phosphoproteomics in drug discovery
TRANSCRIPT
Reviews�GENETO
SCREEN
Drug Discovery Today � Volume 00, Number 00 �November 2013 REVIEWS
Phosphoproteomics in drug discovery
Melody K. Morris1, An Chi1, Ioannis N. Melas2,3 and Leonidas G. Alexopoulos2,3
1Merck & Co., Boston, MA, USA2 ProtATonce Ltd, Athens, Greece3Department of Mechanical Engineering, National Technical University of Athens, Athens, Greece
Several important aspects of the drug discovery process, including target identification, mechanism of
action determination and biomarker identification as well as drug repositioning, require complete
understanding of the effects of drugs on protein phosphorylation in relevant biological systems. Novel
high-throughput phosphoproteomic technologies can be employed to measure these phosphorylation
events. In this review, we describe the advantages and limitations of state-of-the-art phosphoproteomic
approaches such as mass spectrometry and antibody-based technologies in terms of sample and data
throughput as well as data quality. We then discuss how datasets from each technology can be analyzed
and how the results can be and have been applied to advance different aspects of the drug discovery
process.
IntroductionThe pharmaceutical industry is tasked with delivering drugs of
high efficacy and low toxicity. The drug discovery pipeline
needs to become fast and effective in response to the indus-
try-wide challenge regarding a low first-in-human to registration
rate and pressure from strict regulatory requirements, budget
cuts in the healthcare system, vigilant patient foundations and
time constrains caused by patent expiry. Acquiring a better
understanding of how drugs target cells in the human body
is of the utmost importance for increasing drug development
efficiency and decreasing high attrition rates because success
rates from first-in-human studies to registration are only
11% [1].
Although not yet proven, high-throughput phosphoproteo-
mic technologies together with well-established pharmacoge-
nomic and pharmacogenetic measurements hold promise for
improving the drug discovery process because phosphorylation
events are proximal to many disease-causing signaling mechan-
isms. For example, dysregulated phosphosignaling caused by
mutations is a known driving mechanism of several types of
cancer, and inhibitors of kinases such as Raf (e.g. sorafenib),
Corresponding author. Alexopoulos, L.G. ([email protected])
1359-6446/06/$ - see front matter � 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.drudis.
anaplastic lymphoma kinase (ALK; e.g. crizotinib) and epider-
mal growth factor receptor (EGFR; e.g. erlotinib) have proven
efficacy in the treatment of a variety of cancers. Additionally,
kinase inhibition has proved to be an effective method for
inhibiting activation of immune cells important in autoimmune
disease [e.g. Janus kinase (JAK) inhibition for the treatment of
rheumatoid arthritis]. By understanding the phosphosignaling
networks underlying aberrant growth in cancer or immune cell
activation in autoimmunity, drug developers can choose better
targets and better understand how therapeutics will alter cellular
processes.
Several phosphoproteomic technologies have emerged in
recent years that have become invaluable tools for drug dis-
covery and biomarker development. In this review, we discuss
how modern phosphoproteomic technologies can be used to aid
target identification, understand drug mechanism of action
(MOA), construct signaling pathways, predict toxicity and/or
efficacy, as well as their involvement in drug repositioning
when coupled with appropriate computational analyses. For
each technology we describe its distinct advantages and inher-
ent limitations. We then provide examples of how these tech-
nologies have been applied to enhance the drug discovery
process.
2013.10.010 www.drugdiscoverytoday.com 1
REVIEWS Drug Discovery Today � Volume 00, Number 00 �November 2013
DRUDIS-1274; No of Pages 8
Review
s�G
ENETO
SCREEN
Phosphoproteomic technologiesProtein measurements can be divided into two distinct categories:
those that make no a priori assumption about proteins to be mea-
sured [i.e. 2D-PAGE and mass spectrometry (MS) technology]; and
those that are based on a pre-determined set of measured proteins
(i.e. antibody- or aptamer-based approaches). In the first category,
MS approaches promise an unbiased (i.e. hypothesis free) screening
of thousands of phosphoprotein targets. In the second category,
antibody-based technologies exist in several different formats and
varieties that promise a throughput of thousands of samples per day
and improved quantification on pre-determined phosphosignals.
Despite the tremendous improvements in both technologies, inher-
ent limitations exist: MS-based technologies offer great coverage of
the phosphoproteome but require significant sample preparation
and data post-processing time, leading to limited sample through-
put; by contrast, antibody-based formats are restricted by limited
antibody quality and target availability. All commonly used tech-
nologies are presented in Fig. 1 and described below.
