Systems Biology in Immunology - A Computational Modeling Perspective

IY29CH19-Germain ARI 7 February 2011 22:24
ANNUAL
REVIEWS Further Systems Biology in
Click here for quick links to
Annual Reviews content online,
including:
Immunology: A
Other articles in this volume
Top cited articles Computational Modeling
Perspective
Access provided by Linkopings University Library on 10/05/17. For personal use only.
Top downloaded articles

Annu. Rev. Immunol. 2011.29:527-585. Downloaded from www.annualreviews.org
Our comprehensive search
Ronald N. Germain, Martin Meier-Schellersheim,

Aleksandra Nita-Lazar, and Iain D.C. Fraser
Program in Systems Immunology and Infectious Disease Modeling, National Institute of
Allergy and Infectious Disease, Laboratory of Immunology, National Institutes of Health,
Bethesda, Maryland 20892; email: rgermain@nih.gov
Annu. Rev. Immunol. 2011. 29:52785 Keywords

First published online as a Review in Advance on high-throughput analysis, modeling, genomics, proteomics, RNAi
January 4, 2011
The Annual Review of Immunology is online at Abstract

immunol.annualreviews.org
Systems biology is an emerging discipline that combines high-content,
This articles doi:
multiplexed measurements with informatic and computational model-
10.1146/annurev-immunol-030409-101317
ing methods to better understand biological function at various scales.
0732-0582/11/0423-0527$20.00
Here we present a detailed review of the methods used to create com-

This is a work of the U.S. Government and is not putational models and to conduct simulations of immune function. We
subject to copyright protection in the United
provide descriptions of the key data-gathering techniques employed to
States.
generate the quantitative and qualitative data required for such model-
ing and simulation and summarize the progress to date in applying these
tools and techniques to questions of immunological interest, including
infectious disease. We include comments on what insights modeling can
provide that complement information obtained from the more famil-
iar experimental discovery methods used by most investigators and the
reasons why quantitative methods are needed to eventually produce a
better understanding of immune system operation in health and disease.
527
INTRODUCTION zation of microarrays (1), the widespread use

of deep-sequencing methods (2), the advent
Immune responses are symphonies of molec-
of highly multiplexed ow cytometry (3), and
ular and cellular interactions, with each player
the availability of high-throughput proteomics
doing its part to produce the composite
(4). Rather than exploring a single element in
behavior we see as effective host defense, or
depth, investigators employ these latter tech-
when discoordinated, as immunopathology or
nologies to broadly probe the state of biological
immunodeciency. Just as listening separately
systems (gene expression, protein identity,
to the notes played by individual instruments
or substrate modication). This has led to a
fails to capture the ensemble effect achieved
major change in how research is conducted in
when an entire orchestra plays in unison, so
many laboratoriesrather than experiments
too are we limited in our understanding of how
being designed based on preformed hypotheses
the immune system operates when we focus
derived from past training and knowledge of
only on the properties or actions of one or a
the literature, high-throughput methods are

few unconnected components.
being used for unbiased exploration of the
In the nineteenth and early twentieth cen-
properties of a system to then generate novel
turies, biology was largely a study of physiology,
hypotheses (5). It is up to the investigator to
the integrated basis for the functionality of an
sort through the massive amount of new data
organism. However, with the advent of new in-
owing from the various multiplex technolo-
strumentation and technology in the late 1970s,
gies, a process that requires substantial ability
especially recombinant DNA methods, biology
in statistics and/or the capacity to properly use
became progressively more reductionist, with a
algorithms and software developed by experts
focus on individual cells and molecules, and im-
in mathematics and computation. We thus
munology was no exception to this trend. The
have entered a new domain in which the skills
new knowledge acquired by the eld through
of so-called bioinformaticians are becoming
many such detailed studies has been enor-
essential elements in the research efforts of
mously important in developing a parts list of
many laboratories. The technical capacity to
the components involved in immune processes
generate these large-scale, in some cases global,
and in identifying some of the contributions
data sets in turn has led to the emergence of
of these molecular and cellular elements to the
the new discipline of systems biology, which
overall functioning of the system. Nonetheless,
in its simplest form is the old physiology recast
this information still has yielded only limited
in modern guise. It constitutes an attempt by
insights into the way these various elements
the eld to move from the very specic, from
integrate with each other to give rise to
the detail, from the single molecule or gene, to
complex immunological behaviors, especially
a quantitative analysis of such elements as they
into how small quantitative changes in indi-
operate together to give rise to behaviors not
vidual component function affect more global
possessed by any single component aloneso-
properties. This latter issue is of substantial
called emergent properties (6), the symphony
importance, for example, in understanding how
rather than just the notes of the violin or oboe.
polymorphisms linked to disease by large-scale
Many investigators consider the term bioin-
genetic studies inuence immune function.
formatics synonymous with systems biology.
At the same time as new tools were being
But the truth is more complex. Although the
developed for and applied to the ever-ner
statistical analyses of large data sets to look
dissection of the cell, genes, and proteins of the
for trends, to cluster individual components
immune system, another set of technological
into related groups, or to uncover connec-
advances increased the rate of data acquisition
tivity among elements to produce large net-
from a trickle, to a stream, to a river that has
work maps are essential to make use of such
turned into a torrent, with the commoditi-
extensive information, these approaches by
528 Germain et al.

themselves fall short of moving the eld from to a complete organism, the modeler needs the
the mere organization of knowledge to a deeper contributions of informaticians to develop a
understanding of the principles underlying a realistic and valid model structure for further
systems behavior or to an explanation of its computational processing. Conversely, without
mode of operation. The most common output modeling, the mere organization of data does
from informatic manipulation of data elements not add the necessary insight into global system
is a nonparametric graph that shows qualita- performance sought by biologists.
tive interactionsoften referred to as a hair- From the perspective of the practicing im-
ball or ridiculome1 because of the enormous munologist, what does systems biology in all
complexity of such global depictions. In truth, its guises have to offer? Isnt experimentation
these graphs are extremely useful to illustrate not mathematical twiddlingreally the essen-
relationships between elements and to under- tial activity involved in gaining new under-
stand the organization of components into op- standing of how the immune system functions?
erational modules, but they do not allow the And yes, informatics is useful for handling large
investigator to predict how alterations in the data sets, but isnt its major value in the dis-
concentration or efciency of the function of a covery of new interesting molecules or genes
particular element will inuence the overall sys- so one can go back into the lab to study these
tems activity or to discern why and how certain in detail using comfortable experimental tools
properties of the system arise from its elements. and techniques? In this review we argue that
But in the end, this is just what we want from these existing paradigms are changingthat
such a systemic analysis: the ability to fathom the value of traditional experimental studies will
how higher-level function emerges from com- increase dramatically if more quantitative tools
ponents that on their own lack the capacity in are introduced into mainstream immunology
question and to predict how perturbations of research in the form of analytic measurements,
individual elements will change this behavior, formalized model generation, simulations, and
both for the basic insight this provides and for computer predictions, built on a foundation
the potential clinical utility of such information. involving systematic measurements organized
It is the domain of modelers to move from and parsed by informatics approaches.
informatic analyses into this more functional What is the basis for this view? We suggest
realm. Mathematical or computational mod- that too often the interpretation of experimen-
eling is not a new endeavor, especially in tal data is limited by a failure to intuit the com-
immunology, but it is a less widely employed plex, nonlinear behaviors typical of highly con-
and less appreciated aspect of the emerging nected systems with large numbers of feedback
discipline of systems biology as compared to connections (7, 8), which of course perfectly de-
the bioinformatic analysis of data. But we scribes the immune system. If one adds to this
believe that the two are complementary, and the exponential increase and decrease in lym-
indeed, each cannot reach its full potential phocyte cell numbers during adaptive immune
without the other. Computational simulation responses, properties that amplify markedly the
is effective only if the modeler has in hand the inuence of very small differences in the activ-
properly processed and analyzed data necessary ity of molecular circuits or cells (9, 10), the need
to instantiate a model close to biological reality for more formal representations and quantita-
(in terms of element identity, organization, tive analyses of immune function becomes even
and quantitative parameters). At any level of more evident.
resolution, from molecules, to cells, to tissues, We are not talking here about the type of
ad hoc theoretical immunobiology that has ac-
quired a questionable reputation in the past.
1
The term ridiculome comes from a talk given by Andrea Rather, we are referring to the combination
Califano of Columbia University. of rich experimental data sets and existing
www.annualreviews.org Computational Systems Biology 529

knowledge in the eld with newer efforts to suppression, and time-gated function (13)] and
obtain more quantitative measurements of bio- can assist in our understanding of the underly-
chemical or cellular parameters suitable for ing control circuits that regulate behaviors such
computer modeling and simulation. Predic- as switching between tolerant and immune
tions of system behavior from such simula- states (8), the antigen thresholds required for
tions obtained under dened conditions corre- induction of responses, original antigenic sin,
sponding to experimentally testable situations and the choice of CD4 effector fate. In concert
amount to in silico experimentation (11). These with the critical efforts already underway to
predicted outcomes then must be tested at the obtain data on gene expression in immune cells
bench to examine the strength of the underlying in a systematic manner (14) and to quantify
computational model. Through iterative cycles aspects of immune function examined previ-
of such model building, simulation, prediction, ously in a more qualitative manner, we can
experiment, and model renement (when ex- begin to generate a body of models for many
perimental results and prediction disagree as such modules of immune function. These in
they inevitably will), one can develop much turn can be rened by contributions from
more complete and informative models of im- many investigators in the eld, hopefully over
munological processes than those we formulate time approaching the underlying reality more
purely in our imagination or represent as sim- closely and leading eventually to the generation
plied cartoons in reviews. of an integrated supermodel as these smaller
What about the scale of such models? Many pieces prove their worth through rigorous
investigators bemoan the presumed need for experimental testing. It is an opportunity for
completeness to achieve a useful model and all immunologists to contribute to and receive
despair obtaining the data required to reach back from a group undertaking that ultimately
this ultimate goal. Although for bacteria or supports their own specic research interests
yeast it is possible to undertake truly systems- while advancing the entire eld. Rather than
level studies involving the measurement of all being concerned about big science in thinking
gene transcripts or proteins expressed by a cell about systems approaches, we hope that immu-
under various conditions and the systematic nologists will view systems immunology as the
perturbation of each of these elements through new immunophysiology, with opportunities
mutation, this is clearly impractical or impos- for all to participate and to benet.
sible for more complex organisms. However, As discussed above, there are two major
useful models that represent emergent behav- threads in systems biology: informatics and
iors need not involve the entire systemthe modeling. Each has become such a large
complex properties of subcircuits or modules enterprise that we cannot do justice to both
that form key parts of larger networks are valu- here, so we focus on the less commonly used
able to investigate and simulate on their own modeling and simulation aspect. In the body
(7, 12) even if the eventual goal must be to of this review, we discuss the computational
stitch such incomplete models together into a approaches and tools available to translate data
grander scheme that more truly reects overall into models suitable for the simulation and
physiology. To conduct studies at the systems prediction of biological behavior, the key tech-
level, it is merely necessary to at least move up nologies for data acquisition that contribute
the scale from individual component dissection to effective computational modeling, and the
to a consideration of the integrated behavior limitations that must be overcome for their
of sets of connected components (molecules, more effective use in supporting these endeav-
cells, even disparate tissues). Such efforts help ors. In each section, we provide examples of
us organize our thinking about the aspects of how these technologies and tools have begun
the subnetworks structure that give rise to its to contribute to a better understanding of the
specic properties [e.g., amplication, noise immune system. We end with a perspective on
530 Germain et al.

Table 1 Computational approaches and tools for systems biology

Modeling approach Typical applications Limitations Tools
Individual particle- Small subcellular signaling Applies only to small systems (in terms MCell (32), Smoldyn (314),
based stochastic processes, aspects of of space and chemical complexity) ChemCell (315), GetBonNie
bacterial biochemistry (nonspatial) (49)
Particle number Signaling processes with Applies only to small systems (in MesoRD (35), SmartCell (33),
stochastic important stochastic aspects terms of space and chemical GetBonNie (nonspatial)
(due to small system size or complexity), has less detail than
high sensitivity) individual particle simulation
Concentration-based Cellular signaling processes Provides either high spatial resolution Virtual Cell (37), Simmune (36)
spatial, nonstochastic with important spatial or biochemical complexity, has no
aspects stochasticity
Concentration-based, Cellular signaling processes Assumes global biochemical Copasi (46), E-cell (44),
nonspatial, without spatial aspects homogeneity in the simulated system Cellware (45), Systems
nonstochastic Biology Workbench (47),

GetBonNie
what can be expected in the next few years in in the development of specialized computer
this rapidly changing arena. hardware that permit ab initio molecular
dynamics simulations of large peptide-folding
processes (15, 16), the most fundamental
MODELING AND SIMULATION scale of computational models used widely
SCALES, PARADIGMS, in immunology is the scale describing the
TECHNIQUES, AND dynamics of molecular interactions in terms
INTEGRATED APPROACHES of the rates of association, dissociation, and
Systems biology puts a strong emphasis on post-translational modication (PTM). Based
quantitative data and strategies for comprehen- on such molecular event rates, dynamic models
sive measurements of biological parameters. To can be developed that cast the time evolution
understand why the cycle of hypothesis build- of a particular part of cellular biochemistry into
ing, data acquisition, dynamical modeling, and mathematical expressions that describe the rate
subsequent renement of the hypothesis de- of change per unit time of the number of its
mands this emphasis, we begin with a summary molecular species. Depending on the specic
of the different questions that can be addressed question at hand, the focus of such mathemat-
by modeling and simulation, the various types ical descriptions may be on the formation of
of computational models that can be built, and individual multimolecular signaling complexes
the tools available or still needed for construc- (clusters), on the biochemistry of subcellular
tion and simulation of these various models compartments, or on the biochemical behavior
(Table 1). of signaling pathways within entire cells. In
most cases, the equations are too complicated
to be solved on a piece of paper and instead are
Modeling and Simulation at the used in computer simulations that iteratively
Molecular Scale solve them for small time steps to calculate the
Biologists study diverse questions that range evolution of the system over biological time
from the molecular (vibrational modes that af- periods of seconds to hours or days.
fect protein folding or molecular binding) to the Immunologists have long recognized the
organismal scale. Although substantial progress importance of early receptor-receptor liga-
has been made in simulation techniques and tion events in regulating cell function and

differentiation state (1719), and there is in- molecular species with high concentrations em-
creasing evidence for a crucial role of the bio- bedded into spatial domains that go beyond
chemical and conformational properties of the the size of small bacteria. Reaction-diffusion
membrane and proximal actin and myosin ber simulations of large networks of molecules
structures in guiding such events during lym- aimed at investigating more than just a few
phocyte interactions (2025). These insights seconds of signaling dynamics have to give
have prompted numerous theoretical studies up the advantages of particle-based approaches
focusing on the small-scale spatial aspects of in terms of dynamical detail and conceptual
cellular signaling. Stochastic spatially resolved simplicity in favor of the computational ef-
simulations (26) explore the early kinetics of sig- ciency of molecule number- or concentration-
naling processes within small networks of indi- based modeling techniquesthat is, treatment
vidually interacting proteins and lipids. Many of of the bulk or average behavior of a molecular
these approaches utilize Monte Carlo2 meth- species rather than of each individual element.
ods (27, 28) to incorporate the random ther- Such techniques, if they include spatial aspects,
mal uctuations governing molecular Brown- have to divide the relevant cellular space in
ian motion and reactive encounters. With few which the biochemical events are being simu-
exceptions (29, 30), the molecules are treated lated (cytoplasmic domains, intracellular com-
as spatially structureless entities, and only their partments, or membrane regions) into subvol-
approximate radius is (sometimes) taken into umes small enough to justify the assumption
account. Studies that aim at elucidating the co- of local homogenous concentrations, because
operative behavior of receptor complexes in dif- this assumption underlies all concentration-
ferent states of activation frequently introduce or particle numberbased simulations. Individ-
the simplication of assigning those complexes ual molecular Brownian motion simulated in
positions on the nodes of spatial grids where Monte Carlo approaches is replaced by the phe-
only neighboring grid points can inuence each nomenological concept of diffusion, driven by
other to make the simulations computationally concentration differences. Instead of strictly lo-
tractable for large numbers of individual recep- calized bimolecular interactions, mass action-
tors (31). Such methods do establish a model based equations are used to simulate reactions
that corresponds in some degree to data from that modify the chemical composition of the
experimental studies of molecular interactions modeled system (3335).
based on distance-sensitive methods such as Spatially resolved simulations without
Forster resonance energy transfer (FRET), but stochasticity (that is, without explicit treatment
the specic geometry of the grid affects the sim- of the substantial uctuations characteristic of
ulation outcome, and the chosen geometry may systems with small numbers of components
not provide a close approximation of the envi- in which average behavior is not an adequate
ronment in which the actual molecular inter- description) use discretized partial differential
actions occur, limiting the correspondence be- equations describing the effects of diffusion
tween the model and biological system. and of molecular reactions to calculate how
These particle-based methods can be quite molecule concentrations change as a function
detailed and realistic in terms of space (32), of space and time. Without the cost associ-
but they simply are too computationally ex- ated with simulating stochastic effects, such
pensive in applications that include many simulations can afford to calculate extended
concentration time courses for spatial domains
comprising entire cells or even groups of cells
2
The name was chosen with reference to the Monte Carlo embedded in representations of extracellular
gambling casino, due to the element of chance used in the space (3638), provided that the signaling
approach to calculate the probability of particular events
such as molecular movements or interactionsat any time networks are not too complicated. How-
point. ever, for networks that comprise too many
532 Germain et al.

interacting components and/or components distribution in the simulated (sub)cellular do-

that have many functional binding sites me- mains. This in turn places demands on the in-
diating interactions with other molecules, the vestigator that fall into two categories. The rst
resulting systems of equations describing the is the collection of the necessary quantitative
rates of change of the molecular concentrations data to constrain the many parameters used to
in the system may become too large to be dene the model. One needs to know, for exam-
simulated with spatial resolution. Examples of ple, how many molecules of each species exist
signaling networks with great complexity are in what volume and, preferably, in what state of
the pathways downstream of immunorecep- biochemical modication (e.g., phosphorylated
tor or EGF receptor ligation. In fact, these or not, ubiquitinylated or not), their afnities
pathways may be so complex that differential of interaction, and the rates of enzymatic
equationbased approaches trying to simulate transformation. One also needs a reasonably
kinetics down to the activation of transcription complete picture of the molecular interactions
factors face serious problems with regard to just in the (sub)pathway being modeled. For this
the formulation of the reaction kinetics in ways reason, a diverse set of high-throughput
that permit the communication of the models data-acquisition methods is required to collect
assumptions in a human-readable format. the necessary quantitative and organizational
The problem is simple to understandif each data on the components to be included in the
receptor has multiple sites for phosphorylation model, and much of the remainder of this
(e.g., ITAMs), and each phosphorylated site review is devoted to a discussion of the relevant
can bind adaptor molecules that in turn technologies, their uses, and limitations.
recruit kinases or phosphatases subsequently The second category relates to the modeling
interacting with downstream targets, and if and simulation itself. Many modeling efforts
the phosphorylation of each set of sites in the in immunology in particular and biology in
cascade is heterogeneous, then the state of general have relied on expert mathematicians
the system is represented by a combinatorial and computer scientists to develop equations,
expansion that grows rapidly to enormous computer scripts, or code to handle these
numbers. The phenomenon of such combi- formal descriptions and to conduct the cal-
natorial complexity arising in models with culations necessary to simulate a model of
components that have many functional binding interest to the biologist studying the question
sites has been discussed extensively (39, 40). at hand. Although such collaborations between
As an alternative to describing such systems in theorists and experimental biologists reect the
terms of hundreds or thousands of differential inherent multidisciplinary nature of systems
equations, one in principle can describe their biology, they always carry with them the danger
properties in terms of functional molecular of losing in translation important biological
binding sites and their interactions, just as aspects of a model. Moreover, complex realistic
we do in the previous sentence, and leave the models of extensive systems, in particular those
task of unfolding the resulting networks of that cover multiple scales, often are difcult
reactions to the computer simulating their to implement even with expert computational
concentration kinetics (36, 4143). skills. For both these reasons, substantial effort
All the basic simulation techniques dis- has been devoted to developing computer
cussed aboveparticle and concentration software that eases the translation of biological
based, stochastic and deterministic, spatial and thinking into the equations necessary for
nonspatialtake as inputs formal, computer- quantitative simulations. Some approaches
readable descriptions of the quantitative allow the user to dene reactions in tabular
interactions between signaling components form or by connecting nodes of a graphically
and of parameters such as initial concentrations created network of signaling complexes (44
and, for spatially resolved models, their initial 48) and to simulate the dynamics of those

networks. An even more accessible approach into a useful cellular response? The majority
is provided by novel software that performs of immunologists today believe that agonist
an automated generation of reaction networks stimulation differs from stimulation through
from the simple biologist-dened bimolecular nonagonists only with regard to the associa-
interactions that underlie all reaction systems, tion and dissociation rates of the interaction
input via graphical user interfaces offering between the TCR and the pMHC, although
iconic representations of molecules and their other concepts have been proposed (51, 52). A
interactions that aim at resembling classical higher association rate would mean that agonist
signaling diagrams (36, 49) and with which the pMHCs can trigger TCRs with a higher fre-
biologist is comfortable. When combined with quency, whereas a lower dissociation rate would
user-friendly software that also permits devel- give the intercellular TCRpMHC complex a
oping spatially resolved computational models longer lifetime. If the TCR (or its immediate
of cellular behavior (36, 50), such techniques downstream signaling partners) keeps the
remove many of the impediments that in the memory of a previous ligation for a nonnegligi-
past have prevented experimental biologists ble amount of time, both types of kinetic advan-
from exploring the validity of their hypotheses tage of agonists over nonagonists would mean
through quantitative computer simulations. that the TCR is activated over greater periods
Furthermore, the output in these tools can take of time, even between binding events (53), giv-
the form of graphs and charts that resemble ing it the opportunity to activate downstream
the output of bench experiments, allowing the signaling components more effectively. This
biologist to understand easily the predictions memory, however, must not last too long as it
made by the model and simulation and to otherwise would blur the distinction between
relate this output to data that can be derived the kinetics of agonist and nonagonist binding.
from new experiments corresponding to the Full phosphorylation of the ITAMs in the
conditions chosen for the computer simulation. TCR-associated -chain by Lck is the hall-
mark of successful TCR activation. For this
to occur, the binding of the TCR to the
Molecular-Scale Modeling of Immune pMHC ligand has to (transiently) shift the
Receptor Interactions and Proximal balance between the activities of the kinases
Signaling Events directly or indirectly inducing this phospho-
These various modeling approaches for simu- rylation and the phosphatases reverting it in
lating molecular events have been applied in an favor of the kinases. Many modeling efforts
increasing number of laboratories to explore have been directed toward the identication
the basis for immune cell signaling. A challenge of the mechanisms that can achieve this out-
of modelers for decades has been the ability of come with high sensitivity [to allow for re-
lymphocytes to react quickly and strongly to the sponses to low densities of agonist pMHCs on
engagement of even a few immune receptors antigen-presenting cells (APCs)] while retain-
by pathogen-derived ligands while ltering out ing the capacity seen in real cells to distin-
the overwhelmingly more frequent ambient guish between agonists and nonagonists. To ac-
stimuli originating from uninfected host tissue. commodate the latter requirement, researchers
For T cell activation, the two main unanswered introduced the concept of kinetic proofread-
questions are (a) what distinguishes the binding ing through the TCR signaling network (54),
of an agonist peptidemajor histocompatibility according to which ligation of the TCR in-
complex molecule (pMHC) ligand from the duces a sequence of state transitions that are
binding of a nonagonist pMHC ligand, from reverted once the TCR:pMHC bond breaks.
the point of view of the T cell receptor (TCR), Stably bound (or, if the system has memory,
and (b) how does the T cell translate this rapidly rebinding) ligands (see 55 for a quantita-
difference in the nature of the primary stimulus tive exploration) would complete the transition
534 Germain et al.

