Precise virulence inactivation using a CRISPR-associated transposase for combating Enterobacteriaceae gut pathogens

Abstract

Targeted gene manipulation in a complex microbial community is an enabling technology for precise microbiome editing. Here, we introduce BACTRINS, an in-situ microbiome engineering platform designed for efficient and precise genomic insertion of a desired payload and simultaneous knockout of target genes. This system leverages conjugation-mediated delivery of CRISPR-associated transposases (CAST) to achieve RNA-guided genomic integration, allowing precise insertion of a therapeutic payload while neutralizing pathogen virulence without causing cell death. When applied against an Enterobacteriaceae Shiga toxin-producing pathogen in the gut, this system delivers a CRISPR-associated transposase by bacterial conjugation for site-specific inactivation of the Shiga toxin gene and integration of a nanobody therapeutic payload to disrupt pathogen attachment. A single dose of this therapy results in high efficiency Shiga gene inactivation and improved survival in a murine infection model of Shiga-producing pathogen. This work establishes a new type of live bacterial therapeutic capable of reducing gut infections by transforming toxigenic pathogens into commensal protectors.

Ed summary:

A self-transmissible CRISPR-associated transposase system encodes a nanobody payload to treat Shiga toxin infections.

Advances in CRISPR-Cas technologies have created excitement for microbiome editing as a type of bacterial therapeutic, especially against pathogenic infections.¹ CRISPR-associated nucleases (e.g., Cas9, Cas12) have been explored as a strategy for targeted killing of pathogens by generating double-stranded DNA breaks (DSB) of their genomes.^2–10 Additionally, Cas3’s capacity for large genomic deletions and Cas13’s ability to cleave single-stranded RNA (ssRNA) have been harnessed for targeted pathogen killing.^10–12 One major challenge with these targeted killing methods is that pathogen elimination is never 100% effective, and therapy-resistant variants (e.g., present at a frequency of 10⁻⁴)¹³ inevitably emerge to cause secondary infections that are even less treatable. Therefore, more effective approaches against toxigenic pathogens are needed.

Among common bacterial infections, Shiga toxin-producing bacteria are a major cause of food-borne illnesses, responsible for ~300,000 deaths annually.^14,15 With an infectious dose of less than 10 cells, pathogenesis is caused by production of Shiga toxin, one of the most potent bacterial toxins.^16,17 The toxin is encoded by the Shiga toxin gene (Stx), which are primarily found in the genomes of Shiga toxin-producing Escherichia coli (STEC) and Shigella dysenteriae.¹⁸ Two major toxin subtypes of Shiga toxin exist, Stx1 and Stx2, with each consisting of an A and B subunit.¹⁹ Subunit B binds to host epithelial cells and drives uptake of subunit A, which subsequently cleaves host ribosomal protein, thus halting protein synthesis and causing cell death.¹⁹ Presence of Shiga toxin in the digestive tract leads to severe diarrhea, hemorrhagic colitis and hemolytic uremic syndrome (HUS).¹⁴ Elimination of the Shiga toxin and its bacterial carrier is therefore key to combatting its disease burden.

Currently, there are no effective clinical treatments for infections caused by Shiga toxin-producing bacteria, with palliative hydration being the only available care.^20,21 Importantly, use of antibiotics is avoided since antibiotics can increase production and release of the Shiga toxin, reduce gut motility, increase intestinal exposure to the toxin, and worsen clinical outcomes.^8,23 Various experimental treatments have been explored, including antibody therapy, inhibitors, antimicrobial agents, receptor analogs, vaccines, and phage therapy, but they have not gained clinical approval.^20,24–31 CRISPR-associated nucleases have also been used for targeted killing of STEC, but no therapeutic benefits have been demonstrated in animal disease models.^4,32,33 In addition to the relatively high rate of bacteria escaping CRISPR-mediated double strand break through mutations, this strategy would ultimately cause cell death and Shiga toxin release similar to antibiotic treatment, which is clinically contraindicated. Recently the discovery and engineering of CRISPR-associated transposases (CAST) that enable programmable, RNA-guided integration of large genetic payloads (>10 kb) into defined target sequences was described,³⁴ which has been used for targeted integration in various bacteria in vitro.^35–37 In parallel, we demonstrated in situ microbiome engineering by using bacterial conjugation to directly introduce genetic payloads into members of the gut microbiome.³⁸ These innovations, if combined, could produce bacterial ‘gene drives’ that massively expand specific allelic configurations across a population by broad range conjugation and precise genomic integration.

Based on these ideas, we developed a microbial therapeutic system dubbed Bacterial CRISPR-Transposase Reduction of Virulence In Situ (BACTRINS) that eliminates virulence genes in pathogens and reduces pathogenicity without direct killing, which avoids the emergence of resistant mutants. When applied against Shiga toxin-producing Enterobacteriaceae, the system exploits high-efficiency conjugation to mobilize CAST into target Shiga toxin bacteria to insert a disruptive therapeutic payload directly into Stx genes. When the Stx loci are inactivated, the bacteria become non-pathogenic and can further promote protection by expressing the nanobody payload to reduce pathogen attachment to the gut epithelia. We characterized the performance of BACTRINS in vitro and in a mouse model of STEC infection, demonstrating efficient delivery and integration of a therapeutic payload that improved outcomes after a single dose. This platform technology could be used for a variety of applications to introduce and/or remove targeted functions from specific strains of different human and environmental microbiomes.