Forward- and reverse-phase protein microarrays forphosphoprotein measurementsA protein microarray consists of a solid surface, typically a glass
slide or membrane, on top of which antibodies, aptamers, purified
proteins or cell lysates are spotted and then probed with molecules
RPPA
1000
100
10
1
1 10 100 1000
MS
Antibody microarrays
WB
EL
ISA
Num
ber
of s
amp
les
Number of signals (phosphoproteins)
xMAP Tm
HCS∗∗
Flow-Cyt.∗∗
CyTOF TM∗∗
∗∗ Single cell technologies
Data throughput =samples x signals
Drug Discovery Today
FIGURE 1
State-of-the-art phosphoproteomic platforms for high-throughput
measurements of phosphosignaling. We illustrate the difference in sample
and signal throughput of each technology by plotting number of samples
readily assayed by number of phosphorylation sites (phosphosignals) readilymeasured. Reverse-phase protein arrays (RPPA) utilize a single-antibody
format to probe thousands of lysate spots. Protein arrays follow a regular
sandwich ELISA format where lysates are incubated with dozens ofantibodies. High-content screening (HCS) is a single-cell technology that
usually employs an automated fluorescent microscope capable of visualizing
96- or 384-well plates. Fluorescence-based flow cytometry (Flow-Cyt.) can
detect fewer than 12 phosphoproteins at a single-cell level. Mass cytometry(CyTOF) offers greater multiplexability for single-cell analysis via transition
metal ion tags. Mass spectrometry (MS) utilizes affinity phosphorylaton
enrichment approaches with high-performance MS instrumentations
capable of identifying thousands phosphorylation sites. xMAPW is asuspension ELISA microarray on microbeads that utilizes a dual-antibody
format where thousands of samples can be probed in a single day with
dozens of antibodies. Western blot (WB) and ELISAs are the standard low-throughput assays that have been mapped for comparison reasons.
2 www.drugdiscoverytoday.com
interacting with the spots [2–5]. Depending on the immobilized
and probing molecules, two types of phosphoprotein microarray
are the most prominent: (i) forward-phase protein microarrays
(FPPAs – also known as antibody microarrays) with immobilized
phospho-antibodies and (ii) reverse-phase protein microarrays
(RPPAs) with immobilized samples.
Antibody microarrays are ELISA-type assays where the ‘sand-
wich’ antibody format enables high fidelity measurements of
several intracellular signals on a limited number of samples.
Several vendors offer ready-to-use arrays. For example, R&D Sys-
tems offers an array for detecting 43 phosphoproteins, PathScan1
from Cell Signaling can measure 18 phosphosignals from a single
lysate and the RayBioTech Phosphorylation Array (RayBiotech
Inc., Norcross GA, USA) can detect the activation of 71 tyrosine
kinases. In all cases, a handful of samples can be measured by a
simple incubation of samples with the arrays. During incubation,
phosphoproteins in the lysate are pulled down by the capture
antibodies on the spots, and a secondary antibody that carries a
detection motif (fluorophore, Horseradish peroxidase, etc.) binds
to a different epitope of the phosphoprotein and produces a
quantifiable signal (i.e. fluorescent intensity). In most cases, an
image analysis algorithm is required to measure the intensity of
the spot and thus quantify the phosphoprotein. The spotting
procedure together with the imaging step can introduce some
artifacts on the signal quantification (i.e. spot evaporation, donut
shape spot, uneven spotting, ‘growing’ spot size, etc.). Major
bottlenecks for establishing high-throughput assays are sample
preparation and handling and image analysis procedures that
usually require a manual curation step. Recent developments have
established protein arrays in a 96-well plate format (i.e. Meso Scale
Discovery, Rockville, MD USA) but at the expense of multiplex-
ability (usually less than �6 signals are measured per sample).
In general, phosphoprotein antibody microarrays are a well-
established technique with several products available to buy off
the shelf. Owing to the ‘sandwich’ antibody format, antibody
microarrays are ideal for high-quality measurements of several
phosphoproteins. However, their total data throughput (sam-
ples � signals per time for assay run) is facing strong competition
from bead-suspension systems that are based on the same protein
detection principle, are compatible with 96-well plate automation
and offer similar multiplexability.
RPPAs bypass the need for antibody pairs by direct printing
(spotting) of the lysate onto a functionalized glass or membrane.
The single-antibody detection format enables a large variety of
phopshoproteins to be measured, and multiple slides can be
spotted with the same lysate and probed with different antibodies
to enable analysis of multiple phosphoproteins in the same sam-
ples. Several hundred samples can be spotted on a single slide but a
very limited number of proteins can be measured on each slide.
Compared with ready-to-use protein arrays, RPPAs require signifi-
cant technical expertise for protein spotting and protein detection
that can be established in-house or out-sourced. In addition, the
single-antibody detection format is prone to low data quality as a
result of higher background noise and antibody specificity. By
contrast, the sample throughput of RPPAs is unmatched compared
to any other known technology. When thousands of samples need
to be measured, RPPAs offer by far the lowest cost per data point,
but at the expense of data quality and multiplexability.
Drug Discovery Today � Volume 00, Number 00 �November 2013 REVIEWS
DRUDIS-1274; No of Pages 8
Reviews�GENETO
SCREEN
Suspension microarraysSuspension microarrays are ELISA assays performed on the surface
of suspended microparticles (usually microbeads) that are mixed
with the samples. The most commonly used system is the xMAP1
technology by Luminex Corporation where fluorescently coded
magnetic beads can detect up to 500 signals simultaneously – a
number usually referred to as multiplexability. In reality, the
multiplexability for phosphoproteins is constrained by the avail-
ability of antibody pairs and thus not more than a few dozen
phosphoproteins have been measured simultaneously. The prin-
ciple of the assays is based on phosphospecific antibodies that are
coupled on fluorescently barcoded magnetic microbeads and
mixed with the samples in 96- or 384-well plates. Phosphoproteins
of interest bind to the capture antibodies on beads and are detected
with fluorescently labeled secondary antibodies in an instrument
that resembles a flow cytometer where two to three colors are used
to identify the bead fluorophores (captured antibody) and another
color identifies the abundance of phosphoprotein on the bead
surface (secondary antibody). xMAP1 technology offers high data
quality similar to a sandwich ELISA but in a much higher data
throughput as a result of: (i) assay multiplexability and (ii) sample
automation via magnetic manipulation of the suspended bead.
