Cell Sept 2011

Leading Edge
In This Issue
Chromothripsis Moves beyond Cancer
PAGE 889
Chromothripsis, a phenomenon in which numerous genomic rearrangements

are apparently acquired in one single catastrophic event, has been characterized in cancer genomes. Liu et al. now document that patients with developmental disorders also harbor complex genomic rearrangements with features
of chromothripsis. The authors provide evidence that these germline-acquired
chromosome catastrophes occur through DNA replication errors rather than
the shattering and reassembly of a chromosome.
Blocking Myc in Myeloma

PAGE 904
Despite the centrality of Myc in the pathogenesis of cancer, conventional

approaches for direct Myc inhibition have not proven successful. By targeting
chromatin regulatory complexes influencing Myc expression with a chemical
probe of BET bromodomains, Delmore et al. block MYC transcription and demonstrate the efficacy of this compound in
mouse models of multiple myeloma.
Positive ID on a Tumor Target

PAGE 918
ID proteins inhibit differentiation and maintain stem cell fate and are ubiquitinated and degraded in differentiated tissues.
However, in many neoplasms, IDs appear to escape degradation. Williams et al. now identify USP1 as the deubiquitinating
enzyme controlling ID protein stability and show that, by promoting ID stability, USP1 preserves stem cell-like characteristics
of osteosarcoma tumor cells. Their findings suggest that the USP1-ID axis that normally controls bone development is
usurped to propagate osteosarcoma tumor stem cells and point to USP1 as an attractive target for differentiation therapy.
Helicase Merges onto the Single-Stranded Highway

PAGE 931
The eukaryotic replicative helicase complex CMG is loaded onto dsDNA, but it could, in principle, unwind DNA by translocating along ssDNA or dsDNA. By colliding replisomes with strand-specific roadblocks in Xenopus egg extracts, Fu et al. show
that CMG can bypass a lagging strand, but not a leading strand roadblock, strongly supporting a 3-to-5 ssDNA translocation
mechanism. Replication initiation therefore likely involves reconfiguration from a dsDNA to an ssDNA binding mode.
Microtubules Master Transcription

PAGE 942
Sugioka et al. show that extrinsic signals that remodel the cytoskeleton lead to
asymmetric gene expression and cell fate in the dividing C. elegans embryo.
Wnt signaling generates an asymmetry in the astral microtubules that, in turn,
influences the levels of b-catenin in the nuclei of daughter cells as they separate
from each other. The findings point to a direct role of cytoskeletal dynamics in
transcriptional regulation.
Acetylation Winds the Aging Clock

PAGE 969
Lu et al. find that yeast lifespan is regulated by an intrinsic timing mechanism

that acts in parallel with nutrient-sensitive aging pathways. Acetylation of a subunit of AMPK decreases with age, leading to the progressive phosphorylation
of Akt. The energy sensor AMPK therefore acts in a nutrient-independent
manner to regulate longevity.
Cell 146, September 16, 2011 2011 Elsevier Inc. 843
A New Recipe for Making Gradients

PAGE 955
Griffin et al. show that the MEX-5 protein gradient in the C. elegans zygote
arises from a localized phosphorylation/dephosphorylation cycle that regulates
the rate of MEX-5 diffusion. These results present a paradigm for protein concentration gradient formation that does not require localized protein synthesis
or degradation.
Endothelium at the Eye of the Storm

PAGE 980
Cytokine storms are excessive and potentially fatal immune responses to

influenza and other viral infections. Teijaro et al. find that activation of the
sphingosine-1-phosphate receptor (S1P1) suppresses early immune responses, preventing cytokine storms and protecting mice from pathogenic
human influenza virus. S1P1 is expressed in vascular endothelial cells in the
lung, identifying these cells as key mediators of the response to influenza infections and suggesting a potential therapeutic
path for preventing morbidity due to cytokine storms.
Toggling Hunger
PAGE 992
How is hunger neuronally coded? Using a combination of optogenetic, electrophysiological, and pharmacological
approaches, Yang et al. demonstrate that hunger or an appetite-stimulating hormone induces synaptic activity in the neurons
that drive feeding behavior. This activity persists due to a positive feedback loop involving AMPK and neuronal calcium
release until leptin, a hormone signaling satiety, switches it off. The findings reveal a neuronal circuit, switched on and off
by hormonal pulses, that operates as a memory storage device for a physiological state.
A Neuron in Any Other Place Would Smell as Sweet

PAGE 1004
Choi et al. probe how the brain translates sensory stimulation such as odors to behavior. By artificially stimulating clusters of
neurons in a region where olfactory information is processed (the piriform cortex), the authors show that activating the same
arbitrarily chosen neuronal cluster can lead to opposite behavioral responses, depending on the experience (aversive or
attractive) coupled to the neuronal stimulation. The findings highlight the plasticity of responses to olfactory stimuli and
indicate that spatial order in the piriform does not inform odorant identity or behavioral output.
Histone Vocabulary Explodes

PAGE 1016
The diversity of histone modifications and their many functions becomes ever
more complex. Tan et al. now identify 67 histone marks, including lysine crotonylation. This modification is evolutionarily conserved, abundant in core histones, and marks either active promoters or potential enhancers. Their data
also suggest that lysine crotonlyation marks active sex chromosome genes in
male germ cells.
Going Ape Over the Sperm Methylome

PAGE 1029
Genome-wide reference maps of DNA methylation are critical to an understanding of its many roles in gene regulation. Molaro et al. generate full DNA
methylation profiles of human and chimp sperm and compare these to embryonic stem cells. Their findings reveal distinct properties for the epigenetic
reprogramming that occurs in germ cells and somatic cells during mammalian
development and provide insight into the relationships between the evolution of
the genome and the epigenome.
Leading Edge
Select
Human Evolutionary Genomics
The forces that have shaped the evolution of human genomes are now coming into sharp
focus. This issues Select explores recent discoveries that change our view of how our genomes
became what they arefrom the impact of interbreeding between ancient modern humans and
Neanderthals to evidence that rates of de novo mutation may be subject to large variation between
individuals.
Gene Hunter Gatherers
Emerging evidence suggests that

Eurasian genomes have been shaped
by interbreeding of ancient modern
humans with archaic human species,
such as Neanderthals.
Ancient modern humans interbred with archaic humans including Neanderthals,

acquiring genes aiding survival in lands far from their African origin. This is the
larger narrative revealed by the work of Peter Parham and colleagues (Abi-Rached
et al., 2011), who examined the frequency and geographic distribution of polymorphic alleles of human leukocyte antigen (HLA) class I genes in moderns humans and
their occurrence in Neanderthals (found in Europe) and their sister group the
Denisovan (who lived in Asia). HLA genes encode molecules that present antigens
to T cells and natural killer cells and thus have key roles in immune surveillance
against pathogens. The authors show that Neanderthal and Denisovan alleles of
HLA class I genes are widespread in modern humans throughout Eurasia but
rare or absent in Africa. These alleles have the opposite patterns of linkage
and diversity to those spreading from Africa. From this analysis they conclude
that admixture with Neanderthals and Denisovans greatly shaped the immune
systems of ancient modern humans following their migration out of Africa and
may have conferred survival advantages against local pathogens. Thus, it
seems that, having left Africa 200,000 years after the Neanderthal/Denisovan
ancestors, the newcomers reaped the genetic benefits of their predecessors
extensive experience.
Abi-Rached, L., et al. (2011). Science. Published online August 25, 2011. 10.1126/
science.1209202.
African Hotspots
A more recent admixture occurred in the Americas between people of European and African ancestries. By examining
sites of recombination in the genomes of 30,000 African Americans, Hinch et al. (2011) provide a detailed map of crossover frequency with surprising results. At large size scales (>3 Mb), most of the recombination events match up with
those previously reported in studies of European genomes. However, at smaller size scales, the authors find 2,500
hotspots that are predominantly active in West Africans. A closely related study by Berg et al. (2011) similarly demonstrates the existence of these hotspots from a survey of Hap-Map genotype data and subsequent analysis of meiotic
recombination crossover events in sperm from African and European males. Both groups provide evidence that this
striking variability between African and non-African populations is due to different alleles of PRDM9, a gene encoding
a DNA-binding histone methyltransferase, which is expressed during meiosis and involved in recombination. Further
analysis of the enriched hotspots shows that they often contain a 17 bp DNA sequence, which is predicted to bind
the alleles of PRDM9 commonly found in Africans. These findings indicate that recombination frequency and location
are more variable than previously thought. Future work may confirm the prediction that the risk of disease-causing
genomic rearrangements is higher in individuals carrying these particular PRDM9 alleles or may delve further into
why the 17 bp motif is singled out by particular PRDM9 variants.
Berg, I.L., et al. (2011). Proc. Natl. Acad. Sci. USA 108, 1237812383.
Hinch, A.G., et al. (2011). Nature 476, 170175.
Squeezing More out a Genome

