Protein Modeling

Docking in FITTED

Human AKR1D1 and AKR1C4 proteins are known to metabolize substrates resembling cholestenone and coprostanone¹⸴² . First, we must verify that it's even possible for both of our enzymes to bind to our substrates of interest prior to our wet lab testing them. To that end, we used a protein-ligand docking software, FITTED to test this theory³.

The crystal structures of the two proteins we modelled, AKR1D1 and AKR1C4, are shown above on the left and right respectively.

In order to use FITTED's protein-ligand docking features, we needed to first specify an active site for the software to probe for the most probable orientation/location of our ligand. The PDB file we used for AKR1D1 (PDB: 3BUR) was crystalized with testosterone, which we used as the center of the active site.

AKR1D1 crystal structure with testosterone, and testosterone docked with FITTED⁴

However, AKR1C4 (PDB: 2FVL) was not crystalized with a metabolite. In order to identify it's active site, we aligned the 3D structure of AKR1C4 with that of AKR1D1 and found that they have 74% structural similarity. With this information, we concluded that AKR1D1 and AKR1C4 should have similar active site locations.

AKR1C4 and AKR1D1 crystal structure aligned with testosterone docked⁵

In order to determine whether AKR1D1 and AKR1C4 bind with our substrates of interest, we compared the docking scores of each enzyme docked with its desired substrate with the docking score for a known metabolite of each enzyme.

Docking scores in FITTED are comparative within the same run. A lower docking score within the same run indicates a better enzyme-substrate fit.

AKR1D1

AKR1D1 is known to dock with 7α-hydroxy-cholestenone.

AKR1D1 docked with 7α-hydroxy-cholestenone​
FITTED Score: -23.9790

AKR1D1 docked with cholestenone ​
FITTED Score: -25.3300​

AKR1C4

AKR1C4 is known to dock with 7α-hydroxy-cholestenone.

AKR1C4 docked with 7α-hydroxy-cholestenone​
FITTED Score: -21.9373

AKR1C4 docked with cholestenone ​
FITTED Score: -25.6598

AKR1D1’s FITTED score docked with cholestenone was lower than its FITTED score docked with a known metabolite. The docking scores are based on the binding energies and intermolecular interactions between the substrate and the protein detected by FITTED. Due to the stochastic nature of FITTED, docking scores are only comparable between substrates docking in the same run with the same protein. In general, the lower the FITTED docking score, the better its predicted fit is. Therefore, we concluded that AKR1D1 was likely to metabolize cholestenone.

Similarly, AKR1C4’s FITTED score docked with coprostonanone was lower than its score docked with a known metabolite, so we concluded that AKR1C4 was also likely to metabolize its desired subsrate, coprostanone.

Mutating Proteins

Now that we have high confidence that our protein will be able to metabolize cholestenone and coprostanone, we can move on to identify residues to permute in order to optimize binding to our substrates of interest.

Proximity approach

Assuming that residues closest to the active site would have the most profound effect on substrate binding, we used the docked structures we created previously in addition to other known metabolites of AKR1D1 and AKR1C4 to help identify possible residue candidates.

List of all substrates are shown in the table below.

We then loaded the docked structures into pyMOL ⁶ to identify residues within 4 angstroms of the target and known substrates in both AKR1D1 and AKR1C4.

Surface view of AKR1D1 docked with cholestenone.

Surface view of AKR1C4 docked with coprostanone.

Homologs in other species

NCBI Protein BLAST⁷ on both Uniprot⁸ and nr databases was used to identify homologous proteins to AKR1D1 and AKR1C4 in species other than humans (top 20 results) to identify plastic residues as potential candidates for mutation. All other parameters for the blast search were left as default.

Mutant candidates master list

Once acquiring the lists of mutations we aggregated all candidates found by the 2 NCBI⁷ homologous protein and substrate proximity searches into a spreadsheet each for AKR1D1 and AKR1C4. We selected top candidates that were found in both the NCBI and the proximity search.

Finally, we verified that the mutations selected were from function proteins using Uniprot⁸ . Then we generated the mutated sequences with the residue permutation that was found in the homologs.