Glucocorticoid Structure Modelling
The in silico experiments were focussed on modelling the protein of interest, glucocorticoid receptor (GR) to demonstrate ligand-binding theoretically, since it is already known that the GR binds cortisol to its ligand-binding domain (LBD). This experiment was also to show that cortisol binding into the binding pocket is unchanged with the sequence we used, which was modified with a his-tag and underwent codon-optimisation for expression in E. coli.
Methods
Using the same DNA sequence that is going to be used in in vitro experiments, the expressed protein and ligand docking was modelled, starting with the translation from DNA to amino acid sequence using the expasy.org Translate tool by Swiss Institute for Bioinformatics (SIB) (full sequences in results). Next, the amino acid sequence was modelled as a 3-dimensional structure both by swissmodel.expasy.org by SIB, which uses templates for its prediction, and by the trRosetta Server by Yanglab. Furthermore, partial sequences Met560-Leu753 (= Met57-Leu250 respectively) were modelled as well using trRosetta.
Lastly, using the structure predictions as PDB files, docking experiments were conducted with cortisol as a ligand. The cortisol structure was an SDF file (retrieved from drugbank.com DB000741). The ligand docking was run by the ROSIE server by GrayLab using standard settings. BCL for conformer calculations was ticked, as well as using the input position. The results were evaluated in BIOVIA Discovery Studio Visualizer 2021 to highlight certain amino acids and bonds.
Results
DNA-Sequence: Following the optimisation for expression this sequence was translated to an amino acid sequence using expasy.org operated by the SIB.
According to Frank, Ortlund, and Liu et al. 2021 amino acids N564, Q570, R611, Q642, and T739 (here N61, Q67, R108, Q139, and T236, blue markings) are directly interacting with the ligand cortisol. They describe the ligand binding domain’s structure was investigated using dexamethasone, a synthetic hormone with similar structure to cortisol. The amino acids Frank, Ortlund, and Liu et al. 2021 mention as interactive with cortisol are all in the list of the ones interacting with dexamethasone as well, shown by Bledsoe et al. 2002 and highlighted here in grey (M560, L563, N564, L566, G567, Q570, W600, M601, M604, A605, L608, R611, F623, Q642, M646, L732, Y735, C736, T739, I747, L749 (here F), and L753).
Using this sequence a structure model was done using a template-based prediction by swissmodel.expasy.org (SIB) (Fig. 1) and a second model by the tRosetta server by Yanglab (Fig. 2). Due to the addition of the his-tag the model prediction by tRosetta (GrayLab) varies in its tertiary structure compared to the swissmodel, who simply removed the n-terminal tail to fit its templates. Otherwise, they subjectively portray high similarity to each other and show high prediction values, for Fig. 1 – Global Model Quality Estimate: 0.87 and for Fig. 2 – confidence estimation by TM-score: 0.906, which indicate high accuracy of the protein structure model.
Following the structure prediction of the his-tagged glucocorticoid receptor that is expressed in E. coli a docking experiment with the ligand of interest, cortisol, was conducted. The results of this docking with ROSIE server by GrayLab using standard settings indicate an interaction of cortisol with the N-terminus near the his-tag rather than the binding pocket it would usually bind to according to Frank, Ortlund, and Liu et al. 2021 and Bledsoe et al. 2002 (Fig. 4, blue). This binding at the terminus is unexpected, thus another docking without the his-tag was conducted. Here the same server settings were used, however a tRosetta structure prediction with only Met560 - Leu753 (estimated TM-score = 0.926 indicating very high accuracy), which are the first and last amino acid of the dexamethasone-interacting ones, were used for this docking to remove the his-tag (Fig. 5). Here, cortisol binds in closer proximity to the known binding site, though not to the exact same predicted amino acids.
For ligand docking models the documentation describes the interface delta score as the most reliable unit, where the smallest number indicates the best result. For cortisol docking to the his-tagged GR LBD (Fig. 4) this score is -5.325, while the score for docking only the binding site is -4.623, making it a less likely interaction partner than the N-terminus of the his-tagged protein domain.
Discussion
According to ROSIE server documentation, the modelling algorithm assumes a stiff protein backbone that does not move. Since we are expecting a conformational change of the GR upon ligand binding, this server may not have been the optimal methodological choice, and it may explain the results. If cortisol is more likely to bind to the N-terminus than into the pocket of the GR LBD, it may impair the efficiency of the molecular design for the cortisol biosensor.
However, the his-tag is planned to be used for attachment to the electrode before the receptor would enter a milieu with cortisol. Thus, since the his-tag is already bound, and cortisol does bind onto the GR LBD pocket, this might not impair the actual mechanism. This must be shown in vitro for verification and requires more precise and complex in silico experimental setups.