Much information is hidden within our genomes. From the analysis of
individual genomes, Li and Durbin (2011) have devised a clever
approach for inferring the history of human population size, which is
thought to have undergone major bottlenecks at key transition points.
Their analysis relies on the uneven distribution of alleles that are heterozygous between the two copies of our diploid genomes. A high density of
heterozygous sites in a given region indicates that the most recent Heatmap showing the time to the most recent
common ancestor for that locus resides deeper in the past, whereas common ancestor along a chromosomal region.
a region that has a low density of heterozygous sites points to a more
recent common ancestor. By analyzing this distribution across hundreds of thousands of loci, the authors show that
it is possible to infer how effective population size has varied over time. Their results are consistent with the notion
that non-Africans experienced a significant bottleneck between 10 and 60 thousand years ago, coincident with their
founders departure from Africa. Meanwhile, African populations also experienced a bottleneck, but one that was
less severe in both duration and extent. Further analysis suggests that although African and non-African populations
began their separation 60,000 years ago, considerable genetic exchange might have continued up until 20,000
years ago. Future developments of the analysis, with possible adjustments to the parameters used for mutation rate
and generation time, may refine this picture and thus further contribute to a coalescence of views about human prehistory. The approach is also applicable to any diploid genome, thus providing a valuable new tool for the fields of evolution, ecology, and paleobiology.
Li, H., and Durbin, R. (2011). Nature 475, 493496.
Both Parents Are to Blame

De novo mutations are increasingly viewed as a major cause of sporadic disease. Although it has long been assumed
that the male germline is more prone to de novo mutations than is the female germline due to the large number of cell
divisions involved in sperm production, high-throughput sequencing of entire families now makes it possible to test this
supposition. Recent work by Conrad et al. (2011) takes a first step toward determining the degree to which de novo
mutations occur, and their findings suggest that rates could vary greatly between individuals and families. To do
this, they sequenced the full genomes of two pairs of parents and their offspring. Between the two families, the number
of germline de novo mutations in the offspring were of the same order of magnitude (49 and 35, respectively, for the two
offspring tested), whereas a much larger number of non-germline de novo mutations were acquired (although some of
these may have occurred in the culturing of the lymphoblast cell lines used as the tissue source). The biggest surprise is
that in one family, most of the de novo mutations were inherited paternally (92%), whereas in the other, the majority
were maternal (64%), suggesting that individual variance could be quite large. The findings thus offer a tantalizing
glimpse of what might be found in larger-scale studies, which could be especially revealing in establishing the consequences of maternal and paternal age on the prevalence of de novo mutations.
Conrad, D.F., et al. (2011). Nat. Genet. 43, 712714.
Robert P. Kruger
Leading Edge
BenchMarks
Molecular Mechanism
of Protein Folding in the Cell
James E. Rothman1,* and Randy Schekman2,*
1Department
of Cell Biology, Yale University School of Medicine, New Haven, CT 06520-8002, USA
of Cell and Molecular Biology, HHMI, University of California, Berkeley, CA 94720-1302, USA
*Correspondence: james.rothman@yale.edu (J.E.R.), schekman@berkeley.edu (R.S.)
DOI 10.1016/j.cell.2011.08.041
2Department
F.-Ulrich Hartl and Arthur Horwich will share this years Lasker Basic Medical Science Award for the
discovery of the cells protein-folding machinery, exemplified by cage-like structures that convert
newly synthesized proteins into their biologically active forms. Their fundamental findings reveal
mechanisms that operate in normal physiologic processes and help to explain the problems that
arise in diseases of protein folding.
How is it that, even when proteins cannot
fold on their own in vitro, they somehow
fold beautifully in the much more complex
environment of a cell? As with most
breakthrough discoveries in biology, the
answer is remarkably simple: the cell provides special protein machinery, called
molecular chaperones, that surrounds
the folding protein and removes it from
the rest of the cell.
The surfaces of this remarkable machinery are very forgiving. They actually
utilize metabolic energy (i.e., ATP hydrolysis) to alternate their physical chemistry
between hydrophobic states and hydrophilic states, which restarts the folding
when it stalls and expels the protein
when folding is completed. In the case
of the heat shock protein 60 (Hsp60), a
molecular chaperone shaped like a test
tube, access to the cavity is highly selective and limited to only nascent proteins or
those purposely unfolded to repair conformational damage. Analogous specialized test tube environments, also based
on limited access to the internal cavity of
a ring-based protein structure, enable targeted proteolysis (by proteosomes),
making this strategy a general principle
in cellular biology.
These singular discoveries are celebrated in this years Lasker Basic Medical
Research Award to F.-Ulrich Hartl and
Arthur Horwich. In collaborative work,
they discovered the fundamental principle
that protein folding in the cell is generally
a facilitated process, and in independent
and often complementary work, they
and their collaborators established the
pathway and molecular mechanisms involved in this process. Together, the

discoveries of Hartl and Horwich stand
at the apex of decades of work by many
investigators who helped start and extend
the field of protein folding in the cell.
The pathway of protein folding is utterly
general; it is essentially the same in bacteria, fungi, plants, animals, and of course
humans. The full implication of the latter is
that when the pathway is saturated, or its
coupling to the pathway of protein degradation is imperfect, then protein aggregates accumulate in the cell. Depending
on the nature of the protein species
involved, this can result in neurodegenerative disease, such as Alzheimers,
Huntingtons, Creutzfeld Jacobs, and
Parkinsons Diseases, or can affect other
tissues, as in cystic fibrosis.
Freeing Proteins to Fold
The achievements of Hartl and Horwich
must be seen in a historical perspective.
In 1972, Christian Anfinsen received a
Nobel Prize for discovering in the 1950s
that proteins assume their three-dimensional conformations and therefore gain
their catalytic potential, exclusively based
on primary instructions in their intrinsic
amino acid sequence. This insight forged
the link between the genetic code and
protein function, but it emerged from
simple denaturation-renaturation experiments involving a small number of isolated, pure enzymes, which at that time
were rather simple and small. As more
complex genes have been analyzed and
expressed over time, it has become
common knowledge that most complex

