Based on the results of the Testing and Modeling groups, we arrived at the following simple model for how StarCores function.

We were expecting that the free AMPs behaved differently when they are fused to a protein scaffold. The previously available literature that describe free AMPs does not seem to apply for our constructs. With these considerations in mind, we set out to improve on the basic StarCore designs developed in the Design Section. Our method of choice was a combination of synthetic random library, rational mutagenesis and machine learning aided design.

To acquire comprehensive knowledge on the AMP’s behavior in a fusion, we obtained the bacterial self-targeting efficiency of a randomly synthesized library of 0.3 million peptides. Then we built a custom-made machine learning guided, AMP optimization software, AMP Designer (see details in our webpage Paris_Bettencourt/Software ). Using this software, we found a specific engineering method that changes the charge distribution around the positive cluster on AMP. Based on this semi-rational approach, we designed ~ 12,000 variants library that we we attached to mCherry through a linker in order to optimize the already characterized effectiveness of core-AMPs. The common pattern discovered in our top-efficient AMP fusions showed a strong preference on charge-reduced peptide instead of the increased, indicating the important trade-off between AMP’s efficiency alone, and the AMP’s impact on protein expression, folding and assembly.

Design of the AMP Twist Library

At the core of the StarCore library are three AMPs: Ovispirin, X, Y. We chose to fuse each AMP to the fluorescent protein mCherry. This allowed us to easily quantify the protein expression level. It also served as a “simulated core” that allowed us to test AMP activity in the context of a large fusion protein. As we learned in the production section, protein expression and folding are critical to obtaining an effective StarCore.

We chose to focus library diversity on positively charged residues. Charge density is known to critically affect both protein folding and AMP function. Therefore, we systematically generated mutations to move, concentrate and amplify positive charges. These mutations are described in figure 1.

We cloned the DNA sequence library into a T7-based expression vector and transformed it into a strain of self-lysing E. coli. Cells were grown to mid-log phase then induced for StarCore expression and self-lysis. The resulting lysates were added to fresh cultures of E. coli. Peptide expression was measured by mCherry fluorescence and effects on growth were quantified by OD.

Of 12 000 total clones screened, we selected the 200 most effective for sequencing and characterization.

Library design

Roughly 12 000 variants were designed using 5 templates from naturally occurring peptides. 3 types of variation rules were applied:

Due to their short length, AMPs were expressed as a fusion protein to prevent post-transcriptional degradation. The fluorescent reporter mCherry was chosen for this purpose. A long flexible linker was chosen to give the AMPs freedom to interact with the membrane.

Due to their short length, AMPs were expressed as a fusion protein to prevent post-transcriptional degradation. The fluorescent reporter mCherry was chosen for this purpose. A long flexible linker was chosen to give the AMPs freedom to interact with the membrane.

We first used cold fusion assembly to construct a low-copy expression vector containing mCherry. Into this vector, the Twist DNA library was cloned using Golden Gate assembly.

Once our pDuet-mCherry vector obtained, we pursue the cloning procedure by performing Golden Gate assembly of the all the library. This method enabled us to join the 12k variants in a one pot reaction. For this, 60 cycles of digestion/ligation were realized to minimize the level of empty vector. High efficiency electrocompetent cells were used to recover the all library size. After plasmid purification, the plasmids containing the library were transformed once more into expression cells.

To express our constructs, we chose to use XJ BL21(De3) autolysis strain. These cells are suitable for T7 regulated protein expression. The production of the lambda lysozyme is inducible by addition of arabinose. Subsequently, the cells can be efficiently lysed by one freeze/thaw cycle as the bacterial membrane is fragilized by the endotoxin.After transformation, we manually picked around 5k single colonies and grown them overnight. Auto-induction media was preferred to other complete media to decrease the pipetting steps.

We obtained a cell lysate by freeze-thawing our XJ bacterial strains, that are specifically engineering to lyse reliably. It is this lysate that contains our AMPs, and is used in lieu of antibiotics for the killing assay. The lysate was diluted 1:4, and added to 96 well plates containing LB. Once made, the wells were inoculated with E.coli (ask which strain), that was pre- incubated for 2h at 37 C. The OD600 and mcherry fluorescence of the original lysate plates was measured and the OD600 was measured for the killing plates after 8h to determine the killing effect. Blanks including only LB-lysate were used as a negative control. AMP containing-lysate was used, as opposed to only internal expression because many AMPs that are effective when produced in vivo, often display different properties once isolated, probably due to poor stability, half-life, or inability to penetrate lipid membranes to reach their targets.

Fusion proteins allow us to produce a large variety of unnatural AMPs, compare to natural extraction or chemical synthesis of peptides. link between mCherry production and fluorescence As a future direction of this project, we would aim to define a biologically relevant and high throughput assay for screening AMPs. With our method, we can not assess the mode of action of AMPs or its target. We would also like to test the cytotoxicity of the constructs against different cell lines. Developing the use of compartmentalized self replication through the droplet microfluidics can be a possible way.

1.Bechinger, B., and S-U. Gorr. Antimicrobial peptides mechanisms of action and resistance. Journal of dental research 96.3 (2017) 254-260.

2.Tucker et al., Discovery of Next-Generation Antimicrobials through Bacterial Self-Screening of Surface-Displayed Peptide Libraries, Cell (2018)-doi.org/10.1016/j.cell.2017.12.009.