RESULTS

High-performance pathogen editing with BACTRINS

The BACTRINS system is designed to precisely inactivate any gene of interest located on the genome and plasmids of a target strain by insertional disruption with a programmable CAST delivered through high-efficiency bacterial conjugation (Figure 1a). In contrast to strategies that inactivate pathogenic target genes with DSB-generating nucleases, which immediately kill any modified cells due to genome cleavage, BACTRINS does not kill the inactivated cells. Moreover, BACTRINS can carry genetic payloads encoding a variety of functions such as reinforced therapeutic activity, improved gene drive capacity, or enhanced host fitness, which cannot be achieved using other non-DSB-generating genome editing strategies (e.g., base editing).^39–43 When applied to the target Stx gene in Shiga pathogens, BACTRINS will site-specifically introduce the therapeutic payload into the middle of the Stx gene thereby rendering the resulting Stx protein product non-functional. Through inactivation of all Stx loci, BACTRINS-modified bacteria become non-pathogenic and obtain the ability to further propagate the conjugative therapeutic vector, even if the target genes are located on the genome. Since the system does not kill the target pathogen, there is substantially decreased selective pressure to develop resistance phenotypes.

Figure 1. Overview of BACTRINS system for bacterial virulence reduction in situ.

(a) BACTRINS can reduce bacterial pathogenesis by two combined strategies: 1) Elimination of Shiga toxin production by targeted insertion of a transposon into the Stx genes. 2) Use of a payload encoding a therapeutic function, such as a secreted nanobody that disrupts pathogen attachment to gut. (b) Illustration of design of CAST gRNAs bind to conserved regions of Shiga toxin genes in clinically isolated pathogens. The Stx1 sequence has 3 basepair mismatches between the tested strains while the Stx2 sequence is identical in both strains. (c) Map of BACTRINS vectors showing gRNA, CAST genes, transposon payload, origin of replication, origin of transfer, mobilization genes, and antibiotic resistance genes.

We first aimed to demonstrate that BACTRINS can precisely integrate into the Stx gene to disrupt Shiga toxin production. Stx genes from prototypic Shiga-producing species that have been clinically isolated (ATCC 2215^44,45, 13313^46,47, 43895^48,49, and 43889) were analyzed. Strains 2215 and 13313 contained only Stx1, and strain 43889 only contained Stx2. Strain 43895 carries both Stx1 on its genome, thought to be encoded by a cryptic prophage, and Stx2, encoded by the temperate bacteriophage 933W (Figure 1b).⁵² We computationally designed CAST gRNAs that could target highly conserved regions of both Stx1 and Stx2 while also minimizing off-targets across the rest of the genomes (Figure 1b). We hypothesized that disruption of either the A or B subunit of Stx would prevent downstream pathogenesis. As such, a universally conserved region was identified in the A subunit of Stx1 and the B subunit Stx2 (Figure 1b, Suppl. Table 1). These gRNAs are also applicable to most other bacterial strains harboring Shiga toxin genes, targeting sequences with less than 10% base pair variation (Suppl. Figure 1a).

To test the functional targeting capacity of the two Stx1A and Stx2B gRNAs, we separately cloned them into a pGT-B plasmid³⁸ containing the Type I-F Tn6677 CAST machinery (tniQ, cas6, cas7, cas8, tnsA, tnsB, and tnsC) and a 705 base pair transposon payload encoding an inert non-expressing chloramphenicol resistance gene without a promoter or ribosomal binding site³⁶ (Figure 1c). The plasmids were transformed into a donor EcGT2 E. coli strain³⁸ for further mobilization via RK2/RP4-mediated conjugation, yielding BACTRINS.v1 (BTS.v1) (Suppl. Table 2). The donor cells were individually mixed in vitro with each of the four target pathogen strains for conjugative delivery of the BACTRINS vector and transposon payload integration. Successful conjugation could be identified by agar plate selection for a kanamycin resistance gene on the vector, which showed in vitro conjugation rates ranging from 0.003% to 18% (Figure 2a). Differences in conjugation efficiency between strains may be due to presence of restriction modification systems, CRISPR-Cas systems, membrane receptors, and/or incompatible plasmids.^53–56 Successful and precise payload integration was determined via junction qPCR of the integrant product normalized to a reference housekeeping product outside of the integration site (Suppl. Figure 1b, Suppl. Table 3).³⁴ Overall, CAST-mediated payload integration efficiency into the Stx1A or Stx2B targets ranged from 42% to 82% (Figure 2b). While the payload can in theory integrate in either RL or LR (i.e., forward or reverse) orientation, we observed a strong (5 to 39-fold) preference for the RL orientation in all the strains (Suppl. Figure 1c). Given that the lower GI tract is generally anaerobic, we further tested the functionality of BACTRINS under anaerobic conditions in vitro. The conjugation efficiency was reduced by 2–13 fold while the payload integration efficiency remained similar at ~70% (Suppl. Figure 1d, Suppl. Figure 1e). These results demonstrated that BACTRINS.v1 can be delivered via conjugation and can mediate site-specific integration into the Stx1A or Stx2B locus of a variety of Shiga pathogens. We decided to proceed with S. dysenteriae 13313 (Sd13313) and Shiga toxin-producing E. coli 43895 (Ec43895) strains in downstream experiments due to their different Stx subtypes, species differences, and high in vitro efficiencies.

Figure 2. In vitro validation of BACTRINS system.

