks old, male and female) were treated with antibiotics and DSS similarly. After treatment, the mice were allocated to different groups and gavaged with the following: (1) NC101 experiments: 10⁸ CFU of E. coli NC101 or NC101Δpks was administered on days 0, 7 and 14. Concurrently, 10⁹ CFU of Ec^OmpA–ClbS or Ec^OmpA was administered on days 0, 3, 7, 10, 14 and 17 using NC101, NC101Δpks and PBS as controls. (2) Staggered NC101 treatment: NC101 was administered 4 h before the corresponding Ec^OmpA–ClbS or Ec^OmpA dosing using the same schedule as above. (3) C. koseri experiments: 10⁸ CFU of C. koseri ATCC BAA-895 (a pks⁺ strain⁴⁶) was gavaged on days 0, 7 and 14. In parallel, 10⁹ CFU of Ec^OmpA–ClbS or Ec^OmpA was administered 6 times on days 0, 3, 7, 10, 14 and 17. Apc^Min/+ mice were monitored twice weekly and euthanized if they showed impaired mobility or feeding, ≥15% weight loss or poor body condition (body condition score = 2). No adverse effects were observed before the humane end-point. Stools were collected 1 day after oral gavage, snap frozen in liquid nitrogen and transferred to −80 °C. The genomic DNA of stools was extracted using a Quick-DNA Fecal/Soil Microbe Miniprep Kit. Mice were killed at day 28 and colons were collected and cut open longitudinally for tumour counting. Paraffin-embedded sections were prepared for H&E staining to corroborate tumour burden and immunofluorescence staining for different cell types.

For the colonization experiments by quantifying CFUs in stools, C57BL/6J mice were treated with a similar scheme mentioned above. NC101 and NC101Δpks carry a functional lacZ gene while K-12 Ec^OmpA–ClbS and Ec^OmpA strains have a mutated lacZ, a non-essential gene for lactose metabolism. This genotypic difference can distinguish NC101 derivatives (pink colonies) from K-12 derivatives (white colonies) on Mac-Conkey agar plates. At the end-point (day 28), 1 or 2 faeces per mouse were collected, weighed and suspended in PBS buffer (100 mg ml⁻¹) followed by 15 min of incubation. The mixture was homogenized using a micro-tissue homogenizer, and 10-time serially diluted suspensions were plated on MacConkey agar plates. After incubation at 37 °C for 12–16 h, CFUs were determined by normalizing to the weight of faeces. Before bacterial administration in mice, faecal CFU enumeration on MacConkey agar revealed no colony for Enterobacteriaceae. In the same cohort, for the colonization experiments quantifying CFUs in the caecum and colon, the tissues were collected from killed mice at the end-point and subsequently weighed, minced and digested with RPMI-10 with 5% FBS, 10 mM HEPES, 20 μg ml⁻¹ DNase I and 1 mg ml⁻¹ collagenase IV for 1 h at 37 °C with agitation. The cell homogenate was serially diluted and plated on MacConkey agar plates. CFUs were calculated by normalizing to the weight of tissues after overnight incubation at 37 °C.

For the safety experiment, 10–12-week-old mice were orally gavaged with 10⁹ CFU Ec^OmpA, Ec^OmpA–ClbS or PBS. The serum of mice was collected at days 0 (before gavage), 3 and 7 after gavage isolated from submandibular blood. AST and ALT were measured by the Unit for Laboratory Animal Medicine (ULAM) Pathology Core at University of Michigan, Ann Arbor.

For systemic immune responses following oral administration of Ec^OmpA–ClbS, ClbS-specific IgG titres in mouse serum were measured by ELISA. Bacterial lysates were obtained through lysozyme-based lysis followed by sonication and centrifugation. Total protein concentration was quantified using the Bio-Rad DC assay, and ELISA plates were coated overnight at 4 °C with 50 μl per well of lysate in PBS (total protein concentration: 20–100 μg ml⁻¹). Plates were blocked with PBS, 1% BSA and 1 mM EDTA for 1 h at RT. Serum samples were serially diluted in blocking buffer and incubated on coated plates for 2 h. After washes, bound IgG was detected using horseradish peroxidase-conjugated anti-mouse IgG antibody, followed by tetramethylbenzidine substrate development. Reactions were stopped with 2 M sulfuric acid, and absorbance was measured at 450 nm by Molecular Devices SpectraMax M5 Multi-Mode Microplate Reader.

