Results
Phage Spotting
To validate the phage–host receptor panel used in downstream experiments, we performed phage spotting assays in which Lambda, HK97, P2 and Mu were spotted onto E. coli strains that differ only in outer-membrane receptor or LPS composition. In this spot assay, small volumes of phage lysate are applied to a confluent bacterial lawn and lytic infection produces clear zones (plaques) where bacteria are killed; the presence, number, and size of plaques reflect phage infectivity on that host. Each phage was tested on its cognate target strain and on control strains lacking the presumed receptor (ΔLamB or ΔOmpC) or expressing a truncated R1 LPS.
The results in Fig. 1 below confirm the expected phenotype – the WT protein-binding phages (Lambda and HK97) have high spotting efficiency towards the LamB receptor (ΔOmpC) and none towards OmpC receptor (ΔlamB), and the WT glycan-binding phages (P2 and Mu) have high spotting efficiency towards the K12 LPS and much less so towards the truncated R1 LPS. This validates our positive control for the phage spotting, and that these receptor types are suitable for examining how RBP swapping might change host specificity.
The wild-type protein binding phages Lambda and HK97 had high spotting efficiency in the OmpC knock-out E. coli, and low efficiency in lamB knock-out E. coli. Similarly, the wild-type glycan-binding phages P2 and Mu had high spotting efficiency in E. coli expressing the K12 LPS, and low efficiency against E. coli expressing the truncated R1 LPS.
sgRNA Validation
To assess activity of candidate single-guide RNAs (sgRNAs) against our phages, we performed Cas-dependent plaque-reduction spot assays. Bacterial lawns were co-transformed with tCas9 or I-C plasmids expressing a Cas protein and either a sgRNA or an empty sgRNA cassette (empty vector, EV). The tCas9 plasmid expresses S. pyogenes Cas9, regulated by the arabinose-inducible PBAD promoter. I-C expresses a cascade (multiprotein complex) comprised of Cas5c, Cas7c, Cas3, and Cas8c proteins. Expression is regulated by the rhamnose-inducible promoter system (RhaSR-PrhaBAD).
Upon Cas-protein induction, a functional sgRNA was expected to prevent plaque formation by its cognate phage while leaving non-target phages unaffected. Results and interpretation of these assays is shown in Figs. 2 and 3 below.
P2 phages were spotted on E. coli K12, which were transformed with tCas9, which inducibly expresses Cas9, and either a sgRNA for a specific phage or an empty sgRNA cassette (empty vector, EV). We expect plaque formation in all conditions (uninduced P2 EV, P2-1; induced P2 EV) except for induced P2-1, where Cas9 is expressed and the sgRNA is present.
P2, Mu, and HK97 phages were spotted on E. coli K12, which were transformed with the pCas3Rh plasmid, which inducibly expresses the Type I-C CRISPR-Cas components and either a sgRNA for a specific phage or an empty sgRNA cassette (empty vector, EV). We expect plaque formation in all uninduced and EV conditions. When Cas expression is induced and the sgRNA is present, we expect reduced plaque formation. Note: Mup49 and Mup52 refer to two different RBDs for the Mu phage.
Plaques were smaller and fainter in Cas-induced and sgRNA-present conditions than EV or uninduced conditions in both tCas9 and IC plasmid systems, indicating that Cas expression and sgRNA targeting may have partially interfered with phage replication. However, the spotting efficiency was lower than the average spotting efficiency for all phages overall for all conditions, including controls. This suggests that the bacteria are behaving abnormally under induction conditions, since we expect robust plaque formation in uninduced and empty-vector controls.
The inconsistent results from tCas9 and IC plasmids prompted us to test other Cas-protein and sgRNA expression vectors, leading us to select pCas9 and pCRISPR. pCas9 expresses S. pyogenes Cas9 (SpyCas9) under a constitutive promoter, and pCRISPR is a compatible sgRNA expression plasmid with an independent origin of replication and antibiotic resistance marker. The results of the subsequent testing are shown in Figs. 4 & 5 below.
Lambda, HK97, P2, and Mu phages were spotted on E. coli K12, which were transformed with pCas9 expressing Cas9, and either a sgRNA for a specific phage or an empty sgRNA cassette (empty vector, EV). We expect all phages to form plaques except for the phage being targeted by the gRNA. Mup49 & Mup52 are used to denote two different RBDs of the wild-type Mu phage. Results indicate that all sgRNAs did not inhibit plaque formation.
In a subsequent experiment, pCas9 was co-transformed with another plasmid, pCRISPR, in E. coli K12; here, the plasmid expressing Cas machinery was separated from the sgRNA expression vector, shown in Fig. 5 below.
Lambda, HK97, P2, and Mu phages were spotted on E. coli K12, which were co-transformed with pCas9 (expressing Cas9 with an empty sgRNA cassette) and either pCRISPR with a sgRNA for a specific phage or an empty sgRNA cassette (empty vector, EV). We expect all phages to form plaques except for the phage being targeted by the sgRNA. Mup49 & Mup52 are used to denote two different RBDs of the wild-type Mu phage. Positive control 1 and 2 are sgRNA targeting K12 that were previously validated in the Davidson Lab. Results indicate that all sgRNAs did not inhibit plaque formation. However, since positive controls 1 & 2 also did not show reduction in plaque formation, we interpret this as systemic error in the experimental setup, which leads us to new avenues for troubleshooting.