(a) Conjugation efficiency with different combinations of pathogen strains and Stx gene targets using BACTRINS.v1 (BTS.v1). (b) Integration efficiency of transposon payload in pathogen strains and Stx gene targets using BTS.v1. (c) Conjugation efficiency with Sd13313 and Ec43895 using BACTRINS.v2 (BTS.v2). (d) Integration efficiency of transposon payload with Sd13313 and Ec43895 using BTS.v2. All prior statistics are calculated by one-way ANOVA. Statistics for conjugation were performed on log-transformed data. (e) ELISA of Shiga toxin in pathogen supernatants. Statistics are calculated by two-way ANOVA with Bonferroni correction. (*) denotes P_adj<0.05 and (**) denotes P_adj<0.01. (***) denotes P_adj<0.001. (****) denotes P_adj<0.0001.

Conjugation efficiency is a key determinant for whether the BACTRINS system will provide sufficient in vivo Stx disruption in the gut microbiome. We therefore explored the use of a self-mobilizing conjugative plasmid (TP114) that efficiently transfers between proteobacteria in the gut⁵⁷ for BACTRINS.v2 (BTS.v2). We first characterized conjugation rates of the native TP114 plasmid from an E. coli MG1655 donor tagged with mCherry and spectinomycin resistance gene in its glms neutral site (glmS::mCherry-specR) to Sd13313 and Ec43895 recipients, which exhibited higher conjugation efficiencies by 3 to 7-fold than the pGT-B plasmid. Moreover, we confirmed that TP114 could mediate secondary transfers to recipient strains and effectively generate donors (Suppl. Figure 2a). For BTS.v2, the CAST machinery was cloned onto the TP114 plasmid (Figure 1c, Suppl. Table 2). Conjugation efficiency decreased by 6 to 11-fold in TP114 vectors harboring CAST (Figure 2c), while integration rates were comparable to the RP4-mediated conjugative delivery of pGT-B (Figure 2d). Nevertheless, the TP114 plasmid was maintained at 3 and 14 copies in the BACTRINS donor and Ec43895 recipient, respectively (Suppl. Figure 2b). We further tested the use of truncated transposon ends, which previously showed higher CAST integration under certain conditions.³⁶ However, we did not observe that truncated transposon ends improved conjugation or integration efficiency in vitro (Suppl. Figure 2c–d).

To confirm that CAST integration functionally disrupted the Stx protein products and decreased Shiga toxin production, we performed ELISA to measure Shiga toxin levels in Stx-disrupted Sd13313 and Ec43895 strains (Figure 2e). In Sd13313 which has only Stx1, toxin production was eliminated upon CAST integration of the payload into Stx1A. In Ec43895, which contains both Stx1 and Stx2, we found that disruption of the Stx2B locus led to a toxin reduction of 4.6-fold, while disruption of the Stx1A locus did not statistically affect toxin levels. This suggests that Shiga toxin in Ec43895 is mostly produced from Stx2, with a negligible amount being produced by Stx1. Collectively, these results demonstrate that BACTRINS.v2 can be efficiently delivered via conjugation using TP114 plasmid and mediate site-specific integration into Stx loci in Shiga toxin producing pathogens. Consequently, the modified pathogens exhibit significantly reduced or completely abolished Shiga toxin production.

Mouse infection models of Shiga toxin producing pathogens

To test the efficacy of the BACTRINS system in the gut, we established mouse infection models using Sd13313 and Ec43895 strains, which we genomically tagged with carbenicillin and tetracycline resistance genes respectively in their glmS neutral sites to allow plate-based selection and quantification. Consistent with previous studies showing gut colonization resistance by the native murine microbiome,⁵⁸ human Shiga pathogens could only be stably maintained in the mouse gut following an antibiotic cocktail pre-treatment (Suppl. Figure 3a, Methods). Mice colonized with Sd13313 did not display any symptoms or adverse side effects. In contrast, mice colonized with Ec43895 exhibited severe Shiga-associated pathology, including body weight loss of >20% and 100% fatality by 5 days post-infection (Suppl. Figure 3b–c). Histopathology analysis of tissues from these cohorts revealed substantial vacuolation and lipidosis in the liver and necrosis of epithelial cells in the kidney and colon in Ec43895-colonized mice, but not in Sd13313-colonized mice or control animals (Suppl. Figure 4). Clinically, pathogens with Stx2 are documented to cause more severe disease.^16,59 Therefore, we posit that the substantially more severe pathology in the Ec43895 mouse model is likely due to the additional presence of Stx2, consistent with the higher Shiga toxin levels observed in vitro for Ec43895 compared to Sd13313 (Figure 2e), which only carries Stx1.

To further confirm whether Stx1 or Stx2 was primarily responsible for the severe pathogenicity in Ec43895 and verify the key target for BACTRINS, we used BACTRINS to generate Stx-disrupted strains in vitro (Ec43895-Stx1A::Tn, Ec43895-Stx2B::Tn) and introduced them to separate mouse cohorts. While both Stx1A::Tn and Stx2B::Tn strains colonized the gut (Suppl. Figure 5a), the Ec43895-Stx2B::Tn cohort did not exhibit disease phenotype including weight loss and mortality. In contrast, Ec43895-Stx1A::Tn mice showed a similarly high level of mortality as wild-type Ec43895 (Suppl. Figure 5b, Figure 3a). These results suggest that the Stx2B locus is the prime target for attenuating pathogenicity of Ec43895 strain in this mouse model. We further speculated that disruption of Stx2B in Ec43895 would substantially attenuate virulence to become a live attenuated vaccine, which would allow the immune system to develop immune memory against the pathogen.⁶⁰ To test this hypothesis, we challenged mice previously exposed to the attenuated Ec43895-Stx2B::Tn with the wild-type pathogenic Ec43895. Prior to re-infection with wild-type pathogenic Ec43895, mice were re-treated with an antibiotic cocktail for 7 days to re-establish the infection model and tested to confirm the absence of any residual Ec43895-Stx2B::Tn from the prior inoculation. Two subsequent infections with the pathogenic Ec43895 did not result in any pathological phenotype; all the mice survived, and no weight loss was observed (Suppl. Figure 5c–d). These results suggest the possible benefit of in situ pathogen editing and attenuation to boost the immune system against future exposures.