Nowadays, many companies offer xMAP1 phosphoprotein and
proteomic measurements on different formats that range from
ready-to-use kits to full service xMAP1 data collection and
analysis.
The main bottlenecks for data throughput in multiplexed sys-
tems are the cross-reactivity of antibodies and the natural abun-
dance of simultaneously measured phopshoproteins in the lysates.
The presence of many detection antibodies can lead to higher
background noise and low signal-to-noise ratios. Just one ‘dirty’
antibody in the mixture of detection antibodies is sufficient to
bind nonspecifically to noncorresponding beads and severely
affect the data quality. There are two main reasons for cross-
reactivity: (i) the structural homology among proteins and (ii)
the inherent specificity of antibodies. Single-banded Western blots
are used to infer antibody cross-reactivity in multiplexed and RPPA
assays. One method for addressing this cross-reactivity is to probe
the same sample repeatedly with different sets of antibody-coated
beads [6]. Using stringent antibody selection criteria, multiplex-
ability can be as high as 40 phosphoproteins, although typical
experiments use ten to 17 phosphosignals with a good signal-to-
noise ratio [7,8].
Mass spectrometryMS-based phosphoproteomics is another important tool for under-
standing the structure and dynamics of signaling networks as a
result of its inherent multiplex capability, absolute specificity and
wide dynamic range [9–13]. During the past decade, a series of key
technical advances were made for chemical and immuno-affinity
phosphorylaton enrichment approaches and chemical or meta-
bolic labeling quantification strategies. Phosphopeptide enrich-
ment coupled with MS-based stable isotope labeling with amino
acids in cell culture (SILAC), iTRAQ1 or label-free quantification
are now commonly used to quantify differences among control vs
treated or diseased samples in an unbiased manner [14,15].
Although chemical affinity enrichment is applicable for all phos-
phorylated peptides, immunoaffinity enrichment in conjunction
with MS can be used to enrich for phosphopeptides of specific
motifs [16] or specific phosphorylation sites [17–20].
Although MS phosphoproteomics can yield comprehensive
information about phosphorylation in a few samples, sample
preparation is often complex and requires a relatively large
amount of sample depending on complexity of the biological
matrix and abundance of the analyte. This issue leads to challenges
in its use, particularly when analyzing clinical specimens. To
address this limitation, targeted phosphoproteomic analysis using
multiple reaction monitoring (MRM) coupled with automated
sample preparation has shown promise for improving the sensi-
tivity and throughput significantly. The future development of
such high-throughput, sensitive, selective and absolute quantita-
tive MS-based assays could enable this technique to become an
alternative approach in clinical applications when antibody
reagents are not available or not easily generated.
Single cell technologiesTwo primary technologies enable measurement of protein phos-
phorylation in single cells: single cell imaging and flow cytometry.
In fixed-cell imaging and flow cytometry cells are fixed and anti-
bodies conjugated to fluorophores that are used to detect phos-
phoproteins [21]. Tagging different antibodies with multiple
fluorophores enables measurement of up to ten to 15 phosphor-
ylation events in each cell, although typical instruments can
measure up to three colors. Spectral overlap and background
fluorescence limit the number of phopshosignals that can be
measured. Recently, the use of heavy metal isotope tagged anti-
bodies followed by MS quantification has overcome the multiplex
limitation for flow cytometry, enabling measurement of up to 100
phosphosites and/or surface markers with a technique known as
mass cytometry [22,23].
Applications in drug discoveryThe availability of high-throughput phosphoproteomic technol-
ogies that enable the acquisition of a few thousand data points per
run has expanded the impact of this data type on the drug
discovery process. Despite the fact that there is no clear method
of choice for all applications, there is a somewhat clear decision-
making rule regarding what is the most appropriate phosphopro-
teomic platform. MS-based approaches are unbiased and can
detect thousands of phosphorylation sites, making them inher-
ently more amenable to discovering novel targets or biomarkers.
However, MS is limited by sample throughput as a result of the
relatively large amount of samples and the time for sample pre-
paration and post data analysis. Thus, when large numbers of
samples must be analyzed, xMAP1 technology and RPPA offer
greater sample throughput and data quantification. We consider a
few applications of phosphosignaling to aid several aspects of the
discovery process (Table 1). These applications are examples of
support for the entire drug discovery and development pipeline
(Fig. 2).
Target identificationTarget identification is the process of determining the cellular
component (usually a protein) to target with a small molecule
or biologic in order to modulate disease activity. Most pure target
identification phosphoproteomic studies involve identifying
www.drugdiscoverytoday.com 3
REVIEWS Drug Discovery Today � Volume 00, Number 00 �November 2013
DRUDIS-1274; No of Pages 8
TABLE 1
High-throughput technologies for phosphoproteomics
Platf orms Working principle Numberof
phosphosignalsper
samplealiquot
Operation al throughpu ta
Sample r equir ements RefsType Amoun t Preparation
timea
Eli sa Plana r arr ay Antibody (Ab)
1 High Lys ate Low Low
Flowcytometry
Cytometry Ab ~3–10 Med ium Sing le cell
Med ium Low [26]
Mas scytometry
Cytometry with ion-tagged antibodies
Ab, mass , charge
~1–20 Med ium Sing le cell
Med ium Low [43]
Fixed cell imag ing
Imag ing Ab ~3–10 High Sing le cell
Low Low
Lumine x Suspen sion bead arr ay
Ab ~5–40 High Lys ate Low Low [6,7,30,36,
44]
Antibod ymicroarr ays
Plana r arr ay Ab ~1–70 Low Lys ate Low Low [21]
RPP A Plana r arr ay Ab 1b High Lys ate Low High [22,37]
Mas s spe ctrometry
Mas s spe ctrometry
Mass, cha rge
aTypical operational throughput and preparation time estimates are given for laboratories with averagelevels of experience for each technology. Exact values depend greatly on the scientist who carries out theassay.bLimitation of reverse-phase protein microarray (RPPA) measurement of one protein per spot can beovercome by spotting the same lysate on multiple arrays and probing each array with a different antibody.