These data indicate that while native phage-receptor interactions were robustly validated, with protein-binding (Lambda, HK97) and glycan-binding phages (P2, Mu) showing the expected host-range patterns, the CRISPR-Cas interference assays did not provide reproducible evidence of sgRNA-mediated phage inhibition. Inducible tCas9 and Type I-C systems produced smaller, fainter plaques under induction yet showed low overall spotting efficiency and loss of expected control phenotypes, implicating induction conditions, plasmid burden, or other systemic experimental error. That compromised bacterial or phage fitness. As a result, we moved to a constitutive pCas9/pCRISPR configuration to decouple Cas and sgRNA expression and minimize confounding effects; further investigation will be required to conclusively evaluate sgRNA activity.
Gibson assembly of dry lab-generated RBDs
Gibson assembly for the dry lab’s first batch of Mu and P2 RBDs was validated by Sanger sequencing.
sgRNA Troubleshooting
We are in the process of troubleshooting our sgRNAs, this time examining fundamental parts of our experimental design (such as the types of E. coli K12 strains we are using) to see where errors might be occurring. We suspect there might be issues with our bacterial strain having natural chloramphenicol resistance, which would prevent it from uptaking the pCas9 plasmid. By investigating these issues and making necessary troubleshooting steps, we hope to have working sgRNAs in the coming weeks.
Gibson Assembly of New RBD Batches
At the time of writing this wiki, we are in the process of ordering and receiving new batches of RBDs — batch 1 of generated HK97 and Lambda RBDs (arrived), and batch 2 of generated Mu and P2 RBDs (to be ordered). Using existing primers we previously ordered, we are ready to complete Gibson Assembly to clone these RBDs into pETDuet-1.
We will confirm successful Gibson Assembly of the new RBDs into pETDuet-1 by Sanger sequencing.
Recombination of RBDs Using Validated sgRNAs
Immediately following successful sgRNA validation, we will proceed with recombination of our cloned RBDs into lysogenic E. coli K12, validating successful swaps via colony PCR with checking primers. Afterwards, we will conduct the same phage spotting assay shown in Fig. 1, this time with the mutant phages (and wild types as controls). We are working to complete these experiments in the coming weeks.
Immediate Future Experiments
To characterize generated RBDs that enable successful phage targeting of the above bacterial receptors, we will employ adsorption and growth curve assays. These serve to characterize the results of the spotting assay in two ways. First, the adsorption assay interrogates the binding efficiency of the native and designed RBDs to the bacterial receptor it targets. The growth curve quantifies the phage spotting assay using phage titres, allowing for an objective assessment of lytic efficiency of the bacteriophage of interest. A one-step growth curve is achieved by inoculating a sensitized bacterial culture with the phage of interest and sampling the culture at various time-points to perform a plaque assay, through which the changes in phage titres can be quantified and plotted over the infection time[1].
Future Directions in Phage RBD Discovery and Optimization
Advances in structural biology will be critical once a working AI-designed RBD is identified. Cryo-electron microscopy (cryo-EM) is especially well-suited to phage tail complexes because it requires no crystals and can image very large assemblies in a near-native state. Recent cryo-EM studies routinely achieve near-atomic resolution on whole phage particles and tail fibers[2]. X-ray crystallography, by contrast, can deliver the highest atomic detail for isolated RBP domains (especially rigid tailspikes), yielding clear structures of the binding interface[2]. In practice one would use both: cryo-EM on the intact phage (or tail complex) to visualize overall architecture and cryo-electron tomography (cryo-ET) for in situ context, while X-ray diffraction of purified RBD or tailspike fragments provides high-resolution models of the receptor-binding site[2]. Together these structures (“blueprints”) help explain how the RBD docks to its receptor and guide further model design and engineering.
High-throughput library screening will let us rapidly test thousands or millions of AI-generated RBD variants. For example, one could build a phage- or yeast-display library of RBD sequences and perform iterative “biopanning” rounds against the target bacterium or purified receptor. Phage display leverages extremely diverse libraries (often > 10⁹ variants) and enriches binders through repeated selection rounds, quickly winnowing a huge pool down to a few strong candidates[3]. Coupling this with deep sequencing or high-content sorting allows parallel measurement of many variants’ binding. Approaches like surface plasmon resonance (common in antibody engineering) would let us scan the AI-designed library at scale and pick out functional RBDs for further validation[4].
Finally, directed evolution can refine any hits and generate data to improve the AI model. In practice this might mean taking a promising AI-designed RBD, introducing diversity (e.g., by error-prone PCR, DNA shuffling or mutagenesis plasmids), and then selecting for tighter binding or broader host range over multiple rounds. For example, phage-assisted continuous evolution (PACE) or cyclic error-prone phage propagation could iteratively improve binding affinity and specificity. These methods are analogous to antibody maturation and have been used to expand phage host range by mutating tail fiber proteins[2]. Each round of selection generates new sequence–function data (which RBD mutations enhance binding), which could be fed back to retrain the AI model. In this way, iterative cycles of design → experimental selection → data-driven re-design would converge on optimized RBDs that both work in vitro and enrich the training set for future predictions[2].