Figure 3. In vivo characterization of BACTRINS system.

(a) Animal survival of modified Ec43895 where Stx1A or Stx2A are deleted compared to the wild-type control group. (b) Conjugation efficiency of BTS.v2 from donor to Sd13313 pathogen over time. (c) Integration efficiency of BTS.v2 into Stx2A of Sd13313 pathogen over time. (d) Gut colonization of Sd13313 in terms of colony forming units (CFUs) per gram stool over time in untreated, MG1655-treated, or BTS.v2-treated groups. (e) Animal survival exposed to Sd13313 in different experimental groups. (f) Conjugation efficiency of BTS.v2 from donor to Ec43895 pathogen over time. (g) Integration efficiency of BTS.v2 into Stx2A of Ec43895 pathogen over time. (h) Gut colonization of Ec43895 in terms of CFUs per gram stool over time in first 3 days post infection. (i) Animal survival of Ec43895 mouse model treated with BTS.v2 or MG1655 control. For all panels, n=4 per group. (*) denotes P_adj<0.05; (**) denotes P_adj<0.01; all tests are calculated by log-rank (Mantel-Cox) test.

To determine the minimal inoculation dose of Ec43895 needed to produce the severe pathogenic model, we performed a dose-response study. As expected, higher initial doses (i.e., 10⁸ cells) led to faster gut colonization by day 1 and greater weight loss over the course of 5 days, with 100% mortality by day 5. Remarkably, as few as 100 cells of Ec43895 were sufficient to expand massively in the gut to densities of >10⁹ colony forming units (CFU) per gram in fecal matter by day 2 and cause severe weight loss and lethality by day 5 (Suppl. Figure 6a–c). This finding is consistent with clinical observations that foods contaminated with a very small amount of Ec43895 (i.e., <10 CFUs) can cause illness in humans.¹⁶ Together, these results demonstrate the establishment of both asymptomatic and symptomatic mouse infection models of Shiga toxin pathogens and highlight the central role of Stx2B in pathogenesis for subsequent testing of the BACTRINS therapeutics.

BACTRINS-mediated inactivation of Shiga toxin gene in vivo

We used the asymptomatic Sd13313 murine gut colonization model to monitor the kinetics of conjugative delivery of the BACTRINS payload and the subsequent integration into its Stx gene without the confounding disease and lethality effects. Following initial introduction and establishment of Sd13313 (glmS::carbR), a dose of 10⁹ cells of the BACTRINS.v2 donor targeting Stx1A was orally gavaged 8 hours later. Fecal pellets from the mice cohorts were collected regularly over 20 days for pathogen analysis by plating and molecular assays (Methods). We observed that conjugation efficiency of the BACTRINS vector from the MG1655 (glmS::mCherry-specR) donor to Sd13313 (glmS::carbR) recipients reached nearly 100% by day 3 (Figure 3b). By Day 5, integration of the targeting payload into the Stx1A gene reached 100% in recipient pathogens, indicating that efficient delivery and target inactivation was achieved in vivo (Figure 3c). We were unable to detect Shiga toxin via ELISA in fecal samples from both the control and treated groups, likely due to the lower levels of production that were also observed in vitro (Suppl. Figure 7a). Since the native microbiota was depleted, the donor also persisted in the gut throughout the experiment (Suppl. Figure 7b). Notably, mice treated with the Stx1-targeting BACTRINS donor displayed a loss of the pathogen at least 3 days earlier than with Sd13313 alone or with the empty vector control donor (Figure 3d). As before, no pathological symptoms such as weight loss or death were observed in all Sd13313-colonized mice, indicating that the BACTRINS system could be safely administered and did not impart any additional adverse effects (Suppl. Figure 7c, Figure 3e).

Next, we tested the Stx2B-targeting BACTRINS.v2 system in the symptomatic Ec43895 (glmS::tetR)-colonized mice, which showed that the payload could be conjugated from the donor into ~76% of Ec43895 recipients in the gut based on stool analysis at 72 hours (Figures 3f). Integration and disruption of the Stx2B gene in Ec43895 reached 80–90% across the population in the presence of persistent pathogens and donors (Figures 3g–h, Suppl. Figure 8a). Despite these high conjugation and integration efficiencies, we did not observe statistically significant improvement in weight loss or survival compared to the control cohort (Suppl. Figure 8b, Figure 3i). This lack of BACTRINS therapeutic activity against Ec43895 could be due to several factors. First, the kinetics of payload delivery and Stx2B inactivation in Ec43895 may be too slow and tissue damage occurring within the first two days may have caused irreparable lethal damage. Second, Ec43895 infiltration deep into the gut epithelial tissue may prevent conjugative delivery of BACTRINS (not detectable by stool assays) and thereby limit therapeutic efficacy. Third, the remaining 10–20% of Ec43895 that still produce Stx2B might be sufficient to cause severe disease and mortality. Nevertheless, we showed that BACTRINS vectors can be efficiently delivered into Shiga-containing pathogens colonizing the gut and a payload could be integrated site-specially into target Stx loci.

