Biomedical Engineering Reference
In-Depth Information
tion sites of individual helices and strands (which may reflect genuine protein
dynamics).
77,84
The chemical shifts of nuclei in polypeptide regions that sample multiple
conformations with sub-millisecond lifetimes are averaged in a population-
weighted fashion. Berjanskii and Wishart have quantified this observation in
the form of the Random Coil Index (RCI), determined from the reciprocal of
the average secondary shifts at
1
H
a
,
13
C9,
13
C
a
,
13
C
b
and
15
N sites.
85-88
RCI
profiles correlate well with other methods for estimating backbone flexibility,
such as crystallographic B-factors, order parameters derived from nuclear spin
relaxation and molecular dynamics (MD) simulations.
88
They are best suited
for detecting flexible regions in structured proteins;
31,33
for example, RCI
values can be used to distinguish between contributions to B-factors from
internal dynamics and static disorder due to conformational heterogeneity in
crystal samples.
88
3.6 Predicting Dihedral Angles
A careful comparison of chemical shift assignments for the homologous
proteins IRAP and IL-1b demonstrated that similar backbone conformations
give rise to similar secondary chemical shift patterns.
89
Observations of this
kind of 'chemical shift homology'
90
were exploited by the popular TALOS
program, which searches a database for tripeptide fragments with amino acid
sequence and secondary shift patterns that are similar to those of a query
protein, assuming that close matches can be used to estimate values for the
backbone dihedral angles w and y.
29
The PREDITOR approach supplements a
fragment-matching algorithm with information derived from homologous
protein structures.
91,92
Both methods use Ramachandran plots
93
to analyse the
backbone conformations of the ten closest matching fragments, deriving shift-
based predictions of w and y from the mean values of hits within the major
cluster, while ignoring contributions from outliers. These procedures bias the
final predictions towards regions of Ramachandran space that are highly
populated, causing particular problems for glycine sites and residues that
precede a proline. The recently updated TALOS+ package makes acceptable
estimates for a greater proportion of residues,
31
but all three methods return
boundary ranges for w and y that routinely overestimate the accuracy of the
prediction.
The DANGLE algorithm,
81
available as a tool in the CcpNmr Analysis
suite,
94
addressed these issues by using Bayesian inference to calculate the
likelihood of conformations throughout Ramachandran space, paying explicit
attention to the population distributions expected for different residue types,
such as glycine, proline and pre-proline sites.
95
The conditional probability
that a (w,y) conformation within a specified 10u610u bin can produce the
pattern of secondary shifts observed in the query protein is used to assemble a
Global Likelihood Estimate (GLE) diagram for each residue (Figure 3.5).
81
Predicted values for w and y are determined from weighted means over the