sequence to the end point of full TCR activa- and APCs. Goldstein and colleagues (61, 62)
tion, whereas weak ligands would lead only to investigated the relationships between TCR-
progression through part of the sequence be- pMHC dissociation rates and the ability of a
fore dissociating and would cause the signaling single pMHC to engage many TCRs, demon-
system to fall back to its basal state. However, strating that a model with physiological rates for
the questions of which signaling components molecular interactions and diffusion could con-
and mechanisms can account for this behav- rm earlier experimental estimates of substan-
ior and whether the above-mentioned shift to- tial serial engagement (63). Focusing entirely
ward higher kinase activity is part of the proof- on the physical aspects of the T-APC interface,
reading have remained controversial. Based on Chakraborty and coworkers (64) showed that
the notion that close contact between the T the characteristic spatial patternthe immuno-
cell and the APC membrane surface necessary logical synapse (for review see, e.g., 65)seen
for the formation of TCR:pMHC complexes for the distribution of immune and adhesion
would exclude the phosphatase CD45 due to receptors in theory could emerge just based on
its large bulky extracellular domain (thus limit- physical membrane properties and receptor size
ing dephosphorylation of Lck and the ITAMs differences.
in the CD3 and -chains), van der Merwe and Downstream of the immediate receptor-
colleagues (56) suggested that kinetic segrega- proximal ligand discrimination, T cells engage
tion of kinases and phosphatases could be the a complex signaling network, enforcing its cell-
mechanism creating (relatively) kinase-rich do- wide consequences (activation or adaptation)
mains promoting the initiation of the proof- through various feedback mechanisms. Among
reading cascade. To support their model, these these, the competing activation of the phos-
authors performed particle-based Monte Carlo phatase SHP-1 and the kinase Erk was shown
simulations in which the particles, representing to play an important role (66). Based on this ex-
a pMHC and TCR, were conned to move- perimental work, several computational mod-
ment on the nodes of a two-dimensional spatial els were developed. Because many molecules
grid representing the membranes. The simula- involved in these networks are cytosolic, most
tions demonstrated that the segregation model models assume that spatial aspects at this later
can achieve ligand discrimination but that it stage of the cellular response are less impor-
did not reproduce the experimentally observed tant than those during its rst seconds, in spite
ability of T cells to perform this discrimina- of growing evidence that intracellular signal-
tion within less than 1 min, as biochemical ing cascades such as the one activating MAPK
studies show is the case (18). To reproduce involve spatial coordination of molecular in-
these reported timescales, computational mod- teractions with the help of scaffolds (67, 68).
els may have to focus on the smallest biochem- Chan et al. (69) used a simple differential equa-
ical scale, the scale of signaling microclusters tion model to investigate the role of phos-
(57, 58), or even individual TCRs interacting phatase/kinase feedbacks and showed how such
locally with a highly dynamic actin skeleton mechanisms would be able to take advantage
(59). Such models would require experimental of self- (or null) peptides and could also ac-
data that combine high-resolution microscopy count for the phenomenon of antagonism. A
techniques with careful quantitative analysis of far more complex model with explicit simu-
molecular movement and interactions (60). lation of many of the molecular interactions
Whereas the models discussed above focus downstream of the TCR allowed Altan-Bonnet
on specic biochemical aspects of membrane- & Germain (70) to make detailed predictions
proximal TCR signaling, other modeling ef- about the dynamics of SHP-1 and Erk acti-
forts have aimed at elucidating the more ki- vation that could be tested experimentally. A
netic or biophysical aspects of the interactions key nding of this work was that the Erk re-
between the receptors of conjugated T cells sponse of single T cells was digitalon or

offin contrast to the more continuous dis- on different spatial and biological scales have
tribution of Erk phosphorylation levels ob- emerged that treat biological systems in which
served in populations of stimulated cells. Us- the cell is the smallest element of function. The
ing a stochastic simulation based on Gillespies rst uses mainly differential equationbased,
(71) Monte Carlo algorithm, Chakraborty and mathematical models to investigate organism-
colleagues (72) showed subsequently that feed- wide cell population dynamics, in particular
back activation by Ras-GTP of the Ras-specic in response to pathogenic challenges such as
nucleotide exchange factor SOS (73) can lead viral or bacterial infection. These are typically
to such all-or-nothing cellular responses in the compartment models, in which the differential
Erk pathway and a lter-like sharp distinction equations report the change over time in the
between weak and strong signals coming from concentration (number) of particular elements
the TCR. Although postulated originally for (cells, pathogens) in dened organ (spatial)
TCR signaling, the concept of kinetic proof- compartments such as lymph node, liver, or
reading was shown subsequently to be rele- lung, as well as in functional compartments such
vant for B cell receptor ligand discrimination as as central memory versus effector T cells. The
well, and mathematical models were used to ex- second categorytypically focusing on cell
plore its limitations and point out the possibil- population-wide consequences of single cell
ity of low-afnity ligands escaping this quality decisionsexplores the balance between
control (74). cell-autonomous programmatic behavior, on
Although these initial efforts at molecular- the one hand, and cellular cross talk, on the
scale modeling and simulation in immune other hand, in the regulation of processes such
systems have been productive, they fall short as cell activation, differentiation, division, or
of combining all the elements that would make death. The third category takes direct advan-
the work more robust. Careful proteomic tage of recent advances in high-resolution in
studies that provide protein abundances and vivo microscopy and of ever-increasing com-
interaction rates, combined with extensions puter power to study how individual cells react
of ongoing elegant work involving single-cell to external mechanical and chemical stimuli
imaging that reports molecular reorganization with the help of image-guided computational
and signaling behavior in a spatiotemporal models that incorporate aspects of dynamic
manner with high resolution, will allow more single-cell morphology.
complete models to be built and simulated. Operating on the highest, organism-wide
These will more closely reect biological reality scale, the rst category has the longest history in
and hence have more power to predict system mathematical immunologynot only because
behavior and guide experimental advances. the types of experimental data for such stud-
Recent work exploring apoptotic signaling in ies, mainly derived from cell counts in blood
nonlymphoid cells has demonstrated clearly the samples or dissociated organs, have been avail-
possibilities and benets of such efforts (7577). able for many decades, but also because many
models in this category are based on the an-
cient family of predator-prey models (78) de-
Modeling Cellular Behavior scribing how interacting species (such as foxes
Many issues discussed above for the modeling of and rabbits or, in our case, immune cells and
molecular behavior and interaction networks pathogens) inuence each others population
stochastic versus deterministic treatment, dis- sizes. The basic idea behind these models is that
crete simulations versus differential equations, proliferation and death rates of one species de-
spatial versus nonspatial approachesapply pend on the size of the populations of other
similarly to computational models at the level species with which it interacts: When there are
of cellular behavior and cellular interaction a lot of rabbits, foxes nd enough food to pro-
networks. Three main modeling categories liferate quickly. When their appetite for rabbits
536 Germain et al.

starts causing a reduction of their food source, instance, a decrease in the size of one popu-
this in turn causes a decrease in the prolifera- lation could stem from an increase of death
tion of foxes. The smaller number of foxes then or differentiation (leaving one population and
allows the rabbits to recover, starting a new cy- entering another) or reduced proliferation or
cle. The transfer to immunology in principle inux from other populations. Many modeling
is straightforwardwith the interesting twist efforts therefore take advantage of data that
that in HIV infection the roles of the foxes report cell-proliferation histories in addition
and the rabbits are not clearly assigned as both to cell counts. Cell labels such as BrdU and
(immune cells and virus) are sometimes preda- CFSE become diluted by a factor of two every
tor and sometimes prey. In spite of their time a cell divides. Following the uorescence
simplicity, these approaches have contributed intensity of such markers in different cell pop-
much to our understanding of host-pathogen ulations thus allows the experimentalist and
dynamics because of the way they sharpened modeler to distinguish, for instance, between
our intuition with regard to cell population dy- increased proliferation and reduced death (85
namics. Modeling studies of the kinetics of the 88). Careful analysis of the time evolution of
loss of T cells after SIV/HIV infection (79, CFSE intensity proles in single-dividing-cell
80) and the (only incomplete) recovery follow- populations moreover has yielded valuable
ing highly active antiretroviral therapy have insights into the role played by cell-intrinsic
prompted controversial discussions and led to variability and its heritability between mother
experiments that moved the eld from the con- and daughter cells in lymphocyte dynamics
cept of increased T cell proliferation after in- (85, 89).
fection as a homeostatic response following di- Although kinetic studies of cell population
rect virus-induced depletion toward a focus on sizes can provide insights into high-level regu-
uncontrolled broad immune activation (81, 82) latory dynamics and the relative contributions
and gradual loss of central memory T cells (83) of cell-intrinsic properties and external regula-
as the driving force behind the progression to tion, they cannot explore readily the cell biolog-
AIDS. ical foundations for those dynamics: Knowing
Almost as a side effect of the HIV modeling that cell population A will proliferate with a rate
efforts (84), theorists developed differen- that is inversely proportional to the death rate of
tial equationbased population models to population B does not immediately help us un-
understand the regulation of lymphocyte derstand the factors on which a cell of a particu-
homeostasis. The experimental inputs for such lar type will base its decision to die, proliferate,
models typically consist of population sizes differentiate, or change its location (e.g., the
(cell counts) of cells of specic types or differ- balance of previous history versus contempo-
entiation states. As the goal of these studies is raneous exposure to cytokines, chemokines, or
to infer relationships among the proliferation, direct cell-cell contacts). Some degree of detail
differentiation, and death rates of interdepen- of cellular communication can be incorporated
dent cell types by tting the corresponding into differential equation models, for instance,
parameters of equations describing the change by allowing differentiation rates to depend on
over time of the population sizes, the ideal extracellular concentrations of cytokines and
data consist of longitudinal studies measuring modeling the time course of such concentra-
the relevant cell counts at several consecutive tions alongside the cell population sizes (90).
time points. However, longitudinal studies are However, cytokine and chemokine concentra-
difcult to conduct with human subjects and tions may be strongly location dependent just
are expensive with nonhuman primates. More- as are cell densities, and hence opportunities
over, the combined effects of proliferation, for direct cell-cell communication frequently
differentiation, and death are difcult to extract will be distributed nonhomogenously in vivo
from cell-count time series alone because, for (91), a property of the system that is not

captured with ordinary differential equations molecular species or cell types. In many cases,
that are the basis for most such models. such systems have natural representations as
Although differential equationbased ap- networks. This is obvious for molecular signal-
proaches can be extended to include multiple ing processes. The nodes of the models of such
spatial compartments with specic cellular and networks represent concentrations or activa-
molecular composition (92), agent-based mod- tion states of the different molecular species,
els (which simulate individual cells as discrete whereas the links (or edges) encode interactions
agents that interact with other agents and spec- and state transitions (e.g., think of phosphory-
ify their locations and states) can capture such lation of a molecular species). Simulating such
spatial aspects more easily. Since the seminal models then simply means updating the con-
work of Celada & Seiden (93, 94), who devel- centrations (or activation states) according to
oped their simulator ImmSim to study immune the interactions (and rates) associated with the
receptor signalbased cellular behavior with a links. Some modelers introduce the simplica-
bit-string representation for receptor specici- tion that the nodes can be only in two stateson
ties, agent-based modeling in immunology has or off. Such networks, which have been applied
been used in particular to study phenomena that to immunological systems such as in the analy-
involve the formation of specic spatial cellu- sis of TCR activation (102), are called Boolean
lar arrangements. Among the classical examples networks because the rules for update deter-
in this category are the formation of germi- mine the on or off state for the next iteration,
nal centers and the concomitant evolution of based on operations that produce 0 or 1 values
B cell receptor afnities (95, 96), the dynamics and take as input logical combinations of 0 and 1
of pathogen control through granuloma forma- values. For example, a node may switch to on if
tion in the immune response to Mycobacterium the neighbor nodes it is linked to are all on (log-
tuberculosis infection (97), and the maturation ical AND operation). Although such networks
of T cells during their migration through the obviously have far fewer parameters than the
spatially heterogeneous structures of the thy- number that must be provided for a simulation
mus (98). Whereas the majority of agent-based of continuous-state networks, the lack of graded
approaches neglect the inuence of mechani- responses of single nodes seriously limits the
cal interactions between cells and constrain the possible dynamical modes of such networks and
positions of the simulated cells to the nodes of hence how well they reect biological reality.
a spatial grid, recent modeling efforts, build- Many of the above remarks on molecular
ing on pioneering work by Graner & Glazier networks apply to network models of interact-
(99), have begun to incorporate cellular mor- ing cells types. In cases in which the description
phological dynamics into simulations. Among of the state of a cell type in a model involves sev-
many potential applications, the present focus eral parameters (as opposed to just one for con-
of these new approaches has been to reproduce centration or activation), cellular interaction
data from in vivo microscopic observation of networks describe rules for interactions and
the interactions between T cells and APCs with induced transformations between multistate
high delity to permit more accurate estima- entities, sometimes called automata. The sim-
tion of the duration of these interactions and ulation tool Simmune (36) combines molecular
how they are inuenced by the stromal net- interaction networks and cellular state tran-
works within lymph nodes (100, 101). sitions by allowing users to couple particular
biochemical states (concentration thresholds)
to whole-cell responses such as division, death,
Network Models for Molecular movement, or secretion of new molecules.
and Cellular Interactions In addition to the dynamic approaches dis-
In the previous two sections we discuss various cussed above, networks also are used to repre-
modeling approaches for systems of interacting sent more static (or at least slowly changing)
538 Germain et al.

relationships between biological entities. Typ- example applications, and some limitations. An
ically, the goal of those models is to nd the understanding of the tools presently available,
most likely network of interdependencies be- the advances needed to address deciencies
tween, for instance, the transcriptional states in the current methods, and the time frame for
of genes that is compatible with messenger these techniques to be well employed is essen-
(m)RNA expression data obtained under var- tial for an investigator to develop an effective
ious experimental conditions. A major mathe- systems approach to immunological questions
matical/statistical approach used for such net- of interest.
work construction is Bayesian inference. For a
more thorough discussion, we refer the reader
to the bioinformatics literature (103). RNA Expression Analysis
Nucleic acidbased technologies provide the
most comprehensive analysis of cell and tissue
LARGE-SCALE
states. As such they are the most well-developed

DATA-ACQUISITION and available methods for generating the so-
TECHNOLOGIES called parts list of a cell. Transcript proling
To construct models at various biological scales using microarray technology has been the most
(cells, tissues, the whole organism) using the ap- widely used omic technology and has pro-
proaches and tools discussed above, one needs vided signicant insights into immunological
a variety of data sets. For the generation of development, homeostasis, and transcriptional
comprehensive models at any of these scales, dynamics during the immune response to anti-
one must know the entire set of expressed gens and to pathogens. However, this method is
proteins and other molecules (e.g., the pro- an incomplete reporter of cell statethere can
teome, glycome, lipidome), as well as the quan- be a substantial difference between the tran-
titative rules governing their interactions and scriptome and proteome in a cell (104106),
chemical transformations, the gene activity (the and immunologists are well acquainted with the
transcriptome) and epigenetic state, and the existence of stored mRNA that is translated
metabolic state. However, as indicated above, only upon cellular stimulation (107). There-
the construction of complete models is a task for fore, complementary methods are needed to
the distant future, except perhaps in the realm of develop an effective model of cellular behavior
metabolomics. There is nevertheless still great based on the actual set of expressed proteins,
value in generating models based on less com- for example. At the same time, transcriptomic
plete data and focused only on a dened aspect technology can provide an extraordinarily use-
of a cell, tissue, or organ that seek to under- ful and extensive, although incomplete, picture
stand the properties of an ensemble of elements of the differentiation state of a cell or tissue, and
rather than each component alone. In pursuit it is an effective means of monitoring changes
of this goal, an increasing number of large-scale in this state induced by exposure to a stimulus.
and, in some cases, global data-acquisition tech- Both ne-grained models of cellular biochem-
nologies have been developed that enable the istry and coarse-grained models involving tissue
collection of the information necessary to un- and organ function often are built on pathway
dertake useful model building and simulation. maps emerging from the microarray analysis of
Each technology has its benets and limita- transcriptional states and small RNA expres-
tions, and an understanding of these competing sion, in some cases linked to genetic mapping
aspects of the methodologies is key to devising of quantitative trait loci (QTLs) in a technique
an effective and reliable approach to a systems called expression (e)QTL mapping (103, 108).
analysis at any level of biological resolution. This technique, the oldest of the omic-scale nu-
In this section, we summarize these technolo- cleic acid methods, is now being supplanted by
gies, essential aspects of their proper use, some the emergence of next-generation sequencing

(NGS), which promises to revolutionize not regard, through analysis of signatures from
only mRNA transcript proling but also the whole blood, peripheral blood mononuclear
analysis of genome sequence variation, small cells, or specic hematopoietic cell populations.
RNA expression and function, epigenetic mod- The obvious value of high-quality transcript
ication, and protein-nucleic acid interactions. data repositories to biological research has led
It has been 15 years since the rst mi- to numerous efforts to generate comprehensive
croarray techniques were described using databases of mRNA expression maps in most
complementary (c)DNA spotted on lter experimental model systems. Following the
paper (109, 110), and this technology has characterization of transcriptional responses to
been rened at a remarkable rate with the pathogens in innate immune cells (117120),
development rst of cDNA-based and second the analysis of the blood transcriptome of pa-
of oligonucleotide-based, imaging-friendly tients has produced comprehensive catalogs of
microarrays that permit genome-wide prol- gene changes characteristic of autoimmune and
ing of mRNA expression levels under varying inammatory disorders (121126) and specic
experimental conditions. Oligonucleotide responses to viruses (127130), bacteria (131
arrays have progressed from features that had a 133), and parasites (118, 134). The ImmGen
bias toward the 3 region of mRNA transcripts project is attempting to provide the eld with
and were unable to discriminate splice variants a highly qualied catalog of gene expression in
to currently available exon-tiling arrays that nearly all identied hematopoietic cell types in
cover the entire transcript of each gene and the resting state and after dened stimulation
provide signicant insight into the specic with ligands such as those engaging Toll-like
splice variants expressed (111, 112). Moreover, receptors or various cytokine receptors (14).
with the explosion of available whole-genome Although microarray technology will re-
sequences, it has become commonplace to main an important contributor to integrative
generate arrays specic to multiple organisms, biology programs for the foreseeable future, it
including pathogens (113), which permits the has some inherent drawbacks. First, it is de-
simultaneous tracking of expression changes pendent on prior knowledge of a given tran-
in both host and pathogen during an infection. scriptome and thus is limited to organisms with
Microarray technology has made a huge con- fully sequenced genomes. Second, the depen-
tribution to our understanding of the dynamic dency on nucleic acid hybridization contributes
transcriptome in normal and disease states, an unavoidable degree of noise in the transcript
and in clinical applications such as cancer cell prole and a limited dynamic range leading to
proling and disease prognosis, it has shown poor performance in quantifying less abundant
remarkable predictive capacity. Its ability to transcripts. Some of these drawbacks can be ad-
identify signatures in samples from patients dressed using the direct analysis of specic tran-
with acute myeloid and lymphocytic leukemia scripts with the quantitative polymerase chain
that correlated with disease outcome was one reaction (PCR) technique, although this is more
of the rst demonstrations of the power of non- limited in the number of genes that can be ana-
biased large-scale data-gathering approaches lyzed in a single experiment. These limitations
to provide insight that would have been almost can be improved by the use of microuidics to
impossible to obtain from more reductionist parallelize hundreds of reactions in a single as-
methods (114). These initial successes were say (135), multiplexed bead-based assays that
followed with the clear demonstration that ex- permit the simultaneous detection of up to 100
pression proles could guide therapy, through transcripts (136), and the recent development of
the correlation of signatures with clinical the Nanostring technology that can capture di-
outcomes in many tumors, both solid and rectly up to 500 different nonamplied mRNA
hematological (115, 116). The immune system transcripts using a unique molecular bar-coding
has proven to be particularly accessible in this approach (137). However, at the genome-wide
540 Germain et al.

level, all these technologies likely will be super- tion of a quantitative distribution of expressed
seded in the coming years by recent advances mRNA across the genome. This has a much
in nucleic acid sequencing that will provide not greater dynamic range than array technology
only more comprehensive coverage, but also (over ve orders of magnitude), permitting
more quantitative data suited to the use of infor- more effective identication of low-abundance
mation gathered for qualitative descriptions of transcripts, and it essentially eliminates the
the cell or tissue state and for quantitative mod- background noise associated with hybridiza-
eling of changing gene activity correlated with tion. It is also less prone to variation between
alterations in chromatin state and transcription research groups using different array platforms,
factor availability in the nucleus. although it is not without its drawbacks. Data-
storage requirements are signicant, and con-
siderable bioinformatic resources are needed to
Next-Generation Sequencing manage the data and deal with alignment chal-
NGS already has made a signicant contri- lenges such as those associated with repetitive
bution to many elds of biology, accelerating elements (141). One also must deal with se-
the analysis of genomic sequence variation and quencing errors; these are especially important
disease linkage, epigenetics, transcriptomics, in studies of the microbiome in which align-
and small RNA function. Next-generation ment to reference sequences is not possible.
sequencers, initially developed by 454 and The most-developed NGS application has
Illumina, and followed soon after by ABI and been the analysis of nucleic acid bound to a
Helicos, use massively parallel analysis of specic protein, through the use of chromatin
individually amplied DNA fragments (138, immunoprecipitation (ChIP). This global
139). Although a run on the capillary-based method is ideal for developing gene regulatory
sequencers used for the Human Genome network maps for model building, including
Project could provide approximately 100 reads models that link intracellular biochemical
of 800 base pairs (bp), NGS machines produce changes induced by extracellular stimuli to
shorter reads of 35400 bp, but the number of changes in gene expression. In ChIP, an
reads can reach 107 , thus exceeding 1 gigabase antibody recognizing a specic protein (such
of sequence per run. This has an enormous as RNA polymerase, a sequence-specic
impact on the potential experimental applica- transcription factor, or a modied histone) is
tions of sequencing approaches. Genome-wide used to afnity-purify the protein chemically
analyses can be undertaken that do not entail cross-linked to its associated nucleic acid,
sequencing the entire genome, but instead which then is separated from the conjugated
represent it as expressed mRNA, modied protein and subject to deep sequencing (142).
DNA (e.g., methylated), nucleic acid bound to In the case of RNA pol IIbased ChIP-seq, this
a specic protein (e.g., a transcription factor), provides a dynamic complement to RNA-seq
or DNA sensitive to enzymatic degradation (2). as it identies those genes being transcribed
As a direct alternative to microarrays for actively under specic experimental conditions,
mRNA proling, cDNA reverse transcribed and this can be taken a step further by sequenc-
from poly A+ RNA can be deep sequenced di- ing ribosome-protected mRNA to provide a
rectly [RNA-seq (140)] after adapter ligation ei- quantitative measure of translation. Similarly,
ther from one end (single-end sequencing) or applications of ChIP-seq with transcriptional
both (paired-end sequencing), the latter being regulatory elements and chromatin-associated
particularly important in the analysis of splice proteins are becoming commonplace and
variants (this should be facilitated further as the promise to revolutionize our understanding of
length of NGS reads increase). Mapping each how transcription factor binding and epige-
individual read back to a reference genome and netic modications control gene expression on
counting hits at each site lead to the genera- a system-wide level (143, 144).

In the context of disease characterization, signaling to change of cell state and, because
genome-wide association studies also are be- they often target transcription factors, to gene
ing enriched by NGS. The analysis of genomes expression. The cellular processes engaged
of patients for common genetic polymorphisms during small interfering (si)RNA-mediated
that correlate with specic traits, such as com- gene knockdown are part of the endogenous
plex or multifactorial disease, has until now pathway for the processing of miRNAs, which
been done by proling several hundred thou- are expressed initially as pol IIbased primary
sand single-nucleotide polymorphisms (SNPs) transcripts of several hundred nucleotides.
using array or bead-based technology (145). Ul- Processing by the ribonucleases Drosha and
timately, NGS will increase SNP coverage sig- Dicer produces a mature miRNA, a double-
nicantly beyond these most common mark- stranded (ds)RNA duplex of 2023 nucleotides
ers, and efforts such as the 1000 Genomes (nt), which binds to cognate target mRNAs,
Project, established in 2008, promise to ex- primarily within their 3 untranslated region
pand signicantly our knowledge of human ge- (UTR), through a 68-nt seed sequence toward
nomic variation and disease linkage. However, the 5 end of the mature miRNA leading strand.
most complex diseases cannot be explained by This leads to a reduction in protein expression,
a single gene mutation and are instead caused usually through a combination of mRNA
by multiple complex factors. Combined anal- destabilization and translational repression
ysis of transcript proles with genetic varia- (149). Several hundred miRNA genes have
tion has shown that gene expression can be been identied, and many have been classied
inuenced signicantly by polymorphisms in into families based on seed sequence homol-
regulatory elements. Such quantitative traits ogy. Although this would suggest overlap in
based on statistically signicant changes in tran- target mRNAs and some degree of functional
script abundance have been termed eQTLs redundancy, experimental evidence suggests a
(103, 108). Although accelerated data genera- signicant amount of (still poorly understood)
tion by NGS and even newer sequencing tech- functional specicity among family members.
nologies will make this an area of rapid progress Attempts to model the changing expression of
in the next few years, existing eQTL maps will molecules (in particular proteins) in a cell in
need to move beyond the use of Epstein-Barr response to extracellular stimuli will require
virus (EBV)-derived lymphoblastoid cell lines integration of methods such as those detailed
to more relevant tissue-specic sources (146). above for tracking protein-encoding transcripts
Also eQTL mapping can take us only so far with information on the post-transcriptional
and will require integration of metadata, such regulation of translation by miRNAs, along
as the analysis of regulatory small RNAs and with proteomic methods for direct assessment
protein modication of regulatory factors, to of the translation products, as discussed below.
permit a more comprehensive interpretation of Due to their RNA-based structure and
the consequences of genetic sequence variation Watson-Crick base-pairing-dependent mech-
in disease susceptibility. anism, the technology to prole and probe
the function of miRNAs has developed rapidly
through the modication of well-established
miRNA Profiling nucleic acidbased methods. This has led to the
Since their relatively recent discovery in plants rapid generation of miRNA expression proles
and nematodes, micro (mi)RNAs have been across cell types in normal and disease states
shown to have key roles in the regulation of (150). Similarly, the analysis of the function
the genome/proteome in numerous cell types, of specic miRNAs has been facilitated by the
including those of hematopoietic origin (147, development of antisense-based inhibitors (an-
148). As such, they are key elements in at- tagomirs) and chemically synthesized miRNA
tempts to build comprehensive models that link mimics (151, 152). As these reagents are small
542 Germain et al.

