Page 203 - Macromolecular Crystallography

192 MACROMOLECULAR CRYS TALLOGRAPHY

(Jones and Thirup, 1986). Rotamers, the favoured 13.4 Map fitting and refinement
dihedral angles for protein side chains, were first
One of the most difficult steps in X-ray crystallog-
observed from an analysis of well-refined protein
raphy is map interpretation. A major problem is
structures with no dihedral restraints (Ponder and
correctly connecting secondary structure elements.
Richards, 1987).
This topic has been discussed in detail by several
authors (Richardson and Richardson, 1985; McRee,
1993; Kleywegt and Jones, 1997). Most crystallogra-
13.3 Initial fitting to density phers are fortunate enough to learn map fitting from
an experienced mentor. It is critical to have the most
In their classic paper (Brändén and Jones, 1990) the
accurate data and phase information possible. The
late Carl-Ivar Brändén and Alwyn Jones observed
original experimental map should always be kept
that fitting the linear protein sequence into an imper-
for references. A major concern in the later stages of
fect electron density map was a process ‘between
refinement is model bias which is usually addressed
objectivity and subjectivity’. They observed the con-
by calculating omit maps (Bhat and Cohen,
ventional R factor was not as reliable an indicator as
1984) or simulated annealing omit maps (Hodel
previously thought as several published structures
et al., 1992).
were incorrect. Three validation techniques have
One of the first protein refinement programs
practically eliminated grossly incorrect structures,
rotated torsion angles against real-space density gra-
where the sequence is misfolded into the density:
dients (Diamond, 1971). This is the idea used by
the definition of the real-space R factor by Jones’
most automatic and interactive building programs.
group (Jones et al., 1991) with a per residue com-
The current CNS version performs reciprocal-
parison between observed and calculated density;
space torsion angle refinement (Rice and Brünger,
the threading and three-dimensional (3D) sequence
1994). There is debate over the best refinement
profiling methods developed by Eisenberg’s group
program – many favour REFMAC (Murshudov
(Bowie et al., 1991; Lüthy et al., 1992); and the use
et al., 1997). Refinement is discussed in Chapter 11
of statistical cross-validation with the development
of this volume.
of the free R value by Brünger (Brünger, 1992) to
prevent over-fitting of the data.
A nearly complete initial structure roughly fit 13.5 Validation of structures
to the electron density may be available through
homology modelling and molecular replacement, as The process of structure solution is iterative: model
discussed by Delarue in Chapter 7 of this volume. building, refinement, and analysis until one is sat-
A relatively new technique is the automatic inter- isfied with the model. Each lab. or researcher uses
pretation of crystallographic electron density maps their favourite programs and protocols and may
with construction and refinement of preliminary arrive at a model via different paths. Grossly incor-
models. The ARP/wARP software suite is recom- rect structures are detected as discussed in the
mended (www.embl-hamburg.de/ARP) (Perrakis previous section. The validation criteria of atomic
et al., 1999). A main chain model or a more complete models from crystallography is fairly standard-
model with side chains may be output provided ized, as reviewed by Gerard Kleywegt (Kleywegt,
data around 2 Å is available. Attempts to use arti- 2000). The PDB’s automated deposition system
ficial intelligence have periodically been applied uses Janet Thornton’s PROCHECK (Laskowski
to automate map fitting (Feigenbaum et al., 1977; et al., 1993) as the validation module, checking
Glasgow et al., 1993). A promising new method all bonds, angles, dihedrals, and close contacts
usingAItechniqueshasbeendevelopedtoworkopt- and providing a summary report. The SFCHECK
imally with maps around 2.8 Å resolution (Ioerger (Vaguine et al., 1999) program validates the structure
and Sacchettini, 2003). These methods are covered factors.
by Morris, Perrakis, and Lamzin in Chapter 11 of Dihedrals are not generally constrained to
this volume. expected values in refinement programs. Any