Boosting BACTRINS against Shiga toxin with a nanobody therapeutic payload

To further improve the in vivo performance of BACTRINS against Shiga pathogens, we explored the use of payloads that can encode functional genes with therapeutic activity and be integrated into Stx genes with CAST. STEC and enteropathogenic E. coli (EPEC) form attaching and effacing (A/E) lesions on the surfaces of gut enterocytes through interaction of intimin, which is a bacterial outer membrane protein, and a bacterially encoded receptor protein (Tir) that is exported and translocated into the host cell membrane. Disruption of this gut adhesion property can potentially reduce Shiga-associated pathogenesis and help in its clearance.^61,62 Several nanobodies were recently described to be able to disrupt the attachment of Shiga toxin producing bacteria in vitro.^24,63

We hypothesized that a transposon payload able to inactivate StxB while simultaneously expressing a STEC adhesion-disrupting nanobody could improve the therapeutic activity of BACTRINS. The TD4 nanobody was chosen since it is the best documented nanobody that binds Tir to inhibit attachment. This BACTRINS.v3 (BTS.v3) design contained the TD4 payload fused to the synthetic E. coli secretion tag, NSP4, previously shown to have efficient extracellular secretion (Figure 1c).⁶⁴ Truncated transposon ends were used in BTS.v3 with the hope of further improving payload integration in vivo.³⁶ From the base BTS.v3 design, we constructed two variants BTS.v3a and BTS.v3b. BTS.v3a contained the TD4 payload and a non-targeting gRNA, which could conjugate into Ec43895 but not integrate into Stx2B. This variant tested whether expression of the Tir-binding nanobody alone is sufficient to produce a therapeutic effect. BTS.v3b contained the TD4 nanobody and the Stx2B-targeting gRNA, which tested the combined effect of STEC adhesion disruption and Stx2B inactivation (Suppl. Table 2).

With these constructs, we first confirmed their conjugation and integration rates in vitro in Ec43895, which showed similarly high or even better results than the previous BTS.v2 constructs (Suppl. Figure 2c–d). We then compared BTS.v3a, BTS.v3b, BTS.v2t (BTS.v2 with truncated transposon ends) and the donor control in the Ec43895 mouse infection mode (Suppl. Table 2, Methods). Mice were infected with 10² cells of Ec43895 and inoculated with 10¹⁰ cells of BACTRINS after 2 hours. Consistent with our previous findings, the pathogen and donors all effectively colonized the mouse gut, and all BACTRINS versions showed efficient payload delivery by conjugation into Ec43895 by Day 3 (Figure 4a–b, Suppl. Figure 8c). Strikingly, BTS.v3b, which contained the TD4 nanobody payload, showed an integration rate of 100% into Stx2b compared to only 60–70% for BTS.v2t with the inert payload (Figure 4c). ELISA of fecal samples showed up to a 3-fold depletion in Shiga toxin levels by day 3 for BTS.v3b compared to the empty donor control (Figure 4d), confirming functional reduction in Shiga toxin levels in the gut. Weight loss was observed in the control and treated groups upon infection (Suppl. Figure 8d). However, we found that BTS.v2t and BTS.v3a slightly increased survival of Ec43895-infected mice past day 5 to day 6 or 7, while BTS.v3b showed significant delayed lethality and a survival rate of 10% for more than 15 days (Figure 4e). Therefore, expression of the TD4 nanobody or inactivation of Stx2B alone can have some improved benefit to the Shiga model, but their combined effects have significantly higher therapeutic benefit. We subsequently challenged the surviving mice from the BTS.v3b cohort to a follow up round of antibiotic treatment and Ec43895 challenge, and all mice survived the second infection. Thus, this data further corroborates the possible benefit of in situ pathogen attenuation to protect against secondary re-infections. (Suppl. Figure 8e). In all, these results demonstrate the in vivo therapeutic effect of BACTRINS by simultaneous virulence inactivation and therapeutic payload expression.

Figure 4. Optimized BACTRINS system using a functional nanobody payload.

(a) Colonization of Ec43895 pathogen in different treatment groups (n=6–12 per group). (b) Conjugation efficiency of BACTRINS into colonized Ec43895 in different treatment groups (n=6–12 per group). (c) Integration efficiency of different BACTRINS designs (BTS.v2t = targeted inert gene payload; BTS.v3b = targeted TD4 nanobody) into Stx2B of Ec43895 compared to a non-targeted TD4 payload (BTS.v3a) (n=6–12 per group). (d) ELISA measurement of total Shiga toxin levels in stool for mice treated with donor carrying BTS.v3b or a MG1655 control strain (n=12 per group). (*) denotes P_adj<0.05, calculated by two-way ANOVA with Bonferroni correction. (e) Animal survival in different treatment groups compared to the MG1655 control group (n=9–21 per group). (*) denotes P_adj<0.05; (**) denotes P_adj<0.01; (***) denotes P_adj<0.001; (****) denotes P_adj<0.0001; all tests are calculated by log-rank (Mantel-Cox) test.

DISCUSSION

In this work, we presented a therapeutic platform that enables targeted in situ pathogen attenuation by concurrently inactivating virulence genes and integrating a therapeutic payload. This system leverages a self-transmissible CRISPR-associated transposase that encodes a nanobody payload to treat Shiga toxin infections. Using an efficient conjugation vector and carefully designed gRNAs that target conserved Shiga toxin sequences, the BACTRINS system can deliver and integrate a genetic payload into clinically important human Shiga-producing pathogens. The precise integration event effectively inactivates the Stx genes and reduces toxin production. Application of BACTRINS to mice infected with the lethal Ec43895 strain demonstrated reduced virulence and improved the animal’s life span and survival when a nanobody payload was used that could interfere with pathogen attachment to the gut. Additionally, this system can potentially function as a bacterial gene drive to further propagate the virulence inactivation functionality. Edited pathogens become non-pathogenic and the surviving mice are resistant to further infections, similar to the result of vaccination with a live-attenuated pathogen. These results establish BACTRINS as first precise and efficient in vivo microbiome editing platform, using a programmable CAST system against gut pathogens.