1000+ Med ium Lys ate High High [18,19,31–
33,38,41,42]
Above average performan ce Average performan ce Below average performan ce
Discovery and preclinical development Clinicaldevelopment
Clinical trials• Drug repositioning
• Pharmacodynamic (PD) biomarkers
• Compound stratification• Understanding toxicity
• Understanding mechanism of action• Signaling pathway construction
Target ID andinitial validation
• Efficacy biomarkers• Patient stratification biomarkers
Biomarker development
Further target validation
Chemistry developmentExplorationof chemical
space
Compoundchoice
Optimization
Drug Discovery Today
FIGURE 2
Drug discovery and development process. Applications of phosphoproteomic technologies to support drug development are shown. Aspects of preclinical
development are shown in the yellow box, whereas the orange box indicates clinical development. Steps in the drug discovery and development process arelisted in black text with blue text indicating applications aided by phosphoproteomic technologies.
4 www.drugdiscoverytoday.com
Review
s�G
ENETO
SCREEN
Drug Discovery Today � Volume 00, Number 00 �November 2013 REVIEWS
DRUDIS-1274; No of Pages 8
TABLE 2
Specific examples in drug discovery and potential phosphoproteomic solutions–turnaround time for all applications is several monthsto one year
Drug discovery goal Potential phosphoproteomic solution
Target identification Identify phosphotargets associated with disease by comparing normal and diseased samples
Understanding mechanism of action Identify the phosphoproteins that a compound affects in the signaling network of several celltypes and/or patient samples
Signaling network construction Compare signaling networks in normal and diseased states by using a combination of database
knowledge and algorithms developed to construct and compare pathways
Compound stratification Compare the phosphoprotein signaling of drugs from libraries and connect them to their chemical
structures
Drug repositioning Screen drug libraries and identify drugs that affect phosphoprotein signaling in a similar way to
that of approved drugs. Also, use increased understanding of mechanism of action and toxicity to
determine if the drug could be repositioned to a different disease
Toxicity Identify if a new compound is toxic by comparing phosphorylation signatures with compounds withknown toxicity profile
Biomarker discovery Identify phosphoprotein biomarkers that enable more-accurate monitoring of pharmacodynamics
Patient stratification biomarkers Profile phosphoprotein states in patients to determine markers that are predictive of patients that will
respond to various treatments
Reviews�GENETO
SCREEN
phosphotargets associated with disease by comparing normal and
diseased samples (Table 2). MS is the method of choice because it
offers an unbiased (hypothesis free) approach to obtain the opti-
mal target. By examining differences in the phosphorylation state
of normal and diseased cells or tissue, one can identify proteins or
pathways that are dysregulated in disease. In the study by Huang et
al. [24], U87MG glioblastoma cell lines with varying levels of
EGFRvIII mutation were analyzed by quantitative MS to identify
important tyrosine phosphorylation sites [24]. In related work,
Chumbalkar et al. identified phosphoproteomic changes asso-
ciated with a different EGFR mutation frequently observed in
glioblastoma [25]. Owing to the cost and sample requirements
of MS, Du et al. suggested limiting the sites measured to phosphor-
ylation of tyrosine kinases and utilizing a multiplexed bead-based
approach [26].
Protein microarrays have also been used for target identifica-
tion. They can be used to screen a large number of patients or cell
lines under various experimental conditions and uncover key
differences in the signaling mechanisms of normal vs diseased
tissue. Tworkoski et al. reported that commercial antibody micro-
arrays were used to screen 25 melanoma cell strains and identify
potential therapeutic targets [27]. In another study, a custom RPPA
was used to screen 92 key signaling proteins in 118 poor prognosis
B-Cell PrecursorAcute Lymphoblastic Leukemia (BCP-ALL)
patients and identify the ones that were aberrantly activated [28].
Understanding drug MOAA key application of high-throughput phosphoproteomic tech-
nologies is the understanding of how a drug affects signaling
activity by identifying the phosphoproteins that a compound
affects in the signaling network of several cell types and/or patient
samples (Table 2). The very large number of phosphoproteins
detected using MS technology make it ideal for screening the
effects of a drug on thousands of proteins in a few samples [29–
31]. Several groups have used quantitative MS to interrogate the
effect of ABL kinase inhibitors on the phosphoproteome of several
relevant cell types [29–31]. Similarly, the phosphoproteomic
effects of heat shock protein 90 (Hsp90) [32] and mammalian
target of rapamycin (mTOR) [33] inhibitors have been mapped.
