After deciding on exploiting bacterial conjugation for the purpose of engineering a novel antimicrobial, we had to find an appropriate cargo which would be delivered on by the plasmid, This cargo also had to efficiently eliminate pathogens in a strain-specific manner.

Our first idea was to use a potent bacterial toxin expressed under a tandem promoter composed of pathogen-specific promoters. Harnessing bioinformatic tools, we searched for unique metabolic pathways, pathogenicity islands, analysis of ChIP-seq data, and literature reviews on synthetic promoters to identify associated pathogen-specific promoters. Ultimately, we gathered five promoters and arranged them in a tandem:

• HilD: Responsible for expression of prgH in Salmonella enterica subsp. enterica serovar Typhimurium str. ST4/74
• ToxT: Responsible for expression of tcpA in Vibrio cholerea
• CagA: Responsible for transcription of cagA gene in Helicobacter pylori
• Ler: Responsible for expression of LEE1 PI in pathogenic E.coli
• Anr: Responsible for expression of katA in Pseudomonas aeruginosa

Following the identification of promoters, we built the first simple plasmid in silico adding bacteriocin as the killing gene:

However, we ultimately decided not to pursue this idea further due to several challenges. Firstly, “strain-specific” promoters are poorly characterized outside the original target pathogen and E. coli, making it difficult to ensure true strain specificity within complex microbiomes containing hundreds of strains. Characterizing each promoter for use in such microbiomes would be an unrealistic task for our small team. Additionally, the design lacked control over bacteriocin expression, which risks killing not only the targeted pathogen but also surrounding commensal bacteria, leading to unregulated dysbiosis. Finally, the promoters we identified were either inducible only under specific stimuli such as stress or exhibited low expression levels, making them ineffective for pathogen elimination.

Due to the issues outlined in the “build” stage, we did not assemble the actual plasmid and thus did not test it.

Although we never brought this design to life, we learned the following:

• how to make an antimicrobial system pathogen-specific by focusing on pathogenicity islands, virulence factors, and antibiotic resistance genes
• the importance of control in the targeting system to ensure specificity of bacterial killing

After realising the issues with our initial idea, we sought an alternative killing system. This is when we found studies using RNA-guided nucleases, specifically Cas9, as a potential antimicrobial agent.

A 2014 study by Robert J. Citorik, Mark Mimee, and Timothy K. Lu demonstrated the use of the type II CRISPR-Cas system from S. pyogenes, delivered via ϕRGN bacteriophage particles or RP4 conjugative plasmid, to induce cytotoxicity. The system damaged DNA, disrupted toxin-antitoxin systems, and re-sensitized bacteria to antibiotics. However, escape mutants arose from mutations within the CRISPR locus, deletions of tracrRNA, or inactivation of Cas9’s nuclease domain.

Furthermore, a 2019 study by Thomas A. Hamilton et al. demonstrated that IncP RK2 cis-conjugative plasmids could be used as vectors to transfer a TevSpCas9 dual nuclease under the pBAD promoter, from E. coli to Salmonella enterica. The reported S.enterica killing efficiency for some targeted genes (e.g., katG and polA) was as high as 100%. However, the leaky expression of TevSpCas9 caused cytotoxicity in E. coli, selecting for mutated inactive plasmids, which compromised the long-term stability of the system. Notably, it was shown that in biofilms which are characterised by enhanced cell-cell contact, the frequency of conjugation increases to nearly 100%. Whereas typically, biofilms pose a significant challenge to the antimicrobial treatments including bacteriophages and antibiotics. Since many human microbial communities form biofilms, this method could offer faster, more efficient propagation of treatment within natural biofilms.

Also, a 2019 study by Marinelle Rodrigues et al. showed that conjugative delivery of CRISPR-Cas9 could selectively deplete antibiotic-resistant Enterococcus faecalis in the mouse intestine in vivo. In their experiments, following administration of antibiotics for seven days and a day-long pause, mice were gavaged with E.faecalis to model an acute case of infection. Then, mice were gavaged with donor E.faecalis carrying a targeting conjugative plasmid or a non-targeting control. After 23 days, it was found that mice colonized with donor bacteria possessing targeting plasmids had 0.1% of the total levels of erythromycin-resistant E.faecalis detected in the control mice. The specificity of CRISPR-Cas system was found to be absolute - every successful conjugation event led to the loss of erythromycin resistance. Also, it was observed that donor bacteria carrying targeting plasmids were prevented from acquiring antibiotic-resistance plasmids themselves. However, the study identified challenges, including self-targeting lethality from constitutive promoter expression and initially low conjugation frequencies.

