Challenges in Modelling Metalloenzymes
507
resolution structure of cytochrome c 553 to a high-resolution (0.97 Å) one and, on
the other, it also allows to discriminate between various possible protonation states
on ligands bound to the metal [11–13]. Moreover, in favourable cases, the oxidation
state of the metal ion can be determined; however, for higher oxidation states such
analysis can be compounded by partial reduction by photoelectrons released during
data collection [14]. Knowledge of the oxidation state of the metal and protonation
state of its ligands is critical for reliable modelling of metalloenzymes and their catalytic reactions, hence when they are uncertain it may be worthy investing some time
into performing quantum re-refinement, as illustrated by a recent work by Ryde and
co-workers on nitrogenase [13].
4 Homology Modelling
When experimental structures are not available for a particular enzyme of interest,
yet they are known for homologues, then homology modelling may turn out to be
very useful. The field of comparative structure modelling is a mature one with many
stable software packages and servers, e.g. MODELLER, I-TASSER, ROSETTA or
SWISS-MODEL, to name just a few [15–18]. Obviously, the quality of the model
will depend heavily on the availability of structures with high sequence identity to
the target protein. When the sequence identity between the template and the target is
above 60%, one may expect a high-quality model with RMSD to the true structure
on the order of 1 Å [16]. Such models should be accurate enough to meaningfully
use them in studies on ligand binding or reaction mechanisms. However, when the
sequence identity is on the order of 30%, typical RMSD will increase above 3 Å
and the model may have serious errors prohibiting accurate modelling at the atomic
level. Here as an example may serve comparison of structures of thebaine-6-Odemethylase (T6ODM), a mononuclear non-heme iron enzyme using 2-oxoglutarate
as a co-substrate. A homology model of T6ODM was constructed with the use of
the I-TASSER prior to the public release of the actual X-ray crystal structure of
T6ODM (1.85 Å resolution, PDB: 5O9W) [19]. The I-TASSER server identified and
used for modelling several homologous structures, the one with the largest sequence
identity of 31% was for anthocyanidin synthase (ANS; PDB:1GP6). Figure 2 shows
superimposed structures of the best model generated by the I-TASSER server and the
crystal structure. RMSD between the two structures calculated for protein backbone
is 3.2 Å, whereas for all heavy atoms it amounts to 3.9 Å, which falls into the 4.4 ±
2.9 Å range predicted by I-TASSER. As can be noted from the figure, the jelly-roll
core of the protein superimposes very neatly, whereas the least precisely modelled
fragments are those at the entrance to the active site, which are also responsible for
substrate recognition.
507
resolution structure of cytochrome c 553 to a high-resolution (0.97 Å) one and, on
the other, it also allows to discriminate between various possible protonation states
on ligands bound to the metal [11–13]. Moreover, in favourable cases, the oxidation
state of the metal ion can be determined; however, for higher oxidation states such
analysis can be compounded by partial reduction by photoelectrons released during
data collection [14]. Knowledge of the oxidation state of the metal and protonation
state of its ligands is critical for reliable modelling of metalloenzymes and their catalytic reactions, hence when they are uncertain it may be worthy investing some time
into performing quantum re-refinement, as illustrated by a recent work by Ryde and
co-workers on nitrogenase [13].
4 Homology Modelling
When experimental structures are not available for a particular enzyme of interest,
yet they are known for homologues, then homology modelling may turn out to be
very useful. The field of comparative structure modelling is a mature one with many
stable software packages and servers, e.g. MODELLER, I-TASSER, ROSETTA or
SWISS-MODEL, to name just a few [15–18]. Obviously, the quality of the model
will depend heavily on the availability of structures with high sequence identity to
the target protein. When the sequence identity between the template and the target is
above 60%, one may expect a high-quality model with RMSD to the true structure
on the order of 1 Å [16]. Such models should be accurate enough to meaningfully
use them in studies on ligand binding or reaction mechanisms. However, when the
sequence identity is on the order of 30%, typical RMSD will increase above 3 Å
and the model may have serious errors prohibiting accurate modelling at the atomic
level. Here as an example may serve comparison of structures of thebaine-6-Odemethylase (T6ODM), a mononuclear non-heme iron enzyme using 2-oxoglutarate
as a co-substrate. A homology model of T6ODM was constructed with the use of
the I-TASSER prior to the public release of the actual X-ray crystal structure of
T6ODM (1.85 Å resolution, PDB: 5O9W) [19]. The I-TASSER server identified and
used for modelling several homologous structures, the one with the largest sequence
identity of 31% was for anthocyanidin synthase (ANS; PDB:1GP6). Figure 2 shows
superimposed structures of the best model generated by the I-TASSER server and the
crystal structure. RMSD between the two structures calculated for protein backbone
is 3.2 Å, whereas for all heavy atoms it amounts to 3.9 Å, which falls into the 4.4 ±
2.9 Å range predicted by I-TASSER. As can be noted from the figure, the jelly-roll
core of the protein superimposes very neatly, whereas the least precisely modelled
fragments are those at the entrance to the active site, which are also responsible for
substrate recognition.