Confocal imaging

For bacterial interactions, HeLa cells were pre-seeded on number 1.5 cover slides in a 6-well plate before infection. ClbS-displaying E. coli and pks₊ E. coli NC101 were labelled with DiI and BODIPY separately. Subsequently, HeLa cells were cocultured with labelled bacteria (MOI 500) for 15 min. Bacteria were washed and fixed with 4% PFA for 10 min, and HeLa cells were counterstained with phalloidin Alexa 647. The cover slides were mounted on a glass substrate with EverBrite Hardset Mounting Medium without DAPI, and images were acquired using the Nikon A1 inverted confocal microscope in the microscopy core at the University of Michigan. Sequential imaging was performed in the following order to minimize spectral bleed-through: Alexa Fluor 647 (far-red) was imaged first using a 640-nm laser with detection at 650–720 nm, followed by DiI (red) using a 561-nm laser with detection at 570–620 nm and finally BODIPY FL (green) using a 488-nm laser with detection at 500–550 nm.

For organoid imaging, 3 days after organoid infection, the medium was discarded and organoids with Matrigel were washed, fixed in 4% PFA for 30 min at RT, permeabilized with 1% Triton X-100 in PBS for 30 min at RT after washes and blocked using the antibody-staining buffer (0.2% Triton X-100, 3% BSA, PBS) for 2 h at RT. Followed by the blocking step, organoids were incubated with 1 μg ml⁻¹ purified anti-γH2AX (Ser139) antibody in the antibody-staining buffer for 24 h at 4 °C; stained with 2 μg ml⁻¹ anti-mouse IgG (H + L), F(ab’)₂ and Alexa Fluor 647 Phalloidin for 1 h at RT in the dark; mounted using EverBrite Hardset Mounting Medium with DAPI; and sandwiched between a cover glass and poly-L-lysine-coated cover slide. The washes were performed in between before the mounting steps. Samples were imaged by a ZEISS LSM 800 confocal microscope after 24 h. Nuclei (DAPI) were excited at a wavelength of 405 nm detected in the range of 400–493 nm, cell skeleton actin was excited at a wavelength of 640 nm detected in the range of 656–700 nm and DNA damage (Alexa Fluor 488) was excited at a wavelength of 488 nm detected in the range of 493–620 nm. The dynamic range was adjusted to be the same under a channel-mode confocal modality. Images were visualized and analysed using Zen Blue (Zeiss) and Fiji (2.9.0).

For IF staining, after deparaffinization and rehydration, antigen retrieval was performed using Antigen Unmasking Solution, Citrate-Based in a pressure cooker. Slides were heated under full pressure for 3 min, followed by natural cooling to room temperature. After being blocked in antibody staining buffer (1 × PBS, 5% BSA) for 1 h at RT, serial sections (four consecutive slides per sample) were stained with one of four primary antibodies (1:200): anti-E-cadherin, anti-lysozyme, anti-chromogranin A or anti-MUC2 overnight at 4 °C and then stained with Alexa Fluor 647-conjugated goat anti-rabbit IgG (1:500) for 1 h at room temperature. Slides were then incubated with Alexa Fluor 488 anti-phospho-Histone H2AX (S139; 1:100) for 1 h. Nuclei were counterstained with EverBrite Hardset Mounting Medium with DAPI. Immunofluorescence images were acquired using a Nikon Ti2 widefield fluorescence microscope. The following filter sets were used for each fluorophore: DAPI, excitation 365/28 nm, emission 445/50 nm; Alexa Fluor 488, excitation 475/30 nm, emission 530/40 nm; and Alexa Fluor 647, excitation 630/50 nm, emission 695/50 nm. Images were acquired using Nikon NIS-Elements software.

qPCR

Faecal DNA was used as the qPCR template, and qPCR primers for clbP and universal 16S are provided in Supplementary Table 3. qPCR was performed in a 96-well plate format using FastSYBR Low Rox (CoWin Biosciences) on Quantstudio 6 Real-Time PCR System (Thermo Fisher) using the following protocol: 95 °C for 20 s, 40 cycles of 95 °C for 3 s and 60 °C for 30 s, and a final cycle of 95 °C for 15 s, 60 °C for 1 min and 95 °C for 15 s. Relative gene expression was calculated by the 2^−ΔΔCt method.

