Proteomic Analysis of Cellular Systems - Systems Biology Concepts and Insights - page 11

of stimulus [92 94] . Given increasing coverage of the

proteome, even the isoform specific regulation of the pro-
teome can be investigated [95] .

surprisingly low, pointing to technical limitations of the
individual approaches and emphasizing that, despite being
large-scale, all datasets were non-saturating, sampling
different parts of a vast interactome [107] . In addition, it
also emphasizes the fundamentally different, but highly
complementary nature of AP-MS and Y2H [108] : Y2H
data consist of binary combinations of proteins with mutual
affinity, including weak interactions, which can be recorded
as long as they lead to activation in the genetic readout. In
contrast, AP-MS data provide lists of proteins that co-occur
in protein complexes

e

INTERACTION PROTEOMICS
Specific interactions of proteins with other proteins, with
nucleic acids, lipids, carbohydrates, and metabolites or
other small molecules, orchestrate all aspects of life at the
molecular level. The dissection of molecular assemblies
has been a longstanding goal of modern biology, which
requires identification of the constituent partners as the first
step. This is a field at which MS-based proteomics has
excelled from its early days. The ultimate goal is the
delineation of the 'interactome', which is defined as the
sum of all molecular interactions of a biological system.
The size of the interactome of a given organism is a matter
of debate and of how the definition is interpreted, but it is
undoubtedly far more complex than the genome or pro-
teome; current interactome datasets likely merely scratch
its surface [96] .
Mass spectrometry has the unique ability to identify
very small amounts of any protein without prior knowl-
edge, and in principle it can therefore directly unravel the
protein composition of any molecular assembly. Alterna-
tive methods of unbiased interaction detection, such as
phage display [97] or the yeast two hybrid assay (Y2H)
[98] , use genetic readouts that test for direct binding but do
not involve the formation of actual multi-protein
complexes. All approaches in MS-based interaction pro-
teomics are based on the assumption that a molecular
interaction is the result of an affinity that can be exploited
to purify or enrich the assembly from a crude mixture.
Typically, one molecule serves as the 'bait' which is
coupled to an affinity matrix. This can be done via an
antibody or a genetically encoded tag in the case of proteins
or via chemical synthesis in the case of peptides, nucleic
acids or small molecules. Mass spectrometry is then used to
identify the 'prey' proteins that interact with the bait. This
workflow is known as affinity purification followed by
mass spectrometry (AP-MS) ( Figure 1.4 A).
The first application of this methodology was the
identification of the members of protein complexes [99] ,
classically defined as entities that can be purified bio-
chemically. This was fuelled by the development of the
tandem affinity purification (TAP) tag, which resulted in
clean preparations of protein complexes from endogenous
sources by two consecutive purification steps [99,100] .
This technology was mostly used for the generation of the
first large-scale AP-MS interaction datasets of model
organisms such as the budding yeast [101

but
provide no direct topological information. Comparison of
data from both sources therefore requires the conversion of
co-complex members into binary contacts, which can be
done using different models [107] .
Weak or transient interactors tend to be under-repre-
sented in AP-MS datasets because they easily get lost
during washing steps in the sample preparation workflow.
These washing steps are necessary to reduce the number of
proteins that bind non-specifically to the affinity matrix.
Unspecific interactors have been the bane of interaction
proteomics and were originally dealt with by extensive
blacklisting of proteins that were identified across many
different affinity purifications. However, this is a less-than-
ideal solution as it inevitably leads to lower true positive
rates while also failing to remove many false positives.
Virtually all of these drawbacks have been overcome by
the advent of quantitative proteomics: specific interactors
can easily be distinguished from unspecific background
binders by directly comparing their quantities in affinity
purifications vs. controls [109,110] . This paved the way for
second-generation quantitative interaction proteomic
studies ( Figure 1.4 A). Isotopic labeling techniques allowed
the detection of interactions in the presence of high
amounts of background binders, and of molecular assem-
blies which could not be purified extensively. Importantly,
this principle is applicable to any conceivable bait molecule
that can be immobilized on the affinity matrix. For instance,
early highlights include the identification of proteins that
interact with specific post-translational modifications rep-
resented as modified, synthetic peptides [111,112] . Such
assays can be streamlined and used to probe the biological
relevance of large-scale PTM datasets. For instance,
synthetic peptides corresponding to phosphotyrosine sites
with potential key functions as molecular switches have
been synthesized and their cellular interaction partners
have been determined [113,114] .
In a similar fashion, oligonucleotides can be immobi-
lized to identify proteins binding to specific DNA [115] or
RNA sequences [116] . In this way, quantitative interaction
proteomics identified the transcriptional repressor respon-
sible for the difference between a fat and a lean pig geno-
type, which is caused by a single nucleotide mutation
[117,118] .

including indirect binders

e e

104] .
These datasets allowed the first comparisons of AP-MS
data with each other and with previously available large-
scale Y2H datasets [105,106] . The overlap turned out to be

e