If the response to a few well-known phosphorylation sites is of
interest, xMAP1 or protein arrays can be used to identify altera-
tions in cell signaling mechanisms induced by the drug. For
example, xMAP1 technology has been used to map how different
cancer drugs affect the signaling network of hepatocellular carci-
noma cell lines treated with many different growth factors and
cytokines [7,8,34,35].
Construction of signaling networksUnderstanding cellular signaling networks provides crucial insight
of signaling network dysregulation in disease and can aid target
identification as well as understanding of the MOA. Traditionally,
signaling networks are obtained from literature mining and path-
way databases. However, recent advances in phosphoproteomic
technologies have shown significant discrepancies between litera-
ture-derived signaling pathways and experimentally verified sig-
naling activity [36,37]. On this front, computational tools have
immerged to build signaling pathways by modeling phosphopro-
teomic data using reverse engineering or model training
approaches. These techniques typically require phosphoprotein
data gathered after in vitro treatment of cells with multiple activat-
ing stimuli (e.g. cytokines and growth factors) in the presence of
specific small molecule kinase inhibitors. A network is then fitted
to the phosphoprotein data either by inferring links between
protein signals using statistical relationships or by choosing an
optimal network from a suite of possible networks deduced from
canonical signaling networks or protein–protein interaction data
[36,38–41]. The signaling pathways derived from these approaches
have solid phosphoproteomic validation and can be specific to a
disease, donor and/or cell type.
Because this application requires measurement of a defined set
of phosphoproteins in many samples, xMAP1 or RPPA technol-
ogies have been used to measure these dynamic activation states
in cells treated with a library of cytokines or compounds [8].
Signaling networks were then inferred by optimization with
www.drugdiscoverytoday.com 5
REVIEWS Drug Discovery Today � Volume 00, Number 00 �November 2013
DRUDIS-1274; No of Pages 8
Review
s�G
ENETO
SCREEN
reverse-engineering algorithms [40,42] or a logic-based frame-
work [7,35,36,38,39,43]. In the latter case, hypothesized signal-
ing reactions were removed if they were not functional, based on
the data at hand, resulting in a network predictive for the signal-
ing mechanisms of the interrogated cell type.
Drug repositioning and compound stratificationDrug repositioning refers to the process of identifying new or
additional disease indications for drugs that have been proven
safe but not necessarily efficacious in the clinic. Thus, drug repo-
sitioning can also be applied to revive drugs that failed because of
efficacy issues. In all cases, phosphoproteomics can be used to
match drug on- and off-target effects to the causative MOA of the
disease. For example, Weber et al. used MS to propose that erlotinib
and gefitinib activity on the Src family of kinases and/or Bruton’s
tyrosine kinase (Btk) was responsible for the unexpected activity of
these drugs in acute myeloid leukemia [44]. If much higher sample
throughput is needed (i.e. repositioning from drug libraries), a
similar approach can be used with an xMAP1 or a RPPA platform.
Phosphoproteomic technologies can also be used for compound
differentiation and stratification. In these applications, the rela-
tionship between the molecular structure of the drug and its effects
on the phosphoproteome is determined. MS has been used to
determine differences in the effects of full and biased G-protein-
coupled receptor (GPCR) agonists [45], and mass cytometry was
used to compare the effects of several JAK inhibitors on 14 phos-
phorylation sites in 14 peripheral blood mononuclear cell (PBMC)
types after stimulation with 12 cytokines [46].
ToxicityDrug toxicity and especially hepatotoxicity is a major problem in
drug discovery and a common reason for drug withdrawal from
the market. Phosphoproteomic technologies can complement
standard preclinical and clinical toxicity assessments biomarkers
[cytochrome 450 (CYP450), lactate dehydrogenase (LDH) release,
transaminase enzymes, etc.] to define signaling motifs that are
predictive of toxicity. On this front, xMAP1 phosphoproteomic
technologies have been employed to generate phosphorylation
response signatures that predict idiosyncratic toxicity that has
poor prediction capability using standard preclinical tests [47].
Similarly, phosphoproteomic technologies have recently been
used to identify toxicity biomarkers [differentially expressed pro-
teins related to hepatotoxicity, oxidative stress pathways, endo-
plasmic reticulum (ER) stress pathways, inflammatory response
pathways, hepatocarcinogenicity and hepatic fibrosis] in an
attempt to screen potentially toxic candidate compounds out of
the drug discovery process before carrying out extensive in vivo (or
clinical) experiments [48,49].
Pharmacodynamics and target engagement biomarkeridentificationPharmacodynamics (PD) biomarkers are increasingly important in
decision-making and lead compound selection during preclinical
and clinical studies because they link drug–target inhibition in
cells with a biological outcome. In practice, proximal PD biomar-
kers (i.e. target engagement biomarkers) aim to measure the direct
interaction of a drug with its biological target and distal PD
biomarkers aim to capture the drug-dependent regulation of the
6 www.drugdiscoverytoday.com
disease pathway [50]. For kinase drug discovery, protein substrate
phosphorylation sites often represent candidate PD biomarkers. In
some cases, autophosphorylation sites on the target kinases them-
selves are ideal proximal PD/(target engagement) candidates
instead of downstream substrates where the signal can result from
multiple converging pathways and might not accurately reflect the
degree of target kinase inhibition [51].
Although the discovery effort could be conducted using multi-
plex antibody arrays or MS-based approaches, PD biomarker assays
today are often focused on the drug effect on a single phosphor-
ylation endpoint rather than a network of phosphorylation sites.