16S rRNA sequencing

Analyses of 16S rRNA were performed both in-house and by Novogene. Genomic DNA concentration was measured with 1% agarose gels. According to the concentration, DNA was diluted to 1 ng μl⁻¹ using sterile water. The 16S rRNA V4 specific primers are 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 806A (5′-GGACTACHVGGGTWTCTAAT-3′). All PCRs were carried out in 30-μl reactions with 15 μl of Phusion High-Fidelity PCR Master Mix (New England Biolabs), 0.2 μM forward and reverse primers, and about 10 ng of template DNA. Thermal cycling is started with the initial denaturation at 98 °C for 1 min, followed by 30 cycles of denaturation at 98 °C for 10 s, annealing at 50 °C for 30 s and elongation at 72 °C for 60 s. Finally, the solution was kept at 72 °C for 5 min. PCR products were mixed with an equal volume of SYBR Green loading buffer and subjected to electrophoresis on 2% agarose gel for quantification and qualification, followed by purification with the GeneJET Gel Extraction Kit. Sequencing libraries were generated using the NEB Next Ultra DNA Library Prep Kit for Illumina following the manufacturer’s recommendations, and index codes were added. The library quality was assessed on a Qubit 2.0 fluorometer and Agilent Bioanalyzer 2100 system. Finally, the library was sequenced on an Illumina HiSeq 2500 platform, generating 250-bp paired-end reads. For taxa relative abundance outcomes, qualified reads were filtered by QIIME software to obtain clean tags. All tags were compared with the reference database using the UCHIME algorithm to obtain effective tags only, which were used for diversity analysis. For diversity analyses, FASTA files were processed with DADA2 for R, with primers sorted and demultiplexed sequences filtered and trimmed at truncation lengths of 240 bp and 160 bp for forward and reverse reads, respectively, with a minimum quality score of 20. Following merging and removing of chimeric reads, taxonomy was assigned as amplicon sequence variants with the Silva v138 taxonomic database. Using phyloseq for R, amplicon sequence variants not present at least five times in 50% of the samples were pruned, and in all outcomes aside from alpha diversity, reads were rarefied to even sampling depth.

Statistical analysis

All statistical analyses were performed using GraphPad Prism 10 and R (4.3.2). Data were analysed with two-sided unpaired t-test, permutational multivariate analysis of variance (β-diversity), one-way or two-way analysis of variance (ANOVA) for statistical significance, Kruskal–Wallis test with Benjamini–Hochberg false discovery rate correction (Shannon, Chao1 index and relative abundance).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 |. Schematic of protecting intestinal epithelial cells from the genotoxic effects of pks⁺ bacteria by repurposing a toxin-anti-toxin mechanism in colibactin-producing pks⁺ bacteria.

Certain enteric bacteria carry a polyketide synthase (pks) island coding for colibactin. Colibactin can translocate to the nucleus of epithelial cells and induce DNA damage (that is, genotoxicity) and potential tumorigenesis. However, these pks⁺ bacteria express an intracellular resistance protein, ClbS, to cope with colibactin-induced self-DNA damage. Inspired by this toxin-anti-toxin mechanism, we develop a strategy to prevent colibactin-mediated DNA damage and tumorigenesis induced by pks⁺ bacteria through the bacterial surface display of E. coli ClbS, and its homolog proteins in different commensal E. coli strains (for example, K-12, Nissle and NGF-1). Figure created with BioRender.com.

Extended Data Fig. 2 |. Ec^OmpA-ClbS is superior to D-serine, a small molecular inhibitor for colibactin biosynthesis, in suppressing genotoxicity induced by various pks⁺ E. coli.

a, a schematic of the coculture setup. b, D-serine failed to reduce the genotoxicity by DH10B pBAC-pks in HeLa cells while Ec^OmpA-ClbS completely suppressed the genotoxicity. c, Ec^OmpA-ClbS suppressed host DNA damage induce by pks⁺ E. coli (NC101, MIT A2 and MIT A21) and outperformed the inhibitory effect from D-serine (10 mM). D-serine was included in the exponentially growing phase of pks⁺ E. coli isolates as well as in the coculture with HeLa cells. The MOI is 20. Figures show representative data from three independent biological replicates (b, c). Two-sided unpaired t-test (b, c). Data are means ± SD (n = 3, b, c). Panel a created with BioRender.com.

Extended Data Fig. 3 |. Surface display of ClbS in two pks+ E. coli chassis strains lower their intrinsic genotoxicity while exhibiting anti-colibactin activities against other pks+ bacteria.

a, flow cytometry analysis of DNA damage in Hela cells infected with EcN/EcN Δpks and NGF-1/NGF-1 Δpks. The MOI is 50. b, representative flow cytometry histogram for the surface display levels of OmpA-ClbS and OmpA alone in EcN and NGF-1. c, a schematic of bacteria-HeLa coculture. d, surface display of ClbS in two pks+ E. coli chassis strains lower their intrinsic genotoxicity while exhibiting anti-colibactin activities against other pks+ bacteria. Figures show representative data from three independent biological replicates (a, d). The MOI for NC101 or NC101Δpks is 20. The MOI for EcNOmpA-ClbS, EcNOmpA, NGF-1OmpA-ClbS or NGF-1OmpA is 400. Two-sided unpaired t-test (a, d). Data are means ± SD (a, d). Panel c created with BioRender.com.