RNAs themselves, they can be transfected into interaction rarely leads to a >50% reduction in
cells using the same delivery protocols as for expression and more commonly averages ap-
siRNA, and many screening groups now rou- proximately 30%. Ninety percent of conserved
tinely add these reagents targeting all known interactions are single site, although most
miRNAs to their siRNA screening pipelines to mRNAs have at least one further site for a
provide a more direct means to identify the different miRNA that can promote further
miRNAs inuencing the cellular process under repression. Because a 50% change in expres-
study. The advantage of this approach is that sion of a protein may not result in a marked
it can identify miRNAs important for main- phenotype, as evidenced by the rarity of
taining a cellular state that do not necessar- haploinsufcient mutants, this might explain
ily change their expression level between ex- why few phenotypes have been observed in sys-
perimental conditions (and thus would not be tematic screens for miRNA mutants, although
picked up by proling studies). some important immunological exceptions to
Experimental investigations of miRNA ex- this view have emerged recently in cases in
pression and function have been complemented which single miRNAs target multiple key sig-
by computational approaches for miRNA tar- naling molecules (reviewed in 148). The degree
get prediction. Based primarily on the analysis of inuence of an miRNA on a given process
of miRNA conservation across species and the also may be exaggerated in laboratory settings
conservation of seed sequences among orthol- as the described technologies for modulating
ogous 3 UTRs, target prediction algorithms miRNA expression can lead to more dramatic
based on seed matching have improved their alterations than can occur physiologically. Lab-
predictive accuracy and have reduced their false oratory conditions likely are more favorable to
prediction rate to less than 25% (149). How- maintaining the stability of mutant phenotypes,
ever, these algorithms still have a signicant but conditions that better simulate the environ-
false-negative rate as the inuence of sequence ment that shaped evolution may be more suited
context beyond the seed match, both in the to uncovering the importance of miRNAs in
miRNA and the target mRNA, is poorly un- regulating the proteome. This brings us to the
derstood. Thus experimental approaches to the question of the mechanism by which miRNAs
identication of miRNA targets are in need of inuence mammalian cell function. It has been
further development. Immunoprecipitation of proposed that, rather than promoting large
the protein components of the RNA-induced changes in the expression of a few proteins,
silencing complex (RISC-IP), followed by they function on a broader scale to tune the
array deep-sequencing analysis to identify proteome. The high degree of conservation of
the miRNA-bound mRNA, holds signicant miRNA target sites, and the selective avoidance
promise but has not been applied to large-scale of seed sequences in the 3 UTRs of mRNA that
studies (153). The coupling of mRNA proling are coexpressed with a given miRNA, suggests
(by microarray or RNA-seq) with quantitative that precise control of protein concentration
proteomics under conditions of differential has signicant implications for organismal
miRNA expression is another approach to tness. Such subtle changes in protein con-
target identication, and the feasibility of this centrations may sensitize regulatory networks.
method has been demonstrated in two recently These considerations re-emphasize the point
published studies (154, 155). These approaches made above that miRNAs and their activities
are vital to determine the scope of action of are crucial to the development of informative
individual miRNAs and to infer the functional models relating stimulus to response because
consequence of their altered expression on a network properties clearly involve not only the
broader scale. protein interactome we more typically map and
Certain characteristics of miRNA function whose component modications we routinely
are emerging. A single-site miRNA/mRNA consider, but also the inuence of temporally

changing miRNA expression on the balance of specic elements that play a role in signaling or
protein components within such networks. metabolic pathways; maintain cellular home-
Another notable pattern that has emerged ostasis; regulate cell structure, intracellular
from miRNA studies, particularly in immune transport, and capacity for migration; and mod-
cells, is the conservation of target sites for ify gene expression and in other ways contribute
a single miRNA, or miRNA family, within to normal physiology or, when abnormal in
proteins involved in a common process. For form or amount, give rise to disease. Genetic
example, several key signaling components assessments of this type have been valuable es-
involved in the response of monocytic cells pecially in model organisms such as Drosophila
to bacterial endotoxin, such as IRAK1 and in which common phenotypes often have been
TRAF6, are targets for miR146 (156), whereas found to be caused by genes acting in a single
phosphatases shown to downregulate signaling linked signaling pathway, associated with rec-
through the TCR (SHP2, PTPN22, DUSP5, ognizable organelles or structural elements of
and DUSP6) are targets for miR181 (157). the cell, or comprising a linked gene regulatory
As these targets have a coherent inuence on pathway (159). However, this approach tends
signal ow through the pathway (i.e., either to identify genes contributing to core func-
activating or inhibiting the process), these tionality conserved across species rather than
ndings suggest that these miRNAs, which the components and mechanisms responsible
have been shown to change expression in re- for the subtleties of cell-type specicity and
sponse to cellular stimulation, act as negative or context-dependent cellular function. Thus our
positive feedback regulators. This is consistent understanding of pathways remains incom-
with their tuning function, in which they can plete, and the discovery of unknown pathway
adjust thresholds for activation under various components has been hampered by canonical
conditions. They also may have key roles in bias in experimental design and reagent avail-
mediating cross talk between inammatory ability (160). Nonbiased approaches therefore
mediators, as evidenced by recent data suggest- are vital to ll in the gaps in networks to
ing that the anti-inammatory effects of IL-10 provide a more complete framework upon
are promoted through the direct inhibition of which we can base predictive models, while
lipopolysaccharide-induced miR155 expres- at the same time pruning the large parts lists
sion (158). These observations emphasize the generated by global methods of components
importance of developing quantitative models unlinked to a direct test of functional relevance.
to predict accurately the inuence of miRNA The discovery of RNA interference (RNAi)
expression changes on cellular function. and major advances in the understanding of
small RNA biology in the past decade have pro-
vided researchers with an invaluable tool for
RNAi Screening wide-scale and rapid genetic screening that rep-
Although microarrays and NGS report the resents a less biased means of probing the role
genes (or miRNAs) present in a cell or tissue, of various elements in cellular biology (161). As
the state of chromatin, or the binding behavior a research tool, RNAi takes advantage of en-
of transcription factors and can inform us about dogenous RNA processing machinery, which
how these elements change over time, these permits the silencing of mRNA transcripts with
methods are descriptive and correlative, failing small complementary dsRNA sequences. In
to provide a direct link between transcript, Caenorhabditis elegans and Drosophila, this can be
translated product, and biologic function. achieved through the introduction of relatively
Establishing such connections traditionally has long pieces of gene-specic dsRNA (>100 nt);
been the realm of genetic screens in mutant however, as this would induce an interferon re-
cell lines or organisms. Such studies have con- sponse in mammalian cells, the gene-specic
tributed signicantly to the identication of dsRNA must not exceed 21 nt and thus is
544 Germain et al.

introduced as siRNA (162). siRNA transfec- RNA studies is the arrayed screen, in which
tion is the simplest and most direct applica- each well of a microtiter plate contains
tion of RNAi in mammalian cells; however, in siRNA(s) or shRNA(s) against a single gene.
less tractable cell types and in cases in which This permits a straightforward correlation of
longer-term silencing is desirable, the siRNA phenotype to target, but it has the drawback
sequence can be expressed from plasmid or viral that the time and expense required for a
vectors as a short hairpin (sh)RNA. This works genome-wide screen are signicant. An alter-
especially well when the shRNA is designed to native approach that can be used with virally
mimic an endogenous primary miRNA tran- expressed shRNAs is the pooled screen, in
script, with the region that is processed to the which large populations of cells are transduced
mature miRNA replaced with a gene-specic with complex mixtures of viruses expressing
targeting sequence (163). shRNAs against many genes. Clonally isolated
Soon after the technical application of RNAi cells showing the desired phenotype then can
was established for the study of individual be sequenced to determine which shRNA they
genes, its discovery potential on a broader scale express. If the output assay can be designed to
became evident, leading to the development of select for a cell growth or survival phenotype,
genome-scale libraries of reagents for several a more efcient means to identify hits is to
organisms, including human and mouse (161, use unique nucleic acid bar codes within
164). Although the potential of these broad each shRNA vector that can be screened for
screening tools has attracted many researchers, enrichment or depletion by either microarray
the practicalities of genome-wide screening are or sequencing technology. Such pooled screens
challenging and require the acquisition of new are powerful, especially for applications in
techniques and capabilities for most biologists, human primary cells of limited availability,
as well as substantial monetary commitments. but it remains difcult to ensure full genome
Furthermore, despite the discovery potential coverage. These pooled shRNA methods have
of RNAi, the technology has its limitations, been applied productively to the analysis of the
and certain practical criteria must be met for signaling pathways involved in the growth and
a gene to be detected in an RNAi screen: viability of human B cell lymphomas, as one
(a) The expression of the target gene must be example (165167).
reduced by the siRNA to a level that promotes The screening format also is in part
an observable phenotype. (b) The protein must dependent on the cell type chosen, as easy-to-
have a short enough half-life to permit de- transfect cells are more amenable to arrayed
pletion in the time course of the experiment. siRNA screens, whereas less tractable cell
(c) The biological function must not be sup- lines and primary cells may require infection
ported by multiple redundant factors. (d ) The with shRNA-expressing viruses. Achieving
depletion of the target gene must not be toxic satisfactory levels of gene knockdown remains
to the cells under study. Because of the growing a major technical hurdle in assay development
importance of this technique in systems biology and often forces researchers to choose a
research and because it has rather strict qual- nonphysiological cell type for the process they
ity control requirements and interpretive limi- are studying, raising the concern that the phys-
tations, even more so than other methods, we iological relevance of hits identied in a highly
devote some space here to a detailed considera- transfectable cell line may be questionable. In
tion of the different types of siRNA screens and some cases, optimal transfection conditions can
the technical aspects of the effective use of the be identied in more physiological cell lines
methodology. if an efcient and relatively high-throughput
knockdown assay [such as the targeting of
Screening formats and assay design. The a stably expressed reporter gene (168)] can
most commonly used approach in large-scale be developed that permits an investigator to

assess rapidly and conveniently a wide range of and published reports of screens with multiple-
transfection parameters and reagents. assay readouts until recently have been limited
The design and validation of a robust and to subgenome-scale efforts. Three recent re-
specic assay are the most critical, and often ports have combined a mammalian genome-
the most time-consuming, aspects of an RNAi wide siRNA screen with multiparametric im-
screen. This work can be extremely laborious, age analysis. Zerial and colleagues (169) studied
requiring attention to detail and continual assay the endocytosis and trafcking of either a recy-
repetition with minor alterations to optimize cling (transferrin) or late endosome/lysosome
performance; however, careful development of cargo (EGF receptor) in HeLa cells costained
a robust assay pays off in the quality and discov- with nuclear and cytosolic markers. They iden-
ery potential of the screening results. There are tied hits enriched in specic pathways such
several considerations in choosing and design- as Wnt, TGF-, and Notch, and a Bayesian
ing a screening assay, such as whether the read- network analysis led them to predict that nu-
out should be end point or time resolved, single merous kinetic and spatial aspects of endocy-
feature or multiplex, and whether data should tosis are highly regulated. Ellenberg and col-
be collected from populations or single cells. In leagues (170) used a live-cell assay to prole
each case, the latter option clearly would pro- cell-division parameters in HeLa cells express-
duce a richer data set; however, the value of ing GFP-labeled histone 2B. They used an
the type of data collected often is inversely pro- siRNA microarray approach that involves plat-
portional to the ease of assay design and im- ing cells in chambers with up to 384 siRNA,
plementation. Thus a compromise has to be which has the advantage of standardizing cell-
reached, taking into consideration the overall culture conditions across the array (171). Al-
logistics of running the screen, the cost, and the though they collected data only from a single
researchers expectations from the data. Most uorescent channel, they were able to classify
of the initially published genome-wide screens up to 16 morphological phenotypic character-
used relatively simple assays, such as cell lethal- istics based on the shape of the labeled nu-
ity, and to a certain degree the goal primar- clei, and their data analysis identied numerous
ily was to establish a proof of principle for the genes involved in cell division and migration.
RNAi technology. Screens using more complex Boutros and colleagues (172) assessed the effect
assays and readouts are now being published, of siRNA transfection in HeLa cells costained
but there remain relatively few examples in the for DNA and the cytoskeletal proteins actin
literature of screens using live cells, multiplexed and tubulin. Clustering of gene perturbations
readouts, or single-cell-based data collection. by phenotypic similarity allowed the identica-
Continued improvements in uorescent re- tion of novel players in the DNA damage re-
porter technology and high-content imaging sponse. The use of multireadout, high-content
platforms provide the most versatile experi- screening assays raises additional challenges in
mental platforms for data-rich screening as- the development of assay controls, which are
says. High-content assays involve the collection vital to screening success. In addition to low-
of multiple functional parameters of cell state, noise negative controls, it is desirable to identify
such as cell shape, the location of structural multiple positive control siRNAs that provide a
markers and organelles, signaling response graded perturbation of the selected assay read-
readouts, and/or gene expression changes, of- out. If only strong positive controls are selected,
ten through the use of multiple spectrally dis- the evaluation of screen quality is biased, and
tinct uorescent reporters. This opens up the partial phenotypes of biological interest likely
possibility of tracking multiple readouts in a bi- will be missed. Of course in a multiplex assay
ological process in single cells over an extended with data collected over time, the comparative
time course. This approach, although clearly strength of controls for different readouts will
powerful, poses numerous technical challenges, vary, which in itself can provide new biological
546 Germain et al.

insight prior to the screen. Ultimately, the more additional unfamiliar challenges for most
comprehensive the group of positive controls academic researchers. These screens benet
identied during development of the screening greatly from experimental automation and
assay is, the more valuable the control data will robotic workows, which up until recently
be during data analysis. had been restricted primarily to groups in
A popular statistical metric used to evalu- the pharmaceutical and biotechnology in-
ate a screening assay is the Z factor. Originally dustries. A signicant resource for RNAi
developed for chemical screening, it is essen- screeners has been the recent establishment of
tially a measure of how reliably the negative small-molecule screening infrastructure in the
and positive controls can be distinguished in the academic and government sectors (174), such
data set.3 High Z factors of 0.51.0 are consid- that at many research sites, the expansion of
ered good, whereas low Z factors of less than these facilities into RNAi screening has been
0 are indicative of a poor assay. Although 0.5 a natural progression. Prior high-throughput
is considered a minimum for a good chemical standard operating procedures for screening
screening assay, recent data suggest that this is thus have established already the importance of
overly stringent for RNAi screens, which are detailed scheduling and tracking of day-to-day
inherently more noisy, and Z factors of more workow to permit the correlation of data
than 0 are considered acceptable (173). The irregularities with procedural variations in
comparatively higher noise in RNAi screens screen execution. The importance of recording
can be attributed to several factors: poten- key screen parameters to allow the comparison
tial off-target effects (discussed below); longer of RNAi screening data sets between different
(4872 h) siRNA incubation times required to laboratories also is emphasized by an ongo-
achieve target knockdown; cell stress from the ing effort to establish community standards
transfection procedure; and, in contrast to a for the minimal information that should be
chemical library, the existence of a cellular tar- recorded from an RNAi screening experiment
get, and thus a potential impact on cellular be- (http://miare.sourceforge.net/HomePage).
havior, for every reagent in the RNAi library. Although the use of laboratory automation
Once an assay has been established with sat- in RNAi screening has drawn heavily from
isfactory statistical performance criteria, pilot established expertise and technology in the
screens with a few hundred genes are important small-molecule screening arena, the process
to determine if the assay performance from the of the conrmation of RNAi screening hits
validation step is replicated in the screen. Be- and data analysis has posed some unique
cause this often is not the case, this step provides challenges. As mentioned above, RNAi screens
a further opportunity to optimize the process have a degree of inherent noise, as their nucleic
prior to the signicant investment of time and acidbase pairing-dependent mechanism al-
resources required for a genome-wide screen. ways will suffer from a degree of nonspecicity.
So-called off-target effects can occur because of
Whole-genome considerations. One reason the hybridization of an siRNA to an unintended
many published screens to date have used target, especially within the seed sequence, the
relatively simple assays is that the scale of rst 8 nt from the 5 end of the siRNA leading
genome-wide screening poses numerous strand (175). Although this has been addressed
by improvements in sequence-selection al-
gorithms to increase siRNA specicity and
3
The Z factor is calculated from the following equation: chemical modications of the siRNA duplex
Z = 1 (3SD+ + 3SD )/|+ |,
to limit interference by the siRNA passenger
strand, it will remain a factor to consider (176).
where SD+ is the positive control standard deviation, SD
is the negative control standard deviation, + is the positive For practical reasons, primary genome-wide ar-
control mean, and is the negative control mean. rayed screens often use a pool of several siRNA

reagents in the same well. To address the issue Importantly, the statistical methods used so
of whether hits in a primary screen result from far for hit identication in RNAi screening data
an off-target effect from one of the siRNAs in have been tested primarily using single-readout
the pool, investigators commonly rescreen the assays. The development of more complex
primary hits with the individual siRNAs from assays, such as those used in high-content
each scoring pool. Cases in which only one screens, will require the parallel development of
of the siRNAs from the pool reproduces the appropriate statistical methods to evaluate the
phenotype are considered likely off-target hits data and weight hits according to their effects
and are dropped from further analysis. Another on multiple readouts.
similar approach is to rescreen hits with alterna- A key question for many screeners from
tive siRNAs from another source. Incremental their primary screen data is what constitutes
improvements in the efcacy and specicity of a biologically meaningful hit. The setting of a
genome-wide siRNA libraries in time should denitive threshold for classication of genes
provide reagents that reliably diminish all genes as hits can be inadvisable, as genes that score as
in mammalian cells, with enough reagents avail- a positive in an initial primary screen may drop
able per gene to permit effective ltering of off- below the threshold in a replicate screen. A
target phenotypes during the analysis process. better approach may be the relative weighting
of all genes phenotypic contribution such as in
a gene-versus-phenotype relationship, which
Data analysis and hit identification. The is less likely to miss weak but biologically
data-processing and statistical approaches cur- meaningful hits (159). For practical purposes,
rently favored for analyzing RNAi screening a subset of genes may have to be chosen for
data have been reviewed recently (173), so we follow-up through secondary screens, but the
do not cover them in depth here. Briey, the importance of the retention of all the data in
analysis process follows a series of keys steps: a queryable database will be vital to future
(a) data triage and preprocessing, carried out integrative analysis. This type of quantitative
during and/or immediately after completion of data can be lost in screens that assign qualitative
the screen to identify any unexpected patterns scoring of genes as yes/no. An alternative ap-
in the data that might indicate experimental proach to avoid overly stringent ltering of data
or procedural issues; (b) normalization, usually is to assign hits into multiple categories with
carried out within each plate to remove system- varying levels of condence assigned (161).
atic errors caused by day-to-day experimental
drift over the period of time required to com- Secondary screening and hit validation.
plete the screen; (c) calculation of quality met- The validation of gene hits from a primary
rics, such as Z or Z factor, to evaluate assay RNAi screen is vital to provide a measure of
reproducibility and consistency of controls; and condence in the data prior to attempting a
(d ) hit identication. The most popular method predictive bioinformatic analysis of the data
of hit identication is mean k standard devi- set. The calculation of false-positive and false-
ation, which sets a threshold of deviation from negative rates is a popular validation method
the normalized mean beyond which a gene is in high-throughput screening; however, this
considered a hit. This is satisfactory for nor- approach remains difcult in RNAi studies
mally distributed data, but it performs poorly in because of the lack of reliable data relating
assays prone to outliers, in which methods such the quantitative effect of each siRNA on its
as median k median absolute deviation are target (and potential off-targets) to the ob-
more appropriate. Many of these data-analysis served phenotype. It has been common to use
steps can be carried out within freely avail- relatively subjective criteria to generate smaller
able software packages appropriate for RNAi lists of genes for secondary screens, although
screens (177179). some recent studies have demonstrated that
548 Germain et al.

a well-designed primary screen with alterna- build networks that include spurious compo-
tive selection criteria can generate a higher nents, whereas too stringent a set of criteria will
condence list of primary hits (180). As dis- lead to many false negatives, leaving gaps in the
cussed above, rescreening with deconvoluted network construction as compared to biological
siRNA pools and alternate sources of silencing reality.
reagents has been shown to be an excellent
rst step in the hit-validation process. Another Immunology applications. In the immunol-
method that is impractical at the genome-wide ogy and infectious disease arenas, the primary
scale but is possible in secondary screening is to application of genome-wide RNAi to date has
measure the extent of transcript depletion (or, been in the screening for host factors involved
even better, protein expression loss) with each in viral or bacterial infection. Many early stud-
siRNA and then determine the correlation of ies of this type were carried out using Drosophila
knockdown with phenotype. Although expen- cells as a model for infection due to the tech-
sive, transcript/protein proling with multiple nical simplicity of target gene knockdown in
siRNAs per gene also can help validate hits this organism. Because insect cells do not have
or identify unexpected altered gene clusters an interferon response upon challenge with
that may contribute to off-target phenotypes. dsRNA, a long sequence of dsRNA comple-
Rescue of a phenotype by ectopic expression mentary to the target gene can be introduced
of an RNAi-resistant version of a target gene into cells, and processing of this nucleic acid
is considered the ultimate validation strategy; by the RNAi pathway produces a complex mix-
however, this remains difcult to implement in ture of siRNAs that leads to robust target gene
a practical manner for a large number of genes depletion. Drosophila siRNA screens have iden-
(161). tied several signaling components involved in
The use of a secondary assay is a valuable the y systemic response to infection through
means not only to validate hits, but to identify the Toll, IMD, and Jak/STAT pathways, and
different degrees of dependency of the primary screens with live bacteria and viruses have iden-
and secondary assay readouts on functionally tied specic host factors that, in some cases,
diverse genes, which can lead to new biological are conserved in mammalian cells (reviewed
insight as well as inform the subsequent classi- in 181). However, there is a growing consen-
cation of hits. Secondary assays with a small sus that the integrated cellular responses to
number of genes also provide the opportunity pathogens vary signicantly between y and
to introduce a more technically laborious, but human and that screens in mammalian cells are
perhaps more data-rich and potentially insight- required to obtain data more relevant to human
ful, assay. Alternatively, a secondary assay might disease.
be designed specically to remove nonspecic Genome-wide siRNA screens have been
hits from the gene set, such as genes involved carried out recently in human cells to assess
in the general transcription/translation process the response to several viruses and bacterial
identied in a screen using an expressed re- pathogens. Table 2 summarizes the screens
porter as a readout. The need for a secondary that have been published to date. The potential
assay to validate primary screen hits also em- of these studies to identify candidate drug
phasizes the value of more complex multiread- targets for disease treatment or vaccine thera-
out primary assays, as they essentially provide a pies has been questioned on the grounds that
secondary data set across the entire genome. multiple studies analyzing the same pathogen
It should be obvious that there is a critical have produced largely nonoverlapping gene
trade-off in establishing the cutoffs in primary lists. However, considering that there was no
and secondary screens for hitstoo lax a set signicant effort to control for experimental
of standards will lead to many false positives differences between laboratories in these
in the resulting data set, leading to attempts to studies, there are numerous reasons that might

Table 2 Details of siRNA screens for host factors influencing pathogen infection in mammalian cells
siRNA
IY29CH19-Germain
Pathogen reagent Genes Primary screen Total

screened Strain Cell line source targeted assay Scoring method for hits hits Reference
ARI
Inuenza virus PR8 (H1N1); Udorn HBEC Dharmacon 1,745 Infection of 293T Z-score; twofold cutoff 616 225
550
(H3N2) with HBEC viral
supernatant
Inuenza virus PR8 (H1N1) U2OS Dharmacon 17,877 HA detection on 45% reduction in infected 133 316
infected cell cells
7 February 2011
surface
Germain et al.
Inuenza virus Recombinant WSN A549 Qiagen, 19,628 HA replaced with Z-score; twofold cutoff by 295 317
(H1N1) Invitrogen, luciferase at least two independent
22:24
IDT siRNAs/gene
Inuenza virus WSN (H1N1) A549 Qiagen 22,843 NP staining Z-score; twofold cutoff 287 318
HIV IIIB strain of HIV-1 HeLa Dharmacon 21,121 p24 staining Z-score; twofold cutoff 273 319
transgenic
HIV Recombinant HIV-1 HEK 293T Not stated 20,000 Inhibition of Statistically signicant 213 320
luciferase signal reduction by at
expressed by least two independent
recombinant virus siRNAs/gene
HIV HXB2 HIV-1 HeLa Rosetta/ 19,709 HIV LTR-driven 2 SSMD 390 321
transgenic Merck b-gal reporter
West Nile WNV strain 2471 HeLa Dharmacon 21,121 Viral envelope Twofold infection 305 322
virus protein staining reduction
Mycobacterium H37Rv THP1 Dharmacon 18,174 Altered bacterial Z-score; twofold cutoff 275 323
tuberculosis CFU values 21
days postinfection
Mycobacterium H37Rv J774.1 Qiagen 1,032 Altered bacterial Z-score; twofold cutoff 41 324
tuberculosis CFU values 21
days postinfection
Hepatitis C 1b N strain/ Huh-7 Dharmacon 4,000 NS3-NS5B alkaline >60% reduction in viral 9 325
virus NS3+NS5B phosphatase replication
subgenomic replicon reporter
Hepatitis C 1b N strain/ Huh-7 Dharmacon 21,094 NS3-NS5B Storeys false discovery 96 326
virus NS3+NS5B luciferase reporter rate threshold of 0.10
subgenomic replicon
explain noncorrelation: (a) different gene sensi- pairwise overlap was observed between the
tivities of cellular pathways to perturbation due Konig and Karlas screens, which used the same
to variable expression level of genes across the A549 epithelial cell line and variants of the
range of cell types used, (b) variability in siRNA WSN/H1N1 virus, supporting the likelihood
efcacy between different library sources, and that differences in experimental approaches
(c) differences in the specic strains of virus have a signicant inuence on hit correlation
used and additional variations in screening between independent studies. A combination
assay design and experimental protocols. of the screen hits with interaction data between
In an effort to address the disparities in gene u and host proteins was used to generate a
hit lists from the multiple screens that have host-virus network that showed enrichment for
been carried out for HIV infection, Chanda and primary cellular processes required for the viral
colleagues (182) recently carried out a meta- life cycle: endocytosis, nuclear transport, tran-
analysis of the HIV siRNA data sets, incor- scription/translation, and viral assembly and
porating HIV-host protein-protein interaction budding. However, many genes highlighted by

(PPI) data and SNP contribution to HIV sus- this type of analysis are likely to be poor drug
ceptibility. This analysis initially compared all targets as they are necessary for fundamental
1,254 human genes identied in the genome- cellular processes. The selection of genes linked
wide surveys to the 1,434 genes in the NCBI to cellular processes known to be important for
HIV interaction database and found that the the inuenza viral life cycle may be less fruitful
overlap of 257 genes had a high degree of statis- than a more nonbiased approach to the screen-
tical signicance. Network enrichment analysis ing data. For example, greater insight might
of the functional categories over-represented be gained by combining the unltered siRNA
in this gene set highlighted several gene on- data sets with genetic variation data to look for
tology groups as being most important for correlation between screen hits and polymor-
HIV entry and replication, including nuclear phisms linked to infection susceptibility.
pore/transport, GTP-binding proteins, protein This raises an important issue with regard
complex assembly, intracellular transport, pro- to the value of siRNA screening at the systems
teasome function, and RNA splicing. A closer level. Although individual screens in isolation
analysis of the three siRNA screens found that may be limited in their ability to identify spe-
of the 34 genes identied as HIV host factors in cic targets for drug or vaccine development,
at least two screens, 29 were expressed in CD4+ the genome-wide data sets may prove to have
cells and/or tissue and thus would have the op- greater value in properly characterizing the
portunity to inuence the host response to virus broader cellular response and identifying the
in these key HIV target cells. Of the eight genes host and pathogen factors inuencing survival
in this set considered druggable, two proteins, versus pathology in real infection scenarios,
CXCR4 and Akt1, already are targeted in es- particularly when combined with metadata
tablished HIV treatment regimens, supporting from the additional approaches we describe
the drug-discovery potential of these screening below. The value of entire screening data
approaches. sets for use in subsequent integrative analyses
Similarly, a recent review of the data from emphasizes the need for published siRNA
multiple screens that have been carried out screens to adhere to standards on data-set
for inuenza host factors revealed a greater content that have community consensus such
overlap of hits when viewed in terms of gene as those described in the MIARE guidelines
ontology categories than that seen when (http://miare.sourceforge.net/HomePage).
comparing individual genes (183). Of the 1,449 In addition to these studies focused on
genes identied in the published inuenza pathogenic host factors, siRNA screens also
screens, 128 were identied by at least two have contributed signicantly to the iden-
screens. Interestingly, the highest degree of tication of key components involved in

immune cell biology. RNAi screening led to developing useful computational models of
the discovery of several molecules involved molecular pathways and networks.
in the process of store-operated Ca2+ inux Most of our insight into cellular processes
through plasma membrane CRAC channels is based on identifying and characterizing the
and the subsequent activation of the transcrip- interactions between cellular proteins and
tion factor NFAT, which are fundamental to other biomolecules, and discrete modeling of
lymphocyte activation and function. Examples increasingly complex biological phenomena
include the CRAC pore-forming protein Orai will require a detailed understanding not
(184186), the ER Ca2+ depletion sensor Stim only of the potential binding partners of each
(187, 188), and NFAT regulators such as molecule, but also of the context-dependency
NRON and the Dyrk family kinases (189, 190). of the interaction(s) and their kinetic char-
RNAi has demonstrated a specic role for acteristics. Although interactions among
Akt1 in the regulation of intracellular bacterial molecules of diverse chemical composition
growth (191); a screen for T cell inhibitory are important (e.g., proteins with lipids or
molecules in Hodgkins lymphoma identied carbohydrates), at present most efforts are
the immunoregulatory glycan-binding protein devoted to producing initially a comprehensive
Galectin-1 as a key factor in promoting im- view of the protein interactome. There are
mune privilege for neoplastic Reed-Sternberg three principal methods in use to catalog
cells (192); and screens using virally expressed PPIs. Yeast two-hybrid (Y2H) remains the
shRNAs in less tractable immune cell types most practical high-throughput method for
have identied an unexpected reliance of mul- the generation of interactome data; afnity
tiple myeloma cells on the transcription factor purication of protein complexes followed by
IRF4 (193), as well as provided signicant mass spectrometry (AP/MS) is the preferred
insight into chronic antigen receptor signaling approach for the characterization of multipro-
and the aberrant activation of NFB in diffuse tein complexes; and protein complementation
large B cell lymphoma (165167). Further assays (PCAs) are emerging as a means to
recent discoveries using RNAi include the measure interactions in a better-preserved cel-
demonstration of key roles for the kinase RIP3 lular context, with important applications for
in programmed necrosis following viral inam- assessing real-time PPI dynamics (for reviews,
mation and AIM2 in sensing of cytoplasmic see 196198). Although none of these methods
DNA (194, 195). These examples highlight stands alone as an optimal approach, together
the potential of RNAi screening technology they have the potential to generate a compre-
both in the discovery of individual proteins hensive knowledge base in the emerging eld
with essential functions in immune responses of interactomics. In time, they arguably will
and in the broader characterization of the generate a higher condence (and certainly
cellular and organismal response to pathogenic a more complete) data set than literature
challenge. curation, which has its own signicant draw-
backs, such as subjective interpretation of data
quality, narrow focus on canonical pathway
Interactomics components, experimental variability between
If microarray or NGS technologies provide laboratories, and nonstandard nomenclature
us with descriptive lists, and RNAi screen- across the published literature. Although most
ing identies gene products with functional scientists tend to attribute higher intrinsic
relevance, we still need information on the quality to the low-throughput experimental
connectivity among the identied elements data that populate literature-curated databases,
to develop a more complete picture of how recent analyses suggest that the overlap
these components work together to sup- between the most widely used databases is
port biological function as we aim toward surprisingly low, and remarkably >85% of
552 Germain et al.