proteins are notoriously difficult to refold
at concentrations present in cells and at
the bulk concentration of protein that
prevails in cells. Indeed, many proteins
even fold poorly when expressed inside
cells from distant species.
Molecular chaperones provide the solution to this paradox. In many cases, chaperones, such as the Hsp70 chaperone
system, stabilize aggregation-prone (i.e.,
hydrophobic) intermediates. In more difficult cases, tube-like chaperones, such as
the GroEL/Hsp60 chaperone system, funnel the polypeptide into the protected
internal space of giant ring-based structures. These two strategies represent the
basic principles that permit proteins to
fold in the cell, and the establishment of
the mechanisms and pathways involved
represents, in our time, an advance comparable to that made by Anfinsen in his time.
Indeed, these advances are strictly complementary. Anfinsens fundamental chemical conclusionthat a proteins threedimensional shape is determined solely by
its intrinsic chemical compositionis in
no way challenged by Hartls and Horwichs singular advance in cell biology. To
the contrary, molecular chaperones associate with their substrates only during the
folding process, and thus they are purely
facilitators that enable a gene product to
efficiently achieve its final folded state
dictated solely by its intrinsic chemistry.
Landmark Papers Set the Stage
Over the past 40 years, many investigators have identified and characterized
individual components of
that Horwich and Hartl stand
what we now recognize as
out (Figure 1).
the core chaperone protein
It is noteworthy that this
machinery, including phage
remarkable set of experiassembly factors and proments was possible only
teins upregulated by stress.
because of previous mechaAnd, by the late 1980s, all
nistic insights into how prothe proteins central for protein
teins encoded in the nucleus
folding were virtually known
are imported into mitochonin both structural and broad
dria. These insights emerged
phenotypic terms. In parmainly from the laboratories
ticular, John Ellis, Costa
of Walter Neupert and GottGeorgopoulos, and George
fried Schatz in the previous
Lorimer deserve special redecade. Not only was Hartl a
cognition for their genetic
protege of Neupert, but
and biochemical description
much of his early work on
of the protein-folding catathe Hsp60 chaperones was
carried out directly in Neulysts. Georgopoulos was the
perts laboratory, as reflected
first to define the genes for all
Figure 1. Chaperone Pathway for the Folding in the E. coli Cytosol
the major folding catalysts.
in Neuperts coauthorship of
The Hsp70 chaperones DnaK/DnaJ stabilize the newly synthesized protein in
a nonaggregated, folding-competent state. Transferring the protein into the
His work as a graduate student
the landmark papers. Schatz
central cavity of the Hsp60 chaperone GroEL requires the nucleotide
in the early 1970s and then
and collaborators introduced
exchange factor GrpE. GroES binds to GroEL in an ATP-dependent reaction
later as a postdoctoral fellow
the background methods inand displaces the unfolded protein into an enclosed folding cage. The protein
led to the mapping of the
volving DHFR translocation
is allowed to fold inside the cage for 10 s, the time needed for GroEL to
complete one round of ATP hydrolysis (of 7 ATPs). Binding of ATP to the
E. coli genes GroEL, GroES,
and folding.
opposite ring of GroEL induces an allosteric change that triggers the opening
DnaK, DnaJ, and GrpE.
of the folding chamber. Folded protein is released, whereas incompletely
These genes all emerged
Hsp70s and Folding
folded protein re-binds to GroEL for another round of attempted folding in the
GroEL-GroES cage.
from genetic screens for
in the Cytoplasm
mutants that affect l phage
The Hsp70 protein family
assembly and, in the case of
provides a second system of
the latter three genes, play a role in phage Hartl and Horwich isolated and char- molecular chaperones. These proteins
and chromosome replication (Georgo- acterized mutations in mitochondrial are a family of monomeric ATPases,
poulos and Herskowitz, 1971).
chaperone Hsp60/GroEL. Surprisingly, with Georgopoulos bacterial DnaK (an
In an entirely independent develop- mitochondrial proteins were synthesized Hsp70) and its cofactors DnaJ and GrpE
ment, in 1980 Ellis discovered the first and imported successfully, but they failed as the prototype of these chaperones. In
molecular chaperone (Barraclough and to fold. This was the first proof that protein all likelihood, the Hsp70 chaperones
Ellis, 1980). He found that unassembled folding required molecular assistance by supply the most general mechanism of
subunits of the Rubisco enzyme complex binding proteins (chaperones) in the facilitated protein folding, as the GroEL
in chloroplasts were associated with ano- living cell (Cheng et al., 1989). The same system is required for only a small fraction
ther protein that turned out to be a chap- team published a second landmark paper of the cells proteins (Houry et al., 1999).
erone. In 1988, Ellis and Georgopoulos (Ostermann et al., 1989) showing that
Key experiments in the laboratories of
discovered that this chaperone is related mitochondrial Hsp60 from mitochondria Hugh Pelham and Elizabeth Craig first
to the GroEL molecule involved in phage assists the folding of monomeric polypep- associated the Hsp70 proteins with proassembly (Hemmingsen et al., 1988). tide chains and requires ATP binding and tein folding in the cytoplasm and in quality
Ellis coined the term chaperonin for hydrolysis to do so. This work focused on control inspection in the secretory paththe GroE-type folding catalysts that are the folding of dihydrofolate reductase way. The DnaK proteins are related to
required to assist a variety of cellular pro- (DHFR) when its imported into isolated a family of heat-induced proteins, called
cesses within organelles and in the cyto- mitochondria, and it established the prin- the Hsp70s, which were first described
plasm of all types of cells. The following ciple of assisted folding in the cell for sin- as heat shock proteins in Drosophila.
year and in subsequent detailed enzymo- gle polypeptide chains, even those that Rothman and colleagues discovered that
logic studies, George Lorimer reconsti- can fold spontaneously and efficiently as the Hsp70s bind and hydrolyze ATP
tuted the assembly of Rubisco subunits pure proteins. This marked a radical de- (Chappell et al., 1986). With this in mind,
with isolated bacterial chaperonins (Go- parture from the earlier view that protein Pelham was the first to speculate that
loubinoff et al., 1989).
folding in the cell can be a spontaneous Hsp70s promote proper folding or reasThese foundational studies beautifully process. It is on the basis of these papers sembly of proteins by using cycles of
set the stage for Hartls and Horwichs and their future seminal contributions in ATP hydrolysis to bind and release agtwo landmark papers in 1989. In a key establishing the key concepts and molec- gregated, denatured proteins (Pelham,
genetic and biochemical experiment, ular mechanisms in cellular protein folding 1986). Pelham was also the first to
852 Cell 146, September 16, 2011 2011 Elsevier Inc.
demonstrate that BiP, an Hsp70 ATPase

in the endoplasmic reticulum lumen, was
associated with the assembly of secretory
proteins that fold in the ER (Munro and
Pelham, 1986), implicating Hsp70s in the
folding process. The core biochemical
mechanismbinding and release of exposed hydrophobic segmentswas established by Rothman and colleagues
(Flynn et al., 1989). Elizabeth Craig developed a yeast genetic approach to discover the roles of the cytoplasmic Hsp70s
in maintaining the unfolded state that
favors protein translocation (Deshaies
et al., 1988).
Hartl and Horwich Hook Up
the Pathway
As important as these early discoveries
were in the developing the story of in vivo
protein folding, the experiments of Hartl
and Horwich from 1987 to 1997 created
a coherent picture of the physiologic and
biophysical processes that cells use to
fold proteins. In an extraordinary body of
work between 1991 and 1994, Hartl
defined, resolved, and reconstituted the
complete pathway by which molecular
chaperones cooperate to fold proteins.
In a brilliant series of articles published
in the journal Nature, he linked together
the entire pathway, establishing how the
folding peptide chain is recognized cotranslationally by Hsp70, preventing premature misfolding (the negative role)
(Langer et al., 1992a; Frydman et al.,
1994), and then transferred sequentially
from Hsp70 to Hsp60/GroEL, which promotes folding (the positive role). He also
then placed in this context many other
(cochaperone) proteins that act as
regulators of the steps in this pathway
(DnaJ, GrpE, GroES).
For most proteins, Hsp70 is essentially
sufficient. The Hsp70 loads onto the
newly synthesized polypeptide chain as
it emerges from the ribosome and stays
on long enough to allow entire folding
domains to be manufactured, thereby
preventing aggregation. It further imparts
energy to the polypeptide as it releases
it, giving folding reactions an extra kick.
When this does not suffice, the chain
(with Hsp70 re-bound to it) is transferred
to the chamber of a GroEL chaperonin
to do the heavy lifting.
By this time, both GroEL (Hendrix,
1979) and proteosomes (Dahlmann
et al., 1986) were well known to be cylindrical particles based on oligomeric

ring-forming subunits. This of course
gave rise to speculation that the cavities
may have functional significance. But the
cylindrical shape of the particle did not
intrinsically require that folding (or degradation) occur within the particle, as a
cavity is a structural consequence of any
ring-based oligomer. Direct evidence for
this hypothesis came first for protein
folding inside GroEL in 1992 (and then 3
years later for protein degradation inside
proteosomes in 1995).
Electron microscopy images of GroEL,
taken by Hartl and Baumeister (Langer
et al., 1992b), offered the first indication
that folding occurs within the GroEL
cavity, and this hypothesis was confirmed
in 1993 by Horwich (Braig et al., 1993).
Hartl proposed that the chain folds in
the internal microenvironment provided
by the cavity of Hsp60/GroEL and its lidshaped cofactor GroES (Martin et al.,
1993). In essence, a series of ATP-driven
conformational steps change the shape
of the wall of the chamber and cock
and load the two symmetrical half
chambers in this molecule. In other
words, ATP binding and hydrolysis drive
conformational changes that open the
chamber to accept an unfolded domain
and then closes the chamber to reinforce
the folding event. Although at first controversial, this beautiful mechanism is now
well accepted and thoroughly confirmed
by X-ray crystal structures of the Hsp60/
GroEL and related proteins.
In 1994, together with Paul Sigler, Horwich solved the X-ray crystal structure
of GroEL, which was one of the largest
protein complexes crystallized at the
time (Braig et al., 1994). An elegant mutational analysis then confirmed, on a structural basis, that an unfolded protein binds
in the center of the hollow cylindrical
GroEL complex by hydrophobic binding
regions on the apical GroEL domains
facing the central cavity (Fenton et al.,
1994). The ATP-binding sites were also
observed as pockets in the equatorial
domain of each subunit.
More details followed in 1997, when
Horwich and Sigler solved the crystal
structure of the GroEL-GroES complex
(Xu et al., 1997). The structure confirmed
that the inner cavity of GroEL undergoes
a massive conformational change upon
GroES binding. This change results in