There is a growing effort going on to enable the measurement of
the signaling-network-level response to different inhibitor treat-
ments. This will be of particular importance in the near future
where combination therapies including more than one or two
targeted therapeutics are usually used with the aim of modulating
an entire set of dysfunctional signaling pathways in patients.
Patient stratification and personalized medicineThe use of phoshoprotein measurement as a predictive tool for
drug responsiveness of specific patient subpopulations has been a
growing area in cancer research and is complimentary to genetic
and genomic efforts. The main goal is to understand how a drug
affects patient-specific cells and whether a phosphobiomarker can
be identified for prediction of patient-specific drug efficacy. Several
successful examples demonstrate the importance of phosphopro-
tein-based personalized medicine. RPPA has been used to demon-
strate differences in phosphoinositide 3 kinase (PI3K) pathway
activity in different subtypes of breast cancer suggesting the
potential antitumor efficacy from the pathway-specific therapeu-
tics [52]. By contrast, Andersen et al. described drug–kinase target
interactions in detail using MS-based phosphoprofiling followed
by validation studies of a phospho-epitope that can predict drug
sensitivity in a large cohort of human cell lines and in tumors [53].
As shown by studies like these, characterization of human cancer
at its pathway level could serve as the basis for biomarker discovery
and clinical assay development to drive treatment decisions for
individual patients using appropriate pathway-specific drugs.
Concluding remarksPhosphoproteomics is increasingly becoming the method of
choice in several areas of the drug discovery process. Data collec-
tion can be obtained using a diverse set of technologies that can
loosely be categorized as MS-based or antibody-based. Taken
together, While both technologies provide a similar number of
data points, a clear distinction can be made based on the number
of phospho-proteins or samples analyzed by the technologies. MS-
based approaches are unbiased (hypothesis free) and can measure a
few thousand phosphoproteins per sample. Thus, they are inher-
ently more amenable to discovering novel targets or biomarkers.
This is a clear and significant advantage compared with antibody-
based technologies that are typically limited to simultaneous
measurement of fewer than 100 phosphoprotein signals in a single
sample. Nevertheless, there are two main challenges a MS-based
approach faces today. The first consideration is the limited
throughput and relatively large sample input. The second con-
sideration is the significant cost in time and money associated
with a follow up on a highly promising but unknown target or
Drug Discovery Today � Volume 00, Number 00 �November 2013 REVIEWS
DRUDIS-1274; No of Pages 8
Reviews�GENETO
SCREEN
biomarker: traditional follow up requires specific antibodies and
molecular biology tools, which might not be available. Although
MS-based follow up usually utilizing targeted MRM-based
approaches is possible, examples of such applications are rare.
Phopshoproteomics today is inherently difficult to implement
for analyzing clinical samples such as serum or plasma despite the
high-quality level of information they provide. The main disad-
vantage is the complicated experimental procedure that involves
extra steps for cell isolation and cell lysis followed by ELISA or MS
procedures. Such procedures increase the possibility of experimen-
tal errors and might hinder their use as a main stream tool for
clinical assays, diagnostics and/or as decision-making tools in
clinical trials. With respect to a diagnostic tool for patient strati-
fication, phosphoproteomic data need to prove to the community
that the benefits for their use overcome the complicated measur-
ing procedures. As such, current applications for phosphoproteo-
mic data are focusing at the early stage such as guiding dosing for
PD, target engagement, screening compound libraries for drug
repositioning, MOA and pathway construction where phospho-
protein data proximity to cell function has clear and distinct
advantages compared with genomic, transcriptomic or metabo-
lomic approaches.
The increased data throughput offered by phopshoproteomic
technologies necessitates the use of systems biology and bioin-
formatics algorithms. A diverse set of computational tools can
play an instrumental part in analyzing phosphoproteomic data
and link them to genomic, transcriptomic and/or metobolomic
datasets, clinical data and general prior knowledge obtained
from literature. Unfortunately, the community still have not
reached a consensus regarding standardization of phosphopro-
teomic data and thus custom approaches are needed for data
gathering, analysis, storage and sharing [54]. Significant com-
putational expertise is needed to identify the appropriate com-
putational tools for drug discovery. For example, a ‘data-driven’
approach such as cluster-based partial least squares regression
(PLSR) can be used when no prior assumptions need to be made
[55]. When a phosphoprotein signature is to be obtained from
categorized data (e.g. responder/nonresponder, normal/patho-
logical sample, safe/toxic drug, pass/fail clinical trials), a super-
vised machine learning approach is usually considered as the
method of choice. Finally, when data are used to train or
validate a computational model (e.g. literature-derived pathway,
ordinary differential equation models) optimization algorithms
are used [7,36,38,39]. As a step forward, those computational
models should be able to incorporate not only the diverse
datasets obtained from other technologies but also the diverse
knowhow found in literature or imposed by the researcher.
Identifying new targets, constructing normal and pathological
pathways, stratifying patients, identifying MOA, repositioning
drugs and predicting efficacy and toxicity are major hallmarks for
improving the drug discovery pipeline. The advancements in
phosphoproteomic technologies enable new approaches to
appear in drug discovery where important decisions can rely
on measurements of signaling activity. Despite the relative imma-
turity of experimental and computational approaches, the impor-
tance of phosphoproteomic measurement is widely accepted for
drug discovery. Further advances in sample and signal through-
put, as well as increasingly sophisticated computational methods
capable of analyzing diverse cohorts of datasets, will provide
scientists with invaluable tools for boosting the drug discovery
pipeline.