the interactions are supported by only a single The most sensitive screening approach is to
publication (199). A noteworthy advantage of mate each BD strain against a matrix of all avail-
literature curation is that biological function able AD strains to ensure that all pairwise com-
often can be inferred directly from the curated binations are tested. However, this becomes
study; however, this study bias also can hamper practically challenging in large screens in which
novel biological insight. Data-driven high- some form of AD strain pooling strategy often
throughput PPI approaches, although lacking is used, followed by clone selection through re-
a direct link to a biological question, usually porter activation. However, the sensitivity of a
avoid study bias; their systematic methodology screen involving a large pooled library can be
can reduce interexperiment variability; and limited by sampling sensitivity, so a compro-
they permit an estimation of interactome mise approach has become popular in which
coverage as negative data also can be collected. small pools are screened in a matrix and positive
Thus, although the total number of expressed pools are rescreened with individual BD/AD
genes among Drosophila, C. elegans, and human pairs (198).

is not signicantly different, the respective Early versions of the assay were prone
interactomes have been estimated at 65,000, to identifying false-positive interactions, of-
200,000, and 650,000 PPIs, respectively (200), ten from auto-activating BD fusions. However,
and may therefore provide a more meaningful the technique has been improved considerably
barometer for organismal complexity. by careful control of fusion expression levels
and the use of multiple reporters to increase
Yeast two-hybrid screens. The Y2H metho- the biological validity of the identied inter-
dology was rst developed in 1989 (201). It actions. The development of recombination-
involves the creation of yeast strains express- based GATEWAY-type cloning technology
ing bait proteins fused with the DNA bind- also had a signicant impact on the transition
ing domain (BD) of a split transcription factor to genome-scale Y2H screening, as this permit-
(such as Gal4) and the mating of these yeast ted more efcient parallelization of open read-
with strains expressing prey proteins tagged ing frame (ORF) cloning to generate genome-
with the transcription factors activation do- scale BD and AD strain collections. Building on
main (AD). The interaction of a bait-prey two- these technical advances, genome-wide screens
hybrid then is detected through the expression have been carried out for yeast (203, 204), Plas-
of a reporter gene driven by a promoter respon- modium falciparum (205), C. elegans (206), and
sive to the chosen reconstituted transcription Drosophila (207), and more recently, approx-
factor. The original lacZ-based colorometric imately one-third of the human genome has
identication of interactions remains a popu- been screened in this manner (208, 209).
lar reporter in Y2H assays, but for large-scale Despite this enormous experimental effort,
screening, protocols have evolved to incorpo- the limited degree of overlap observed in large-
rate some method of positive selection such that scale Y2H data sets involving proteins from the
only cells hosting an interaction can survive same organism remains a concern. Although
in selective media (202). Examples of positive some have argued that this points to a high
selection reporters include HIS3 and URA3, false-positive rate, recent analysis suggests it
which permit growth in histidine- or uracil- is more likely because of poor sampling sensi-
lacking media, respectively. The use of positive tivity (210). The improvement of Y2H screen-
selection reporters also can be combined with ing efciency is thus a continuing challenge but
competitive inhibitors and additional reagents may be attained with developments in accessory
that allow screeners to titrate for interac- techniques, such as NGS for improved clone
tion strength or convert the reporter to neg- sequencing throughput. The ltering of false
ative selection to identify inhibitors of a given positives is another important aspect of Y2H
interaction. data validation. Technical validation of whether

an identied interaction can occur can be ad- be expressed heterologously to introduce the
dressed with orthogonal technologies, such as afnity tag, and it is difcult to ensure that
afnity purication of the two proteins in a the expression level of the tagged protein is
complex from cells. However, whether an in- comparable to the endogenous untagged form.
teraction actually does occur with a functional In yeast, this can be addressed by using homol-
role in the organism is more difcult to ad- ogous recombination at the native locus, and
dress. Bioinformatic methods have been devel- accordingly, genome-wide yeast strains have
oped that use known attributes of the proteins been generated expressing an afnity-tagged
to derive a statistical condence score, which version of every gene for AP/MS studies (211,
has been shown to correlate well with biologi- 212). A similar resource for mammalian cells is
cal relevance in Drosophila studies (207). in the distant future, but a technique has been
In addition to these data-validation chal- developed using BAC transgenomics, which
lenges, several inherent limitations of the permits the tagging of mammalian genes at
methodology require that Y2H data sets be their native locus within a bacterial articial
complemented by additional approaches. Y2H chromosome construct, which then is trans-
interactions must take place in the yeast nu- duced stably into a cell line, allowing the tagged
cleus and thus often lack the appropriate spatial gene to be expressed from its native promoter
subcellular context; only binary interactions are (213). The generation of BAC libraries tagging
detected (missing those that require accessory all human and mouse genes in this manner
proteins in a larger complex); and the technique would provide an invaluable resource for sys-
rarely detects the numerous interactions that tematic interaction analysis, and the potential
depend on PTM of one or more components. of this approach is emphasized in a recent study
of mitotic interaction complexes in human cells
Affinity purification/mass spectrome- (214).
try. AP/MS technology can address some AP/MS promises to generate vital in-
deciencies of Y2H as it involves the im- teractome data, especially for proteins that
munoprecipitation of a protein complex in a participate in numerous heterogeneous and
seminative state followed by the MS-based often multiplex complexes. Coupled with
identication of the complex constituents (see compartment-specic cell-fractionation ap-
below). This usually is achieved by expressing proaches, this could provide insight into how
the bait protein of interest with an afnity tag promiscuous proteins are shared in the con-
that can be recognized by a specic antibody to text of the whole-cell PPI network. However,
permit the isolation of the bait in complex with because of the technical challenges in afnity
associated proteins. Initial studies with simple tagging large numbers of baits, AP/MS-based
afnity tags were prone to false negatives due studies of mammalian interactomes have not
to the stringent wash conditions required for reached the scale of Y2H studies. Some excel-
effective bait purication. The development lent focused studies have been carried out on
of tandem afnity purication (TAP) tags has specic modules known to have key roles in
addressed this by using a protease cleavable normal and disease states, such as the TGF-
tag sequence with two afnity tags (such as (215) and TNF/NFB (216) subnetworks,
protein A and a calmodulin-binding peptide). and these studies have highlighted the abil-
Two rounds of afnity purication with ity of afnity-purication strategies to identify
different tags signicantly reduce nonspecic the dynamic interactions involved in signal re-
binding and permit the use of less stringent lay through such subnetworks. A broader study
purication conditions in each step, which has of over 400 disease-relevant human proteins
been shown to be more sensitive in identifying expressed heterologously from a CMV pro-
lower-afnity interactions (196). One technical moter in HEK293 cells identied over 6,000
drawback is that the bait protein often must interactions, less than might be expected if the
554 Germain et al.

predicted 650,000 total interactome is accurate Immunology applications. In addition to

(217). However, the initial interaction data set the characterization of cellular networks
in this study was >24,000, which may suggest among host proteins, PPI techniques have key
overly stringent ltering of putative false pos- applications in the infectious disease eld in de-
itives during the informatic analysis. This em- termining the mechanisms of immune response
phasizes the importance of evaluating the pro- evasion through the interaction of pathogen-
cess of data triage, although this undoubtedly derived proteins with the host proteome. Our
will improve as the number of available high- knowledge of immune cell signaling, partic-
quality data sets increases. ularly in cells of the innate immune system,
has been informed to a considerable extent by
Protein complementation assays. PCAs the mechanisms used by pathogens, especially
provide another approach to the high- those with type III secretion systems, to disrupt
throughput generation of PPI data. This is the key signaling pathways and establish infection.
most physiological of the described technolo- Because pathogens often express relatively
gies as interactions can be detected in the cell few proteins for direct interaction with the
that normally expresses the protein of interest, host proteome, global screens to determine
with PTM intact and without the problems as- host protein binding partners for pathogenic
sociated with the afnity demands of the TAP proteins have been and will continue to be
method. However, like Y2H, the approach is highly informative in efforts to develop models
best suited to the detection of directly interact- of host-pathogen interactions. As viruses are
ing proteins, whereas TAP can identify molec- dependent on interactions within host cells for
ular machines composed of multiple elements their survival and pathogenicity, elucidation
without the need for tagged elements to interact of virus-host interactomes as well as intraviral
directly with each of these other components. interactions are of particular interest. Again,
PCAs use a similar bait-prey premise as Y2H because of its amenability to high-throughput
(hence the alternative mammalian two-hybrid studies, Y2H has been employed much more
nomenclature), but the tags expressed on the than the other techniques described above.
proteins represent two domains of an enzymatic Intraviral Y2H screens have been carried out
or uorescent reporter, which provides a sig- for numerous viruses, especially herpesviruses
nal only when a detected PPI brings the do- (VZV, KSHV, HSV-1, mCMV, and EBV) due
mains into sufciently close proximity. Exam- to their larger genomes providing more ORFs
ples of reporters include DHFR, -lactamase, (70 to 170) for screening (219222). Analysis
and GFP (197). A recent analysis of interac- of the intraviral network in this family has sug-
tions between >5K yeast ORFs by this method gested a high degree of conservation of core in-
suggested that the condence scores within the teractions that support the viral life cycle, such
interaction data were improved over the other as the involvement of the HSV-1 UL33, UL31,
two technologies, with better enrichment of and UL34 orthologs across all herpesviruses
compartment-specic interactions that might (regardless of sequence conservation) in the
be inaccessible to Y2H and AP/MS (218). capsid envelopment and nuclear egress (223).
With an ever-improving availability of cu- Large-scale viral-host screens against
rated ORFs for human and (to a lesser extent) human proteins thus far have been limited to
mouse genes, and the use of recombination hepatitis C virus (HCV) (224), EBV (219), and
technologies that permit the straightforward inuenza (225). The HCV screen used 27 baits
shuttling of ORFs into expression platforms screened against two human cDNA libraries
for all the above approaches, there are fewer and identied 314 interactions. This was
technical limitations to the use of PPI tech- expanded with literature curation to a network
nologies to carry out genome-scale analyses in involving 11 HCV proteins with 420 human
mammalian cells. interactors. The NS3 and NS5A proteins of

HCV were found to be most highly connected, analysis of these data has emphasized the low
and the Jak/STAT and TGF- networks were coverage obtained with this technique. Future
the most signicantly enriched in the data set. efforts must not only focus on genome-wide
The EBV screen was more comprehensive, analysis, but also integrate a combination of
including 85 of the 89 known EBV proteins the approaches described to obtain a more
screened against 105 106 human AD strains comprehensive evaluation of host-pathogen in-
covering most of the human genome. This teractomes and facilitate their use in building
identied 173 high-condence interactions useful models of infection.
between 40 EBV proteins and 112 human
proteins. The inuenza screen tested 10 viral
proteins against approximately 12,000 human PROTEOMICS
ORFs and identied 135 pairwise interactions As emphasized in the preceding section on
involving 87 human proteins. In all these the interactome, system-wide analysis of the
virus-host screens, there was a signicant proteome has been progressing at a slower pace
enrichment for human proteins previously than that involving nucleic acidrelated aspects
shown to be highly connected in the human of biology, and fewer tools exist to study the
interactome. Targeting of such hub proteins by proteome than the genome. At the roots of
viruses could represent a conserved mechanism this discrepancy lie the natural properties of
used by viruses to hijack the cellular communi- nucleic acids and proteins. The physicochem-
cation machinery to promote viral replication. ical properties of nucleic acids permit their
Recent theories of scale-free cellular network recognition and replication by complementary
structure suggest that signal ow should be base pairing, which provides the means of
relatively robust to the perturbation of single expanding the amount of material in hand from
proteins, but selective targeting of multiple rare sources; DNA is also a relatively stable
highly connected nodes by pathogen infection molecule, and RNA often can be converted
could have a signicant impact. into DNA for experimental manipulation.
A bioinformatic analysis of all published In contrast, no widely useful method exists
host-pathogen interactions showed that of for amplifying proteins from cellular sources.
>10K reported interactions, >98% involve Furthermore, although the folded structure of
viruses, and >77% are from HIV studies (226). a protein is inherent in its primary sequence,
This emphasizes the lack of data addressing protein function depends more on tertiary
bacterial-host interactomes, which could be structure and especially on numerous PTMs,
particularly informative if focused on secreted such as phosphorylation, as compared with
effectors that promote virulence. Interaction nucleic acid products of the cell (although we
networks based on Y2H data were published are beginning to appreciate how RNA folding
recently for anthrax, Francisella, and Yersinia and not just sequence contribute greatly to the
and interestingly came to a similar conclusion performance of this class of biomolecules, and
as the viral screens with regard to preferential DNA methylation is certainly an important
interaction with hub proteins (227). However, postsynthetic modication). Because proteins
the depth of human genome coverage in the cannot be amplied in a way resembling PCR
screened library was not reported in this study, for nucleic acids, methods for measurements
so it is difcult to assess whether the hub pref- of proteins and their properties rely either
erence arises from an unbiased screening pool. on precise, sensitive physical measurements
These examples emphasize both the value (MS) or on signal amplication with antibodies
of the data obtainable from host-pathogen in- (ELISA-based assays, protein microarrays,
teraction screens and the current lack of com- ow cytometry, confocal imaging). Proteomic
prehensive data sets. Almost all the published and protein chemical studies are essential in
large-scale studies use Y2H, and comparative developing quantitative models because they
556 Germain et al.

report on the actual molecular constituents of (229), used for the ionization of solid samples
the cell, not just on the coding potential that bound to specic matrices.
is determined using transcriptomic methods, These ion sources can be coupled with
they provide information on the biochemical different mass analyzers. Mass analyzers
state of the proteins and not just their existence, sort the ions based on the m/z by applying
and they can be used to quantify the number electromagnetic elds, but the details of their
of protein molecules in a cellular compartment construction and applied physics can vary,
and hence their effective concentrations. They leading to differences in performance. The
also can ascertain the association/dissociation most commonly used mass analyzers for
rates for molecular pairs and higher-order protein analysis are quadrupoles, ion traps
complexes and provide information on catalytic [quadrupole/three-dimensional or linear trap
rates. These latter parameters are critical for quadrupole (LTQ)], time-of-ight (TOF), and
ne-grained modeling of such systems as Fourier-transform ion cyclotron resonance
signaling networks. Some of the same methods devices (an excellent summary of the properties
used for proteomic studies are also the basis of different types of mass spectrometers can be
for other large-scale studies of biomolecules found in 230). Protein identication usually is
in cells and uids, especially MS, which is a achieved by tandem MS (MS/MS), in which
primary tool for metabolomic and lipidomic the m/z of peptides derived from enzymatic
investigations. (usually tryptic) digestion of the proteins in
the source sample are determined in the MS
mode by the rst-stage spectrometer, and then
Mass Spectrometry peptides corresponding to particular m/z are
MS has long been a method of choice for sequenced after energetic fragmentation. The
the identication and characterization of sin- spectrum determination of the fragmented
gle proteins, either from pure preparations peptides is performed by an additional mass
or from SDS-PAGE (sodium dodecyl sulfate analyzer (the second MS in the name of this
polyacrylamide gel electrophoresis) bands or method). Therefore, many instruments used
spots. Any soluble protein in principle can in proteomics are hybrids combining different
be analyzed and unambiguously characterized types of mass analyzers (e.g., triple quadrupole,
based on peptide mass and sequence determi- quadrupole/TOF, or quadrupole/ion trap).
nation. The basis of MS is the measurement The instrument selection for an experiment
of the mass-to-charge ratio (m/z) of ions in depends strongly on the application, which
the gas phase. Therefore, a mass spectrome- for proteomic studies falls primarily into three
ter must contain an ion source for conversion groups: the determination of protein expression
of the sample into gas-phase ions, a mass an- levels (globally and in specic cellular compart-
alyzer for the separation of ions based on the ments), the identication and quantication of
m/z, and a detector measuring the number of PTMs, and the characterization of PPIs. Re-
ions in the sample (usually with amplication, cently, a new type of mass analyzer, called the
as the number of ions passing through the in- Orbitrap, has been developed (231) and imple-
strument is quite small). Initially, MS was de- mented in a hybrid instrument with two other
veloped for small molecules, and the methods mass analyzers, a linear ion trap and a C-trap
and instrumentation for small-molecule analy- (232). Its superior sensitivity, accuracy, and
sis are still in wide use. More recently, however, resolution (233), combined with MS/MS capa-
protein analysis using MS has been enhanced bility and robustness, make the LTQ-Orbitrap
markedly by the development of soft ioniza- ideal for high-sensitivity, high-throughput
tion techniques, such as electrospray ionization peptide analysis (234), and these properties
(228), used for the ionization of liquid samples, quickly have made it an instrument of choice
and matrix-assisted laser desorption ionization for many proteomic studies (235237).

The fragmentation method used for se- and accurate peptide sequence identication
quencing depends on the application. The (240).
most popular fragmentation method, collision- Typically, the ion spectra generated by any
induced dissociation (CID), is effective for proof these instruments from the peptide input
tein identication. The peptide ions in the gas are scanned periodically (the time spent on
phase are allowed to collide with molecules of one MS spectrum, or dwell time, is counted
a neutral gas (e.g., helium or argon). The col- in milliseconds). Of the entire set of peptides
lision causes bond breakage in the peptide and injected into the instrument at any moment,
causes the peptide ion to fragment into shorter only three to ve of the most abundant ions
fragment ions. By analyzing the m/z of the en- detected in this short period of scan time
tire set of fragments from a given peptide, one are chosen for fragmentation and therefore
can deduce the amino acid sequence due to the identied. To enable more distinct peptides
characteristic masses of individual amino acids, to be identied, one can exclude automatically
the loss of which, during fragmentation at dif- ions of high intensity present in several consec-
ferent positions, leads to a ladder of fragment utive spectra in subsequent scans and the next
masses based on the identity and position of most abundant ions sequenced. However, this
each amino acid in the peptide. Although some exclusion approach does not assure that all the
amino acids have the same mass, the derived se- peptides present as ion peaks in the MS spectra
quence can be matched to a protein sequence will be sequenced, and therefore many peptides
database, and many such ambiguities are represent in the sample remain unidentied.
solved by this informatics alignment process. This sampling issue is a major problem in MS,
This is not always the case, however, and hence especially in samples such as serum in which
multiple peptides from a single protein usually albumin and immunoglobulins are present at
are needed for the unambiguous identication orders-of-magnitude-greater concentrations
of the parent molecule. PTMs of the side chains than other elements of interest, such as cy-
of the amino acids also can complicate the se- tokines or shed surface proteins that might
quence calling, but the characteristic masses serve as biomarkers. Without depletion of the
of such modications can be used to identify predominant species or extensive fractionation,
these sites within a protein. Labile PTMs, such nearly all the major peptides in each spectra
as serine and threonine phosphorylation, re- will derive from the predominant species and
quire lower energy to dissociate them from the be resequenced, preventing the identication
peptide than that needed to break the peptide of the rarer entities. Even with efforts to enrich
bond. Therefore, when CID is used, the phos- for rare species, it is often the case that the
phate groups are removed, leaving the pep- repeat analysis of the same sample will yield
tide intact and unidentied. The peptides with different results for overall content, simply
PTMs can be analyzed effectively with elec- because of incomplete sampling in each run.
tron capture dissociation, using free electrons For protein phosphorylation measurements (a
(238) and especially electron transfer dissocia- situation in which the species of interest can be
tion, using radicals such as anthracene, to cleave rare indeed relative to the overall concentration
the peptide backbone, leaving the PTMs intact of the parent molecule or other components
(239). Higher-energy C-trap dissociation, us- in the sample), enrichment methods, such as
ing much higher voltages to create fragment precipitation with antiphosphotyrosine anti-
ions, has been developed for the LTQ-Orbitrap bodies or the use of phosphopeptide afnity
and resulted in the improved identication of to metals and metal oxides (immobilized metal
small (100200 Da) fragment ions. This frag- afnity chromatography and metal oxide afn-
mentation method is particularly effective com- ity chromatography), have been developed,
bined with CID in achieving both good relative permitting the isolation of phosphopeptides
quantication with chemical tags (see below) from more abundant nonphosphorylated
558 Germain et al.

peptides in the sample. Enrichment methods representing the different experimental condi-
of this type also are typically coupled with tions of interest are labeled with distinct mass
high-performance liquid chromatography for tags; once a peptide species from a given pro-
further separation of the peptides directly tein is identied, a comparison of the ion cur-
prior to injection into the mass spectrometer. rent peaks for the two isotopic versions from
Two-dimensional PAGE or chromatographic the two cell/animal states allows the relative
methods such as strong cation exchange can quantication of this species in the two states.
be used for additional sample prefractionation. To achieve the necessary isotopic tagging, one
These additional fractionation steps further can label the proteins metabolically, in vivo,
decrease the number of peptides entering the for example, by SILAC (stable-isotope labeling
mass spectrometer at a given time and increase with amino acids in culture) (242) or postextrac-
the probability that the less abundant peptides tion in vitro, for example, by iTRAQ (isobaric
will be detected, sequenced, and identied. tags for relative and absolute quantication)
In contrast to experiments in which ions for (243, 244).

identication are selected in a largely random SILAC relies on the incorporation of heavy
manner based on overall abundance, with the elements into proteins through the culture of
addition of the procedures outlined above to cells in a medium containing heavy or light
detect low-abundance species, instruments amino acids (usually arginine or lysine labeled
with single-reaction monitoring (SRM) and with stable heavy isotopes). After standard pro-
multiple-reaction monitoring (MRM) capacity cessing, the heavy and light labeled samples are
are used in quantitative experiments for repro- mixed and yield tryptic peptides, which display
ducible detection of dened (predetermined) mass differences of several m/z and can be dis-
sets of peptides. As noted above, only the tinguished in the MS mode, in which the m/z of
most intense ions from the MS spectra are whole peptides are displayed. The comparison
likely to be chosen for fragmentation in the of peak intensities gives the ratio of labeled to
MS/MS mode without user intervention. In unlabeled samples. The use of medium labels
SRM/MRM, the peptides from a list generated facilitates the relative quantication of three
by the user are sequenced preferentially even samples in one analysis. The caveats of SILAC
if more intense contaminants are present include incomplete labeling (multiple cell divi-
in the sample. For successful SRM/MRM, sions are required to replace amino acids with
the m/z of the peptides of interest as well their labeled analogs) and the limitation of rou-
as characteristic, noncomplementary, stable tine application to cultured cells [the labeling of
fragment ions (transitions) have to be chosen whole organisms, for example, mice, by feeding
carefully from an initial experiment conducted them labeled amino acid mix has been reported
in discovery mode. The dwell time of each MS but it is not in wide use, probably because of the
peak increases with the list of transitions, so the high cost and unknown incorporation through-
experimental conditions have to be designed out the body (245)].
carefully, but they ensure reproducible data sets Chemical postprocessing protein-
that can be compared among analyses (241). modication methods such as iTRAQ are
The mere identication of species using MS used widely to label samples from virtually any
is not enough to support the development of source. The iTRAQ reagent in its original
quantitative models for simulation; identica- form consists of four reactive mass labels of
tion methods help establish the parts list for varying isotopic composition, used to modify
model building, but one still needs quantita- primary amines in peptides. The tag molecules
tive parameters to be used during simulation if consist of the same amine reactive group, and
dynamic modeling is undertaken. In the most each of the four different isotopic labels is
commonly used relative comparative meth- balanced with a different linker so that the
ods (Table 3), proteins from cells or animals four tags are isobaric, adding the same mass

IY29CH19-Germain
ARI
560
7 February 2011
Table 3 Examples of quantitative mass spectrometry methods
Germain et al.
Label/standard
22:24
Method name Description Relative/absolute used? Limitations Reference(s)