These studies made us wonder whether we could utilise a CRISPR-Cas system as our killing cargo. From the studies it was evident to us that there were two significant problems with this approach. First, the conjugation efficiency in natural settings was reported to be insufficient to make it into an antimicrobial treatment. Second, Cas9 was clearly unsuitable for this task due to significant off-target effects, self-cytotoxicity, and inability to cut more than a single target.

To address these concerns, we explored alternative Cas endonucleases and eventually identified Cas12a as a more promising option. Using Cas12a offers several advantages including ability to multiplex that was demonstrated in the 2017 Zhang et al. study. This is enabled by the ability of Cas12a to process its own crRNA array producing multiple guiding crRNA each targeting it to a unique DNA sequence.

A single plasmid encoding DNase-dead Cas12a has been reported to be able to alter transcription of up 20 genes. In addition, Cas12a has a higher DNA recognition fidelity than Cas9 which reduces off-target cleaving. After considering all this information we came to the conclusion that Cas12a was a perfect candidate for the killing cargo delivered on our conjugative plasmid.

The next step was to design targets for Cas12a. Here we were able to use what we have learned about pathogen-specific elements from working on strain-specific promoters. First, in each pathogenic strain we identified a target: pathogenicity islands, toxin-antitoxin systems, antibiotic resistance genes, and virulence factors. Then in each sequence we manually found two Cas12a PAMs (TTTN) followed by 23nt long target sequences. To assess off-target effect we used BLAST sequence search to confirm that those sequences are not found in E.coli donor or in commensal microbiome.

Overall, we were able to design twenty-one sequences to target eleven genes from nine pathogens and two fluorescent proteins:

Target Genes:

The final synthetic array:

Next, we built our Cas12a plasmid in silico.

After several modifications, we used the pRPF144 plasmid backbone, which contains:

• ColE1: High-copy origin of replication for gram-negative bacteria
• OriV: Origin of replication for gram-positive bacteria
• RepA and RepB: Proteins for plasmid replication in gram-positive bacteria
• OriT and OriJ: Origins of transfer for RP4 conjugation apparatus

The unique synthetic target array had to be synthesized de novo, which posed a significant challenge due to its repetitive sequences, particularly the direct repeat (DR) spacers between the actual targets. After getting in touch with multiple companies, we identified ThermoFisher GeneArt as the only company capable of performing the synthesis. Fortunately, they kindly agreed to assist us in synthesizing the array.

Finally, the ddCas12a gene was obtained from pAC-ddCpf1plasmid. To introduce the mutation that reverts it to the active form we opted for to split the gene into two overlapping fragments. After amplifying the four fragments by PCR we put them together via Gibson reaction. However, this step turned out to be the experimental bottleneck because of low yields of the fully assembled plasmid. Consequently, after transformation and plating only a couple of colonies had grown. However, subsequent plasmid sequencing did not detect the appropriate arrangement of plasmids.

After experimenting with the Gibson assembly protocol, we were finally able to produce a plate with multiple colonies harbouring the OneRing plasmid.

To evaluate the killing efficiency of the plasmid we cultured several bacterial strains that can be targeted by Cas12a. We let our donor E.coli to conjugate with them delivering the plasmid. Then we measure elimination rate of the recipient bacteria to assess the plasmid’s effectiveness.

Unfortunately, we have not yet been able to demonstrate killing efficiency due to severe delays in oligos synthesis and shipment coupled with extensive experimental trouble shooting.

Although we have not yet tested this system, during design and build stages we learned a lot about how to make the kind of killing system we want to achieve. We also realised that for the future experiments we would have to simplify the experimental pipeline by constructing a unified backbone including the AsCas12a gene and make only the synthetic target array changeable by adding a multiple insertion site. Furthermore, after manually constructing the synthetic arrays we reckon we can generate an algorithm which would effectively search for pathogen-specific cutting sites, which would allow to highly specialise the system for any required group of pathogens.