the burial of the hydrophobic regions
and formation of a large hydrophilic
chamber in which proteins up to about
60 kDa in size are free to fold. This structure also identified the mechanism of
ATP hydrolysis, allowing the Horwich
laboratory to work out the nucleotide
reaction cycle (Rye et al., 1997).
Perspective
Our current understanding of proteinfolding mechanisms is rather precise.
However, it has not yet revealed how to
control this process when it goes awry
in diseases of protein folding, such as
Alzheimers and Huntingtons diseases.
A new field of proteostasis seeks to
understand the balance of protein folding,
misfolding, and proteolysis that governs
normal and abnormal cell physiology.
Drugs that target the folding process
and that stabilize a proper folded state
for misfolded proteins, such as mutant
forms of CFTR (Cystic Fibrosis Transmembrane Conductance) in cystic fibrosis, show promise in the treatment of
a variety of genetic diseases. If we are
able to harness our understanding of
protein chaperones in the treatment of
diseases of protein folding, it will be
because of the pioneering efforts of Hartl,
Horwich, and others who elevated the
field to its current level of molecular
precision.
REFERENCES
Barraclough, R., and Ellis, R.J. (1980). Biochim.
Biophys. Acta 608, 1931.
Braig, K., Simon, M., Furuya, F., Hainfeld, J.F., and
Horwich, A.L. (1993). Proc. Natl. Acad. Sci. USA
90, 39783982.
Braig, K., Otwinowski, Z., Hegde, R., Boisvert,
D.C., Joachimiak, A., Horwich, A.L., and Sigler,
P.B. (1994). Nature 371, 578586.
Chappell, T.G., Welch, W.J., Schlossman, D.M.,
Palter, K.B., Schlesinger, M.J., and Rothman, J.E.
(1986). Cell 45, 313.
Cheng, M.Y., Hartl, F.U., Martin, J., Pollock, R.A.,
Kalousek, F., Neupert, W., Hallberg, E.M.,
Hallberg, R.L., and Horwich, A.L. (1989). Nature
337, 620625.
Dahlmann, B., Kopp, F., Kuehn, L., Reinauer, H.,
and Schwenen, M. (1986). Biomed. Biochim. Acta
45, 14931501.
Deshaies, R.J., Koch, B.D., Werner-Washburne,
M., Craig, E.A., and Schekman, R. (1988). Nature
332, 800805.
Fenton, W.A., Kashi, Y., Furtak, K., and Horwich,

A.L. (1994). Nature 371, 614619.
Flynn, G.C., Chappell, T.G., and Rothman, J.E.
(1989). Science 245, 385390.
Frydman, J., Nimmesgern, E., Ohtsuka, K., and
Hartl, F.U. (1994). Nature 370, 111117.
Georgopoulos, C.P., and Herskowitz, I. (1971). In
The Bacteriophage Lambda A, D. Hershey, ed.
(Cold Spring Harbor, NY: Cold Spring Harbor
Laboratory Press), pp. 553564.
Goloubinoff, P., Gatenby, A.A., and Lorimer, G.H.
(1989). Nature 337, 4447.
Hemmingsen, S.M., Woolford, C., van der Vies,

S.M., Tilly, K., Dennis, D.T., Georgopoulos, C.P.,
Hendrix, R.W., and Ellis, R.J. (1988). Nature 333,
330334.
Hendrix, R.W. (1979). J. Mol. Biol. 129, 375392.
Houry, W.A., Frishman, D., Eckerskorn, C., Lottspeich, F., and Hartl, F.U. (1999). Nature 402,
147154.
Martin, J., Mayhew, M., Langer, T., and Hartl, F.U.

(1993). Nature 366, 228233.
Munro, S., and Pelham, H.R.B. (1986). Cell 46,
291300.
Ostermann, J., Horwich, A.L., Neupert, W., and
Hartl, F.U. (1989). Nature 341, 125130.
Pelham, H.R.B. (1986). Cell 46, 959961.
Langer, T., Lu, C., Echols, H., Flanagan, J., Hayer,

M.K., and Hartl, F.U. (1992a). Nature 356, 683689.
Rye, H.S., Burston, S.G., Fenton, W.A., Beechem,

J.M., Xu, Z., Sigler, P.B., and Horwich, A.L. (1997).
Nature 388, 792798.
Langer, T., Pfeifer, G., Martin, J., Baumeister, W.,

and Hartl, F.U. (1992b). EMBO J. 11, 47674778.
Xu, Z.H., Horwich, A.L., and Sigler, P.B. (1997).

Nature 388, 741750.
Leading Edge
BenchMarks
Artemisinin: Discovery
from the Chinese Herbal Garden
Louis H. Miller1,* and Xinzhuan Su1
1Laboratory of Malaria and Vector Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health,
Rockville, MD 20852, USA
*Correspondence: lmiller@niaid.nih.gov
DOI 10.1016/j.cell.2011.08.024
This years Lasker DeBakey Clinical Research Award goes to Youyou Tu for the discovery of artemisinin and its use in the treatment of malariaa medical advance that has saved millions of lives
across the globe, especially in the developing world.
The future benefits of many seminal discoveries in basic biomedical sciences
are not always obvious in the short run.
But for a handful of others, the impact
on human health is immediately clear.
Such is the case for the discovery by
Youyou Tu and colleagues of artemisinin
(also known as Qinghaosu) for treatment
of malaria. Artemisinin has been the frontline treatment since the late 1990s and
has saved countless lives, especially
among the worlds poorest children.
The Promise of Project 523
The story of artemisinin began in the unlikely
atmosphere of the Cultural Revolution in
China as a government initiative to aid the
North Vietnamese in their war with the United States. During the war, malaria caused
by
chloroquine-resistant
Plasmodium
falciparum was a major problem that
spurred research efforts on both sides of
the battlefield. In the US, these efforts culminated in the discovery of mefloquine, a
compound that was curative in a single
dose against chloroquine-resistant parasites (Trenholme et al., 1975). The North Vietnamese, however, lacked a research infrastructure and thus turned to China for help.
Under the instructions of Chairman Mao
and Premier Zhou, a meeting was held on
May 23, 1967 in Beijing to discuss the
problem of drug-resistant malaria parasites. This led to a secret nationwide program called project 523, involving over
500 scientists in 60 different laboratories
and institutes (Zhang et al., 2006).
Although the projects short-term goal
was to produce antimalarial drugs that
could immediately be used in the battlefield (by 1969 three treatments were
established), the projects long-term goal

was to search for new antimalarial drugs
by screening synthetic chemicals and by
searching recipes and practices of traditional Chinese medicine.
Because this work was considered
a military secret, no communication about
the research to the outside world was allowed, and in any case, during the tumult
of the Cultural Revolution, publication in
scientific journals was forbidden. For
these reasons, no one outside of project
523 knew about the work. Yet within
the project, information flowed freely
between the members of the various
research groups, and findings were presented at their frequent joint meetings.
Without a publication record, who
should be credited with the discovery of
artemisinin? The answer to this question
was not generally known when we (X.S.
and L.H.M.) began, in 2007, to delve into
the history of the discovery. Our findings
left no doubt that the major credit must
go to Youyou Tu, who was a principle
investigator at the Institute of Chinese Meteria Medica, China Academy of Chinese
Medical Sciences (CACAMS). In January
1969, Professor Tu led a team in screening
the literature and recipes of traditional
Chinese medicine under project 523. She
was chosen to present the work of project
523 for the first time in October 1981 in
Beijing to a World Health Organization
(WHO) visiting study group on chemotherapy of malaria (Tu, 1981).
From Ancient Recipe to Modern
Drug
During their search, Youyou Tu and
colleagues investigated more than 2,000
recipes of Chinese traditional herbs,

