Team:UCL/Model/Structural
Structural Modelling
Problem
We were faced with two major decisions: which encapsulin to use and how long should the linker between the encapsulin and the DARPin be. We started our journey with the following three Protein Data Bank files (PDBs):
We did not have to worry about the DARPin-to-HER2 interaction or about the initial encapsulin assembly, as these have been previously investigated. However, to fuse the DARPins onto the encapsulin particles, we had to work out the length of linker between the encapsulin monomer and the DARPin to avoid blocking the DARPin binding site and allow the monomer assembly. Additionally, differences in monomer size and surface structure between different encapsulins define how DARPins can arrange on the surface without clashes. In case the DARPins were too crowded on top of both types of encapsulin nanoparticles, we would use monomers both with and without DARPins (see BBa_K3111503).
Solution - MEND
Myxococcus xanthus encapsulin (Fig. 2) we characterised in this project (BBa_K3111001) is 'bumpy'. We assumed that any additions to the 'trough' monomers would result in severe steric clashes, thus impeding the nanoparticle assembly. Therefore, our initial hypothesis was that the rounder Thermotoga maritima encapsulin would be more suitable for this project (Fig. 1).
The first stage was to build the protein model. As none of our team members or supervisors had much experience in structural modelling, we discussed the task with Prof. Paul Dalby and Dr Cheng Zhang, a post-doctoral researcher in his lab. They advised to create a model of two encapsulin planes (a di-pentamer) and add a DARPin to each monomer to see whether the DARPins physically fit on the encapsulin without steric clashes within the same plane and between adjacent planes. Additionally, Dr Zhang suggested using Foldit Standalone (Rosetta GUI) to achieve a more native geometry of manually added features, such as a GSS-linker and a Strep tag.
Easier said then done. We began by figuring out how to bind a single DARPin to one monomer of the encapsulin di-pentamer in PyMol; we called our mini-project "MEND" - Model of Encapsulin aNd DARPin. Firstly, we manually added a linker sequence to both the monomer and the DARPin. Then, we aligned the linkers, removed one of them, and, after a short struggle, we had two PDBs combined into a single object and bonded (Fig. 4).
Unfortunately, even after exporting the model as a single object, only PyMol recognised the bond made between the PDBs, and even PyMol rendered a gap in cartoon representation (Fig. 5) while Rosetta refused to recognise the bond completely (Fig. 6).
Following the lack of result from continuous endeavours to fix the issue ourselves, or by consulting with the computational biologists at London Synthetic Biology Showcase, or by talking to fellow iGEMers at the Newcastle meetup, we received an email from Dr Zhang recommending Miss Valentina Spiteri, a doctoral student of Prof. Stephen Perkins.
Miss Spiteri has been extremely helpful and so far the most knowledgeable regarding model building. We have thoroughly investigated the original and the resulting PDBs, uncovered and fixed some faulty amino acids, weird numbering and chain identification. Additionally, we ran an energy minimisation protocol with Nanoscale Molecular Dynamics (NAMD) in case the linker geometry was implausible. Nevertheless, PyMol kept rendering a gap at the boundary between the original PDB files. However, as a result we had a much neater model (Fig. 7).
Miss Spiteri then suggested using molecular dynamics simulation software, such as GROMACS, with a single encapsulin monomer-DARPin pair to reach a more native linker conformation and a more thorough energy minimisation. The result would then be replicated over the other nine monomers, and the interactions within the di-pentamer could be coarse-grained simulated.
While we have been learning GROMACS and getting access to the UCL supercomputers, we consulted again about the overall modelling strategy with Prof. Andrew Martin. He questioned the validity and real value in such large scale modelling and pointed out that in vivo the flexible linker is likely to maintain the extended geometry we obtained via building it in PyMol. Hence, if we could physically fit the DARPins on the surface of the encapsulin in PyMol, that would be sufficient evidence that the linker is long enough. Nonetheless, since we have a PDB of questionable quality, a proper energy minimisation of a single encapsulin monomer-DARPin pair could be useful before testing the di-pentamer or the full-capsule models.
We prepared our first PDB (of T. maritima encapsulin) for GROMACS, which required some cleanup even before we could run the energy minimisation (Fig. 8). Luckily, both the GROMACS tutorials and Dr Zhang provided brilliant guidance. When we finally ran our energy minimisation, we found a considerable reduction in potential energy (Fig. 9), but no visible change in terms of structure (Fig. 10). The likelihood is that GROMACS fixed the sidechain rotation geometries, thus we obtained a potentially higher quality PDB file, yet it was of no use for our rather visual modelling.
Additionally, Prof. Martin pointed us to experts in the tasks we were considering. Most importanltly, efficiently building a model of a full encapsulin as well as the membrane interactions between our delivery vehicle and HER2. He also advised on cytotoxic cargo options and respective people to consult with.
We began questioning our initial claim about the clashes between DARPins on an M. xanthus encapsulin. So, we decided to check. Given the time it took to prepare the model for energy minimisation versus the outcome value, we skipped this step for the model of M. xanthus encapsulin. With the help of Mr. Sam Ireland and his software "atomium" we obtained the full encapsulin models (Figs. 11 and 12). Figure 12 indicates the artificial clashes we obtained by replicating the encapsulin-DARPin hexamer we created with DARPins at arbitrary positions. However, the space available between the DARPins is large, thus we expect no clashes to occur in vivo.
These figures have shown that the DARPins on the surface of the T. maritima encapsulin resolve themselves much more readily than those on top of the M. xanthus. Although we thought there is sufficient space for the latter to assemble into the full capsule, we wanted to minimise the risks to our time-constrained project. Additionally, we figured that the density of DARPins on the surface of either encapsulin might impede the rotation of DARPins for the efficient binding to HER2 receptors. Therefore, we decided to create and test two T. maritima encapsulin-DARPin types: one made only from encapsulin-DARPin dimers (BBa_K3111502), and the other made of a mixture of encapsulin monomers and the mentioned dimers (BBa_K3111503).
Finally, we realised that the linker should also allow sufficient flexibility for the DARPin to interact with the HER2 receptor on the membrane, i.e. the DARPin-HER2 binding site should be able to face outwards parallel to encapsulin surface. We modelled this by manually bending the linker. Without breaking the bond geometries, the shortest linker to allow this rotation without clashes was 8 amino acids long (Fig. 13).
In conclusion, structural modelling helped us answer our queries: we would use the T. maritima encapsulin, and the DARPin will be tethered by an 8aa long GSS linker.