SILAC Metabolic isotope labeling; Relative Yes Practically limited to cell 242
multiplexing up to three samples per culture
analysis
iTRAQ/TMT Chemical postprocessing labeling Relative Yes Sample handling prior to 243, 244
method; up to eight samples per labeling can introduce error
analysis
AQUA Use of synthetic heavy-labeled standard Absolute Yes Expensive, some peptides are 327
peptides difcult to synthesize
QConCAT Use of biologically expressed Absolute Yes Expression in E. coliPTMs 250
heavy-labeled standard peptides are different
Direct comparison Comparison of two or more separate Relative No; label-free Prone to error 328
analyses on the same platform
Spiking with standard Comparison of samples spiked with Absolute No; label-free Needs careful computational 329
known quantities of standard peptides approach
Targeted proteomics SRM approach with crude synthetic Absolute Yes Limited to several dozens of 247
combined with labeled peptide libraries peptides
peptide standards
Intensity combined with Data-mining computational approach Absolute No; label-free Currently not applicable to 248
spectral counting and SRM PTMs; needs adaptation to
platform used
Abbreviations: AQUA, absolute quantication; iTRAC, isobaric tags for relative and absolute quantication; PTM, post-translational modication; SILAC, stable-isotope labeling with amino
acids in culture; SRM, single-reaction monitoring; TMT, tandem mass tags.
to each peptide. Therefore, the peptides from on the careful examination of the protein se-
the four samples labeled with different tags quence and often is done empirically based on
and then mixed before fractionation elute from results obtained in nonquantitative MS/MS ex-
the chromatographic gradient together and periments that determine the set of detectable
are observed as one peak in the MS spectra. peptides generated from the protein of interest.
The quantication is performed in the MS/MS The synthetic peptide has exactly the same se-
mode, together with peptide identication, quence as the enzymatically generated peptide
when the coeluted tagged peptides are frag- but differs in molecular mass because of stable
mented. The fragmentation breaks bonds isotope incorporation. The retention time and
within the peptide as well as between the fragmentation of the synthetic peptide are de-
peptide and mass tag in a way that separates termined by liquid-chromatography MS/MS,
the tag from the mass-balancing component and the fragment ions (transitions) for SRM are
bound to the labeled amino acid. This yields chosen. Then the known quantities of AQUA
peaks in the MS/MS spectra corresponding peptides are added to the sample mixture, and
to the tags in a low-molecular-mass range the SRM experiments designed to detect the
(between 100 and 120 Da), and these can be paired sample analyte-AQUA peptide are per-
compared to provide a measure of the relative formed. This way, the concentrations of analyte
quantity of the species in the various labeled peptides can be determined based on a standard
samples. Recently developed iTRAQ 8-plex curve derived from AQUA peptide additions,
(244) allows the parallel quantication of up to and whole protein concentrations can be calcu-
eight samples. lated from the peptide values.
For accurate modeling of cellular processes, Because the chemical synthesis of AQUA
the relative quantication of the type obtained peptides is expensive and time-consuming, and
by methods such as SILAC and iTRAQ in their not all peptide sequences can be synthesized
standard mode is not sufcient, and absolute and resolubilized successfully, the method is
quantication of the total number of protein not perfect. An alternative approach is the pro-
molecules or modied proteins in a sample duction of isotopically labeled reference pep-
is required. Accurate absolute quantication tides in E. coli (250). In this method, termed
methodologies in MS-based proteomics are QconCAT, the bacteria express the concate-
recent and are still being tested (246249). nated peptides of interest from a synthetic gene
These methods usually combine spectral and are grown in a medium that allows isotopic
counting, calibration with internal standards, labeling of the synthesized material. The arti-
and advanced analytical software to avoid er- cial protein is puried and digested to yield
rors coming from the differential ionization of peptides that can be used to spike the sam-
peptides and comparisons across experiments ple as internal standards and employed in SRM
(Table 3). Spectral counting (a frequency- as with AQUA peptides (251). The caveats of
based analysis approach using the number QconCAT are problems with expression, solu-
of observed spectra) alone is not reliable bilization, and obtaining equimolar amounts of
and is referred to rather as semiquantitative. all the peptides from the concatenate during en-
Typically, absolute quantication methods zymatic digestion. Recently, this issue has been
rely on the addition of heavy-isotope labeled addressed through an equalizer peptide (EtEP)
peptides to the biological sample to serve as strategy, which allows equalization of internal
internal standards of known concentration. standard peptides with a single peptide, whose
In the AQUA (absolute quantication) concentration has been determined exactly by
method (249), isotope-labeled peptides corre- amino acid analysis (252). Amino acid analysis,
sponding to dened tryptic fragments of pro- a gold standard for the accurate absolute quan-
tein species of interest are chemically synthe- tication of peptides, cannot be done routinely
sized. Peptide selection for synthesis is based for all standard peptides used for quantication

in MS; for this reason one equalizer peptide and multiplexing of up to eight samples, the lack
typically is used as a reference. This method of limitations connected with antibody avail-
combines the benets of AQUA (chemical syn- ability (as for the afnity methods described
thesis of peptides) and ConCAT (equalized below), and measurement accuracy are all rea-
peptides kept in solution to avoid problematic sons it is one of the most successful proteomic
resolubilization). methods.
MS in systems biology is also a tool
for metabolomic analysis of small-molecule
metabolites from biological samples, especially Antibody-Based Technologies
for biomarker discovery. The technology
Antibodies are a staple in a large set of afnity-
differs in its details from the ones used for
based protein-analysis methods. The high
proteomics because of the size difference in
specicity of properly selected and validated
the analytes. The most often used instrument
antibodies makes them an ideal tool for
combination is gas chromatography coupled

the detection and quantication of proteins
with MS, although in some cases the molecules
solubilized from virtually any kind of sample,
can be infused directly into the mass spectrom-
including solid tissue, but the downside is that
eter. An excellent review of developments in
all antibody-based methods are limited by
MS-based metabolomics has been published
the availability of selective and properly vali-
recently (253).
dated antibodies for the molecular species in
MS as a systems biology tool has several
question.
weaknesses. It is low throughputanalyses of
complex samples require laborious preprocess-
ing and/or long high-performance liquid chro- Population-based array methods. Immuno-
matography gradients and time-consuming blot (Western blot) and ELISA (enzyme-linked
data analysis. The common bottom-up ap- immunosorbent assay) are two examples of
proach to protein and PTM identication and widely used, low-throughput assays, although
quantication, based on proteolytic digests, recently several groups have begun to develop
requires abundant starting material (for T ways of conducting these assays in a high-
cells, which are small and do not have a lot of throughput mode suitable for data gathering
cytoplasm, often a minimum of 107 cells per for quantitative modeling, in combination with
analysis). In addition, even with near-complete other methods, such as ow cytometry and live-
digestion efciency, the peptides in the digest cell assays (75). Immunoblot relies on the sep-
may not be ionized optimally or could be too aration of proteins in the gel by mass and/or
large to be detected (the typical detection charge, transfer to the membrane, and detec-
range for MS is m/z 400 to 2,000), so only tion with antibodies. ELISA usually utilizes
a fraction of the sample is eventually analyzed. multiwell plates with immobilized antibodies
Finally, the results are measurements of popula- to capture proteins, which then are detected
tion averages, and there is no current possibility directly (if labeled) or indirectly with another
of single-cell measurements. The high number set of antibodies. Detection in both methods
of cells needed for each analysis has made often involves signal enhancement; chemilu-
experiments on primary cells a challenge, and minescence or uorescence is used commonly.
most often cultured cells have been analyzed; For both these assays, the required quantities of
only recently have quantitative analyses of sample per assay point are typically small and
material from primary cells begun to appear can be derived from a modest number of cells
(237, 254). (104 105 cells), but detection of low-abundance
Nevertheless, the advantages of MS are hard analytes is hard and multiplexing is limited. The
to overestimate. Comparisons of thousands of measurements are also average abundances in
proteins or phosphorylation sites in one analysis cell populations, not single-cell results.
562 Germain et al.

Several techniques have been developed to solid surface of various compositions. The
increase the multiplexing ability of afnity- immobilized fraction can consist of capture
based assays. Many of them are based on bead molecules, such as puried recombinant lig-
technology. In bead-based ow assays (such as ands, peptides, antibodies, antibody fragments,
Luminex), the target proteins are captured from antibody mimicking aptamers, or, in the
solution with immobilized antibodies on mi- case of reverse-phase protein microarrays, a
croparticle beads that are labeled, usually with complex solution of sample cell lysates. When
different amounts of a single uorochrome to capture molecules are immobilized, the sample
allow identication. The captured proteins are solution is applied over the spots. When the
detected by a second antibody set and quanti- lysate is immobilized, it is probed with soluble
ed in a modied ow cytometer. The labeling capture molecules. The latter provides an
of beads displaying one specic antibody with advantage in miniaturization: A single spot
unique dye signature allows for parallel quan- can contain as little protein as from a single
tication of up to 100 different proteins in a cell. However, the stability of microarrays has
single experiment. High selectivity (two anti- to be considered: Spotted antibodies usually
bodies per antigen) is combined here with fairly are more stable and less affected by prolonged
high throughput. storage than immobilized whole lysates.
More sophisticated antibody-based assays The variability of applied material captured
have much in common with these two proto- in individual spots is a challenge in printing
typic methods and are the result of efforts to all microarrays, and the engineering efforts to
improve throughput and accuracy and to lower circumvent this problem are visible in the au-
detection limits. The nanouidic proteomic tomation area: Noncontact robots delivering
immunoassay separates proteins by isoelectric precise, nano- to picoliter droplets are supe-
focusing in a short length of a nanocapillary rior to contact printers, for which carryover
tube (255). Resolved proteins are captured by and pin alignment are an issue (256, 257).
photochemically activated molecules attached The stability and, to some extent, the variabil-
to the capillary wall and then detected with spe- ity issues are prevented in a recently devel-
cic antibodies. Immobilized protein-antibody oped protein microarray method called nucleic
complexes are detected with chemilumines- acidprogrammable protein array (NAPPA), in
cence reagents owing through the tube. The which functional proteins are translated in situ
nanouidic proteomic immunoassay can be from DNA printed on the spots with the use
automated in principle, it is highly sensitive of an in vitro transcription-translation system
(proteins from as few as 25 cells, and fewer than (IVTT) (258). The coding region of the cDNA
500 molecules of HRP gave good signal-to- corresponding to a protein to be expressed is
noise ratio), and only a small sample volume is engineered into an RNA expression plasmid
needed, but the technology so far has been low and fused to a coexpressed GST tag. Plasmid
throughput (up to 12 samples per day, using DNA is cross-linked to a psoralen-biotin con-
the prototype setup) and expensive, and the an- jugate with the use of ultraviolet light and ar-
alyzed proteins must be soluble. Probably the rayed onto glass slides together with avidin and
most popular emerging technology of the afn- polyclonal GST antibody. At this stage, the ar-
ity type is the protein microarray. There are ray is dried and can be stored at room tem-
many variations of microarrays, and consider- perature. To activate and use the array, one
able efforts are made continuously to augment adds the IVTT (such as reticulocyte lysate with
the existing protocols because of the promise T7 polymerase to generate synthetic RNA)
of extremely high throughput, automatization, to the array. The transcribed RNA is trans-
and multiplexing. Protein microarrays are lated by the IVTT, and the expressed proteins
similar to DNA microarrays. Proteins are are immobilized immediately in their spots
spotted automatically and immobilized on a by the capture of the GST tag by anti-GST

antibody. Independently, similar alternative Phosphoow, a ow-cytometric approach to

strategies have been developed: MIST (multi- determine phosphoepitope levels, is an essen-
ple spotting technique) spots the IVTT from E. tial tool in single-cell proteomics for mapping
coli on a printed PCR template; in DAPA (DNA cell signaling networks (260). This method in-
array to protein array) proteins translated on the volves cell xation and permeabilization prior
cDNA array diffuse through a membrane in- to staining with antiphosphoprotein antibodies.
fused with cell-free extract to a surface with cap- A carefully developed procedure with specic
ture molecules; and high-density peptide and reagents compatible with ow cytometry and
protein chips can be produced by the imme- the use of combinations of phospho-specic
diate immobilization of proteins synthesized antibodies labeled with different uorophores
on stalled ribosomes with puromycin-grafted provides a rapid and efcient means to measure
oligonucleotides (259). Such approaches en- the levels of a variety of intracellular phos-
sure reproducible, fresh, high-yield protein mi- phoepitopes (261). The number of distinct
croarrays for each experiment. Also, the expres- phosphoepitopes depends only on the current
sion clones can be stored in a large collection by capabilities of ow cytometry [as of now, 17
using cassettes, permitting facile examination of markers can be measured simultaneously (3)].
the proteins encoded by interesting clones af- To increase sample throughput, investigators
ter their expression in live cells using a mam- developed a cell-based multiplexing approach,
malian expression vector containing the same called uorescent cell bar coding, especially for
insert. Protein microarrays can be used to quan- phosphoow (262), and this method is based on
tify protein abundance in the samples as well as a principle similar to that used for multiplexing
to explore protein function, for example, kinase in bead-based cytokine assays. In uorescent
activity measured by substrate phosphorylation cell bar coding, samples are labeled with uo-
levels. rophore to variable intensity, which is achieved
Immunoassays, like MS, measure averages by treatment with different concentrations of
in the population of molecules obtained from the reactive form of the uorophore. In this
cells. Because even synchronized cells of the way, each sample gets a unique signature of
same origin, not to mention tissue samples, are uorescence intensity. With one uorophore,
most often heterogeneous and their response four to six different samples can be analyzed
to stimuli can be diverse, measurements at the together. Because multiple uorophores can be
single-cell level provide complementary and used to label the cells, the number of different
crucial information. uorescent signatures increases geometrically
(1636 samples can be monitored with the
Single-cell flow-cytometric methods. Flow combination of two markers). The method has
cytometry permits multiparameter analysis, been reported to increase the resolution when
single-cell resolution, and close to real-time used with 96 samples simultaneously, at the
measurements under limited conditions. Ini- same time reducing antibody consumption and
tially used for the detection of cell surface mark- acquisition time.
ers on live-cell populations, this technology Still the overlapping uorophore spectra re-
now is widely used to measure intracellular pro- quire compensation, and the experiments be-
tein and DNA concentrations and cellular pro- come more difcult to perform with each ad-
cesses in xed cells. The multiplexing ability of ditional color. Alternatives for uorophores in
ow methods, which depend on uorescently multiparameter detection include surface en-
labeled antibodies, has been limited by the fac- hanced Raman scattering, which has been im-
tors connected to antibody use in addition to plemented in ow cytometry and microscopy
sample acquisition throughput, and much ef- in several variants (263265), and a newly
fort has been concentrated on overcoming this reported technology called mass cytometry
limitation. (CyTOF) (266). CyTOF replaces uorescent
564 Germain et al.

labels with specially designed multiatom ele- unbiased assessment of the multidimensional
mental (lanthanide) antibody tags. These labels data (262).
are detected by an inductively coupled plasma Beyond the identication and quantica-
TOF MS that recently has become commer- tion of specic molecular moieties in individ-
cially available. The inductively coupled plasma ual cells, ow-based methods also are useful
source produces transient ion clouds from cells to assess molecular interactions using meth-
or beads, and the ions from the generated ods such as FRET or protein complementation,
supersonic plasma jet are detected by a fast TOF described above in the section on interactome
ion detector. The signal is amplied and digi- determination.
tized, so even low-intensity signals can be de-
tected. The availability of many stable isotopes
for the antibody tags facilitates the simultane- Microscopic Methods
ous detection of many target molecules in the Although ow cytometry can reveal much
samplethe authors demonstrate the method about the identities and amounts of compo-
as applied to 20 antigens in model leukemia nents of a cell, and even something about local
cell lines and acute myeloid leukemia patient molecular interactions, it does not provide the
samples, but the method can be broadened means to resolve topographically where in the
to analyze many more antigens on individual cell the labeled molecules reside, and it cannot
cells in a single experiment. The initial device trace changes in the location or number of
and methods do not permit cell sorting, how- molecules in a single cell over time (it only
ever, because particles are disintegrated com- provides a single time point for data collection
pletely during the analysis, but techniques are from each cell); cannot relate signaling events
being developed to allow elution of the labels to changes in cell shape, dynamics, or motility;
and timed collection of the eluted material to and cannot examine cells in complex environ-
maintain correspondence to the cell of ori- ments. For all these aspects of biology that one
gin in the ow stream, so this limitation may might wish to incorporate in models and sim-
change in the near future. Other limitations in- ulations, microscopic imaging is the method
clude the analysis of only a subset (presently of choice, with implementations including
30% but soon to be >70%) of the applied cell ultrahigh-resolution electron microscopic
sample, slower cell throughput than conven- tomography to dene structural details for
tional ow cytometers, and reduced sensitiv- models of such processes as neurotransmitter
ity on the very low end of expression. Never- release in a synapse (267, 268), super-resolution
theless, the ability to analyze dozens of target light imaging to better understand the location
antigens with no concern of spectral overlap of molecules in a single cell (269), total internal
or the need for complex compensation meth- reection uorescence imaging to investigate
ods makes this an extremely promising new components involved in membrane-related
method that may help generate highly quantita- events with counterligands displayed in planar
tive data on the single-cell level in multiplexed membrane surfaces (270), live-cell confocal
form for use in modeling studies. However, the and two-photon imaging in two- and three-
greatly increased number of parameters exam- dimensional environments (271, 272), and in-
ined makes data analysis much more complex; travital imaging using two-photon instruments
even downloading the les from the instru- to assess cell behavior and signaling in complex
ment pushes the limits of current computers. tissues and organs (273). Especially through the
New congurations of processors in special- use of methods of uorescence microscopy that
ized devices are being developed to handle the allow specic molecules to be tagged with light-
data throughput, and new computational algo- emitting proteins (274), it is possible to collect
rithms for automatic clustering of the many pa- quantitative spatiotemporal information about
rameters being measured will enable rapid and protein expression, molecular movements, and

cytoskeletal changes, among other parameters, often considered to be inherently confocal,

in living cells over prolonged time frames. two-photon imaging does have a lower axial
Coupled with the use of emerging microuidic (z-dimension) resolution as compared to con-
methods that parallelize the treatment of focal designs, and for isolated cells, the latter is
individual members of a cell population with often preferable if bleaching and phototoxicity
the same stimulus and allow for the simulta- are not limitations in the experiment.
neous imaging of this cell population, one can Resolution in uorescence microscopy until
develop extensive data sets on multiple target recently has been constrained by the diffrac-
molecules through color multiplexing and tion limit, which is dependent on the numer-
with additional information about intercellular ical aperture of the lens and the wavelength
heterogeneity within the cell population in of light used. More recently, so-called super-
terms of both molecular expression and cell resolution methods have been introduced that
behavior in response to the stimulus (275, 276). allow computational techniques to discrimi-
Fluorescence microscopy is by far the most nate point sources below this diffraction limit.
commonly employed technique in systems Two such high-resolution techniques are pho-
studies. The broad spectrum of uorophores toactivated localization microscopy (279) and
emitting light of different colors allows the stochastic optical reconstruction microscopy
simultaneous detection of several molecules. (280). Photoactivated localization microscopy
Basic epi-uorescence microscopy is affordable resolves molecules separated only by a few
and requires simple instruments, but the blur- nanometers by serial photoactivation and the
ring of the image due to the light emission from subsequent bleaching of many sparse subsets of
different depths of the sample is a constant photoactivatable uorescent protein molecules.
challenge, especially for correct quantitative Each source spot at each photoconversion cycle
measurements. This problem is overcome by is treated as a Gaussian intensity distribution,
confocal microscopy, in which only the light and the center of the Gaussian peak is calcu-
from the focal plane is collected. The resulting lated, improving the resolution from approxi-
sharp images allow the reconstruction of the mately 200 to 2040 nM. The method works
three-dimensional image by stacking a series well for separating the images of molecules
of optical x-y sections in the z plane (z stacks). densely packed within cell samples and is con-
Confocal microscopy can be used for the stantly being improved with the introduction
quantitative measurements of cell motility of new photochromes and methods for per-
and dynamic intracellular events in time-lapse forming the excitation, collection, and bleach
experiments (277), although there are limita- cycles at increasingly faster speeds, thus allow-
tions. High-energy excitation photons from ing the method to image molecules in motion
the lasers used on confocal microscopes can (281). In stochastic optical reconstruction mi-
lead to phototoxicity and photobleaching, and croscopy, which is closely related, the uores-
scattering limits the thickness of the sample cent image is constructed from high-accuracy
to 80 m (277). The alternative technology is localization of individual uorescent molecules
multiphoton laser scanning uorescence mi- that are switched on and off using different color
croscopy (278), which uses lower-energy light light.
in short pulses, compensating the energy by the More examples and a detailed review of
simultaneous absorption of multiple photons modern microscopic technologies used for
for excitation of the sample. Two-photon imaging the immune system can be found in
microscopy, in which two photons of infrared Reference 282. Here we highlight some exam-
light are enough to cause light emission from ples of the applications of the instruments and
the uorophore, yields superb results in thicker techniques mentioned above. The recruitment
(up to 1 mm) samples of live, explanted immune of molecules to the immunological synapse
tissues with limited photodamage. Although has been elucidated in three-dimensional detail
566 Germain et al.

using digital uorescence microscopy (283, ne-grained models of cellular behavior, espe-
284), which led to subsequent modeling of cially signaling pathways leading to gene acti-
the molecular patterns during immunological vation that can be assessed by the nucleic acid
synapse formation (64) and speculation about technologies also described here, but the use of
the role of the cSMAC as a site of TCR sig- this information for model building and simu-
naling inactivation rather than activation, as lation still is impeded by the absence of many of
originally proposed (285, 286). Intravital two- the quantitative parameters that can constrain
photon microscopy has yielded data on cell dy- the substantial degrees of freedom in such mod-
namics in vivo that have served as the basis for els, particularly those involving the dynamics
several computational exercises examining is- of molecular interactions and of enzymatic
sues such as the search pattern of T cells within processes. Only a few of the required numbers
lymph nodes and the role of stromal elements exist in a validated form in the literature, in part
in this behavior (287), the relationship between because until recently it was unclear to investi-
signal intensity and the transition between the gators why time and resources should be spent
motile and static states seen during the initi- determining afnity constants and catalytic
ation of T cell adaptive responses (288290), rates. With an increased emphasis on modeling
the possible roles of guided versus random walk in immunology, such numbers have become
patterns in ensuring the effective activation of much more important, although still sparse. A
T cells during a single passage through a lymph number of tools exist for acquiring the neces-
node (291), and the validity of older models of sary data, although few have been developed
migration between the dark and light zones of as robust, multiplexed methods suitable for the
the germinal center in afnity-based selection scale of typical systems biology model building.
(292). The binding constants necessary for
FRET has been used not only in ow stud- detailed molecular interaction models are pos-
ies, but also in microscopic imaging to assess sible to obtain with the use of surface plasmon
PPIs within the live-cell environment. Many resonance (SPR) technology, implemented in
features have been studied using this method, instruments such as Biacore (GE) or ProteOn
for example, differences in TCR recruitment to (BioRad). The phenomenon of SPR occurs
the immunological synapse and CD8-TCR in- when the plasmons (electron charge density
teraction after T cell induction with different waves) on the planar surface of the metal in
APCs (293), the characterization of the TCR a special chip (such as gold) are excited by a
chain connecting peptide motif allowing bind- change in the boundary of this surface with a
ing to CD8 and efcient signal initiation (294), medium of different refractive index (such as a
and the formation of a complex leading to B buffer). This occurs when the mass of material
cell antigen receptor signaling upon binding of at the buffer-chip interface changes, as will be
multivalent antigens (295). the case when an applied substance binds to
molecules immobilized on the chip surface.
Because the resultant resonance change is not
Surface Plasmon Resonance Methods affected by sample color or opacity, SPR can
The methods described above allow the be used to determine the interactions between
generation of a catalog of expressed molecules, many different types of ligands and their
especially proteins, their quantication, and binding partners; to search for novel PPIs,
the description of their chemical modication; DNA-protein, and DNA-RNA interactions;
the detection of molecular interactions; and the and to determine drug targets (296). At present,
generation of time- and space-resolved data sets a major limitation of this method is the substan-
involving molecular interactions, modication, tial amount of effort required to produce pure
and intracellular relocation. All these types of protein in a form suitable either for linkage to
information are critical for the generation of the chip surface or for addition as the analyte

to a derivatized chip. The generation of such tially in origin and function. Luber et al. (237)
puried reagents is typically a slow process examined the differences in protein abundance
involving one molecule at a time, and the pro- in some of the commonly described DC subsets
duction of a reliable data set can take months. using a label-free quantitative MS approach
This is clearly inadequate for use in parameter- with an in-house quantication algorithm.
izing complex models with many components. They proled protein-abundance differences
However, improvements in chip design, espe- among conventional DC subsets, which they
cially the multiplexing capacity of microuidic dened as conventional CD8+ , CD4+ , and
versions of these instruments, allow many double-negative DCs separated by ow cytom-
assays to be performed in parallel. There is the etry, to a depth of more than 5,000 proteins and
prospect of combining some of the technology requiring only 1.52.0 106 puried cells.
of NAPPA for in vitro synthesis of proteins The results were validated with surface markers
from expression clones with such microuidic of known abundance in DC subtypes and com-
multiplexed SPR devices to enable the analysis parison with the results of previous microarray
of dozens of molecular interactions in a short studies. Based on the latter comparisons, the
(several weeks) period without the need for in- authors concluded that this approach is a reli-
dividual molecule purication, albeit with less able method to study closely related cell types
precision than would be the case for carefully in vivo. The analysis revealed the mutually ex-
puried individual components. However, the clusive expression of some pattern-recognition
nature of molecular models is such that small pathways (the differential expression of mem-
errors in such measurements typically have bers of the NLR, Toll-like receptor, and RLH
little effect on model performance, but they do signaling pathways in response to u virus and
need to be in the right overall numerical range; Sendai virus), which might explain differences
sensitivity testing of the model can determine in response to different pathogens. These new
which parameters are most critical to simulated data provided insight into the regulation of
outcomes, and these can point to specic pattern-recognition receptors in infected cells.
interactions that might require remeasurement A more focused approach was used by Segura
by the more precise methods involving highly et al. (297), who isolated the plasma membranes
puried material. This combination of mul- of mouse spleen CD8+ and CD8 DCs by
tiplexed SPR with NAPPA-style synthesis of immunoafnity methods and analyzed their
relevant molecules can open the door to the protein composition using MS. The results
collection of interaction parameters needed were only semiquantitative because only the
for ne-grained modeling at a pace and to spectral counting method was used. Never-
an extent that will allow a substantial reduc- theless, the comparative analysis suggested
tion in the degrees of freedom in molecular that many known or potential pathogen-
models. recognition receptors and immunomodulatory
molecules are expressed differentially, and
these results were conrmed subsequently by
Applications in Immunology ow cytometry and Western blot analyses. At
Progress in proteomic instrumentation and this point in time, however, these proling
technology has made possible many novel stud- experiments are mainly cataloging exercises
ies of primary immune system cells. One area that yield parts lists for future studies of a more
that has yielded especially intriguing results is integrated and functional nature.
the study of the dendritic cell (DC) proteome. Phosphorylation in immune cell signaling
DCs, a heterogeneous population of profes- has not been analyzed frequently by MS using
sional APCs, consist of a number of distinct primary cells because of the difculty of obtain-
subtypes, distinguishable by multiple surface ing these cells in sufcient number. Most exam-
markers. These DC subtypes differ substan- inations of phosphorylation sites involved in T
568 Germain et al.

