Biology Reference
In-Depth Information
strategies tailored for individual complexes typically make
use of high-affinity antibodies directed against a specific
complex member, or use other affinity matrices such as
DNA, RNA, metabolites, or drugs, inspired by the specific
biochemical properties of the complex. These approaches
have the important advantage of targeting endogenous,
natural forms of the complexes, but they are not readily
amenable to proteome scale.
Proteome-scale co-complex interactome mapping
employs a variety of strategies that attach epitope tags to
bait proteins. These include DNA engineering as well as
post-translational protein engineering
[101
to design the appropriate infrastructure that will allow users
to distinguish intuitively between these two types of edge
[119]
. Curation, storage, representation and analysis of co-
complex interactome data are key challenges in computa-
tional systems biology.
Regardless, the emerging picture of mass spectrometry-
derived co-complex interactome maps is that of a modular
organization. Protein complexes appear to assemble from
a limited number of core modules, with small sub-
complexes as well as individual proteins binding to the core
modules and to each other. Modular organization creates
the possibility of achieving functional diversity through
combinatorial effects while maintaining highly interde-
pendent central parts of molecular machines invariable
[120
106]
. The
purified protein complexes are then systematically treated
with proteases to release peptides, which then are frac-
tionated by one- or two-dimensional liquid chromatog-
raphy. Amino acid sequences are then imputed based on the
mass and charge of the resulting peptides using mass
spectrometry readouts. If applied rigorously and with
attention to statistical significance, proteins containing
these peptide sequences can be derived with a low false
discovery rate. Proteins may also go undetected for
a number of reasons, leading to false negatives in co-
complex interactome maps. Quantitative estimation of
dataset quality using a framework analogous to the one
implemented for binary interactome maps
[36]
has now
been implemented once for a D. melanogaster co-complex
interactome map, which demonstrated high sensitivity
[107]
.
How should we interpret the long lists of proteins that
typically are the readouts of the mass spectrometry anal-
yses of co-purified proteins? All successful bioinformatics
approaches to assign co-complex memberships to co-
purified proteins rely on network analyses, considering
each protein as a node and each co-purification relationship
as an edge. Algorithms can isolate subnetworks that are
highly interconnected or completely interconnected, and
then compute affiliation to one subnetwork compared to
overall frequency to determine the most likely co-complex
associations
e
123]
.
e
DRAWING INFERENCES FROM
INTERACTOME NETWORKS
Combining analyses of network topology with exogenous
data integration can help make sense of complex systems.
This ability is well illustrated in a famous study of high-
school friendships across the USA
[124]
. In the topological
structure of networks where nodes are students and edges
are friendships, communities of tightly linked high-school
friends emerge (see Chapter 9). When nodes in these
topological communities are colored with ethnicity infor-
mation, the extent of ethnic segregation in each high school
is revealed (
Figure 3.4
). Similarly, binary and co-complex
protein
protein interactome network maps provide 'scaf-
fold' information about cellular systems. When inter-
actome maps are analyzed in terms of topology and
integrated with orthogonal functional information, the
resulting knowledge allows investigators to imagine novel
hypotheses and answer basic questions of biology
(
Figure 3.4
)
[125,126]
.
e
110]
. High-quality datasets are
obtained from multiple redundant purifications over
a single search space, which may encompass whole
genomes, as was done for S. cerevisiae, Escherichia coli,
Mycoplasma
[91,108
e
Refining and Extending Interactome Network
Models
A major aim of the analysis of interactome network maps is
to obtain better representations of the underlying inter-
actome itself, since available maps are imperfect and
incomplete. Given topological and exogenous biological
data, which proteins are most likely to interact with a given
protein of interest for which few or no protein interactions
have been described? Which binary interactions and co-
complex associations are the most reliable, and therefore
worthy of mechanistic follow-up?
When there are multiple sources of experimental
evidence supporting a particular protein interaction, the
evidence can be combined to generate a confidence score
for this interaction. This integration can be restricted to
pneumonia,
and
D.
melanogaster
[87,91,104,107,111,112]
,
or
selected
pathways
and
subnetworks as was done for human
[113
118]
. The more
redundant the dataset, the more reliable complex prediction
will be, leading to ever finer granularity of the resolution of
the map.
When two proteins belong to the same protein complex
they may not necessarily be in direct physical contact
(
Figure 3.3
). Hence, edges in a network representation of
co-complex interactomes have a very different meaning
than edges in a network representation of binary inter-
actomes. Most literature-curation databases are struggling
e