Currently, various CRISPR and phage-based anti-pathogen technologies are in various stages of development.^1,7,11,65,66 However, BACTRINS has several distinct advantages compared to these killing-based technologies. While CRISPR-based methods rely on irreparable nucleic acid damage and phage-based strategies primarily exploit bactericidal activity to eliminate pathogens, BACTRINS inactivates virulence factors without directly killing the target bacteria. This strategy reduces the evolutionary pressure that can lead to the emergence of resistance, as pathogen clearance is driven by more complex mechanism involving the immune system to respond to the attenuated pathogen. Additionally, BACTRINS has the potential advantage of amplifying therapeutic effects across the population as secreted nanobodies could spread and target pathogens that have not yet been edited.

The use of mouse infection models has limitations in completely recapitulating all clinical manifestations of human Shiga-producing pathogens. Human gut microbes and pathogens generally have a difficult time persisting in the murine gut due to colonization resistance properties of the native microbiome, which has been well documented.⁶⁷ Therefore, antibiotic pre-treatment is necessary to enable pathogen establishment, but could potentially introduce confounders not present in humans. The resulting level of the pathogen in the murine gut may be high (e.g., 10⁹ CFUs/g) compared to what is found clinically in people during an infection, where the bacterial load decreases rapidly over the course of the disease.^17,68 While the Sd13313 strain causes illness in humans, our mouse model did not show any pathological signs of disease in mice. Nevertheless, the asymptomatic Sd13313 infection model was used to evaluate the in vivo efficiency of payload delivery and integration. In contrast, the Ec43895 strain caused much more severe clinical symptoms in our mouse model than in the human population. We observed 100% mortality of mice by day 5 with Ec43895, which is much more extreme of an outcome than in infected people.⁶⁹ As such, the observed survival benefit from BACTRINS in the Ec43895 mouse model constitutes significant protection, even if not preventing death in all cases, and is likely have a therapeutic effect for Ec43895 infections in humans, which shows less severe pathology and mortality.

This work represents a foray into in vivo pathogen editing enabled by high-performance CAST systems. Future studies could benefit from improvements such as better functional characterization of the TD4 nanobody, more potent nanobody payloads⁷⁰, catalytically faster CAST variants⁷¹, protection of the donor strain during gut transit^72,73 and better conjugation machinery or alternative gene delivery systems such as phage.^7,74 Furthermore, studies would be needed to determine the optimal dosing and timing of treatment to achieve sufficient therapeutic effects in human patients. Other studies using E. coli-based strategies have required high dosing and removal of genotoxic properties.^75–77 Future strategies using microbial encapsulation or oral formulation methods could provide a dual advantage: reducing required dosages while enhancing local gut concentrations of donor cells.^72,73 Such strategies could help mitigate potential adverse effects due to the administration of high levels of bacteria in clinical settings. Although the current experimental design uses antibiotic resistance markers for selection, these markers were utilized solely to enable tracking and calculate efficiencies in preclinical settings. Any therapeutic BACTRINS product would prioritize marker-free approaches, such as sequencing-based methods⁷⁸ or fluorescence-based reporters,³⁸ to prevent the spread of antibiotic resistance. Additionally, biocontainment strategies, such as auxotrophic systems or inducible kill switches, could be incorporated to mitigate the unintended dissemination of edited bacteria.⁷⁹

While this study demonstrates in vivo use in two strains of Shiga-toxin producing bacteria, further validation on an expanded set of pathogenic strains may be needed for clinical advancement. However, considering the broad spectrum of phyla that our in vivo microbiome editing technology was able to infiltrate³⁸ and the diverse range of species in which CAST systems have been shown to function,^35–37 we believe that the BACTRINS platform can potentially be adapted to target other gut pathogens by modifying the plasmid and regulatory elements. Thus, this system could be more broadly applied to other gut pathogens or endogenous pathobionts with known toxin genes, such as Clostridium difficile, Clostridium perfringens, or Bacteroides fragilis. The exploration of diverse payloads with different capabilities such as immune modulation, wound-healing, or metabolic functions could also create additional therapeutic opportunities for gut microbiome editing.⁸⁰ We anticipate that precise single-step knockout of genes and their replacement with functional variants will be a crucial strategy for modulating the microbiome in humans and livestock with numerous potential biomedical and agricultural applications.

MATERIALS AND METHODS

Sequence analysis and gRNA design

Human pathogenic strains annotated as containing Shiga toxin genes were purchased from ATCC. 2215 has the complete ATCC reference number of BAA-2215, is classified as E. coli, has the strain designation 2006–3008, and is of serotype O103:H11. 13313 is classified as S. dysenteraiae and has the strain designation NCTC 4837 strain Newcastle. 43895 is classified as E. coli, has the strain designation CDC EDL 933, and is of serotype O157:H7. 43889 is classified as E. coli, has the strain designation CDC B1409-C1 [1271–84], and is of serotype O157:H7. The sequences of these commercially available strains were downloaded from ATCC and analyzed for the presence of Stx1 and Stx2. The Stx1A and Stx2B sequences were then aligned using Geneious software. gRNAs for conserved Stx1A and Stx2B sequences were designed as previously described^34,36, so that integration ~50bp downstream would disrupt the coding region of the Shiga toxin gene. Both gRNAs are 32bp in length and use the PAM motif, CC; a non-targeting gRNA sequence was used as a negative control (Suppl. Table 1).