cell signaling thus have utilized the leukemic The use of various types of protein arrays
Jurkat T cell line (298). However, recent in- has facilitated many recent systems-type
creases in instrument sensitivity and improved immunological studies. Merbl et al. (301)
enrichment methods now allow the analysis combined an antigen-microarray device and
of primary cells, and reports of phosphopro- a genetic algorithm analysis to investigate
teomics in nontransformed T cells have be- large-scale patterns in the antibody reper-
gun to appear. Iwai et al. (254) examined the toire reecting the state of metastatic or
tyrosine phosphorylation network in T cell re- nonmetastatic tumors in C57BL/6 mice. The
ceptor signaling in cells from diabetes-prone analysis of immunoglobulin (Ig)G and IgM au-
and normal mice, identifying and quantifying toantibodies binding to over 300 self-antigens
over 100 tyrosine phosphorylation sites using revealed informative antibody patterns re-
iTRAQ labeling. More frequently, phospho- sponding to the growth of the tumor cells
specic ow cytometry has been utilized to ana- and distinguishing between animals bearing
lyze phosphorylation-based signaling in the im- metastatic and nonmetastatic tumors. NAPPAs
mune system cells. Hale et al. (299) applied were used successfully (302) to identify Pseu-
phosphoow to examine the changes in cy- domonas aeruginosa proteins that trigger an
tokine signal transduction through the STAT adaptive immune response in cystic brosis.
family of transcription factors during clinical Out of 262 P. aeruginosa outer membrane
progression of systemic lupus erythematosus and exported proteins examined, antibodies to
(SLE) in human patients. The measurements 12 were detected in a statistically signicant
of signaling responses at the single-cell level in number of patients and conrmed by ELISA
ve immune cell types to a panel of 10 cytokines as potential candidates for future vaccine
thought to play essential roles in SLE led to a development. Ceroni et al. (303) used the same
highly multiplexed view of cytokine signal pro- technology to investigate the adaptive immune
cessing in adaptive and innate immune cells. response to infection with Varicella zoster virus.
Robust changes, including an increase in the T All 69 virus proteins were studied in sera from
cell response to IL-10 and cessation of STAT 68 infected individuals, and the IgG antibodies
molecule responses to multiple cytokines, were against many viral antigens were detected and
found during progression of SLE, providing ev- validated by Western blot. These antibodies
idence for negative feedback regulation and the were present only in the subset of VZV-positive
existence of a cytokine-driven oscillator, which patients, demonstrating the complexity of the
may regulate periodic changes in SLE activity. humoral immune response to the viral infection
In early rheumatoid arthritis, 15 signaling ef- and validating the NAPPA approach for the
fectors were examined by phosphoow in pe- development of diagnostic tests. Protein arrays
ripheral blood mononuclear cells, identifying were also used for systemic, large-scale pro-
patterns of effector activation by phosphoryla- ling of the specicity of antibody responses
tion characteristic for rheumatoid arthritis (dis- against autoantigens in a panel of autoimmune
tinct from the patterns observed in osteoarthri- diseases (304) and allergies (305, 306).
tis), but the same in early rheumatoid arthritis
as in the late rheumatoid arthritis (300). For
example, the ratio of p-AKT to p-p38 was el- CONCLUDING REMARKS
evated signicantly in patients with rheuma- Above we attempt to present one major aspect
toid arthritis and may be used for diagnostic in the still emerging eld of systems biology,
purposes. To date, these data have not been namely the use of dynamic computational
used to derive models suitable for simulation models for simulating biological behavior. In
and testing of intracellular signaling, but rather particular, we emphasize the importance of
may serve primarily as tools for biomarker integrating high-throughput data-collection
development. methods and highly quantitative experimental

techniques with mathematical approaches to in the central Erk signaling pathway (307309),
develop an enhanced understanding of how for the role of modest differences in transcrip-
the immune system operates at various scales. tion factor concentration in controlling
It is apparent that much more has been ac- developmental cell fate of lymphocytes (310
complished to date at the level of omic-scale 312), or for small differences in the ratio of
data collection than in the development and regulatory versus effector cells in determining
use of models and simulation. This is not to the presence or absence of autoimmune disease
say that important contributions have not come (313) speaks to the need to develop more quan-
from the modeling efforts we and many others titative methods to integrate our ne-grained
have undertaken, but rather that the number knowledge of immune components into larger
and power of the models created to date are schemes that better reect system output.
rather limited and this approach is not yet the A variety of activities and events will impact
primary way high-content data are used to de- the rate at which modeling and simulation are
velop new insights into immune system func- integrated into the immunological mainstream.
tion. In part, this limitation comes from the First, nothing succeeds like success, and as
recent emergence of the experimental capacity various groups produce models and simulations
to collect the necessary high-quality, extensive that make useful predictions about biology and
data sets required for computational analysis. these predictions are validated by experiment,
It also arises from a gap in the skill sets of ex- others will seek to adopt such approaches in
perimental biologists and the only recent de- their own work. Second, the dissemination
velopment of software platforms that empower of improved tools for constructing complex
these investigators to undertake complex model models, conducting simulations, and sharing
building. Finally, it reects a cultural bias that the results, together with greater access to ever-
places primary emphasis on the discovery na- cheaper mass computing power, will empower
ture of biological research and accords model- the community to work together to build and
ing less value. utilize better and bigger models. Finally, the
Indeed, a primary goal of this review has changes already in progress in the way students
been to lay out the rationale for the critical im- are educated, with the inclusion of more statis-
portance going forward of incorporating more tics, mathematics, and computer programming
computational methods into biological studies. at various stages of secondary, college, and
The overall complexity of all biological systems, postgraduate education, will make it easier for
especially the immune system with its hun- new investigators to incorporate these methods
dreds of cell types and regulatory and effector into their research activities, just as the familiar-
molecules, necessitates such an approach if we ity of newly minted MDs and PhDs with omic
are to place this multitude of components into a technologies has led over the past decade to
scheme that helps us understand and eventually the facile inclusion of such methods into their
predict how they interact to yield biological projects. In our group, we have constructed a
behavior. We have a rather limited ability to research enterprise utilizing a multidisciplinary
visualize the nonlinear ways in which concate- team approach that brings together expertise
nated feedback circuits so characteristic of the in each of the major areas discussed in this
immune system affect its output. This demands review as necessary for effective computational
the use of formal mathematical treatments to systems biology endeavors (Figure 1), believ-
associate modest differences in component ing that it can serve as a model for how to
concentrations, binding afnities, enzymatic move forward in this direction. We therefore
activity, cell replication and death rates, and the close on an optimistic note, suggesting that
like with the effects of gene polymorphism and within a generation, modeling and simulation
mutation or drug treatments on system perfor- will have become a mainstream component of
mance. The recent evidence for digital behavior biological research, comfortable for many if
570 Germain et al.

Computer modeling/ Genomics

simulation
Collection and analysis of
Creation of software tools for data on gene expression,
constructing and simulating miRNA, epigenetic
complex multiscale biological modifications, discovery of
processes gene regulatory networks;
in connection with CMB, Bioinformatics
experiments to connect
signaling to gene expression Development and
prior to and in follow-up to application of statistical
Cell/molecular biology modeling such connections tools for extracting new data
from literature, analyzing
HTS to discover new nodes HTS data, microarray and
and edges (molecules and next-gen sequencing data,
interactions) in modular construction of statistical
inference network models

networks; testing of
predictions from models
using RNAi and related
technologies Immunology
Predictive Wet lab experiments at the
Proteomics models cell and organism levels to
explore immune behavior
Protein modifications, and feed data into as well as
number of molecules, test emerging models of
ka /kd /kcat ... immune function and
for parameterizing models host/pathogen interactions
Figure 1
Team approach to modeling. The gure illustrates the various technical, data-gathering, and biological components of an integrated
research approach to computational systems biology with a focus on ne-grained, dynamic modeling and simulation of processes such
as cell signaling. Abbreviations: CMB, cell and molecular biology; HTS, high-throughput screening.
not most investigators. This will result in the high-quality predictive analysis of how small-
movement from cartoon representations of and large-scale aspects of the immune system
biological systems to instantiated quantitative will behave when perturbed experimentally, by
models shared by the eld and capable of genetic variation, or by medical intervention.
DISCLOSURE STATEMENT
The authors are not aware of any afliations, memberships, funding, or nancial holdings that
might be perceived as affecting the objectivity of this review.
ACKNOWLEDGMENTS
The authors wish to thank many colleagues, especially Gregoire Altan-Bonnet, for discussions and
research interactions that were critical to our developing the perspective espoused in this review.
We apologize to those in the eld whose papers and contributions were not cited explicitly due
to space limitations. This work was supported by the Intramural Research Program of NIAID,
NIH.

LITERATURE CITED
1. Hoheisel JD. 2006. Microarray technology: beyond transcript proling and genotype analysis. Nat. Rev.
Genet. 7:20010
2. Wold B, Myers RM. 2008. Sequence census methods for functional genomics. Nat. Methods 5:1921
3. Perfetto SP, Chattopadhyay PK, Roederer M. 2004. Seventeen-colour ow cytometry: unravelling the
immune system. Nat. Rev. Immunol. 4:64855
4. Aebersold R, Mann M. 2003. Mass spectrometrybased proteomics. Nature 422:198207
5. Benoist C, Germain RN, Mathis D. 2006. A plaidoyer for systems immunology. Immunol. Rev. 210:229
34
6. Noble D. 2002. Modeling the heart: from genes to cells to the whole organ. Science 295:167882
7. Alon U. 2007. Network motifs: theory and experimental approaches. Nat. Rev. Genet. 8:45061
8. Herzenberg LA, Black SJ. 1980. Regulatory circuits and antibody responses. Eur. J. Immunol. 10:111
9. Germain RN. 2001. The art of the probable: system control in the adaptive immune system. Science
293:24045
10. Callard R, Hodgkin P. 2007. Modeling T- and B-cell growth and differentiation. Immunol. Rev. 216:119
29
11. Meier-Schellersheim M, Fraser ID, Klauschen F. 2009. Multiscale modeling for biologists. Wiley Inter-
discip. Rev. Syst. Biol. Med. 1:414
12. Davidson EH. 2009. Network design principles from the sea urchin embryo. Curr. Opin. Genet. Dev.
19:53540
13. Alon U. 2007. An Introduction to Systems Biology: Design Principles of Biological Circuits. Boca Raton, FL:
Chapman & Hall/CRC
14. Heng TS, Painter MW. 2008. The Immunological Genome Project: networks of gene expression in
immune cells. Nat. Immunol. 9:109194
15. Shaw DE, Deneroff MM, Dror RO, Kuskin JS, Larson RH, et al. 2008. Anton, a special-purpose machine
for molecular dynamics simulation. Commun. ACM 51:9197
16. Klepeis JL, Lindorff-Larsen K, Dror RO, Shaw DE. 2009. Long-timescale molecular dynamics simula-
tions of protein structure and function. Curr. Opin. Struct. Biol. 19:12027
17. Varma R, Campi G, Yokosuka T, Saito T, Dustin ML. 2006. T cell receptor-proximal signals are
sustained in peripheral microclusters and terminated in the central supramolecular activation cluster.
Immunity 25:11727
18. Huse M, Klein LO, Girvin AT, Faraj JM, Li QJ, et al. 2007. Spatial and temporal dynamics of T cell
receptor signaling with a photoactivatable agonist. Immunity 27:7688
19. Batista FD, Arana E, Barral P, Carrasco YR, Depoil D, et al. 2007. The role of integrins and coreceptors
in rening thresholds for B-cell responses. Immunol. Rev. 218:197213
20. Delon J, Kaibuchi K, Germain RN. 2001. Exclusion of CD43 from the immunological synapse is medi-
ated by phosphorylation-regulated relocation of the cytoskeletal adaptor moesin. Immunity 15:691701
21. Faure S, Salazar-Fontana LI, Semichon M, Tybulewicz VL, Bismuth G, et al. 2004. ERM proteins
regulate cytoskeleton relaxation promoting T cell-APC conjugation. Nat. Immunol. 5:27279
22. Gomez TS, Billadeau DD. 2008. T cell activation and the cytoskeleton: You cant have one without the
other. Adv. Immunol. 97:164
23. Badour K, Zhang J, Shi F, Leng Y, Collins M, Siminovitch KA. 2004. Fyn and PTP-PEST-mediated
regulation of Wiskott-Aldrich syndrome protein (WASp) tyrosine phosphorylation is required for cou-
pling T cell antigen receptor engagement to WASp effector function and T cell activation. J. Exp. Med.
199:99112
24. Sohn HW, Tolar P, Pierce SK. 2008. Membrane heterogeneities in the formation of B cell receptor-Lyn
kinase microclusters and the immune synapse. J. Cell Biol. 182:36779
25. Treanor B, Depoil D, Gonzalez-Granja A, Barral P, Weber M, et al. 2010. The membrane skeleton
controls diffusion dynamics and signaling through the B cell receptor. Immunity 32:18799
26. Andrews SS, Bray D. 2004. Stochastic simulation of chemical reactions with spatial resolution and single
molecule detail. Phys. Biol. 1:13751
27. Saxton MJ. 2007. Modeling 2D and 3D diffusion. Methods Mol. Biol. 400:295321
572 Germain et al.

28. Kerr RA, Bartol TM, Kaminsky B, Dittrich M, Chang JC, et al. 2008. Fast Monte Carlo simulation
methods for biological reaction-diffusion systems in solution and on surfaces. SIAM J. Sci. Comput.
30:312649
29. Tolle DP, Le Novere N. 2010. Meredys, a multi-compartment reaction-diffusion simulator using mul-
tistate realistic molecular complexes. BMC Syst. Biol. 4:24
30. Gruenert G, Ibrahim B, Lenser T, Lohel M, Hinze T, Dittrich P. 2010. Rule-based spatial modeling
with diffusing, geometrically constrained molecules. BMC Bioinforma. 11:307
31. Chan C, George AJ, Stark J. 2001. Cooperative enhancement of specicity in a lattice of T cell receptors.
Proc. Natl. Acad. Sci. USA 98:575863
32. Sosinsky GE, Deerinck TJ, Greco R, Buitenhuys CH, Bartol TM, Ellisman MH. 2005. Development
of a model for microphysiological simulations: small nodes of ranvier from peripheral nerves of mice
reconstructed by electron tomography. Neuroinformatics 3:13362
33. Ander M, Beltrao P, Di Ventura B, Ferkinghoff-Borg J, Foglierini M, et al. 2004. SmartCell, a framework
to simulate cellular processes that combines stochastic approximation with diffusion and localisation:
analysis of simple networks. Syst. Biol. 1:12938

34. Schaff J, Fink CC, Slepchenko B, Carson JH, Loew LM. 1997. A general computational framework for
modeling cellular structure and function. Biophys. J. 73:113546
35. Hattne J, Fange D, Elf J. 2005. Stochastic reaction-diffusion simulation with MesoRD. Bioinformatics
21:292324
36. Meier-Schellersheim M, Xu X, Angermann B, Kunkel EJ, Jin T, Germain RN. 2006. Key role of local
regulation in chemosensing revealed by a new molecular interaction-based modeling method. PLoS
Comput. Biol. 2:e82
37. Schaff JC, Slepchenko BM, Loew LM. 2000. Physiological modeling with virtual cell framework. Methods
Enzymol. 321:123
38. Schneider IC, Haugh JM. 2005. Quantitative elucidation of a distinct spatial gradient-sensing mechanism
in broblasts. J. Cell Biol. 171:88392
39. Hlavacek WS, Faeder JR, Blinov ML, Perelson AS, Goldstein B. 2003. The complexity of complexes in
signal transduction. Biotechnol. Bioeng. 84:78394
40. Hlavacek WS, Faeder JR. 2009. The complexity of cell signaling and the need for a new mechanics. Sci.
Signal. 2:pe46
41. Lok L, Brent R. 2005. Automatic generation of cellular reaction networks with Moleculizer 1.0. Nat.
Biotechnol. 23:13136
42. Hlavacek WS, Faeder JR, Blinov ML, Posner RG, Hucka M, Fontana W. 2006. Rules for modeling
signal-transduction systems. Sci. STKE 2006:re6
43. Feret J, Danos V, Krivine J, Harmer R, Fontana W. 2009. Internal coarse-graining of molecular systems.
Proc. Natl. Acad. Sci. USA 106:645358
44. Takahashi K, Ishikawa N, Sadamoto Y, Sasamoto H, Ohta S, et al. 2003. E-Cell 2: multi-platform E-Cell
simulation system. Bioinformatics 19:172729
45. Dhar P, Meng TC, Somani S, Ye L, Sairam A, et al. 2004. Cellware: a multi-algorithmic software for
computational systems biology. Bioinformatics 20:131921
46. Hoops S, Sahle S, Gauges R, Lee C, Pahle J, et al. 2006. COPASI: a COmplex PAthway SImulator.
Bioinformatics 22:306774
47. Hucka M, Finney A, Sauro HM, Bolouri H, Doyle J, Kitano H. 2002. The ERATO Systems Biology
Workbench: enabling interaction and exchange between software tools for computational biology. Pac.
Symp. Biocomput. 2002:45061
48. Shapiro BE, Levchenko A, Meyerowitz EM, Wold BJ, Mjolsness ED. 2003. Cellerator: extending a
computer algebra system to include biochemical arrows for signal transduction simulations. Bioinformatics
19:67778
49. Hu B, Fricke GM, Faeder JR, Posner RG, Hlavacek WS. 2009. GetBonNie for building, analyzing and
sharing rule-based models. Bioinformatics 25:145760
50. Moraru II, Schaff JC, Slepchenko BM, Blinov ML, Morgan F, et al. 2008. Virtual Cell modelling and
simulation software environment. IET Syst. Biol. 2:35262

51. Minguet S, Schamel WW. 2008. A permissive geometry model for TCR-CD3 activation. Trends Biochem.
Sci. 33:5157
52. Ma Z, Janmey PA, Finkel TH. 2008. The receptor deformation model of TCR triggering. FASEB J.
22:10028
53. Germain RN. 1997. T-cell signaling: the importance of receptor clustering. Curr. Biol. 7:R64044
54. McKeithan TW. 1995. Kinetic proofreading in T-cell receptor signal transduction. Proc. Natl. Acad. Sci.
USA 92:504246
55. Dushek O, Das R, Coombs D. 2009. A role for rebinding in rapid and reliable T cell responses to antigen.
PLoS Comput. Biol. 5:e1000578
56. Burroughs NJ, Lazic Z, van der Merwe PA. 2006. Ligand detection and discrimination by spatial relo-
calization: a kinase-phosphatase segregation model of TCR activation. Biophys. J. 91:161929
57. Bunnell SC, Hong DI, Kardon JR, Yamazaki T, McGlade CJ, et al. 2002. T cell receptor ligation induces
the formation of dynamically regulated signaling assemblies. J. Cell Biol. 158:126375
58. Campi G, Varma R, Dustin ML. 2005. Actin and agonist MHC-peptide complex-dependent T cell
receptor microclusters as scaffolds for signaling. J. Exp. Med. 202:103136

59. Dushek O, Mueller S, Soubies S, Depoil D, Caramalho I, et al. 2008. Effects of intracellular calcium and
actin cytoskeleton on TCR mobility measured by uorescence recovery. PLoS One 3:e3913
60. Cairo CW, Das R, Albohy A, Baca QJ, Pradhan D, et al. 2010. Dynamic regulation of CD45 lateral
mobility by the spectrin-ankyrin cytoskeleton of T cells. J. Biol. Chem. 285:11392401
61. Wofsy C, Coombs D, Goldstein B. 2001. Calculations show substantial serial engagement of T cell
receptors. Biophys. J. 80:60612
62. Coombs D, Kalergis AM, Nathenson SG, Wofsy C, Goldstein B. 2002. Activated TCRs remain marked
for internalization after dissociation from pMHC. Nat. Immunol. 3:92631
63. Valitutti S, Muller S, Cella M, Padovan E, Lanzavecchia A. 1995. Serial triggering of many T-cell
receptors by a few peptide-MHC complexes. Nature 375:14851
64. Qi SY, Groves JT, Chakraborty AK. 2001. Synaptic pattern formation during cellular recognition. Proc.
Natl. Acad. Sci. USA 98:654853
65. Padhan K, Varma R. 2010. Immunological synapse: a multi-protein signalling cellular apparatus for
controlling gene expression. Immunology 129:32228
66. Stefanova I, Hemmer B, Vergelli M, Martin R, Biddison WE, Germain RN. 2003. TCR ligand discrim-
ination is enforced by competing ERK positive and SHP-1 negative feedback pathways. Nat. Immunol.
4:24854
67. Kolch W. 2005. Coordinating ERK/MAPK signalling through scaffolds and inhibitors. Nat. Rev. Mol.
Cell Biol. 6:82737
68. Levchenko A, Bruck J, Sternberg PW. 2000. Scaffold proteins may biphasically affect the levels of
mitogen-activated protein kinase signaling and reduce its threshold properties. Proc. Natl. Acad. Sci. USA
97:581823
69. Chan C, Stark J, George AJ. 2004. Feedback control of T-cell receptor activation. Proc. Biol. Sci. 271:931
39
70. Altan-Bonnet G, Germain RN. 2005. Modeling T cell antigen discrimination based on feedback control
of digital ERK responses. PLoS Biol. 3:e356
71. Gillespie DT. 1976. A general method for numerically simulating the stochastic time evolution of coupled
chemical reactions. J. Comput. Phys. 22:40334
72. Das J, Ho M, Zikherman J, Govern C, Yang M, et al. 2009. Digital signaling and hysteresis characterize
ras activation in lymphoid cells. Cell 136:33751
73. Margarit SM, Sondermann H, Hall BE, Nagar B, Hoelz A, et al. 2003. Structural evidence for feedback
activation by Ras.GTP of the Ras-specic nucleotide exchange factor SOS. Cell 112:68595
74. Hlavacek WS, Redondo A, Metzger H, Wofsy C, Goldstein B. 2001. Kinetic proofreading models for
cell signaling predict ways to escape kinetic proofreading. Proc. Natl. Acad. Sci. USA 98:7295300
75. Albeck JG, Burke JM, Aldridge BB, Zhang M, Lauffenburger DA, Sorger PK. 2008. Quantitative analysis
of pathways controlling extrinsic apoptosis in single cells. Mol. Cell 30:1125
76. Albeck JG, Burke JM, Spencer SL, Lauffenburger DA, Sorger PK. 2008. Modeling a snap-action,
variable-delay switch controlling extrinsic cell death. PLoS Biol. 6:283152
574 Germain et al.

77. Albeck JG, MacBeath G, White FM, Sorger PK, Lauffenburger DA, Gaudet S. 2006. Collecting and
organizing systematic sets of protein data. Nat. Rev. Mol. Cell Biol. 7:80312
78. Volterra V. 1926. Variazioni e uttuazioni del humero dindividui in specie animali conviventi [Variations
and uctuations of the number of individuals in animal species living together]. Mem. Acad. Lincei 2:31
113
79. Mohri H, Bonhoeffer S, Monard S, Perelson AS, Ho DD. 1998. Rapid turnover of T lymphocytes in
SIV-infected rhesus macaques. Science 279:122327
80. De Boer RJ, Mohri H, Ho DD, Perelson AS. 2003. Turnover rates of B cells, T cells, and NK cells in
simian immunodeciency virus-infected and uninfected rhesus macaques. J. Immunol. 170:247987
81. Grossman Z, Meier-Schellersheim M, Sousa AE, Victorino RM, Paul WE. 2002. CD4+ T-cell depletion
in HIV infection: Are we closer to understanding the cause? Nat. Med. 8:31923
82. Brenchley JM, Douek DC. 2008. HIV infection and the gastrointestinal immune system. Mucosal Im-
munol. 1:2330
83. Okoye A, Meier-Schellersheim M, Brenchley JM, Hagen SI, Walker JM, et al. 2007. Progressive CD4+
central memory T cell decline results in CD4+ effector memory insufciency and overt disease in chronic
SIV infection. J. Exp. Med. 204:217185

84. Grossman Z, Paul WE. 2000. The impact of HIV on naive T-cell homeostasis. Nat. Med. 6:97677
85. Hawkins ED, Hommel M, Turner ML, Battye FL, Markham JF, Hodgkin PD. 2007. Measuring lym-
phocyte proliferation, survival and differentiation using CFSE time-series data. Nat. Protoc. 2:205767
86. De Boer RJ, Ganusov VV, Milutinovic D, Hodgkin PD, Perelson AS. 2006. Estimating lymphocyte
division and death rates from CFSE data. Bull. Math. Biol. 68:101131
87. Yates A, Chan C, Strid J, Moon S, Callard R, et al. 2007. Reconstruction of cell population dynamics
using CFSE. BMC Bioinforma. 8:196
88. Luzyanina T, Roose D, Schenkel T, Sester M, Ehl S, et al. 2007. Numerical modelling of label-structured
cell population growth using CFSE distribution data. Theor. Biol. Med. Model. 4:26
89. Dowling MR, Hodgkin PD. 2009. Modelling naive T-cell homeostasis: consequences of heritable cellular
lifespan during ageing. Immunol. Cell Biol. 87:44556
90. Yates A, Callard R, Stark J. 2004. Combining cytokine signalling with T-bet and GATA-3 regulation in
Th1 and Th2 differentiation: a model for cellular decision-making. J. Theor. Biol. 231:18196
91. Busse D, de la Rosa M, Hobiger K, Thurley K, Flossdorf M, et al. 2010. Competing feedback loops
shape IL-2 signaling between helper and regulatory T lymphocytes in cellular microenvironments. Proc.
92. Regoes RR, Barber DL, Ahmed R, Antia R. 2007. Estimation of the rate of killing by cytotoxic T
lymphocytes in vivo. Proc. Natl. Acad. Sci. USA 104:1599603
93. Celada F, Seiden PE. 1992. A computer model of cellular interactions in the immune system. Immunol.
Today 13:5662
94. Celada F, Seiden PE. 1996. Afnity maturation and hypermutation in a simulation of the humoral
immune response. Eur. J. Immunol. 26:135058
95. Kesmir C, De Boer RJ. 2003. A spatial model of germinal center reactions: cellular adhesion based sorting
of B cells results in efcient afnity maturation. J. Theor. Biol. 222:922
96. Figge MT, Garin A, Gunzer M, Kosco-Vilbois M, Toellner KM, Meyer-Hermann M. 2008. Deriving
a germinal center lymphocyte migration model from two-photon data. J. Exp. Med. 205:301929
97. Ray JC, Flynn JL, Kirschner DE. 2009. Synergy between individual TNF-dependent functions deter-
mines granuloma performance for controlling Mycobacterium tuberculosis infection. J. Immunol. 182:3706
17
98. Efroni S, Harel D, Cohen IR. 2007. Emergent dynamics of thymocyte development and lineage deter-
mination. PLoS Comput. Biol. 3:e13
99. Graner F, Glazier JA. 1992. Simulation of biological cell sorting using a two-dimensional extended Potts
model. Phys. Rev. Lett. 69:201316
100. Beltman JB, Maree AF, de Boer RJ. 2007. Spatial modelling of brief and long interactions between T
cells and dendritic cells. Immunol. Cell Biol. 85:30614
101. Beltman JB, Henrickson SE, von Andrian UH, de Boer RJ, Maree AF. 2009. Towards estimating the
true duration of dendritic cell interactions with T cells. J. Immunol. Methods 347:5469

