Once our team decided to fuse viral antigens (VP3) to B. subtilis spore coat proteins, we had numerous concerns about the viability of this approach. What will the VP3 protein look like when it is fused to CotB? Will it still retain the same general shape and folding as VP3 proteins in the actual Deformed Wing Virus (DWV) virion? Could there be any interactions between vitellogenin and our fusion protein? Will our protein remain stable when bound to the cell wall?
A huge limiting factor for our dry lab team was how little was understood about honeybee immunology. While we know that vitellogenin will bind the B. subtilis cell wall and membrane, not much is understood about what may happen to our viral antigen, in particular, upon being ingested by the queen; thus, we were limited in our efforts to undertake other tasks such as engineering our VP3 protein sequence to maximize its immunogenicity, etc.
First, we needed to obtain structures for all the proteins involved in this project. While we were able to obtain the crystal structure for VP3 generating... , there were no experimentally-determined structures available for our CotB or vitellogenin. Hence, we resorted to predicting the 3D structure of our fusion protein using AlphaFold generating... .
We were also unable to find an isolated amino acid sequence for our VP3 protein. Hence, we used a DWV genome map generating... along with the sequence for the DWV polyprotein (RefSeq accession: GCF_000852585.1) to determine the VP3 protein sequence.
AlphaFold model of VP3-RLinker-CotB construct. CotB is shown in blue, VP3 in orange, and the rigid linker is shown in green.
The prediction of our protein was really good except for the C-terminal region of the CotB side, which received a pLDDT score < 50. We believe that this is likely an intrinsically-disordered region (i.e., a region lacking secondary structure) or a transmembrane domain. In either case, the exact structure of this domain would need to be verified using experimental techniques such as crystallography or cryo-EM, which is out of the scope of our project.
In order to determine the likely orientation of our fusion protein on the surface of B. subtilis, we made use of the OPM PPM3 server generating... provided by the University of Michigan. The orientations of all three versions of our fusion protein (with the rigid, photocleavable, and thrombin-cleavable linker) were determined. They are shown in the images below, where VP3 is shown in orange and CotB is shown in blue.
VP3-RLinker-CotB
VP3-PhoCl-CotB
VP3-ThrCl-CotB
We are least certain about the attachment point of the fusion protein with the ThrCl linker as it seems to interact with the membrane at a residue in the linker and not in the potential IRD/transmembrane region we had identified earlier. However, the orientations of the fusion proteins with the first two linkers seem more plausible.
Additionally, we also ran molecular docking simulations to determine if there will be any binding between vitellogenin and our fusion protein. While not much is known about Vg interactions with peptide antigens (we only know that it binds the peptidoglycan of our B. subtilis), we believed that running this simulation may reveal potentially useful information on the binding between Vg and our viral antigen.
To run our simulations, we used lightdock, a macromolecular docking framework written in python generating... , generating... . Many of the other docking tools our team had surveyed (HDOCK, AutoDock Vina, etc) did not offer an intuitive way to simulate the binding between a membrane-bound protein and a ligand, but lightdock allowed our team to accomplish this with relative ease. Here are the steps we took to conduct this simulation, following lightdock’s tutorial on membrane-associated protein docking.
For the structure of our membrane-bound receptor, lightdock recommends using a coarse-grained (MARTINI) representation already determined through MD simulations, and many proteins have already been modeled in this manner and their CG models deposited on databases like MemProtMD. However, we were unable to find our novel fusion protein (VP4-RLinker-CotB) on such databases (for obvious reasons), nor were we able to find the model of our CotB protein (for expressing it on the surface of a vegetative cell is a poorly-studied technique as well). Additionally, we lacked the computing resources to simulate this ourselves; hence, we created an approximated version using lightdock’s membrane builder tool. We estimated the anchor point of our proteins using the OPM PPM3 results from our previous work and increased the radius offset to 60.00 Å since these settings generated an orientation that was more similar to our OPM results.
The structure of vitellogenin, our anchor protein, was already available in the AlphaFold database (Entry ID: Q868N5). Both this and the approximated membrane were passed to lightdock’s lightdock3_setup.py script to prepare for the simulation.
Lightdock uses the DFIRE scoring function. In order to calculate the docking result for each swarm, we wrote a script that iterates through each swarm, runs the lightdock tool on that swarm, and then cleans up any unnecessary files. In the script, we used lightdock’s built-in ant_thony.py script to parallelize these operations.
After the simulation script had finished, we generated the structures for the predictions, cluster them by swarm, and filtered for structures with the highest docking scores. This was also accomplished using a script that used ant_thony.py under the hood. The best swarm had a docking score of 34.408 and is shown below.
Docking result with the highest score, VP3 section in orange and Vg colored N->C