For conservation analysis of all shiga toxin sequences, E. coli and S. dysenteraiae genomes were accessed through NCBI genome portal (https://www.ncbi.nlm.nih.gov/datasets/genome/, access date: May 25, 2024) and all genomic coding sequences (CDS) from chromosome-level or complete genomes were obtained. All coding sequences were then compared again reference sequences of Shiga toxins (Stx1A, Stx1B, Stx2A, Stx2B) by BLASTN v2.7.1⁸¹ with ‘-evalue 10 -perc_identity 90’ and query coverage threshold greater than 90%. Blast hits were considered as shiga toxin variants and multisequence alignment was performed on variants of each Shiga genes (Stx1A, Stx1B, Stx2A, Stx2B) using MUSCLE v5⁸² to calculate mismatch rate at each position.

Plasmid and strain construction

The versions of BACTRINS were constructed as follows (Suppl. Table 2). The pGT-B construct containing CAST machinery, an inert payload, and kanamycin resistance was obtained from the Sternberg lab at Columbia University.³⁶ The inert payload consists of the 705 base pair chloramphenicol resistance that is not expressed due to lack of a promoter. gRNAs were cloned into the construct using BsaI restriction sites. The vector was digested with BsaI, and gRNA sequences were generated by phosphorylation and annealing of the two ordered complementary oligonucleotides. The digested vector and hybridized oligoduplex were then ligated and transformed into NEB turbo competent cells. The constructs with correct gRNA sequences were then transformed into the EcGT2 E.coli donor strain containing Δasd::mCherry-specR modifications.^38,83 The TP114 construct was obtained from the DSMZ-German Collection of Microorganisms and Cell Cultures GmbH (DSM No. 4246).⁵⁷ gRNAs were first cloned into the pGT-B construct as described above. The TD4-NSP4 nanobody payload was also first cloned into the pGT-B construct. The TD4-NSP4 payload sequence was ordered as a geneblock based on previously reported sequences^24,64 and then cloned into the payload region using NEB Gibson Assembly Master Mix. Any edits in the CAST machinery were then cloned into the TP114 vector using double recombinase insertion of DNA (DROID)⁸⁴. All PCR reactions were done with Q5 DNA polymerase (NEB). The completed construct with kanamycin and chloramphenicol resistance was then electroporated into an MG1655 E. coli strain that had spectinomycin resistance and mCherry expression on the genome. The specR and mCherry were cloned using λ red recombineering⁸⁵ into a neutral site upstream of the glmS locus. The same recombineering was used to clone carbenicillin resistance onto the genome of strain 13313, and tetracycline or erythromycin resistance was cloned onto the genome of strain 43895 all upstream of the glmS locus. Ec43895-Stx1A::Tn and Ec43895-Stx2B::Tn strains were generated in vitro using BTS.v1 and validated via PCR.

In vitro conjugation

Donor and recipient strains were grown overnight at 240RPM at 37°C in 5mL of Luria Broth (LB) supplemented with their appropriate antibiotics; the EcGT2 donor was also supplemented with diaminopimelic acid (DAP). For studies under anaerobic conditions, the EcGT2 donor strain was grown aerobically prior to conjugation due to poor anaerobic growth. The recipient pathogenic strain was grown anaerobically. 1.5mL of cultures were spun down at 6500g for 4 minutes and washed with 1mL of phosphate buffered solution (PBS). This wash was repeated two more times and resuspended in 1mL of PBS. OD600 was used to determine cell count. ~8×108 cells (OD: 1) of the donor and recipient were added to the same tube and spun down at 6500g for 4 minutes. The supernatant was removed, and the pellet was resuspended in 10 μl of PBS. The 10 μl was then spotted on LB overnight at 37C; conjugations with the EcGT2 donor were supplemented with DAP. The following day, bacterial growth was scraped into 1mL of PBS and homogenized with a pipette. 10x serial dilutions were made in PBS, and 5 μl of each sample was plated both with and without the antibiotics on LB. Plates were incubated overnight at 37C. Conjugation was calculated by determining the ratio of colony forming units (CFUs) with and without antibiotic selection for the plasmid. Secondary conjugation of strain 43895 was determined using the cloned strains with different antibiotic resistance.

Integration

Integration was calculated within the transconjugant population, or the pathogenic recipients that received the plasmid. Transconjugants from non-confluent spots were scraped from into 1mL of PBS. Optical density measurements at 600 nm were taken of scraped colonies that had been resuspended and approximately 3.2 × 10⁸ cells (the equivalent of 200 μl of OD600 = 2.0) were transferred to new tubes. Cells were pelleted by centrifugation at 4,000g for 5 min and resuspended in 80 μl of H2O, before being lysed by incubating at 95 °C for 10 min in a thermal cycler. The cell debris was pelleted by centrifugation at 4,000g for 5 min, and 10 μl of lysate supernatant was removed and serially diluted with 90 μl of H2O to generate 10 fold dilution for qPCR analysis. qPCR primers were designed for a reference product, forward (RL) integration product, and reverse (LR) integration product (Suppl. Table 3). These PCR products were ~200bp in size with a target melting point of 60C. The reference product amplified a product within 500bp of the targeted integration site. For integrated product amplification, one primer bound outside of the integration site and one primer bound within the payload. qPCR was run on a Biorad C1000 thermocycler using KAPA SYBER FAST mastermix. Thermocycling conditions were as follows: polymerase activation and DNA denaturation (95°C for 3 min), 40 cycles of amplification (95°C for 3 s, 60°C for 20 s), and terminal melt-curve analysis (65–95°C in 0.5°C per 5 s increments). All biological samples were run in triplicate, and the average quantification cycles (Cqs) were calculated. qPCR integration was calculated by taking the difference of 2^{delta Cq} between the reference and integration products as previously described³⁴. Total integration was calculated by summing the integration of the two products. This method of using a reference site outside of the integration product was validated using clones that were fully integrated, and melting curves were validated for each PCR product.