AcknowledgementsM.M. and A.C. prepared this review while they were employees at
Merck. L.G.A. and I.N.M. would like to acknowledge funding
support by the European Union Seventh Framework Program
(FP7/2007–2013) under grant agreement no. 305397, and from the
European Social Fund (ESF) and Greek national funds through the
Operational Program ‘Education and Lifelong Learning’ of the
National Strategic Reference Framework (NSRF) – Research
Funding Program: ERC. Investing in knowledge society through
the European Social Fund.
References
1 Kola, I. and Landis, J. (2004) Can the pharmaceutical industry reduce attrition rates?
Nat. Rev. Drug Discov. 3, 711–716
2 Berrade, L. et al. (2011) Protein microarrays: novel developments and applications.
Pharm. Res. 28, 1480–1499
3 Ray, S. et al. (2010) Label-free detection techniques for protein microarrays:
prospects, merits and challenges. Proteomics 10, 731–748
4 Hall, D.A. et al. (2007) Protein microarray technology. Mech. Ageing Dev. 128,
161–167
5 Wolf-Yadlin, A. et al. (2009) Dissecting protein function and signaling using protein
microarrays. Curr. Opin. Chem. Biol. 13, 398–405
6 Poetz, O. et al. (2010) Sequential multiplex analyte capturing for phosphoprotein
profiling. Mol. Cell. Proteomics 9, 2474–2481
7 Mitsos, A. et al. (2009) Identifying drug effects via pathway alterations using an
integer linear programming optimization formulation on phosphoproteomic data.
PLoS Comput. Biol. 5, e1000591
8 Alexopoulos, L.G. et al. (2010) Networks inferred from biochemical data reveal
profound differences in Toll-like receptor and inflammatory signaling between
normal and transformed hepatocytes. Mol. Cell. Proteomics 9, 1849–1865
9 Choudhary, C. and Mann, M. (2010) Decoding signalling networks by mass
spectrometry-based proteomics. Nat. Rev. Mol. Cell Biol. 11, 427–439
10 Gstaiger, M. and Aebersold, R. (2009) Applying mass spectrometry-based
proteomics to genetics, genomics and network biology. Nat. Rev. Genet. 10,
617–627
11 Kaake, R.M. et al. (2010) Profiling of protein interaction networks of protein
complexes using affinity purification and quantitative mass spectrometry. Mol. Cell.
Proteomics 9, 1650–1665
12 Pan, C. et al. (2009) Global effects of kinase inhibitors on signaling networks
revealed by quantitative phosphoproteomics. Mol. Cell. Proteomics 8, 2796–2808
13 Mann, M. (2006) Functional and quantitative proteomics using SILAC. Nat. Rev.
Mol. Cell Biol. 7, 952–958
14 Olsen, J.V. et al. (2010) Quantitative phosphoproteomics reveals widespread full
phosphorylation site occupancy during mitosis. Sci. Signal. 3, ra3
15 Harsha, H.C. et al. (2008) Quantitative proteomics using stable isotope labeling with
amino acids in cell culture. Nat. Protocols 3, 505–516
16 Mandell, J.W. (2003) Phosphorylation state-specific antibodies: applications in
investigative and diagnostic pathology. Am. J. Pathol. 163, 1687–1698
17 Zhang, G. and Ghosh, S. (2002) Negative regulation of Toll-like receptor-mediated
signaling by tollip. J. Biol. Chem. 277, 7059–7065
18 Ross, A.H. et al. (1981) Phosphotyrosine-containing proteins isolated by affinity
chromatography with antibodies to a synthetic hapten. Nature 294, 654–656
19 Levine, L. et al. (1989) Antibodies and radioimmunoassays for phosphoserine,
phosphothreonine and phosphotyrosine. Serologic specificities and levels of the
phosphoamino acids in cytoplasmic fractions of rat tissues. J. Immunol. Methods 124,
239–249
20 Rush, J. et al. (2005) Immunoaffinity profiling of tyrosine phosphorylation in cancer
cells. Nat. Biotechnol. 23, 94–101
www.drugdiscoverytoday.com 7
REVIEWS Drug Discovery Today � Volume 00, Number 00 �November 2013
DRUDIS-1274; No of Pages 8
Review
s�G
ENETO
SCREEN
21 Maecker, H.T. et al. (2012) Standardizing immunophenotyping for the Human
Immunology Project. Nat. Rev. Immunol. 12, 191–200
22 Tanner, S.D. et al. (2013) An introduction to mass cytometry: fundamentals and
applications. Cancer Immunol. Immunother. 62, 955–965
23 Bandura, D.R. et al. (2009) Mass cytometry: technique for real time single cell
multitarget immunoassay based on inductively coupled plasma time-of-flight mass
spectrometry. Anal. Chem. 81, 6813–6822
24 Huang, P.H. et al. (2007) Quantitative analysis of EGFRvIII cellular signaling
networks reveals a combinatorial therapeutic strategy for glioblastoma. Proc. Natl.
