Challenges in Modelling Metalloenzymes
509
entrance to the active site, nor made direct contacts with the substrate—ectoine.
Later on, several other crystal structures were published for EctD (PDB: 4NMI, [28]
4MHR, 4MHU and 4Q5O [23]), yet in neither of them the loop region is resolved.
However, crystal structure obtained for EctD - 5-hydroxyectoine complex confirmed
the previous modelling study successfully identified the substrate-binding site of
EctD.
6 Prediction of pK a
One of the important steps in preparation of the model of the protein is determination
of its protonation state at the selected pH. For modelling of enzymes, the latter will
usually be optimal pH for the catalytic reaction. The titration curve of an amino acid
that is part of a protein can significantly shift with respect to that known from water
solution. This happens due to: (i) desolvation of the residues when transferred from
the aqueous solution to less polar interior of the protein, which generally destabilizes the charged state and (ii) electrostatic interactions inside the protein which may
either stabilize or destabilize the charged state of a particular residue [29]. It should
be underlined that with the change of pH protonation state of the protein residues may
change and this in turn affects electrostatic environment of the residue under consideration. As a result, one has to know pH-dependent titration curves of all residues
within the protein to determine their probability of protonation. The ionization state
of the protein has a crucial role not only for its stability or strength of protein–protein interactions but also for ligand binding and catalysis, which is arguably most
important from the enzymology perspective.
Prediction of pK a values for protein residues and hence their protonation state at
a given pH is a necessary step before any further modelling can be conducted, be it
MM, MD or QM. Indeed, even in the case when the initial geometry is derived from
the crystal structure, the ambiguity of the electron density, especially in the vicinity
of the heavier metal sites, makes initial geometry optimization, e.g. by means of
MD or QM/MM/MD methods, a necessary step. Hence, the protonation state of all
ionisable residues should be determined as accurately as possible.
Although the protonation state of individual residues can be assessed experimentally in solution with the help of NMR, [30] IR or in crystal with neutron diffraction,
[31] for computational methods, it is still a challenging task. This is due to the fact
that even small differences in free energy of protonation give rise to significant pK a
shifts (1.364 kcal/mol of energy at 298.15 K corresponds to a change of pK a by a
unit) [32].
The majority of computational methods developed for this purpose in the last
20 years concentrate on predicting pK a shifts of ionisable groups from the known
reference state (such as pK a of isolated amino acid in water or Ala-Ala-X-AlaAla pentapeptides) [33]. Presumably, the most accurate and reliable methods to
approach this problem are those based on QM or QM/MM computational models.
However, they are still too demanding computationally for routine treatment of whole
509
entrance to the active site, nor made direct contacts with the substrate—ectoine.
Later on, several other crystal structures were published for EctD (PDB: 4NMI, [28]
4MHR, 4MHU and 4Q5O [23]), yet in neither of them the loop region is resolved.
However, crystal structure obtained for EctD - 5-hydroxyectoine complex confirmed
the previous modelling study successfully identified the substrate-binding site of
EctD.
6 Prediction of pK a
One of the important steps in preparation of the model of the protein is determination
of its protonation state at the selected pH. For modelling of enzymes, the latter will
usually be optimal pH for the catalytic reaction. The titration curve of an amino acid
that is part of a protein can significantly shift with respect to that known from water
solution. This happens due to: (i) desolvation of the residues when transferred from
the aqueous solution to less polar interior of the protein, which generally destabilizes the charged state and (ii) electrostatic interactions inside the protein which may
either stabilize or destabilize the charged state of a particular residue [29]. It should
be underlined that with the change of pH protonation state of the protein residues may
change and this in turn affects electrostatic environment of the residue under consideration. As a result, one has to know pH-dependent titration curves of all residues
within the protein to determine their probability of protonation. The ionization state
of the protein has a crucial role not only for its stability or strength of protein–protein interactions but also for ligand binding and catalysis, which is arguably most
important from the enzymology perspective.
Prediction of pK a values for protein residues and hence their protonation state at
a given pH is a necessary step before any further modelling can be conducted, be it
MM, MD or QM. Indeed, even in the case when the initial geometry is derived from
the crystal structure, the ambiguity of the electron density, especially in the vicinity
of the heavier metal sites, makes initial geometry optimization, e.g. by means of
MD or QM/MM/MD methods, a necessary step. Hence, the protonation state of all
ionisable residues should be determined as accurately as possible.
Although the protonation state of individual residues can be assessed experimentally in solution with the help of NMR, [30] IR or in crystal with neutron diffraction,
[31] for computational methods, it is still a challenging task. This is due to the fact
that even small differences in free energy of protonation give rise to significant pK a
shifts (1.364 kcal/mol of energy at 298.15 K corresponds to a change of pK a by a
unit) [32].
The majority of computational methods developed for this purpose in the last
20 years concentrate on predicting pK a shifts of ionisable groups from the known
reference state (such as pK a of isolated amino acid in water or Ala-Ala-X-AlaAla pentapeptides) [33]. Presumably, the most accurate and reliable methods to
approach this problem are those based on QM or QM/MM computational models.
However, they are still too demanding computationally for routine treatment of whole