Overview
The main goal of modelling the Cholesterlock fusion protein is to simulate a linker for the protein. To achieve this, our team collaborated with external researchers to identify the necessary conditions that the linker must satisfy. Firstly, it needs to function properly at the pH level of the digestive tract. Secondly, it must remain linear and fit into NPC1L1 without interfering with the function of the cholesterol-modifying domain or the Fc fragment. Lastly, we need to determine whether both the cholesterol and the linker can fit inside NPC1L1, which is the protein responsible for cholesterol uptake. To find linkers that meet these conditions, we used our linker web app to generate a list of linkers to test. Afterward, we built and parameterized these models and ran them through the molecular dynamics software via the university's supercomputer. After running these tests, we obtained movies that demonstrated the proteins' stability and behaviour over time. We used these movies to compare the linkers, and ultimately pick one for the final Cholesterlock model.
Challenges with Glycosylation and Bonding
Model refinement has brought up two major challenges for the team: missing bond parameters and incomplete glycosylation sites. The bond between Cholesterol and Glycine - the connecting amino acid on the section of the hedgehog protein we’re using - is a covalent bond occurring between the OHL oxygen atom on Cholesterol and the backbone carbon on Glycine.
The bond being covalent presents a challenge: MD parameters don’t recognize covalent bonds in a system, meaning the bond is missing its length and angles. To resolve this, the team created a patch file to predefine the bond parameters so the system will recognize it. To do this, we searched in the topology file of the system to find other occurrences of C and OHL binding together, and copied these bond parameters. When you copy in the instances of the C and OHL bond, you can test whether it is a reliable match by using the Tanimoto similarity chart. Whatever instance you decide to copy, continue to use this for the rest of the bond definition.
Tanimoto similarity chart: Tanimoto similarity charts can aid in identifying existing compounds with similar bonds, making it easier to select suitable candidates for reuse or modification.
Once the bond was resolved, the glycosylation for our PDB file was incomplete. To fix this, we referenced the Oligosaccharides on our IgG fragment from the protein databank. Using COOT, we manually built this glycan structure from the protein databank onto the glycosylation site of the protein using the alignment tools provided in COOT.
First Run through ARC: (C-MYC)
Our first run of CholesterLock was using a linker with a c-Myc tag. As we obtained results in the form of a movie of our protein with the current linker sequence over a period of time, we discovered that although the linker is flexible enough to allow cholesterol to move as required, it is hydrophobic enough to attach to cholesterol. As a result, this binding of cholesterol and the linker inhibits the binding between NPC and cholesterol as shown by the legs of the linker sequence folding within.
Our solution was to change a couple amino acids within the chain, however before doing that, we are also testing the same for using HA and His 6 tag, which are showing good preliminary results and they would be tested further for our potential linker sequence.
Second Run through ARC: (HIS6)
The run with this linker had a stable appearance. We already have a tag on the front of the protein - double confirmation as HIS6 is used as a detection mechanism. As the protein appeared stable in the initial dynamics run, we plan to procure further testing on it.
To ensure that the function of hedgehog is not affected by IgG and the linker, we plan to perform another important simulation which involves testing the models with the entire hedgehog, to verify that the C terminal of hedgehog folds onto the N terminal domain and not the linker because if these interactions occur, the system would stretch out and not undergo autoproteolysis.
CholesterLock with Linkers from LINKED
To test what kinds of linkers would work best in CholesterLock we selected 15 linker sequences of varying hydrophobicity and flexibility. The 15 linkers were incorporated with IgG and hedgehog and folded using Colabfold. Thirty simulations were conducted with both the full length, and processed versions of each CholesterLock-linker combination.
To describe our final model development method for our autoprocessed protein we first ran our IgG, Linker and Hedgehog amino acid sequence through ColabFold to provide the basis of protein folding for our model. This model would then be put into COOT where model standardisation would be completed and OXT terminal end added to the hedgehog segment which is significant for the next step. The model is then exported and brought into Chimera where cholesterol is added and bound to the OXT terminal using the built-in bond builder tools. These steps primarily complete the initial protein visualisation and model creation process. Next we move on to prepping the model for molecular dynamics where we clean up the protein pdb file by removing unnecessary elements and standaring the names of molecules for our forcefield. To perform molecular dynamics there are additional files needed such as forcefield files for the specification of molecule bonds as well as autopsf files which are protein structure files.
To create our protein structure files we used VMD and performed ionisation and solvation for the models to neutralise the charges and to create a waterbox to incase the protein for the simulation. Additionally, VMD was used to add our patch file for the specification of the bond between cholesterol and hedgehog. To perform NAMD we created a standardised configuration file to be used for each protein only requiring slight modifications per run. Once modified our model and files are finally ready to be run through NAMD. By running our model through NAMD we are provided with the necessary files to run further analysis into the interactions of residues with our protein. Our tested linkers from LINKED were chosen to provide a spread of linkers with varying hydrophobicity, and that were either rigid, flexible, or context dependent. The subsequent folding of CholesterLock provided us with a general idea of which properties will work best and to provide insight for future investigation.
