Biomedical Engineering Reference
In-Depth Information
models based on free-state chemical shifts were to high-resolution NMR
structures of the individual domains (1.3
˚
); furthermore, the experimental
CSPs were in general smaller than typical prediction errors for the shifts of
individual nuclei.
173
It appears that considering the chemical shifts for a large
number of nuclei simultaneously can overcome problems caused by errors in
prediction at individual sites. Subsequently, HADDOCK was revised to
incorporate a final quantitative shift-scoring procedure derived using SHIFTX
predictions; preliminary results led to the conclusion that
1
H
a
,
13
C
a
and
15
N
CSPs (but not those for
1
H
N
or
13
C
b
) provided useful discriminating power
above that achieved by the standard qualitative HADDOCK approach.
172
The
CS-HADDOCK method also achieved high accuracy models in test cases
when C
a
RMSD values for the unbound and bound components were ,3.5
˚
,
but struggled with larger conformational changes.
172
CamDock and CS-HADDOCK can cope with limited degrees of molecular
rearrangement when proteins interact, but both are likely to fail when the
chemical shifts observed in the complex are not consistent with generating
compact globular folds when the partners are modelled alone, as might occur
with interleaved homo-oligomers. CS-ROSETTA has been adapted to address
this problem by simultaneously modelling the folding and docking of
oligomeric systems.
174
Assuming that the aggregation number is already
defined (e.g., by analytical ultracentrifugation or electrospray mass spectro-
metry under gentle ionisation conditions
175
), symmetry restraints are applied
from the start, first to extended protomer chains in random orientations and
then during refinement with the customary fragment replacement and full-
atom MD/chemical shift-driven protocols. Initial blind structure prediction
tests with CS-ROSETTA proved successful in 75% of cases, with the lowest
energy models from converged clusters possessing C
a
RMSD values ,3
˚
from reference structures, which typically agreed well when cross-validated
with independent RDC data.
174
The folding-and-docking protocol proved
successful with various oligomer topologies, including a-helical bundles,
interlocking b-sandwiches and interleaved a/b motifs containing as many as
192 amino acids; symmetry was thought to play a crucial role in reducing the
number of the degrees of freedom that required sampling.
174
More recently, the automated ROSETTAOligomers method has been
introduced, aiming to predict solution structures for oligomeric systemsby
supplementing chemical shift data with sparse NOEs and domain orientation
restraints from backbone RDCs.
176
It begins by assuming that the protomers
do not intertwine, calculating structures for the monomeric state from
chemical shift and sparse NOE data, and then using RDC restraints to dock
the subunits together; if the resulting structures fail to converge, the procedure
is automatically restarted using the folding-and-docking protocol.
ROSETTAOligomers converged on correct solutions in 80% of test cases,
returning models with C
a
RMSD values ,2.5
˚
from reference structures for
dimers containing up to 304 amino acids.
176
When applied to data for the
tetrameric p53 oligomerisation domain, the approach selected a model withD2