compiling 640 recipes that might have
some antimalarial activity. They tested
in a rodent malaria model more than
200 recipes with Chinese traditional
herbs and 380 extracts from the herbs.
Among the promising results, extracts
from Artemisia annua L. (Qinghao), a
type of wormwood native to Asia, were
shown to inhibit parasite growth by
68%. Follow-up studies, however, only
achieved 12% to 40% inhibition. Professor Tu reasoned that the low inhibition
could be due to a low concentration of the
active ingredient in the preparation and
began to improve the methods of extraction. After reading the ancient Chinese
medical description, take one bunch of
Qinghao, soak in two sheng (0.4 liters)
of water, wring it out to obtain the
juice and ingest it in its entirety in The
Handbook of Prescriptions for Emergency
Treatments by Ge Hong (283343 CE)
during the Jin Dynasty, she realized that
traditional methods of boiling and hightemperature extraction could damage
the active ingredient. Indeed, a much
better extract was obtained after switching from ethanol to ether extraction at
lower temperature.
However, the extract was still toxic.
Professor Tu then further removed from
the extract an acidic portion that contained no antimalarial activity, leaving a
neutral extract with reduced toxicity and
improved antimalarial activity. The neutral
extract, termed extract number 191, was
tested in the mouse malaria, Plasmodium
berghei, and achieved 100% inhibition
in October 1971. She presented her findings at a 523 meeting held in Nanjing
on March 8, 1972, providing

the parasitemia and became
some critical parameters for
the primary proponent for the
other teams to quickly obtain
use of artemisinin derivatives
pure artemisinin crystals.
in combination therapy, which
Although Tus team struggled
is now the standard treatment
to obtain high-quality crystals
worldwide. In 2010, he was
from the plant in the following
honored by the Canadian
months, two teams (Zeyuan
Gairdner Award for this imLuo, Yunnan Institute of
portant work.
Drug Research and the late
Project 523 developed, in
Zhangxing Wei, Shandong
addition to artemisinin, a numInstitute of Chinese Tradiber of products that are used
tional Medicine), using the
in combination with artemisiinformation and methods she
nin, including lumefantrine,
used, soon obtained pure
piperaquine, and pyronaricrystals from A. annua L. that
dine. Their success reflects
were highly active against
the unique spirit of collaborarodent malaria parasites.
tion from a large number of
Tests in humans by Guoqiao
scientists and institutions inLi, Guangzhou University of
volved in the search for antiChinese Traditional Medimalarial drugs.
cine, using the artemisinin
crystals from Yunnan Institute
An Ongoing Battle
of Drug Research showed
against Resistance
good activity against malaria
An important unanswerable
infection.
question is how effective arteFigure 1. The Antimalarial Drug Artemisinin (Qinghaosu)
Interestingly, the paper demisinin or its derivatives will
Since the discovery of artemisinin from Artemisia annua L., a plant used in
scribing artemisinins X-ray
be in the future. We can pertraditional Chinese medicine, by Youyou Tu and colleagues, many derivatives
have been synthesized, including dihydroartemisinin, which is more active
crystal structure, pharmacolhaps gain perspective from
than artemisinin. To protect this important antimalarial drug, combination
ogy, and efficacy against nonthe history of other antimatherapy with another antimalarial drug is the only treatment used today. The
severe and severe cerebral
larial drugs. The initial morfuture synthesis of new antimalarial drugs may be possible, originating from
the endoperoxide bridge that is required for artemisinins antimalarial activity
malaria listed no specific autality from malaria was ex(Charman et al., 2011).
thors, who were identified
tremely high throughout the
instead as the Qinghaosu
world before the introduction
Antimalarial Coordinating Reof the last herbal medicine,
search Group (1979). The paper showed (Jiang et al., 1982; Li et al., 1984). Artemi- quinine. Wherever it was introduced,
that artemisinin is a sesquipene lactone sinin works quickly within hours com- there was a marked decrease in mortality.
with an endoperoxide, and that the endo- pared to mefloquines slow parasite clear- When the price of quinine decreased
peroxide is required for its antimalarial ance, but because of its short half-life, in Italy, leading to the increased use
activity (Figure 1). In 1985, Klayman, artemisinin requires another drug in com- of quinine, the mortality markedly deworking in the US at the Walter Reed bination to obtain a cure. Patients recov- creased. It is not surprising that when
Army Institute of Research (WRAIR), ered so quickly after taking artemisinin P. falciparum became resistant to the
described the isolation of the same com- that they would not continue treatment synthetic antimalarial drug chloroquine,
pound and its structure from Artemisia after they felt better and thus were ulti- mortality in children rose dramatically.
annua (sweet wormwood), which grew mately not cured. Such incomplete treat- Once the basis of resistance to chloroalong the shores of the Potomac River. ment may promote drug resistance. Lis quine was identified as mutations in the
Klayman pointed out that there are group also developed suppositories con- gene encoding the putative chloroquine
few naturally occurring endoperoxides taining artemisinin to treat cerebral ma- transporter in the food vacuole, PfCRT
described in plants. Although numerous laria that are now being used in field (Fidock et al., 2000), it became evident
hydroxoperoxides had been tested at clinics in Africa. Shortening the time to that the mutations had not originated in
WRAIR, none were found to have antima- treatment by the use of suppositories Africa but were introduced from Southlarial activity (Klayman, 1985).
improves survival.
east Asia (Wootton et al., 2002).
Two clinical studies headed by ProAfter learning of this important disA major challenge to public health
fessor Gouqiao Li compared artemisinin covery by the Chinese, Nick White, who officials in African countries is to decide
and mefloquine. These studies were the was working in Thailand as a professor when to change their antimalarial drug
first to suggest that combination therapy at Oxford, began the study of artemisinin policy. Such decisions require a careful
should be considered to prevent recur- derivatives. He confirmed its rapid activity review of efficacy given that the choices
rence and development of resistance and the need for a partner drug to clear available are limited. However, during
these transitions, deaths (particularly of

children) occur before the decision to
change regimens is made. After chloroquine, the next antimalarial drug to be
introduced to Africa was sulfadoxine-pyrimethamine (Fansidar), which is a synergistic antimalarial combination. After
its introduction, Fansidar resistance, due
to mutations in both P. falciparum dihydrofolate reductase and dihydropteroate
synthetase, has been widely reported.
Likewise, resistance to mefloquine has
occurred in Asia, where it was introduced
as a stand-alone antimalarial drug. Unfortunately, mefloquine resistance has also
undermined the efficacy of its combination treatment with Fansidar, called Fansimef. Additionally, mefloquine is expensive
to produce and cannot fit the financial
needs of the African population.
The insistence that artemisinin derivatives must be used in combination therapy (Figure 1) may protect it from resistance, at least for some time. This will
require careful evaluation of the efficacy
of the added drug to protect artemisinin.
The one hope is that the use of artemisinin
combination therapy will remain effective
and that mortality associated with a
change in drugs will be prevented.
We also face the question of what drug
to adopt next if resistance to artemisinin
becomes a problem. Resistance to artemisinin derivatives may be arising on the
Thai-Cambodian border (Dondorp et al.,
2009), evident in the longer time it takes
for treatment to clear parasites from the
blood than has been reported previously
or from other areas of Thailand. However,
a similar concentration of artemisinin as
originally used for the sensitive parasites
can kill the resistant parasites in vitro
and in vivo. Although there are disagreements as to whether this indicates resistance, the prudent approach is to assume
that it is early resistance and to try to limit
the spread of these parasites. This would
mirror earlier epidemiological patterns of
malarial drug resistancefor both chloroquine and pyrimethamine-sulfidoxine,
resistance originated in Southeast Asia
and later spread to Africa. Vigilance is
particularly warranted given that resistance to antimalarial drugs can develop
rapidly. The one exception has been
quinine, in use for hundreds of years,
and resistance has only developed slowly
with the level of drug required for cure
increasing with time. Our hope is that

