Page 119 - Macromolecular Crystallography

108 MACROMOLECULAR CRYS TALLOGRAPHY

model(s), EPMR (Kissinger et al., 1999, 2001) to explored, with very promising results. It is well
do molecular replacement and CNS (Brünger et al., known that many proteins can exist in several struc-
1998) to refine the model(s). In addition, the authors tural states, such as open or closed, depending on
use their own local version of ARP/wARP, called the presence of one of their substrates. The RMSD
Shake-and-wARP, to attempt automatic building of between the two forms is such that if one tries to
the model in the electron density map. Clearly, solve the MR with the wrong form, MR will fail,
many combinations of such protocols are possible, while it might succeed if one uses the other form. Of
by using different software at each step of the whole course, one might try to use both, but it may well be
process. that all accessible structural states of a given protein
are not deposited in the PDB.
Hinge-bending as well as shear movements have
7.9.3 G. Labesse and L. Martin (GTBio, Lyon,
June 2004) been documented over the years, with RMSD
between the two forms of the same proteins
In this case, the @TOME meta server (Douguet spanning a very large range. While hinge-bending
and Labesse, 2001) is used to detect low struc- movements can, in principle, be tackled by dividing
tural homology three-dimensional models, as a the protein into individual domains and then solv-
consensus between six different ‘threading servers’. ing the MR problem for each of the domains,
Then, after model building by Modeller (Sali and other types of movements are very difficult to
Blundell, 1993), MolRep (Vagin and Teplyakov, 1997; deal with, especially as they are very collective
2000) is run, followed by refinement using Refmac and involve the coordinated movement of many
(Murshudov et al., 1997). In some instances, this atoms. Recently, however, it has become appar-
suite of programs was able to find the MR solution ent that most structural transitions observed in
using a model with sequence identity as low as 20% the PDB can be modelled quite accurately by the
(G. Labesse, personal communication). low-frequency Normal Modes derived from a sim-
plified representation of the protein, namely the
Elastic Network Model (Krebs et al., 2002). The
7.9.4 MrBump (Keegan and Wynn, 2007)
Elastic Network Model (Fig. 7.4) is a simplified
This is the most complete program, downloadable representation of a protein where each residue is
from CCP4 web site www.ccp4.ac.uk/MrBUMP. represented by a single point that is linked to its
Sequence multialignment is performed by MAFFT spatial neighbours (within a given radius, usually
or ClustalW. The model is edited using four differ- taken as 8–12 Ångstroms) by a spring of constant
ent protocols, including CHAINSAW and the one strength (Tirion, 1996); it works surprisingly well
available in MOLREP. Then either Phaser or Mol- despite its simplicity and is by construction most
Rep is used for MR and finally Refmac is used apt to model collective movements (Delarue and
for refinement. The approach is intended more to Sanejouand, 2000). In fact, it has been shown that
be exhaustive than fast, but this is not a problem most movements can be modelled with, on aver-
with current computers. The philosophy is sum- age, only two normal modes (Krebs et al., 2002).
marized by the author’s following statement: ‘It is As a result of advances in the computation of
concluded that exploring a range of search models such modes, there is virtually no limit to the size
automatically can be valuable in many cases’. of the macromolecular assembly for which low-
frequency normal modes can be calculated (Tama
and Sanejouand, 2001).
7.10 Making the most of your model
using normal modes: pushing towards Recently, two independent applications of these
the systematic exploration of structural ideas to MR have been described (Fig. 7.5):
diversity
• One is intended to produce structural diversity
A new way to take into account the structural in the model by grid-sampling the amplitudes of
variability of protein structures is being actively (at most) two of the lowest-frequency normal modes;