102. Saez-Rodriguez J, Simeoni L, Lindquist JA, Hemenway R, Bommhardt U, et al. 2007. A logical model
provides insights into T cell receptor signaling. PLoS Comput. Biol. 3:e163
103. Schadt EE, Monks SA, Drake TA, Lusis AJ, Che N, et al. 2003. Genetics of gene expression surveyed
in maize, mouse and man. Nature 422:297302
104. Unwin RD, Whetton AD. 2006. Systematic proteome and transcriptome analysis of stem cell populations.
Cell Cycle 5:158791
105. Gerling IC, Singh S, Lenchik NI, Marshall DR, Wu J. 2006. New data analysis and mining approaches
identify unique proteome and transcriptome markers of susceptibility to autoimmune diabetes. Mol. Cell
Proteomics 5:293305
106. de Sousa Abreu R, Penalva LO, Marcotte EM, Vogel C. 2009. Global signatures of protein and mRNA
expression levels. Mol. Biosyst. 5:151226
107. Stamou P, Kontoyiannis DL. 2010. Posttranscriptional regulation of TNF mRNA: a paradigm of signal-
dependent mRNA utilization and its relevance to pathology. Curr. Dir. Autoimmun. 11:6179
108. Brem RB, Yvert G, Clinton R, Kruglyak L. 2002. Genetic dissection of transcriptional regulation in
budding yeast. Science 296:75255

109. DeRisi J, Penland L, Brown PO, Bittner ML, Meltzer PS, et al. 1996. Use of a cDNA microarray to
analyse gene expression patterns in human cancer. Nat. Genet. 14:45760
110. Lockhart DJ, Dong H, Byrne MC, Follettie MT, Gallo MV, et al. 1996. Expression monitoring by
hybridization to high-density oligonucleotide arrays. Nat. Biotechnol. 14:167580
111. Gardina PJ, Clark TA, Shimada B, Staples MK, Yang Q, et al. 2006. Alternative splicing and differential
gene expression in colon cancer detected by a whole genome exon array. BMC Genomics 7:325
112. Srinivasan K, Shiue L, Hayes JD, Centers R, Fitzwater S, et al. 2005. Detection and measurement of
alternative splicing using splicing-sensitive microarrays. Methods 37:34559
113. Uttamchandani M, Neo JL, Ong BN, Moochhala S. 2009. Applications of microarrays in pathogen
detection and biodefence. Trends Biotechnol. 27:5361
114. Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, et al. 1999. Molecular classication of
cancer: class discovery and class prediction by gene expression monitoring. Science 286:53137
115. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, et al. 2000. Distinct types of diffuse large B-cell
lymphoma identied by gene expression proling. Nature 403:50311
116. Winnepenninckx V, Lazar V, Michiels S, Dessen P, Stas M, et al. 2006. Gene expression proling of
primary cutaneous melanoma and clinical outcome. J. Natl. Cancer Inst. 98:47282
117. Boldrick JC, Alizadeh AA, Diehn M, Dudoit S, Liu CL, et al. 2002. Stereotyped and specic gene
expression programs in human innate immune responses to bacteria. Proc. Natl. Acad. Sci. USA 99:972
77
118. Chaussabel D, Semnani RT, McDowell MA, Sacks D, Sher A, Nutman TB. 2003. Unique gene expression
proles of human macrophages and dendritic cells to phylogenetically distinct parasites. Blood 102:67281
119. Huang Q, Liu D, Majewski P, Schulte LC, Korn JM, et al. 2001. The plasticity of dendritic cell responses
to pathogens and their components. Science 294:87075
120. Nau GJ, Richmond JF, Schlesinger A, Jennings EG, Lander ES, Young RA. 2002. Human macrophage
activation programs induced by bacterial pathogens. Proc. Natl. Acad. Sci. USA 99:15038
121. Batliwalla FM, Baechler EC, Xiao X, Li W, Balasubramanian S, et al. 2005. Peripheral blood gene
expression proling in rheumatoid arthritis. Genes Immun. 6:38897
122. Batliwalla FM, Li W, Ritchlin CT, Xiao X, Brenner M, et al. 2005. Microarray analyses of peripheral
blood cells identies unique gene expression signature in psoriatic arthritis. Mol. Med. 11:2129
123. Bennett L, Palucka AK, Arce E, Cantrell V, Borvak J, et al. 2003. Interferon and granulopoiesis signatures
in systemic lupus erythematosus blood. J. Exp. Med. 197:71123
124. Bomprezzi R, Ringner M, Kim S, Bittner ML, Khan J, et al. 2003. Gene expression prole in multi-
ple sclerosis patients and healthy controls: identifying pathways relevant to disease. Hum. Mol. Genet.
12:219199
125. Burczynski ME, Peterson RL, Twine NC, Zuberek KA, Brodeur BJ, et al. 2006. Molecular classica-
tion of Crohns disease and ulcerative colitis patients using transcriptional proles in peripheral blood
mononuclear cells. J. Mol. Diagn. 8:5161
576 Germain et al.

126. Pascual V, Allantaz F, Arce E, Punaro M, Banchereau J. 2005. Role of interleukin-1 (IL-1) in the
pathogenesis of systemic onset juvenile idiopathic arthritis and clinical response to IL-1 blockade.
J. Exp. Med. 201:147986
127. Lempicki RA, Polis MA, Yang J, McLaughlin M, Koratich C, et al. 2006. Gene expression proles in
hepatitis C virus (HCV) and HIV coinfection: class prediction analyses before treatment predict the
outcome of anti-HCV therapy among HIV-coinfected persons. J. Infect. Dis. 193:117277
128. Reghunathan R, Jayapal M, Hsu LY, Chng HH, Tai D, et al. 2005. Expression prole of immune
response genes in patients with Severe Acute Respiratory Syndrome. BMC Immunol. 6:2
129. Simmons CP, Popper S, Dolocek C, Chau TN, Grifths M, et al. 2007. Patterns of host genome-wide
gene transcript abundance in the peripheral blood of patients with acute dengue hemorrhagic fever.
J. Infect. Dis. 195:1097107
130. Zaas AK, Chen M, Varkey J, Veldman T, Hero AO 3rd, et al. 2009. Gene expression signatures diagnose
inuenza and other symptomatic respiratory viral infections in humans. Cell Host Microbe 6:20717
131. Ardura MI, Banchereau R, Mejias A, Di Pucchio T, Glaser C, et al. 2009. Enhanced monocyte response
and decreased central memory T cells in children with invasive Staphylococcus aureus infections. PLoS One
4:e5446
132. Jacobsen M, Repsilber D, Gutschmidt A, Neher A, Feldmann K, et al. 2007. Candidate biomarkers for
discrimination between infection and disease caused by Mycobacterium tuberculosis. J. Mol. Med. 85:61321
133. Thompson LJ, Dunstan SJ, Dolecek C, Perkins T, House D, et al. 2009. Transcriptional response in
the peripheral blood of patients infected with Salmonella enterica serovar Typhi. Proc. Natl. Acad. Sci. USA
106:2243338
134. Grifths MJ, Sha MJ, Popper SJ, Hemingway CA, Kortok MM, et al. 2005. Genomewide analysis of
the host response to malaria in Kenyan children. J. Infect. Dis. 191:1599611
135. Wang J, Ramakrishnan R, Tang Z, Fan W, Kluge A, et al. 2010. Quantifying EGFR alterations in the
lung cancer genome with nanouidic digital PCR arrays. Clin. Chem. 56:62332
136. Flagella M, Bui S, Zheng Z, Nguyen CT, Zhang A, et al. 2006. A multiplex branched DNA assay for
parallel quantitative gene expression proling. Anal. Biochem. 352:5060
137. Geiss GK, Bumgarner RE, Birditt B, Dahl T, Dowidar N, et al. 2008. Direct multiplexed measurement
of gene expression with color-coded probe pairs. Nat. Biotechnol. 26:31725
138. Margulies M, Egholm M, Altman WE, Attiya S, Bader JS, et al. 2005. Genome sequencing in microfab-
ricated high-density picolitre reactors. Nature 437:37680
139. Shendure J, Porreca GJ, Reppas NB, Lin X, McCutcheon JP, et al. 2005. Accurate multiplex polony
sequencing of an evolved bacterial genome. Science 309:172832
140. Wang Z, Gerstein M, Snyder M. 2009. RNA-Seq: a revolutionary tool for transcriptomics. Nat. Rev.
Genet. 10:5763
141. Hawkins RD, Hon GC, Ren B. 2010. Next-generation genomics: an integrative approach. Nat. Rev.
Genet. 11:47686
142. Park PJ. 2009. ChIP-seq: advantages and challenges of a maturing technology. Nat. Rev. Genet. 10:66980
143. Farnham PJ. 2009. Insights from genomic proling of transcription factors. Nat. Rev. Genet. 10:60516
144. Wang Z, Schones DE, Zhao K. 2009. Characterization of human epigenomes. Curr. Opin. Genet. Dev.
19:12734
145. Hinds DA, Stuve LL, Nilsen GB, Halperin E, Eskin E, et al. 2005. Whole-genome patterns of common
DNA variation in three human populations. Science 307:107279
146. Cookson W, Liang L, Abecasis G, Moffatt M, Lathrop M. 2009. Mapping complex disease traits with
global gene expression. Nat. Rev. Genet. 10:18494
147. Chen CZ, Li L, Lodish HF, Bartel DP. 2004. MicroRNAs modulate hematopoietic lineage differentia-
tion. Science 303:8386
148. Baltimore D, Boldin MP, OConnell RM, Rao DS, Taganov KD. 2008. MicroRNAs: new regulators of
immune cell development and function. Nat. Immunol. 9:83945
149. Bartel DP. 2009. MicroRNAs: target recognition and regulatory functions. Cell 136:21533
150. Landgraf P, Rusu M, Sheridan R, Sewer A, Iovino N, et al. 2007. A mammalian microRNA expression
atlas based on small RNA library sequencing. Cell 129:140114

151. Meister G, Landthaler M, Dorsett Y, Tuschl T. 2004. Sequence-specic inhibition of microRNA- and
siRNA-induced RNA silencing. RNA 10:54450
152. Krutzfeldt J, Rajewsky N, Braich R, Rajeev KG, Tuschl T, et al. 2005. Silencing of microRNAs in vivo
with antagomirs. Nature 438:68589
153. Karginov FV, Conaco C, Xuan Z, Schmidt BH, Parker JS, et al. 2007. A biochemical approach to
identifying microRNA targets. Proc. Natl. Acad. Sci. USA 104:1929196
154. Baek D, Villen J, Shin C, Camargo FD, Gygi SP, Bartel DP. 2008. The impact of microRNAs on protein
output. Nature 455:6471
155. Selbach M, Schwanhausser B, Thierfelder N, Fang Z, Khanin R, Rajewsky N. 2008. Widespread changes
in protein synthesis induced by microRNAs. Nature 455:5863
156. Taganov KD, Boldin MP, Chang KJ, Baltimore D. 2006. NF-B-dependent induction of microRNA
miR-146, an inhibitor targeted to signaling proteins of innate immune responses. Proc. Natl. Acad. Sci.
USA 103:1248186
157. Li QJ, Chau J, Ebert PJ, Sylvester G, Min H, et al. 2007. miR-181a is an intrinsic modulator of T cell
sensitivity and selection. Cell 129:14761

158. McCoy CE, Sheedy FJ, Qualls JE, Doyle SL, Quinn SR, et al. 2010. IL-10 inhibits miR-155 induction
by Toll-like receptors. J. Biol. Chem. 285:2049298
159. Friedman A, Perrimon N. 2007. Genetic screening for signal transduction in the era of network biology.
Cell 128:22531
160. Fraser ID, Germain RN. 2009. Navigating the network: signaling cross-talk in hematopoietic cells. Nat.
Immunol. 10:32731
161. Boutros M, Ahringer J. 2008. The art and design of genetic screens: RNA interference. Nat. Rev. Genet.
9:55466
162. Elbashir SM, Harborth J, Lendeckel W, Yalcin A, Weber K, Tuschl T. 2001. Duplexes of 21-nucleotide
RNAs mediate RNA interference in cultured mammalian cells. Nature 411:49498
163. Zeng Y, Cullen BR. 2003. Sequence requirements for micro RNA processing and function in human
cells. RNA 9:11223
164. Moffat J, Sabatini DM. 2006. Building mammalian signalling pathways with RNAi screens. Nat. Rev.
Mol. Cell Biol. 7:17787
165. Bidere N, Ngo VN, Lee J, Collins C, Zheng L, et al. 2009. Casein kinase 1 governs antigen-receptor-
induced NF-B activation and human lymphoma cell survival. Nature 458:9296
166. Davis RE, Ngo VN, Lenz G, Tolar P, Young RM, et al. 2010. Chronic active B-cell-receptor signalling
in diffuse large B-cell lymphoma. Nature 463:8892
167. Lam LT, Davis RE, Ngo VN, Lenz G, Wright G, et al. 2008. Compensatory IKKactivation of classical
NF-B signaling during IKK inhibition identied by an RNA interference sensitization screen. Proc.
168. Carralot JP, Kim TK, Lenseigne B, Boese AS, Sommer P, et al. 2009. Automated high-throughput
siRNA transfection in raw 264.7 macrophages: a case study for optimization procedure. J. Biomol. Screen.
14:15160
169. Collinet C, Stoter M, Bradshaw CR, Samusik N, Rink JC, et al. 2010. Systems survey of endocytosis by
multiparametric image analysis. Nature 464:24349
170. Neumann B, Walter T, Heriche JK, Bulkescher J, Ere H, et al. 2010. Phenotypic proling of the human
genome by time-lapse microscopy reveals cell division genes. Nature 464:72127
171. Neumann B, Held M, Liebel U, Ere H, Rogers P, et al. 2006. High-throughput RNAi screening by
time-lapse imaging of live human cells. Nat. Methods 3:38590
172. Fuchs F, Pau G, Kranz D, Sklyar O, Budjan C, et al. 2010. Clustering phenotype populations by genome-
wide RNAi and multiparametric imaging. Mol. Syst. Biol. 6:370
173. Birmingham A, Selfors LM, Forster T, Wrobel D, Kennedy CJ, et al. 2009. Statistical methods for
analysis of high-throughput RNA interference screens. Nat. Methods 6:56975
174. Austin CP, Brady LS, Insel TR, Collins FS. 2004. NIH Molecular Libraries Initiative. Science 306:1138
39
175. Birmingham A, Anderson EM, Reynolds A, Ilsley-Tyree D, Leake D, et al. 2006. 3 UTR seed matches,
but not overall identity, are associated with RNAi off-targets. Nat. Methods 3:199204
578 Germain et al.

176. Jackson AL, Burchard J, Leake D, Reynolds A, Schelter J, et al. 2006. Position-specic chemical modi-
cation of siRNAs reduces off-target transcript silencing. RNA 12:1197205
177. Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, et al. 2004. Bioconductor: open software
development for computational biology and bioinformatics. Genome Biol. 5:R80
178. Pelz O, Gilsdorf M, Boutros M. 2010. web cellHTS2: a web-application for the analysis of high-
throughput screening data. BMC Bioinforma. 11:185
179. Rieber N, Knapp B, Eils R, Kaderali L. 2009. RNAither, an automated pipeline for the statistical analysis
of high-throughput RNAi screens. Bioinformatics 25:67879
180. Major MB, Roberts BS, Berndt JD, Marine S, Anastas J, et al. 2008. New regulators of Wnt/-catenin
signaling revealed by integrative molecular screening. Sci. Signal. 1:ra12
181. Cherry S. 2008. Genomic RNAi screening in Drosophila S2 cells: What have we learned about host-
pathogen interactions? Curr. Opin. Microbiol. 11:26270
182. Bushman FD, Malani N, Fernandes J, DOrso I, Cagney G, et al. 2009. Host cell factors in HIV
replication: meta-analysis of genome-wide studies. PLoS Pathog. 5:e1000437
183. Watanabe T, Watanabe S, Kawaoka Y. 2010. Cellular networks involved in the inuenza virus life cycle.
Cell Host Microbe 7:42739

184. Gwack Y, Srikanth S, Feske S, Cruz-Guilloty F, Oh-hora M, et al. 2007. Biochemical and functional
characterization of Orai proteins. J. Biol. Chem. 282:1623243
185. Zhang SL, Yeromin AV, Zhang XH, Yu Y, Safrina O, et al. 2006. Genome-wide RNAi screen of Ca2+
inux identies genes that regulate Ca2+ release-activated Ca2+ channel activity. Proc. Natl. Acad. Sci.
USA 103:935762
186. Vig M, Peinelt C, Beck A, Koomoa DL, Rabah D, et al. 2006. CRACM1 is a plasma membrane protein
essential for store-operated Ca2+ entry. Science 312:122023
187. Liou J, Kim ML, Heo WD, Jones JT, Myers JW, et al. 2005. STIM is a Ca2+ sensor essential for
Ca2+ -store-depletion-triggered Ca2+ inux. Curr. Biol. 15:123541
188. Roos J, DiGregorio PJ, Yeromin AV, Ohlsen K, Lioudyno M, et al. 2005. STIM1, an essential and
conserved component of store-operated Ca2+ channel function. J. Cell Biol. 169:43545
189. Willingham AT, Orth AP, Batalov S, Peters EC, Wen BG, et al. 2005. A strategy for probing the function
of noncoding RNAs nds a repressor of NFAT. Science 309:157073
190. Gwack Y, Sharma S, Nardone J, Tanasa B, Iuga A, et al. 2006. A genome-wide Drosophila RNAi screen
identies DYRK-family kinases as regulators of NFAT. Nature 441:64650
191. Kuijl C, Savage ND, Marsman M, Tuin AW, Janssen L, et al. 2007. Intracellular bacterial growth is
controlled by a kinase network around PKB/AKT1. Nature 450:72530
192. Juszczynski P, Ouyang J, Monti S, Rodig SJ, Takeyama K, et al. 2007. The AP1-dependent secretion of
galectin-1 by Reed Sternberg cells fosters immune privilege in classical Hodgkin lymphoma. Proc. Natl.
Acad. Sci. USA 104:1313439
193. Shaffer AL, Emre NC, Lamy L, Ngo VN, Wright G, et al. 2008. IRF4 addiction in multiple myeloma.
Nature 454:22631
194. Cho YS, Challa S, Moquin D, Genga R, Ray TD, et al. 2009. Phosphorylation-driven assembly of the
RIP1-RIP3 complex regulates programmed necrosis and virus-induced inammation. Cell 137:111223
195. Burckstummer T, Baumann C, Bluml S, Dixit E, Durnberger G, et al. 2009. An orthogonal proteomic-
genomic screen identies AIM2 as a cytoplasmic DNA sensor for the inammasome. Nat. Immunol.
10:26672
196. Bauer A, Kuster B. 2003. Afnity puricationmass spectrometry: powerful tools for the characterization
of protein complexes. Eur. J. Biochem. 270:57078
197. Michnick SW, Ear PH, Manderson EN, Remy I, Stefan E. 2007. Universal strategies in research and
drug discovery based on protein-fragment complementation assays. Nat. Rev. Drug Discov. 6:56982
198. Parrish JR, Gulyas KD, Finley RL Jr. 2006. Yeast two-hybrid contributions to interactome mapping.
Curr. Opin. Biotechnol. 17:38793
199. Cusick ME, Yu H, Smolyar A, Venkatesan K, Carvunis AR, et al. 2009. Literature-curated protein
interaction datasets. Nat. Methods 6:3946
200. Stumpf MP, Thorne T, de Silva E, Stewart R, An HJ, et al. 2008. Estimating the size of the human
interactome. Proc. Natl. Acad. Sci. USA 105:695964

201. Fields S, Song O. 1989. A novel genetic system to detect protein-protein interactions. Nature 340:24546
202. Walhout AJ, Vidal M. 2001. High-throughput yeast two-hybrid assays for large-scale protein interaction
mapping. Methods 24:297306
203. Ito T, Chiba T, Ozawa R, Yoshida M, Hattori M, Sakaki Y. 2001. A comprehensive two-hybrid analysis
to explore the yeast protein interactome. Proc. Natl. Acad. Sci. USA 98:456974
204. Uetz P, Giot L, Cagney G, Manseld TA, Judson RS, et al. 2000. A comprehensive analysis of protein-
protein interactions in Saccharomyces cerevisiae. Nature 403:62327
205. LaCount DJ, Vignali M, Chettier R, Phansalkar A, Bell R, et al. 2005. A protein interaction network of
the malaria parasite Plasmodium falciparum. Nature 438:1037
206. Li S, Armstrong CM, Bertin N, Ge H, Milstein S, et al. 2004. A map of the interactome network of the
metazoan C. elegans. Science 303:54043
207. Giot L, Bader JS, Brouwer C, Chaudhuri A, Kuang B, et al. 2003. A protein interaction map of Drosophila
melanogaster. Science 302:172736
208. Rual JF, Venkatesan K, Hao T, Hirozane-Kishikawa T, Dricot A, et al. 2005. Towards a proteome-scale
map of the human protein-protein interaction network. Nature 437:117378

209. Stelzl U, Worm U, Lalowski M, Haenig C, Brembeck FH, et al. 2005. A human protein-protein inter-
action network: a resource for annotating the proteome. Cell 122:95768
210. Venkatesan K, Rual JF, Vazquez A, Stelzl U, Lemmens I, et al. 2009. An empirical framework for binary
interactome mapping. Nat. Methods 6:8390
211. Gavin AC, Aloy P, Grandi P, Krause R, Boesche M, et al. 2006. Proteome survey reveals modularity of
the yeast cell machinery. Nature 440:63136
212. Krogan NJ, Cagney G, Yu H, Zhong G, Guo X, et al. 2006. Global landscape of protein complexes in
the yeast Saccharomyces cerevisiae. Nature 440:63743
213. Poser I, Sarov M, Hutchins JR, Heriche JK, Toyoda Y, et al. 2008. BAC TransgeneOmics: a high-
throughput method for exploration of protein function in mammals. Nat. Methods 5:40915
214. Hutchins JR, Toyoda Y, Hegemann B, Poser I, Heriche JK, et al. 2010. Systematic analysis of human
protein complexes identies chromosome segregation proteins. Science 328:59399
215. Barrios-Rodiles M, Brown KR, Ozdamar B, Bose R, Liu Z, et al. 2005. High-throughput mapping of a
dynamic signaling network in mammalian cells. Science 307:162125
216. Bouwmeester T, Bauch A, Ruffner H, Angrand PO, Bergamini G, et al. 2004. A physical and functional
map of the human TNF-/NF-B signal transduction pathway. Nat. Cell Biol. 6:97105
217. Ewing RM, Chu P, Elisma F, Li H, Taylor P, et al. 2007. Large-scale mapping of human protein-protein
interactions by mass spectrometry. Mol. Syst. Biol. 3:89
218. Tarassov K, Messier V, Landry CR, Radinovic S, Serna Molina MM, et al. 2008. An in vivo map of the
yeast protein interactome. Science 320:146570
219. Calderwood MA, Venkatesan K, Xing L, Chase MR, Vazquez A, et al. 2007. Epstein-Barr virus and virus
human protein interaction maps. Proc. Natl. Acad. Sci. USA 104:760611
220. Lee JH, Vittone V, Diefenbach E, Cunningham AL, Diefenbach RJ. 2008. Identication of structural
protein-protein interactions of herpes simplex virus type 1. Virology 378:34754
221. Rozen R, Sathish N, Li Y, Yuan Y. 2008. Virion-wide protein interactions of Kaposis sarcoma-associated
herpesvirus. J. Virol. 82:474250
222. Uetz P, Dong YA, Zeretzke C, Atzler C, Baiker A, et al. 2006. Herpesviral protein networks and their
interaction with the human proteome. Science 311:23942
223. Fossum E, Friedel CC, Rajagopala SV, Titz B, Baiker A, et al. 2009. Evolutionarily conserved herpesviral
protein interaction networks. PLoS Pathog. 5:e1000570
224. de Chassey B, Navratil V, Tafforeau L, Hiet MS, Aublin-Gex A, et al. 2008. Hepatitis C virus infection
protein network. Mol. Syst. Biol. 4:230
225. Shapira SD, Gat-Viks I, Shum BO, Dricot A, de Grace MM, et al. 2009. A physical and regulatory map
of host-inuenza interactions reveals pathways in H1N1 infection. Cell 139:125567
226. Dyer MD, Murali TM, Sobral BW. 2008. The landscape of human proteins interacting with viruses and
other pathogens. PLoS Pathog. 4:e32
227. Dyer MD, Neff C, Dufford M, Rivera CG, Shattuck D, et al. 2010. The human-bacterial pathogen protein
interaction networks of Bacillus anthracis, Francisella tularensis, and Yersinia pestis. PLoS One 5:e12089
580 Germain et al.