a similar course will be seen for the other
herbal medicine, artemisinin.
The Future of Antimalarial
Treatments
Efforts are ongoing to make other antimalarial compounds based on the structure
of artemisinin and its mechanism of
action. It is known that artemisinin
requires hemoglobin digestion and the
release of iron containing heme, which
induces oxidative stress (Klonis et al.,
2011). As noted by Klayman (1985), there
are few natural products with an endoperoxide, and this peroxide also offers an
opportunity to make new antimalarial
drugs (Charman et al., 2011).
But who is going to develop these new
drugs? Artemisinins discovery was
spurred on by war, and it can be hoped
that more peaceful motivations will drive
the development of future malaria treatments. Yet, malaria has not been a major
target of drug companies, which have
historically focused their efforts on more
profitable treatments for diseases prevalent in affluent countries. Filling this
gap, joint public-private companies,
such as the Medicines for Malaria Venture
(MMV), may prove a successful model
funding drug development. In addition,
modern approaches, such as highthroughput screening a large number of
compounds, may also provide new leads
for antimalarial drugs.
Another question that is raised with any
treatment or control measure is, how
effective is it in reducing mortality? In
many areas of the world, two measures
of malaria control were introduced simultaneously: artemisinin combination therapy and bednets treated with long-acting
pyrethroid insecticides. As a result, many
areas in Africa have shown a reduced
incidence of malaria, although a careful
analysis of the data fails to identify the
intervention that led to reduced disease
and, as a consequence, reduced death
(OMeara et al., 2010). In recent years,
there has been a disturbing increase in
pyrethroid resistance by malaria vectors
in Africa (Ranson et al., 2011). A study in
Dielmo, Senegal, where transmission is
continuous at a high level because of a
stream that runs through the village from
an underwater spring, showed a great
reduction of disease after the introduction
of control methods, but this reduction was

followed by a recent rebound (Trape et al.,
2011), perhaps as a result of insecticide
resistance. If such resistance becomes
widespread, people will be completely
dependent on artemisinin combination
therapy for preventing disease and
death given that many have lost some of
their immunity as a result of the marked
reduction in malaria. As long as the
mosquito species that harbor the parasite
remain in Africa, such recurrences can be
expected.
Although the challenges of combating
malaria remain daunting, the discovery
of artemisinin by Youyou Tu and her
many colleagues in the Chinese scientific
community offers hope and is a great
achievement in the history of modern
medicine.
ACKNOWLEDGMENTS
We thank Drs. Susan Pierce, LIG, NIAID and
Thomas E. Wellems, LMVR, NIAID for suggestions
on the manuscript and for suggestions on the figure drawn by Alan Hoofring. This work has been
supported by the Intramural Research Program of
the National Institutes of Health, National Institute
of Allergy and Infectious Diseases. We thank
Dr. Jian Li of Xiamen University for assistance in
translating this article into Chinese (translation
can be found online at http://www.cell.com/
LaskerAward-Chinese).
REFERENCES
Charman, S.A., Arbe-Barnes, S., Bathurst, I.C.,
Brun, R., Campbell, M., Charman, W.N., Chiu,
F.C.K., Chollet, J., Craft, J.C., Creek, D.J., et al.
(2011). Proc. Natl. Acad. Sci. USA 108, 44004405.
Dondorp, A.M., Nosten, F., Yi, P., Das, D., Phyo,
A.P., Tarning, J., Lwin, K.M., Ariey, F., Hanpithakpong, W., Lee, S.J., et al. (2009). N. Engl. J. Med.
361, 455467.
Fidock, D.A., Nomura, T., Talley, A.K., Cooper,
R.A., Dzekunov, S.M., Ferdig, M.T., Ursos, L.M.,
Sidhu, A.B., Naude, B., Deitsch, K.W., et al.
(2000). Mol. Cell 6, 861871.
Jiang, J.-B., Li, G.-Q., Guo, X.-B., Kong, Y.C., and
Arnold, K. (1982). Lancet 2, 285288.
Klayman, D.L. (1985). Science 228, 10491055.
Klonis, N., Crespo-Ortiz, M.P., Bottova, I., AbuBakar, N., Kenny, S., Rosenthal, P.J., and Tilley,
L. (2011). Proc. Natl. Acad. Sci. USA 108, 11405
11410.
Li, G.Q., Arnold, K., Guo, X.B., Jian, H.X., and Fu,
L.C. (1984). Lancet 2, 13601361.
OMeara, W.P., Mangeni, J.N., Steketee, R., and
Greenwood, B. (2010). Lancet Infect. Dis. 10,
545555.
Qinghaosu Antimalarial Coordinating Research

Group. (1979). Chin. Med. J. (Engl.) 12, 811816.
Ranson, H., Nguessan, R., Lines, J., Moiroux, N.,
Nkuni, Z., and Corbel, V. (2011). Trends Parasitol.
27, 9198.
Trape, J.F., Tall, A., Diagne, N., Ndiath, O., Ly, A.B.,
Faye, J., Dieye-Ba, F., Roucher, C., Bouganali, C.,
Badiane, A., et al. (2011). Lancet Infect. Dis.
Published online August 17 2011. 10.1016/

S1473-3099(11)70194-3.
Trenholme, C.M., Williams, R.L., Desjardins, R.E.,
Frischer, H., Carson, P.E., Rieckmann, K.H., and
Canfield, C.J. (1975). Science 190, 792794.
Tu, Y.Y. (1981) Fourth Meeting of the WHO
Scientific Working Group on the Chemotherapy
of Malaria: TDR/CHEMAL-SWG (4)/(QHS)/81.3
(Beijing, China).
Wootton, J.C., Feng, X., Ferdig, M.T., Cooper,

R.A., Mu, J., Baruch, D.I., Magill, A.J., and Su,
X.Z. (2002). Nature 418, 320323.
Zhang, J.-F., Zhou, K.D., Zhou, Y.Q., Fu, L.S.,
Wang, H.S., and Song, S.Y. (2006). A detailed
chronological record of project 523 and the
discovery and development of qinghaosu (artemisinin) (Guangzhou, China: Yangcheng Evening News Publishing Company), ISBN 7-80651539-9.
Leading Edge
Previews
SIP-ing the Elixir of Youth
William Mair,1,2,3,4,* Kristan K. Steffen,1,2,3 and Andrew Dillin1,2,3,*
1The
Salk Institute for Biological Studies, La Jolla, CA 92037, USA

Hughes Medical Institute, La Jolla, CA 92037, USA
3The Glenn Foundation for Medical Research, La Jolla, CA 92037, USA
4Harvard School of Public Health, Boston, MA 02115, USA
*Correspondence: wmair@hsph.harvard.edu (W.M.), dillin@salk.edu (A.D.)
DOI 10.1016/j.cell.2011.08.026
2Howard
AMP-activated protein kinase (AMPK) is a conserved cellular fuel gauge previously implicated in
aging. In this issue, Lu et al. (2011) describe how age-related deacetylation of Sip2, a subunit of
the AMPK homolog in yeast, acts as a life span clock that can be wound backward or forward to
modulate longevity.
From the age of Hippocrates, medicine
has long appreciated the importance of
preserving physiological homeostasis for
a healthy life. Indeed, the ancient Greeks
believed that a failure to maintain proper
balance between the four humors, or
liquids of the blood, was at the root of
human disease. Despite modern medicines departure from humorism, the idea
that a cellular collapse in homeostasis
might underlie late onset diseases still
prevails. In this issue, Lu and colleagues
(Lu et al., 2011) present new data supporting this ancient theory.
One fundamental cellular process requiring tight homeostatic control during
aging is the balance between energy
generation and usage. In eukaryotes, a
key orchestrator of energy homeostasis
is AMPK, a serine/threonine kinase that
functions as a cellular energy sensor, activating catabolic processes to generate
ATP when energy levels are low while
simultaneously shutting down energy requiring anabolic processes (Steinberg
and Kemp, 2009). The importance of
maintaining energy homeostasis throughout life is underscored by the fact that
AMPK activity mediates life span in yeast,
worms, and flies. Uncovering a novel
mode of AMPK regulation in yeast, the
work by Lu et al. (2011) demonstrates that
progressive deacetylation of the AMPK
subunit Sip2 with age limits replicative life
span (defined as the number of times that
a single cell can divide) and that blocking
this change can slow aging.
The S. cerevisiae homolog of AMPK is
a heterotrimeric complex made up of
both a single a and g subunit, Snf1 and
Snf4 (sucrose nonfermenting), respec-
tively, and one of three alternative