Acad. Sci. U. S. A. 104, 12867–12872
25 Chumbalkar, V. et al. (2011) Analysis of phosphotyrosine signaling in glioblastoma
identifies STAT5 as a novel downstream target of (EGFR. J. Proteome Res. 10,
1343–1352
26 Du, J. et al. (2009) Bead-based profiling of tyrosine kinase phosphorylation identifies
SRC as a potential target for glioblastoma therapy. Nat. Biotechnol. 27, 77–83
27 Tworkoski, K. et al. (2011) Phosphoproteomic screen identifies potential therapeutic
targets in melanoma. Mol. Cancer Res. 9, 801–812
28 Accordi, B. et al. (2010) Functional protein network activation mapping reveals new
potential molecular drug targets for poor prognosis pediatric BCP-ALL. PLoS ONE 5,
e13552
29 Li, J. et al. (2010) A chemical and phosphoproteomic characterization of dasatinib
action in lung cancer. Nat. Chem. Biol. 6, 291–299
30 Bantscheff, M. et al. (2007) Quantitative chemical proteomics reveals mechanisms
of action of clinical ABL kinase inhibitors. Nat. Biotechnol. 25, 1035–1044
31 Breitkopf, S.B. et al. (2010) Proteomics analysis of cellular imatinib targets and their
candidate downstream effectors. J. Proteome Res. 9, 6033–6043
32 Sharma, K. et al. (2012) Quantitative proteomics reveals that Hsp90 inhibition
preferentially targets kinases and the DNA damage response. Mol. Cell. Proteomics 11
M111.014654
33 Huber, A. et al. (2009) Characterization of the rapamycin-sensitive
phosphoproteome reveals that Sch9 is a central coordinator of protein synthesis.
Genes Dev. 23, 1929–1943
34 Saez-Rodriguez, J. et al. (2011) Comparing signaling networks between normal
and transformed hepatocytes using discrete logical models. Cancer Res. 71,
5400–5411
35 Melas, I.N. et al. (2012) Construction of large signaling pathways using an adaptive
perturbation approach with phosphoproteomic data. Mol. Biosyst. 8, 1571–1584
36 Saez-Rodriguez, J. et al. (2009) Discrete logic modelling as a means to link protein
signalling networks with functional analysis of mammalian signal transduction.
Mol. Syst. Biol. 5, 331
37 Klamt, S. et al. (2007) Structural and functional analysis of cellular networks with
CellNetAnalyzer. BMC Syst. Biol. 1, 2
8 www.drugdiscoverytoday.com
38 Morris, M.K. et al. (2011) Training signaling pathway maps to biochemical data with
constrained fuzzy logic: quantitative analysis of liver cell responses to inflammatory
stimuli. PLoS Comput. Biol. 7, e1001099
39 Mitsos, A. et al. (2012) Non linear programming (NLP) formulation for quantitative
modeling of protein signal transduction pathways. PLoS ONE 7, e50085
40 Sachs, K. et al. (2005) Causal protein-signaling networks derived from
multiparameter single-cell data. Science 308, 523–529
41 Kholodenko, B. et al. (2012) Computational approaches for analyzing information
flow in biological networks. Sci. Signal. 5, 1
42 Bansal, M. et al. (2007) How to infer gene networks from expression profiles. Mol.
Syst. Biol. 3, 78
43 Melas, I. et al. (2011) Combined logical and data-driven models for linking
signalling pathways to cellular response. BMC Syst. Biol. 5, 107
44 Weber, C. et al. (2012) Dual phosphoproteomics and chemical proteomics analysis
of erlotinib and gefitinib interference in acute myeloid leukemia cells. J. Proteomics
75, 1343–1356
45 Christensen, G.L. et al. (2010) Quantitative phosphoproteomics dissection of seven-
transmembrane receptor signaling using full and biased agonists. Mol. Cell.
Proteomics 9, 1540–1553
46 Bodenmiller, B. et al. (2012) Multiplexed mass cytometry profiling of cellular states
perturbed by small-molecule regulators. Nat. Biotechnol. 30, 858–867
47 Cosgrove, B.D. et al. (2010) Cytokine-associated drug toxicity in human hepatocytes
is associated with signaling network dysregulation. Mol. Biosyst. 6, 1195–1206
48 Amacher, D.E. (2010) The discovery and development of proteomic safety
biomarkers for the detection of drug-induced liver toxicity. Toxicol. Appl. Pharmacol.
245, 134–142
49 Van Summeren, A. et al. (2012) Proteomics in the search for mechanisms and
biomarkers of drug-induced hepatotoxicity. Toxicol. In Vitro 26, 373–385
50 Wagner, J.A. (2008) Strategic approach to fit-for-purpose biomarkers in drug
development. Annu. Rev. Pharmacol. Toxicol. 48, 631–651
51 Paweletz, C.P. et al. (2011) Identification of direct target engagement biomarkers for
kinase-targeted therapeutics. PLoS ONE 6, e26459
52 Creighton, C. et al. (2010) Proteomic and transcriptomic profiling reveals a link
between the PI3 K pathway and lower estrogen-receptor (ER) levels and activity in
ER+ breast cancer. Breast Cancer Res. 12, R40
53 Andersen, J.N. et al. (2010) Pathway-based identification of biomarkers for targeted
therapeutics: personalized oncology with PI3K pathway inhibitors. Sci. Transl. Med.
2, 43ra55
54 Saez-Rodriguez, J. et al. (2011) Setting the standards for signal transduction research.
Sci. Signal. 4, pe10
55 Janes, K.A. and Yaffe, M.B. (2006) Data-driven modelling of signal-transduction
networks. Nat. Rev. Mol. Cell Biol. 7, 820–828