228. Fenn JB, Mann M, Meng CK, Wong SF, Whitehouse CM. 1989. Electrospray ionization for mass
spectrometry of large biomolecules. Science 246:6471
229. Karas M, Hillenkamp F. 1988. Laser desorption ionization of proteins with molecular masses exceeding
10,000 Daltons. Anal. Chem. 60:2299301
230. Han X, Aslanian A, Yates JR 3rd. 2008. Mass spectrometry for proteomics. Curr. Opin. Chem. Biol.
12:48390
231. Makarov A. 2000. Electrostatic axially harmonic orbital trapping: a high-performance technique of mass
analysis. Anal. Chem. 72:115662
232. Makarov A, Denisov E, Kholomeev A, Balschun W, Lange O, et al. 2006. Performance evaluation of a
hybrid linear ion trap/orbitrap mass spectrometer. Anal. Chem. 78:211320
233. Makarov A, Denisov E, Lange O, Horning S. 2006. Dynamic range of mass accuracy in LTQ Orbitrap
hybrid mass spectrometer. J. Am. Soc. Mass. Spectrom. 17:97782
234. Yates JR, Cociorva D, Liao L, Zabrouskov V. 2006. Performance of a linear ion trapOrbitrap hybrid
for peptide analysis. Anal. Chem. 78:493500
235. Macek B, Waanders LF, Olsen JV, Mann M. 2006. Top-down protein sequencing and MS3 on a hybrid
linear quadrupole ion trapOrbitrap mass spectrometer. Mol. Cell Proteomics 5:94958
236. Venable JD, Wohlschlegel J, McClatchy DB, Park SK, Yates JR 3rd. 2007. Relative quantication of
stable isotope labeled peptides using a linear ion trapOrbitrap hybrid mass spectrometer. Anal. Chem.
79:305664
237. Luber CA, Cox J, Lauterbach H, Fancke B, Selbach M, et al. 2010. Quantitative proteomics reveals
subset-specic viral recognition in dendritic cells. Immunity 32:27989
238. Zubarev RA, Horn DM, Fridriksson EK, Kelleher NL, Kruger NA, et al. 2000. Electron capture disso-
ciation for structural characterization of multiply charged protein cations. Anal. Chem. 72:56373
239. Syka JE, Coon JJ, Schroeder MJ, Shabanowitz J, Hunt DF. 2004. Peptide and protein sequence analysis
by electron transfer dissociation mass spectrometry. Proc. Natl. Acad. Sci. USA 101:952833
240. Olsen JV, Macek B, Lange O, Makarov A, Horning S, Mann M. 2007. Higher-energy C-trap dissociation
for peptide modication analysis. Nat. Methods 4:70912
241. Wolf-Yadlin A, Hautaniemi S, Lauffenburger DA, White FM. 2007. Multiple reaction monitoring for
robust quantitative proteomic analysis of cellular signaling networks. Proc. Natl. Acad. Sci. USA 104:5860
65
242. Ong SE, Blagoev B, Kratchmarova I, Kristensen DB, Steen H, et al. 2002. Stable isotope labeling by
amino acids in cell culture, SILAC, as a simple and accurate approach to expression proteomics. Mol.
Cell Proteomics 1:37686
243. Ross PL, Huang YN, Marchese JN, Williamson B, Parker K, et al. 2004. Multiplexed protein quantitation
in Saccharomyces cerevisiae using amine-reactive isobaric tagging reagents. Mol. Cell Proteomics 3:115469
244. Choe L, DAscenzo M, Relkin NR, Pappin D, Ross P, et al. 2007. 8-Plex quantitation of changes in
cerebrospinal uid protein expression in subjects undergoing intravenous immunoglobulin treatment
for Alzheimers disease. Proteomics 7:365160
245. Kruger M, Moser M, Ussar S, Thievessen I, Luber CA, et al. 2008. SILAC mouse for quantitative
proteomics uncovers kindlin-3 as an essential factor for red blood cell function. Cell 134:35364
246. Picotti P, Bodenmiller B, Mueller LN, Domon B, Aebersold R. 2009. Full dynamic range proteome
analysis of S. cerevisiae by targeted proteomics. Cell 138:795806
247. Picotti P, Rinner O, Stallmach R, Dautel F, Farrah T, et al. 2009. High-throughput generation of
selected reaction-monitoring assays for proteins and proteomes. Nat. Methods 7:4346
248. Vogel C, Marcotte EM. 2009. Absolute abundance for the masses. Nat. Biotechnol. 27:82526
249. Kirkpatrick DS, Gerber SA, Gygi SP. 2005. The absolute quantication strategy: a general procedure
for the quantication of proteins and post-translational modications. Methods 35:26573
250. Pratt JM, Simpson DM, Doherty MK, Rivers J, Gaskell SJ, Beynon RJ. 2006. Multiplexed absolute
quantication for proteomics using concatenated signature peptides encoded by QconCAT genes. Nat.
Protoc. 1:102943
251. Mirzaei H, McBee JK, Watts J, Aebersold R. 2008. Comparative evaluation of current peptide production
platforms used in absolute quantication in proteomics. Mol. Cell Proteomics 7:81323

252. Holzmann J, Pichler P, Madalinski M, Kurzbauer R, Mechtler K. 2009. Stoichiometry determination

of the MP1-p14 complex using a novel and cost-efcient method to produce an equimolar mixture of
standard peptides. Anal. Chem. 81:1025461
253. Koal T, Deigner HP. 2010. Challenges in mass spectrometry based targeted metabolomics. Curr. Mol.
Med. 10:21626
254. Iwai LK, Benoist C, Mathis D, White FM. 2010. Quantitative phosphoproteomic analysis of T cell
receptor signaling in diabetes prone and resistant mice. J. Proteome Res. 9:313545
255. ONeill RA, Bhamidipati A, Bi X, Deb-Basu D, Cahill L, et al. 2006. Isoelectric focusing technology
quanties protein signaling in 25 cells. Proc. Natl. Acad. Sci. USA 103:1615358
256. Borrebaeck CA, Ekstrom S, Hager AC, Nilsson J, Laurell T, Marko-Varga G. 2001. Protein chips based
on recombinant antibody fragments: a highly sensitive approach as detected by mass spectrometry.
Biotechniques 30:112630, 1132
257. Zhu H, Snyder M. 2003. Protein chip technology. Curr. Opin. Chem. Biol. 7:5563
258. Ramachandran N, Hainsworth E, Bhullar B, Eisenstein S, Rosen B, et al. 2004. Self-assembling protein
microarrays. Science 305:8690

259. Tao SC, Zhu H. 2006. Protein chip fabrication by capture of nascent polypeptides. Nat. Biotechnol.
24:125354
260. Irish JM, Kotecha N, Nolan GP. 2006. Mapping normal and cancer cell signalling networks: towards
single-cell proteomics. Nat. Rev. Cancer 6:14655
261. Krutzik PO, Nolan GP. 2003. Intracellular phospho-protein staining techniques for ow cytometry:
monitoring single cell signaling events. Cytometry A 55:6170
262. Krutzik PO, Nolan GP. 2006. Fluorescent cell barcoding in ow cytometry allows high-throughput
drug screening and signaling proling. Nat. Methods 3:36168
263. Goddard G, Martin JC, Naivar M, Goodwin PM, Graves SW, et al. 2006. Single particle high resolution
spectral analysis ow cytometry. Cytometry A 69:84251
264. Shachaf CM, Elchuri SV, Koh AL, Zhu J, Nguyen LN, et al. 2009. A novel method for detection of
phosphorylation in single cells by surface enhanced Raman scattering (SERS) using composite organic-
inorganic nanoparticles (COINs). PLoS One 4:e5206
265. Watson DA, Brown LO, Gaskill DF, Naivar M, Graves SW, et al. 2008. A ow cytometer for the
measurement of Raman spectra. Cytometry A 73:11928
266. Bandura DR, Baranov VI, Ornatsky OI, Antonov A, Kinach R, et al. 2009. Mass cytometry: technique
for real time single cell multitarget immunoassay based on inductively coupled plasma time-of-ight
mass spectrometry. Anal. Chem. 81:681322
267. Coggan JS, Bartol TM, Esquenazi E, Stiles JR, Lamont S, et al. 2005. Evidence for ectopic neurotrans-
mission at a neuronal synapse. Science 309:44651
268. Shoop RD, Esquenazi E, Yamada N, Ellisman MH, Berg DK. 2002. Ultrastructure of a somatic spine
mat for nicotinic signaling in neurons. J. Neurosci. 22:74856
269. Smith CS, Joseph N, Rieger B, Lidke KA. 2010. Fast, single-molecule localization that achieves theo-
retically minimum uncertainty. Nat. Methods 7:37375
270. Sohn HW, Tolar P, Brzostowski J, Pierce SK. 2010. A method for analyzing protein-protein interactions
in the plasma membrane of live B cells by uorescence resonance energy transfer imaging as acquired
by total internal reection uorescence microscopy. Methods Mol. Biol. 591:15983
271. Ding JB, Takasaki KT, Sabatini BL. 2009. Supraresolution imaging in brain slices using stimulated-
emission depletion two-photon laser scanning microscopy. Neuron 63:42937
272. Park C, Hwang IY, Kehrl JH. 2009. Intravital two-photon imaging of adoptively transferred B lympho-
cytes in inguinal lymph nodes. Methods Mol. Biol. 571:199207
273. Bajenoff M, Germain RN. 2007. Seeing is believing: a focus on the contribution of microscopic imaging
to our understanding of immune system function. Eur. J. Immunol. 37(Suppl. 1):S1833
274. Cabantous S, Terwilliger TC, Waldo GS. 2005. Protein tagging and detection with engineered self-
assembling fragments of green uorescent protein. Nat. Biotechnol. 23:1027
275. Gerber D, Maerkl SJ, Quake SR. 2009. An in vitro microuidic approach to generating protein-
interaction networks. Nat. Methods 6:7174
582 Germain et al.

276. Tay S, Hughey JJ, Lee TK, Lipniacki T, Quake SR, Covert MW. 2010. Single-cell NF-B dynamics
reveal digital activation and analogue information processing. Nature 466:26771
277. Kim HD, Guo TW, Wu AP, Wells A, Gertler FB, Lauffenburger DA. 2008. Epidermal growth factor
induced enhancement of glioblastoma cell migration in 3D arises from an intrinsic increase in speed but
an extrinsic matrix- and proteolysis-dependent increase in persistence. Mol. Biol. Cell 19:424959
278. Denk W, Strickler JH, Webb WW. 1990. Two-photon laser scanning uorescence microscopy. Science
248:7376
279. Betzig E, Patterson GH, Sougrat R, Lindwasser OW, Olenych S, et al. 2006. Imaging intracellular
uorescent proteins at nanometer resolution. Science 313:164245
280. Rust MJ, Bates M, Zhuang X. 2006. Sub-diffraction-limit imaging by stochastic optical reconstruction
microscopy (STORM). Nat. Methods 3:79395
281. Fuchs J, Bohme S, Oswald F, Hedde PN, Krause M, et al. 2010. A photoactivatable marker protein for
pulse-chase imaging with superresolution. Nat. Methods 7:62730
282. Sarris M, Betz AG. 2009. Shine a light: imaging the immune system. Eur. J. Immunol. 39:1188202
283. Monks CR, Freiberg BA, Kupfer H, Sciaky N, Kupfer A. 1998. Three-dimensional segregation of
supramolecular activation clusters in T cells. Nature 395:8286
284. Lee KH, Holdorf AD, Dustin ML, Chan AC, Allen PM, Shaw AS. 2002. T cell receptor signaling
precedes immunological synapse formation. Science 295:153942
285. Cemerski S, Das J, Giurisato E, Markiewicz MA, Allen PM, et al. 2008. The balance between T cell
receptor signaling and degradation at the center of the immunological synapse is determined by antigen
quality. Immunity 29:41422
286. Cemerski S, Das J, Locasale J, Arnold P, Giurisato E, et al. 2007. The stimulatory potency of T cell
antigens is inuenced by the formation of the immunological synapse. Immunity 26:34555
287. Beltman JB, Maree AF, Lynch JN, Miller MJ, de Boer RJ. 2007. Lymph node topology dictates T cell
migration behavior. J. Exp. Med. 204:77180
288. Henrickson SE, Mempel TR, Mazo IB, Liu B, Artyomov MN, et al. 2008. In vivo imaging of T cell
priming. Sci. Signal. 1:pt2
289. Henrickson SE, Mempel TR, Mazo IB, Liu B, Artyomov MN, et al. 2008. T cell sensing of antigen dose
governs interactive behavior with dendritic cells and sets a threshold for T cell activation. Nat. Immunol.
9:28291
290. Zheng H, Jin B, Henrickson SE, Perelson AS, von Andrian UH, Chakraborty AK. 2008. How antigen
quantity and quality determine T-cell decisions in lymphoid tissue. Mol. Cell. Biol. 28:404051
291. Cahalan MD, Parker I. 2008. Choreography of cell motility and interaction dynamics imaged by two-
photon microscopy in lymphoid organs. Annu. Rev. Immunol. 26:585626
292. Hauser AE, Junt T, Mempel TR, Sneddon MW, Kleinstein SH, et al. 2007. Denition of germinal-center
B cell migration in vivo reveals predominant intrazonal circulation patterns. Immunity 26:65567
293. Yachi PP, Ampudia J, Zal T, Gascoigne NR. 2006. Altered peptide ligands induce delayed CD8-T cell
receptor interaction: a role for CD8 in distinguishing antigen quality. Immunity 25:20311
294. Mallaun M, Naeher D, Daniels MA, Yachi PP, Hausmann B, et al. 2008. The T cell receptors -chain
connecting peptide motif promotes close approximation of the CD8 coreceptor allowing efcient signal
initiation. J. Immunol. 180:821121
295. Tolar P, Sohn HW, Pierce SK. 2005. The initiation of antigen-induced B cell antigen receptor signaling
viewed in living cells by uorescence resonance energy transfer. Nat. Immunol. 6:116876
296. Aslan K, Lakowicz JR, Geddes CD. 2005. Plasmon light scattering in biology and medicine: new sensing
approaches, visions and perspectives. Curr. Opin. Chem. Biol. 9:53844
297. Segura E, Kapp E, Gupta N, Wong J, Lim J, et al. 2010. Differential expression of pathogen-recognition
molecules between dendritic cell subsets revealed by plasma membrane proteomic analysis. Mol. Immunol.
47:176573
298. Kim JE, White FM. 2006. Quantitative analysis of phosphotyrosine signaling networks triggered by
CD3 and CD28 costimulation in Jurkat cells. J. Immunol. 176:283343
299. Hale MB, Krutzik PO, Samra SS, Crane JM, Nolan GP. 2009. Stage dependent aberrant regulation of
cytokine-STAT signaling in murine systemic lupus erythematosus. PLoS One 4:e6756

300. Galligan CL, Siebert JC, Siminovitch KA, Keystone EC, Bykerk V, et al. 2009. Multiparameter phospho-
ow analysis of lymphocytes in early rheumatoid arthritis: implications for diagnosis and monitoring drug
therapy. PLoS One 4:e6703
301. Merbl Y, Itzchak R, Vider-Shalit T, Louzoun Y, Quintana FJ, et al. 2009. A systems immunology
approach to the host-tumor interaction: large-scale patterns of natural autoantibodies distinguish healthy
and tumor-bearing mice. PLoS One 4:e6053
302. Montor WR, Huang J, Hu Y, Hainsworth E, Lynch S, et al. 2009. Genome-wide study of Pseudomonas
aeruginosa outer membrane protein immunogenicity using self-assembling protein microarrays. Infect.
Immun. 77:487786
303. Ceroni A, Sibani S, Baiker A, Pothineni VR, Bailer SM, et al. 2010. Systematic analysis of the IgG antibody
immune response against varicella zoster virus (VZV) using a self-assembled protein microarray. Mol.
Biosyst. 6:160410
304. Robinson WH, DiGennaro C, Hueber W, Haab BB, Kamachi M, et al. 2002. Autoantigen microarrays
for multiplex characterization of autoantibody responses. Nat. Med. 8:295301
305. Harwanegg C, Laffer S, Hiller R, Mueller MW, Kraft D, et al. 2003. Microarrayed recombinant allergens
for diagnosis of allergy. Clin. Exp. Allergy 33:713

306. Hiller R, Laffer S, Harwanegg C, Huber M, Schmidt WM, et al. 2002. Microarrayed allergen molecules:
diagnostic gatekeepers for allergy treatment. FASEB J. 16:41416
307. Miletic AV, Sakata-Sogawa K, Hiroshima M, Hamann MJ, Gomez TS, et al. 2006. Vav1 acidic region
tyrosine 174 is required for the formation of T cell receptor-induced microclusters and is essential in T
cell development and activation. J. Biol. Chem. 281:3825765
308. Nolz JC, Gomez TS, Billadeau DD. 2005. The Ezh2 methyltransferase complex: actin up in the cytosol.
Trends Cell Biol. 15:51417
309. Nolz JC, Gomez TS, Zhu P, Li S, Medeiros RB, et al. 2006. The WAVE2 complex regulates actin
cytoskeletal reorganization and CRAC-mediated calcium entry during T cell activation. Curr. Biol.
16:2434
310. Spooner CJ, Cheng JX, Pujadas E, Laslo P, Singh H. 2009. A recurrent network involving the transcrip-
tion factors PU.1 and G1 orchestrates innate and adaptive immune cell fates. Immunity 31:57686
311. Laslo P, Spooner CJ, Warmash A, Lancki DW, Lee HJ, et al. 2006. Multilineage transcriptional priming
and determination of alternate hematopoietic cell fates. Cell 126:75566
312. Georgescu C, Longabaugh WJ, Scripture-Adams DD, David-Fung ES, Yui MA, et al. 2008. A gene
regulatory network armature for T lymphocyte specication. Proc. Natl. Acad. Sci. USA 105:201005
313. Monteiro JP, Farache J, Mercadante AC, Mignaco JA, Bonamino M, Bonomo A. 2008. Pathogenic
effector T cell enrichment overcomes regulatory T cell control and generates autoimmune gastritis.
J. Immunol. 181:5895903
314. Andrews SS, Addy NJ, Brent R, Arkin AP. 2010. Detailed simulations of cell biology with Smoldyn 2.1.
PLoS Comput. Biol. 6:e1000705
315. Plimpton SJ, Slepoy A. 2005. Microbial cell modeling via reacting diffusing particles. J. Phys. Conf. Ser.
16:3059
316. Brass AL, Huang IC, Benita Y, John SP, Krishnan MN, et al. 2009. The IFITM proteins mediate cellular
resistance to inuenza A H1N1 virus, West Nile virus, and dengue virus. Cell 139:124354
317. Konig R, Stertz S, Zhou Y, Inoue A, Hoffmann HH, et al. 2010. Human host factors required for
inuenza virus replication. Nature 463:81317
318. Karlas A, Machuy N, Shin Y, Pleissner KP, Artarini A, et al. 2010. Genome-wide RNAi screen identies
human host factors crucial for inuenza virus replication. Nature 463:81822
319. Brass AL, Dykxhoorn DM, Benita Y, Yan N, Engelman A, et al. 2008. Identication of host proteins
required for HIV infection through a functional genomic screen. Science 319:92126
320. Konig R, Zhou Y, Elleder D, Diamond TL, Bonamy GM, et al. 2008. Global analysis of host-pathogen
interactions that regulate early-stage HIV-1 replication. Cell 135:4960
321. Zhou H, Xu M, Huang Q, Gates AT, Zhang XD, et al. 2008. Genome-scale RNAi screen for host factors
required for HIV replication. Cell Host Microbe 4:495504
322. Krishnan MN, Ng A, Sukumaran B, Gilfoy FD, Uchil PD, et al. 2008. RNA interference screen for
human genes associated with West Nile virus infection. Nature 455:24245
584 Germain et al.

323. Kumar D, Nath L, Kamal MA, Varshney A, Jain A, et al. 2010. Genome-wide analysis of the host
intracellular network that regulates survival of Mycobacterium tuberculosis. Cell 140:73143
324. Jayaswal S, Kamal MA, Dua R, Gupta S, Majumdar T, et al. 2010. Identication of host-dependent
survival factors for intracellular Mycobacterium tuberculosis through an siRNA screen. PLoS Pathog.
6:e1000839
325. Ng TI, Mo H, Pilot-Matias T, He Y, Koev G, et al. 2007. Identication of host genes involved in
hepatitis C virus replication by small interfering RNA technology. Hepatology 45:141321
326. Tai AW, Benita Y, Peng LF, Kim SS, Sakamoto N, et al. 2009. A functional genomic screen identies
cellular cofactors of hepatitis C virus replication. Cell Host Microbe 5:298307
327. Gerber SA, Rush J, Stemman O, Kirschner MW, Gygi SP. 2003. Absolute quantication of proteins and
phosphoproteins from cell lysates by tandem MS. Proc. Natl. Acad. Sci. USA 100:694045
328. Foster LJ, de Hoog CL, Zhang Y, Xie X, Mootha VK, Mann M. 2006. A mammalian organelle map by
protein correlation proling. Cell 125:18799
329. Mallick P, Schirle M, Chen SS, Flory MR, Lee H, et al. 2007. Computational prediction of proteotypic
peptides for quantitative proteomics. Nat. Biotechnol. 25:12531


IY29-Frontmatter ARI 4 February 2011 21:56
Annual Review of
Immunology
Contents Volume 29, 2011
Innate Antifungal Immunity: The Key Role of Phagocytes

Gordon D. Brown p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Stromal CellImmune Cell Interactions
Ramon Roozendaal and Reina E. Mebius p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p23

Nonredundant Roles of Basophils in Immunity
Hajime Karasuyama, Kaori Mukai, Kazushige Obata, Yusuke Tsujimura,
and Takeshi Wada p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p45
Regulation and Functions of IL-10 Family of Cytokines
in Inammation and Disease
Wenjun Ouyang, Sascha Rutz, Natasha K. Crellin, Patricia A. Valdez,
and Sarah G. Hymowitz p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p71
Prevention and Treatment of Papillomavirus-Related Cancers
Through Immunization
Ian H. Frazer, Graham R. Leggatt, and Stephen R. Mattarollo p p p p p p p p p p p p p p p p p p p p p p p p 111
HMGB1 Is a Therapeutic Target for Sterile Inammation and Infection
Ulf Andersson and Kevin J. Tracey p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 139
Plasmacytoid Dendritic Cells: Recent Progress and Open Questions
Boris Reizis, Anna Bunin, Hiyaa S. Ghosh, Kanako L. Lewis, and Vanja Sisirak p p p p p 163
Nucleic Acid Recognition by the Innate Immune System
Roman Barbalat, Sarah E. Ewald, Maria L. Mouchess, and Gregory M. Barton p p p p p p 185
Trafcking of B Cell Antigen in Lymph Nodes
Santiago F. Gonzalez, Sren E. Degn, Lisa A. Pitcher, Matthew Woodruff,
Balthasar A. Heesters, and Michael C. Carroll p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 215
Natural Innate and Adaptive Immunity to Cancer
Matthew D. Vesely, Michael H. Kershaw, Robert D. Schreiber,
and Mark J. Smyth p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 235
Immunoglobulin Responses at the Mucosal Interface
Andrea Cerutti, Kang Chen, and Alejo Chorny p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 273
HLA/KIR Restraint of HIV: Surviving the Fittest
Arman A. Bashirova, Rasmi Thomas, and Mary Carrington p p p p p p p p p p p p p p p p p p p p p p p p p p p 295
v
IY29-Frontmatter ARI 4 February 2011 21:56
Mechanisms that Promote and Suppress Chromosomal Translocations

in Lymphocytes
Monica Gostissa, Frederick W. Alt, and Roberto Chiarle p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 319
Pathogenesis and Host Control of Gammaherpesviruses: Lessons from
the Mouse
Erik Barton, Pratyusha Mandal, and Samuel H. Speck p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 351
Genetic Defects in Severe Congenital Neutropenia: Emerging Insights into
Life and Death of Human Neutrophil Granulocytes
Christoph Klein p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 399
Inammatory Mechanisms in Obesity
Margaret F. Gregor and Gokhan S. Hotamisligil p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 415
Human TLRs and IL-1Rs in Host Defense: Natural Insights

from Evolutionary, Epidemiological, and Clinical Genetics
Jean-Laurent Casanova, Laurent Abel, and Lluis Quintana-Murci p p p p p p p p p p p p p p p p p p p p 447
Integration of Genetic and Immunological Insights into a Model of Celiac
Disease Pathogenesis
Valerie Abadie, Ludvig M. Sollid, Luis B. Barreiro, and Bana Jabri p p p p p p p p p p p p p p p p p p p 493
Systems Biology in Immunology: A Computational Modeling Perspective
Ronald N. Germain, Martin Meier-Schellersheim, Aleksandra Nita-Lazar,
and Iain D.C. Fraser p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 527
Immune Response to Dengue Virus and Prospects for a Vaccine
Brian R. Murphy and Stephen S. Whitehead p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 587
Follicular Helper CD4 T Cells (TFH )
Shane Crotty p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 621
SLAM Family Receptors and SAP Adaptors in Immunity
Jennifer L. Cannons, Stuart G. Tangye, and Pamela L. Schwartzberg p p p p p p p p p p p p p p p p p 665
The Inammasome NLRs in Immunity, Inammation,
and Associated Diseases
Beckley K. Davis, Haitao Wen, and Jenny P.-Y. Ting p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 707
Indexes
Cumulative Index of Contributing Authors, Volumes 1929 p p p p p p p p p p p p p p p p p p p p p p p p p p p 737

Cumulative Index of Chapter Titles, Volumes 1929 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 744
Errata
An online log of corrections to Annual Review of Immunology articles may be found at

http://immunol.annualreviews.org/errata.shtml
vi Contents

Systems Biology in Immunology - A Computational Modeling Perspective

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Systems Biology in Immunology - A Computational Modeling Perspective

Загружено:

Авторское право:

Доступные форматы

IY29CH19-Germain ARI 7 February 2011 22:24

Top downloaded articles

Our comprehensive search

Ronald N. Germain, Martin Meier-Schellersheim,

Annu. Rev. Immunol. 2011. 29:52785 Keywords

The Annual Review of Immunology is online at Abstract

INTRODUCTION zation of microarrays (1), the widespread use

the literature, high-throughput methods are

528 Germain et al.

www.annualreviews.org Computational Systems Biology 529

530 Germain et al.

Table 1 Computational approaches and tools for systems biology

nonstochastic Biology Workbench (47),

www.annualreviews.org Computational Systems Biology 531

532 Germain et al.

interacting components and/or components distribution in the simulated (sub)cellular do-

www.annualreviews.org Computational Systems Biology 533

534 Germain et al.

www.annualreviews.org Computational Systems Biology 535

536 Germain et al.

www.annualreviews.org Computational Systems Biology 537

538 Germain et al.

states. As such they are the most well-developed

www.annualreviews.org Computational Systems Biology 539

540 Germain et al.

www.annualreviews.org Computational Systems Biology 541

542 Germain et al.

www.annualreviews.org Computational Systems Biology 543

544 Germain et al.

www.annualreviews.org Computational Systems Biology 545

546 Germain et al.

www.annualreviews.org Computational Systems Biology 547

548 Germain et al.

www.annualreviews.org Computational Systems Biology 549

Pathogen reagent Genes Primary screen Total

porating HIV-host protein-protein interaction budding. However, many genes highlighted by

www.annualreviews.org Computational Systems Biology 551

552 Germain et al.

genes among Drosophila, C. elegans, and human pairs (198).

www.annualreviews.org Computational Systems Biology 553

554 Germain et al.

predicted 650,000 total interactome is accurate Immunology applications. In addition to

www.annualreviews.org Computational Systems Biology 555

556 Germain et al.

www.annualreviews.org Computational Systems Biology 557

558 Germain et al.

In contrast to experiments in which ions for (243, 244).

www.annualreviews.org Computational Systems Biology 559

Table 3 Examples of quantitative mass spectrometry methods

Method name Description Relative/absolute used? Limitations Reference(s)

www.annualreviews.org Computational Systems Biology 561

combination is gas chromatography coupled

562 Germain et al.

www.annualreviews.org Computational Systems Biology 563

antibody. Independently, similar alternative Phosphoow, a ow-cytometric approach to

564 Germain et al.

www.annualreviews.org Computational Systems Biology 565

cytoskeletal changes, among other parameters, often considered to be inherently confocal,

566 Germain et al.

www.annualreviews.org Computational Systems Biology 567

568 Germain et al.

www.annualreviews.org Computational Systems Biology 569

570 Germain et al.