b subunits, Snf-interacting proteins (Sip)
1 and 2 and Gal83. Previous work has
shown that one b subunit in particular,
Sip2, plays a pivotal role in regulating
Snf1 activity with age (Ashrafi et al.,
2000). Sip2 inhibits the Snf1 complex by
sequestering Snf4 to the plasma membrane. As cells age, Sip2 shuttles from
the plasma membrane to the cytosol,
progressively activating Snf1, resulting in
limited replicative life span. In the current
study, Lu et al. (2011) expand upon this
model to demonstrate that the acetylation
status of Sip2 is the critical mechanism
controlling Snf1 activity during aging.
Using tandem mass spectrometry, the
authors identify residues on Sip2 that
become progressively deacetylated in
older cells. This change in Sip2 acetylation
status seems causal in yeast replicative
aging, as Sip2 mutants that are always
deacetylated are short lived, whereas
Sip2 acetylation mimics live longer than
wild-type cells. Acetylation of Sip2 is
antagonistically regulated by the acetyl
transferase NuA4 and the deacetylase
Rpd3. Strikingly, Sip2 is the primary
longevity target of these enzymes; the
short life span of NuA4 mutants is rescued
by Sip2 acetylation mimics, whereas the
extended life span of RPD3 mutants is
not further enhanced by increasing Sip2
acetylation. Together, these data suggest
that progressive deacetylation of Sip2
with time acts analogously to a longevity
clock that marks the age of the cells
and that interventions that drive acetylation or deacetylation of Sip2 can wind
the clock backward or forward, respectively (Figure 1).
The mechanism by which the ticking of

this clock modulates cellular aging hinges
upon changes to Snf1 activity; as Sip2
becomes deacetylated with age, it loses
its ability to bind to Snf1 and repress the
complex. The resulting hyperactivation
of Snf1 leads to accumulation of the
stress-associated disaccharide trehalose, increased sensitivity to reactive oxygen species, and limited life spanall
of which can be reversed by preserving
Sip2 acetylation status.
Changes to energy balance can have
profound physiological consequences
for the cell and have been shown previously to even alter the rate at which organisms age. In species ranging from yeast
to primates, reducing available energy
increases organismal life span and protects against age-related diseases (Fontana et al., 2010). Given its role in sensing
energy status, AMPK has therefore been
suggested to be a central mediator of
life span by dietary restriction. However,
in the study by Lu et al., the effects of
Sip2 acetylation on life span appear to
be independent of glucose availability of
the cell, and as such, the authors suggest
a role for Sip2 in determining the intrinsic
rate at which cells age, rather than the
ability to change that rate in response to
extrinsic factors. Interestingly, however,
both the intrinsic aging pathway described by Lu et al. and the dietary restriction-induced extrinsic pathway, regulated by target of rapamycin 1 (TORC1) in
yeast, converge upon a shared downstream mediator, Sch9.
Sch9, itself a kinase, has homology to
Akt and functions analogously to mammalian S6 kinase. No stranger to the
aging field, inhibition of S6K

et al., 2007) and CRTC-1
increases longevity in yeast,
(Mair et al., 2011). However,
worms, flies, and mammals
as is the case in yeast (Zhang
(Fontana et al., 2010).
et al., 2010), different heteroAlthough Sch9 is perhaps
trimeric combinations of
best known as a substrate for
AMPK have alternating funcTORC1, Lu et al. show that
tions in mammals (Steinberg
the Snf1 complex can phosand Kemp, 2009), and which
phorylate Sch9 independently
of these might be involved in
and that the short life span of
life span modulation is yet to
Sip2 acetylation mutants can
be investigated.
be rescued by deleting SCH9.
If the work from Lu et al. in
Sch9 is known to regulate
yeast does translate to mammany cellular processes that
mals, perhaps dysregulation
influence aging, including
of this ancestral energy hoautophagy and translation,
meostasis control mechanism
both central nodes in the
represents a conserved basis
maintenance of cellular hofor aging, the prevention of
meostasis. The data prewhich will take us one step
sented by Lu et al. indicate
closer to the elixir of life.
Figure 1. The Deactylation Life Span Clock
that, at least in yeast, Sch9
Deacetylation of the yeast AMPK subunit Sip2 with time acts as an intrinsic
REFERENCES
may be crucial for modulating
aging clock, which drives cellular aging by a mechanism that is distinct from
the
effects
of
dietary
restriction.
As
yeast
cells
age,
Sip2
becomes
deacetyboth intrinsic and extrinsic
Apfeld, J., OConnor, G., McDonagh,
lated, removing its repression of the AMPK homolog Snf1 and inducing cellular
aging, emphasizing the imaging via Sch9 phosphorylation. The life span clock can be wound backward
T., DiStefano, P.S., and Curtis, R.
portance of identifying the
via Sip2 acetylation mimics or by inactivating the deacetylase Rpd3,
(2004). Genes Dev. 18, 30043009.
promoting longevity. These findings describe a novel level of AMPK regulation
downstream targets of Sch9Ashrafi, K., Lin, S.S., Manchester,
with age and pose new questions about the links between energy homeostasis
mediated longevity.
J.K., and Gordon, J.I. (2000). Genes
and longevity and the potential of AMPK as a target to treat age-related
The study by Lu and colDev. 14, 18721885.
disease.
leagues describes a novel
Fontana, L., Partridge, L., and Longo,
V.D. (2010). Science 328, 321326.
role for acetylation of a nonhistone protein in mediating yeast life span antagonize AMPK. Finally, little is currently Greer, E.L., Dowlatshahi, D., Banko, M.R., Villen,
and provides an exciting link between known about the relationship between J., Hoang, K., Blanchard, D., Gygi, S.P., and
AMPK and S6K that may be critical AMPK and S6K in mammals, but in at least Brunet, A. (2007). Curr. Biol. 17, 16461656.
for mediating intrinsic cellular aging. one cancer model, stimulation of AMPK Lu, J.-Y., Lin, Y.-Y., Sheu, J.-C., Wu, J.-T., Lee,
However, it also leaves some questions results in reduced, rather than enhanced, F.-J., Chen, Y., Lin, M.-I., Chiang, F.-T., Tai,
unanswered. Foremost of these is the phosphorylation of S6K (Taliaferro-Smith T.-Y., Berger, S.L., et al. (2011). Cell 146, this issue,
969979.
extent to which this mechanism will trans- et al., 2009).
late to higher organisms. Indeed, several
The effects of yeast AMPK activity Mair, W., Morantte, I., Rodrigues, A.P.C., Manning,
lines of evidence suggest that, despite on life span also contrast with what is G., Montminy, M., Shaw, R.J., and Dillin, A. (2011).
Nature 470, 404408.
a high degree of conservation between known in multicellular organisms. In yeast,
Snf1 and AMPK, there has been diver- inhibiting AMPK activity is beneficial Steinberg, G.R., and Kemp, B.E. (2009). Physiol.
gence in their regulation. Whereas mam- for life span, whereas inhibiting AMPK Rev. 89, 10251078.
malian AMPK senses cellular energy shortens life span in both C. elegans Taliaferro-Smith, L., Nagalingam, A., Zhong, D.,
status via nucleotide binding to its g (Apfeld et al., 2004) and Drosophila Zhou, W., Saxena, N.K., and Sharma, D. (2009).
subunit, this does not seem to be the melanogaster (Tohyama and Yamaguchi, Oncogene 28, 26212633.
case for Snf1 (Steinberg and Kemp, 2010). Conversely, activating AMPK Tohyama, D., and Yamaguchi, A. (2010). Biochem.
2009). Furthermore, unlike yeast, mam- increases C. elegans life span (Apfeld Biophys. Res. Commun. 394, 112118.
mals only have two AMPK b subunits, et al., 2004), dependent upon downstream Zhang, J., Olsson, L., and Nielsen, J. (2010). Mol.
and as yet, neither has been shown to transcriptional regulators FOXO (Greer Microbiol. 77, 371383.
Leading Edge
Previews
A New Shield for a Cytokine Storm
Akiko Iwasaki1,* and Ruslan Medzhitov1,2,*
1Department
of Immunobiology
Hughes Medical Institute
School of Medicine, Yale University New Haven, CT, 06520, USA
*Correspondence: akiko.iwasaki@yale.edu (A.I.), ruslan.medzhitov@yale.edu (R.M.)
DOI 10.1016/j.cell.2011.08.027
2Howard
Highly virulent influenza virus infection results in excessive cytokine production, recruitment of
leukocytes, and immune-mediated pulmonary injury. Teijaro et al. (2011) now demonstrate that
sphingosine-1-phosphate receptor 1 ligands suppress all features of flu-inflicted pathological
inflammation and place the endothelium at the center of this regulatory network.
Recognition and rapid clearance of pathogens by the innate immune system provide the first line of defense in metazoan
organisms. However, excessive activation of the innate immune system in
response to pathogens can lead to pathological inflammatory consequences. In
the case of highly virulent 1918 and avian
H5N1 influenza virus infections, early
recruitment of inflammatory leukocytes
to the lung, followed by excessive early
cytokine responses (known as a cytokine
storm), is considered to be the key contributor to morbidity and mortality of the
infection (Tscherne and Garca-Sastre,
2011). Likewise, the virulence of the pandemic 2009 H1N1 swine flu is associated with viral replication in the lower
respiratory tract and more severe pulmonary damage compared to seasonal flu
(Tscherne and Garca-Sastre, 2011).
However, the cell types that are responsible for the initiation and amplification of
the cytokine storms that follow virulent
influenza infection remain unclear. In this
issue, Teijaro et al. (2011) have uncovered
an unexpected role of endothelial cells in
coordinating the inflammatory sequelae
via sphingosine-1-phosphate signaling.
Sphingosine-1-phosphate is a metabolite of sphingolipid and is a ligand for a
family of five G protein-coupled receptors, S1P15 (Rosen et al., 2009). Differential expression of these receptors
enables control of angiogenesis, heart
development, and immunity in a highly
specific manner. The sphingosine analog
FTY720, a well-known immunomodulator
that has recently been approved for the
treatment of multiple sclerosis, is phosphorylated in vivo by sphingosine kinase
2 to produce a ligand for the S1P re-
ceptors S1P1, 35. Its mode of action is

