3.6 Dissecting A3D
Performance
and Applicability
When compared with intrinsically disordered proteins or short
peptides, which are assumed to be highly exposed to solvent, the
aggregation propensity of a globular protein is difficult to approxi-
mate, because the specific contribution of each amino does not
depend exclusively of its sequential neighbors. In fact, globular
proteins exhibit diverse structural environments that should be
taken into account for a confident prediction; for example, protein
cores, which are mainly composed of hydrophobic amino acids and
represent regions with the potential to trigger protein aggregation
if exposed. From another point of view, protein interfaces involved
in protein–protein interactions and oligomerization are established
by the same type of contacts than those conforming protein cores.
Indeed, it has been demonstrated that in some disease-related
proteins such as transthyretin, destabilization and consequent dis-
assembly of the tetrameric form, induce their aggregation and
spread of the amyloid disorder [15]. As a consequence, it is not
surprising that these regions represent one of the major origin of
lineal predictors’ failure, because they assume that they are disor-
dered and unprotected, whereas in reality they are protected and
contribute little to the aggregation of properly folded globular
proteins. The structural-based correction of the A3D algorithm
allows to discriminate between buried and exposed APRs, increas-
ing one order of magnitude the ratio between true positive and false
positive predictions [33]. To illustrate this, we present the 3D
structure of the oligomerization domain of thehumanp53 protein,
colored with the A3D scale; in its monomeric state (Fig.4a), where
it exhibits a clear APR; and its tetrameric state (Fig.4b), where the
formation of the oligomeric state masks the threatening region and
turns it undetectable for A3D.
Apart, from discarding the contribution of sequential APRs
when they are buried, A3D correctly predicts structural aggrega-
tion prone regions that are not evident from the linear sequence
(Fig.3a and c)[38–40] because in globular proteins close residues
in the sequence are not necessarily contiguous in space and indeed
they are usually amino acids from distant regions that coalesce to
build up a structural APR.
In solution, globular proteins are not frozen in terms of struc-
ture, but fluctuating between different conformations. It has been
reported that even for highly stable proteins, transient conformers
might expose APRs for a time that is enough to trigger their
aggregation [41–43]. Consequently, for those cases in which the
static run does not unveil a significant structural APR, it is worth
using the dynamic A3D approach, in order to uncover novel poten-
tial aggregation prone regions derived from protein fluctuations
[44]. This is the case of the DNA-binding domain ofhumanp53;
whereas the static version of A3D does detect any significant struc-
tural APR (Fig.4c), the dynamic mode brings to light several of
them (Fig.4d).Predicting the Aggregation of Protein Structures 437