Extended Data Fig. 4 |. Representative images of immunofluorescence staining of colonic epithelial cell types and colibactin-induced DNA damage in mice.

Immunofluorescence staining was performed on colon tissues from the NC101/Ec^OmpA group, which exhibited the highest DNA damage. E-cadherin for colonocytes, MUC2 for goblet cells, lysozyme for Paneth cells, chromogranin A for enteroendocrine cells and γH2AX for cells with DNA damage. Scale bar = 20 μm. Figures show representative data from two independent biological replicates.

Extended Data Fig. 5 |. NC101 can induce tumorigenesis in a colitis-associated mouse colorectal cancer model (DSS/Apc^Min/+).

a, timeline for the DSS/Apc^Min/+ mice treated with pks⁺ or NC101Δpks. Apc^Min/+ mice were pretreated with 2 g/L streptomycin in drinking water for three days for bacterial colonization, followed by 2% DSS in water for one week to induce colitis. Next, the mice were infected with NC101 or the isogenic mutant Δpks (10⁸ CFU/mouse) weekly for three weeks. b, Representative tumours in the colon on day 28 and quantification of colonic tumor counts (n = 5 mice per treatment group). Figures show representative data from two independent biological replicates (b, c). Unpaired t-test, two sided. c, body weight changes over the course of different treatments. Data are means ± SEM (b, c). Panel a created with BioRender.com.

Extended Data Fig. 6 |. Evaluation of Ec^OmpA-ClbS to target pks⁺ C. koseri in vitro and in vivo.

To test whether our engineered Ec^OmpA-ClbS strain can neutralize colibactin-induced damage from non–E. coli sources, we assessed its protective activity against Citrobacter koseri ATCC BAA-895, a taxonomically distinct pks⁺ bacterium. a, Genotoxic indexes of HeLa cells cocultured with C. koseri and Ec^OmpA-ClbS or Ec^OmpA for 4 hr (n = 3 technical repeats, N = 2 biological replicates). The MOI of C. koseri and HeLa cells is 50. The MOI of Ec^OmpA-ClbS or Ec^OmpA and HeLa is 400. b, Timeline for the DSS/Apc^Min/+ mice treated with C. koseri in combination with Ec^OmpA-ClbS or Ec^OmpA. c, Representative tumors in the colon on day 28 and quantification of colonic tumor counts. (n = 5 mice per treatment group). d, Representative image of H&E-stained colon Swiss roll sections. Scale bar: 800 μm (whole Swiss roll section); 50 μm (zoomed-in section). Data are means ± SD using two-sided unpaired t test (a, c). Panel b created with BioRender.com.

Extended Data Fig. 7 |. Landscape of microbiota composition and diversity in mice treated with NC101, NC101/Ec^OmpA or NC101/Ec^OmpA-ClbS.

a, Alpha diversity indices (Chao1 index and Shannon index) at days 11 and 28. In box plots, the bounds of the box indicate the 25th–75th percentiles, the center line indicates the median, and whiskers extend from the minimum to the maximum values. b, Principal coordinates analysis (PCoA) plots based on Bray–Curtis dissimilarity and unweighted UniFrac showing distinct clustering of fecal microbiota composition among NC101, NC101/OmpA-ClbS, and NC101/OmpA groups at days 14 and 28 (n = 5 per group). Percent variation explained by each axis is indicated. C, stacked bar plots of the top 10 bacterial families in fecal microbiota at days 14 and 28. d, relative abundance (%) of Enterobacteriaceae (family level) and Escherichia-Shigella (genus level) of fecal microbiota in different groups at days 11 and 28. Figures show representative data from one independent biological replicate (a, d). Data are presented as mean ± S.D. Statistical significance was determined by Kruskal–Wallis test (two sided) with original false discovery rate (FDR) multiple-comparison correction of Benjamini and Hochberg (a, d) separately in two different time points (Days 11 and 28).

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1038/s41564-025-02177-3.

Data availability

All data generated during this study are available within the article. The 16S rRNA sequencing data have been deposited in the BioProject database (ID PRJNA1335836) and made publicly available before publication. The relevant DNA sequences, materials and detailed information of reagents are provided in Supplementary Tables 1–5. Source data are provided with this paper.