4. Discussion
The ability of a protein to make and break interactions with another
protein in order to bind to it temporarily and perform a certain
function allows for transient association between the proteins. The
binding event is mostly brought about by non-covalent bonds at the
interface between the interacting proteins which also determine the
nature and strength of interaction. The implication of such an event can
range from no structural change (due to weak interaction) to long range
(allosteric communication) of the perturbation signal. The focus of our
work was to study how direct associations play a role in perturbing the
connectivity of residues in globular proteins. The effect of the
perturbation at, around and away from the interface is studied by
constructing a protein structural network (PSN) of the connectivity
within the protein and analysing the alteration of the network.
A collection of structures from the ProPairs database was used for the
analysis. After filtering structures based on their crystallographic
properties, 895 protein chains have been identified that transiently
associate with other proteins and have the same oligomeric state in the
bound complex as well as the unbound form. The structure of the protein
chain obtained as a part of a protein complex, which is the bound form
of the protein, is compared with the structure of the same protein chain
when it is not bound to the interacting partner but yet in the same
oligomeric state. At the local level, the change in network parameters
such as number of edges, hubs and centrality measures are studied. The
global comparison is made in terms of the topological change, as in,
their backbone deviation (RMSD) and in terms of their structural network
dissimilarity (NDS).
Graph spectral comparison methods are used in computing the
dissimilarity between the networks which involves spectral decomposition
to obtain the eigen vectors and eigen values of the PSNs. Few case
studies with high network variation were identified using the global and
basic network comparison. A major contribution to the NDS in these cases
arises from its component, EWCS which is mostly responsible to compute
the change in local clustering of residues. The Fiedler vectors (Fv)
between a pair of PSNs can be examined to identify the sites with high
variations in the clustering of nodes. The eigen vector corresponding to
the second smallest eigen value is called the Fiedler vector. This
vector can provide meaningful information on the algebraic connectivity
of the network and can be used in partitioning the network into
clusters. This is illustrated with a case study.
The first step is to identify the Fv from the spectra of the pair of
graphs. Figure 3A shows the aligned Fv between the bound and the unbound
form of the DLD protein. The absolute difference between the aligned
vectors is computed to find regions of the protein that are not in
agreement. Figure 3B shows the difference between the Fv and highlights
the region with variation between the vectors. The sites with the most
variation, having the highest absolute difference are shown as sticks in
Figure 3C. The cartoon diagram of the chain A of DLD is coloured based
on the absolute difference between the vectors and chain B is shown as
grey surface. The interacting partner E3BP protein is shown using yellow
surface representation. Side chains of the top five residues with
highest absolute difference, yellow in the bound form and red in the
unbound form, are shown using spheres.
Any alteration of network parameters close to the site of interface is
expected as the interfacial sites make new interactions with the binding
partner. However, more often, alterations are also observed far from the
site of binding due to allostery which is the transmission of the
perturbation. The path of this allosteric signal can be analysed by
drawing the shortest path between the site of perturbation to the site
of significant network alteration. The change in shortest path between
the site of binding to the site of perturbation is analysed in the case
study and discussed in the Supplementary Figure 4. A new edge between
spatially proximal nodes GLU 437 and ASP 350 in the bound form reduces
the shortest path when compared to the several possible short paths in
the unbound form.
Most of the variability observed in the dataset occurs at the
non-interfacial sites. This is also evident from the variation of degree
and strength at, around and away from the interface observed in
Supplementary Figure 2. The network variation obtained only by
considering the structure network of non-interface sites says that in
almost 89% of the cases the network dissimilarity is greater than 50%
of the NDS scored from all residues. This result also suggest that most
of the network away from the binding site is affected by the
perturbation, but all sites are not perturbed proportionally. As viewed
from the absolute difference between Fv of the DLD protein case study
only five nodes of the entire length of the long protein were strongly
affected to cluster differently. The effect on all other sites are
feeble and is an effect of subtle changes in the local conformation of
sidechains. Which shows that most residues predominantly still remain in
the same topology, any interactions that are broken are counteracted by
other interactions being made. Hence there is predominantly a
rearrangement of interactions that is being observed. However, in about
60% of the 285 cases that are identified as enzymes, in the working
dataset, a net loss in connectivity is observed. Hence when the
interacting protein is an enzyme, the structure of the bound form may be
less compact than the unbound form which may serve the process of
functioning to catalyse several different reactions.
The case studies have been chosen such that their analysis has certain
clinical relevance and an impact on understanding their mechanism. The
alteration of the network in all the case studies can be related to
several human disease conditions like antibacterial resistance,
diabetes, toxin induced cell death and Latic acidosis. We also related
to known information about the mechanisms of their function. Hence, the
analysis of the structure network is a necessary and beneficial tool in
the analysis of structural excursions. The development of such tools
that can analyse the impact of protein-protein interactions will help in
understanding allostery mechanism and the network analysis of protein
structures for stability engineering and docking studies.