Animal studies and sample collection

All animal experiments were performed in compliance with Columbia University Medical Center IACUC protocol AC-AABJ5552. Female mice (C57BL/6J) were purchased from the Jackson Laboratory at 6 weeks of age. Within each experiment, mice originated from the same vivarium and were within 5 grams of each other in body weight.

For studies with antibiotic pre-treatment, mice were given ad libitum access to an antibiotic cocktail for seven days. The antibiotic cocktail consisted of Ampicillin (0.5g/L) (Sigma-Aldrich A9518), Gentamycin (0.5g/L) (Thermo AC455310050), Metronidazole (0.5g/L) (Thermo Fisher AC210340050), Neomycin sulfate (0.5g/L) (Thermo Fisher AAJ6149914), Vancomycin (0.25g/L) (Thermo Fisher AAJ6279006), and Sucrose (20g/L) (Thermo Fisher A15583). These antibiotics were dissolved in tap water and sterile filtered through 0.22 μm filter prior to administration. After the 7 days with antibiotic water, mice fecal samples were collected and plated on LB aerobically to ensure no background growth. Mice were then given ad libitum access to water without antibiotics for ~30 hours prior to gavage of strains. After antibiotic treatment and prior to gavage, all mice were mixed evenly across new cages.

Strains for gavage were grown overnight, shaking at 240 RPM at 37°C in LB with selective antibiotics. Cultures were spun down at 4000g for 5 minutes, supernatant was removed, and the same volume of PBS was added to resuspend the cells. This wash step was repeated two more times. OD₆₀₀ of the washed culture was taken and used to calculate dilutions for the gavage dose. Mice were then gavaged with the appropriate number of cells in 200 μl of sterile PBS. The following conditions were as follows for each in vivo experiment: For the development of the mouse model, 10⁸ of cells were gavaged to each mouse. For the treatment characterization with 13313 and 43895, mice were gavaged with 10⁴ cells of the pathogen and then gavaged with 10⁹ cells of the donor 8 hours later. The fully integrated strains were also gavaged with 10⁴ of cells. For the improved treatment with 43895, mice were gavaged with 10² cells of the pathogen and gavaged with 10¹⁰ cells of donor 2 hours later. For re-infection experiments, mice were given the same antibiotic pre-treatment as before. Mice were gavaged with 10⁴ of the wild-type pathogen tagged with antibiotic resistance, and colonization of the pathogen was confirmed 6 hours later by plating fecal samples.

Fecal sample analysis

Fecal samples were collected and stored at −80°C. ~10–25mg of the fecal sample was taken and the weight recorded. 250 μl of PBS was added to the tube and a pestle was used to homogenize the sample. 750 μl of PBS was then added and the tube was vortexed for 15 seconds. The tube was then centrifuged at 100g for 30 seconds to pellet the bulk fecal matter. 100 μl was removed and serially diluted 10x in PBS. 5 μl of each serial dilution was then spot plated on LB with appropriate antibiotics and incubated at 37°C overnight. Colonization of the donors and pathogens was determined by plating on LB with selective antibiotics for that strain. Colonies were counted and CFU/g was calculated based on dilutions and fecal sample weight. Conjugation was calculated as before by determining the ratio of colony forming units (CFUs) with and without antibiotic selection for the plasmid. Integration was determined as before in the transconjugant pathogenic population.

ELISA

Shiga toxin levels were determined using an ELISA kit for simultaneous detection of Stx1 and Stx2 in a single test (Creative Diagnostics DEIASL162). For in vitro studies, pathogens were grown in LB with appropriate antibiotics shaking at 240 RPM at 37C overnight. Cultures were spun down at 4000g for 5 minutes and supernatant was used for the ELISA. For in vivo studies, samples were normalized by taking the sample weight (mg) and adding 20x that volume of PBS (μl). 50 μl of sample was mixed with 200 μl of diluent, and 100 μl of the diluted sample was used in the ELISA. The absorbance (450–620nm) of the ELISA was determined using a Biotek Synergy H1 plate reader. Samples were diluted as necessary to not exceed an absorbance of 3.

Histopathology

Tissue samples of liver, kidney and colon were submitted for histopathology analysis by Columbia core facility using focal, multifocal, and diffuse microscopy. Findings of vacuolation, lipidosis, and necrosis were graded by presence and severity.

Statistical Analysis

Group means, fold change, standard errors/deviations were calculated in Microsoft Excel. For multiple comparisons, one-way analysis of variance (ANOVA) was performed with plotted statistical values in Graphpad Prism. For mixed-effects analysis, two-way ANOVA with Bonferroni correction was performed with plotted statistical values in Graphpad Prism. For survival curve comparison, log-rank (Mantel-Cox) test was performed with plotted statistical values in Graphpad Prism. A value of p<0.05 was considered significant, with P values marked in corresponding texts and figures. Error bars represent standard error of the mean, unless otherwise noted.

Data Availabity:

All raw experimental data for the figures are available in the SD.xlsx file provided. Raw data of supplementary material are available upon request.