through sequestration of lymphocytes in
lymph nodes away from peripheral sites
of inflammation. Prolonged exposure to
FTY720 causes S1P1 endocytosis and
degradation, impairing the ability of
lymphocytes to respond to endogenous
S1P to egress out of lymph nodes (Cyster,
2005).
The study by Teijaro et al. (2011) followed up on previous findings by the
same group that local intratracheal instillation of S1P ligands reduces cytokine
responses following respiratory infections
with a mouse-adapted influenza virus
(Marsolais et al., 2008, 2009) or human
2009 H1N1 influenza isolate (Walsh
et al., 2011). These studies also revealed
that the S1P1-specific agonists do not
affect the generation of adaptive immune
responses (CD8 T cells and neutralizing
antibodies) and do not alter viral replication in vivo (Marsolais et al., 2008). Using
S1P1-specific agonists, the current
study shows that stimulation of S1P1
alone recapitulates much of the suppressive phenotype of AAL-R, which is a
broad-spectrum agonist of S1P receptors. Notably, intratracheal instillation of
CYM-5442 (an S1P1 agonist) on hours 1,
13, 25, and 37 of influenza virus challenge
significantly dampened type I interferon
(IFN), cytokine, and chemokine release
into the bronchoalveolar lavage (BAL)
and resulted in a significant decrease in
infiltration of monocytes, macrophages,
neutrophils, and natural killer (NK) cells
into the lung. In addition, oral administration of another S1P1 agonist, RP-002, at
1 and 25 hr after infection resulted in
increased survival of mice challenged
with 2009 H1N1 human isolate (Figure 1).
To address the mechanism of action,

the authors examined the expression of
S1P1 using S1P1-eGFP knockin mice
(Cahalan et al., 2011). As reported previously, in addition to lymphocytes, a high
level of S1P1 expression was found on
lymphatic and vascular endothelial cells
isolated from the lung (Cahalan et al.,
2011). However, influenza-induced cytokine responses and cellular recruitment
to the lung were blocked by CYM-5442
in RAG2-deficient mice (devoid of T and
B lymphocytes), suggesting that S1P1
stimulation dampens innate immune responses in a manner that is independent
of its well-known function in lymphocyte
sequestration.
Next, the authors demonstrated that
production of chemokines CCL2, CCL5,
and CXCL10 from vascular and lymphatic
endothelium following influenza infection
was significantly reduced by CYM-5442
treatment in vivo. Intratracheal instillation
of recombinant CCL2 restored monocyte,
macrophage, and NK cell recruitment
to the lung of mice treated with CYM5442. Surprisingly, CCL2 was not sufficient to restore IFN or cytokine responses
in the lung. Further, depletion of CD11b+
cells (including monocytes, macrophages, neutrophils, and NK cells) using
antibody treatment only resulted in reduction in IFN-g secretion but did not affect
the levels of IFN-a, CCL2, IL-6, or TNF-a
in the lung. These data indicate that
inflammatory leukocyte recruitment is
not sufficient for cytokine storm in the
face of CYM-5442 treatment and is not
required for the production of the majority
of the cytokines in response to influenza
infection. Finally, type I IFNs were
placed upstream of the cytokine storm
targets specifically endotheduring influenza infection, as

lial cells for replication in birds
IFNabR-deficient mice failed
(Feldmann et al., 2000). Thus,
to secrete type I IFN, chemoit would be interesting to exkines, or cytokines despite
amine whether S1P1 agonists
the normal recruitment of
inflammatory leukocytes.
will be effective in the case of
The results of this study
H5N1 influenza. Finally, the
reveal an important role of
possible role of endothelial
S1P1 as a regulator of inflamS1P1 in other conditions
mation but also raise a host of
associated with excessive
questions. First, how does
cytokine production would
S1P1 agonism result in global
be an exciting question for
suppression of cytokines?
future studies.
This may occur both at the
cell-intrinsic level within the
REFERENCES
lung endothelial cells and at
the cell-extrinsic level in the
Cahalan, S.M., Gonzalez-Cabrera,
recruited leukocytes. S1P1
P.J., Sarkisyan, G., Nguyen, N.,
receptor is coupled to GaI,
Schaeffer, M.T., Huang, L., Yeager,
and receptor engagement
A., Clemons, B., Scott, F., and
triggers a multitude of downRosen, H. (2011). Nat. Chem. Biol.
7, 254256.
stream signaling pathways,
Cyster, J.G. (2005). Annu. Rev. Imincluding PI3K and Rac actimunol. 23, 127159.
vation and promoting cell
Feldmann, A., Schafer, M.K.,
survival, motility, and barrier
Garten, W., and Klenk, H.D. (2000).
functions. S1P1 also activates
Figure 1. Suppression of Inflammation by S1P1 Agonists
J. Virol. 74, 80188027.
Influenza
virus
infects
the
lung
epithelial
cells
and
subsequently
alveolar
the MAP kinase and phosmacrophages and other leukocytes. This results in type I IFN, cytokine, and
Manicassamy, B., Manicassamy, S.,
pholipase C pathways and
chemokine secretion by various cell types both resident and recruited. S1P1
Belicha-Villanueva, A., Pisanelli, G.,
intracellular mobilization of
ligands, CYM05442 and RP-002, engage S1P1 on vascular and lymphatic (not
Pulendran, B., and Garca-Sastre,
shown) endothelial cells and block chemokine secretion from endothelial cells,
calcium signaling, resulting
A. (2010). Proc. Natl. Acad. Sci.
which,
in
turn,
block
recruitment
of
leukocytes
to
the
lung.
In
addition,
S1P
1
in cell proliferation and cytoUSA 107, 1153111536.
signaling suppresses type I IFN, cytokine, and chemokine secretion from other
kine secretion (Rosen et al.,
cells, resulting in reduced immunopathology. However, S1P1 agonists do not
Marsolais, D., Hahm, B., Edelmann,
2009). However, it remains
impair dendritic cell activation, thereby enabling T cell and B cell responses to
K.H., Walsh, K.B., Guerrero, M.,
be initiated in the draining lymph node.
unclear whether any of these
Hatta, Y., Kawaoka, Y., Roberts, E.,
Oldstone, M.B., and Rosen, H.
pathways directly inhibit cyto(2008). Mol. Pharmacol. 74, 896903.
kine and chemokine production by endothelial cells or, alternatively, et al., 2009). Whether endocytosis and Marsolais, D., Hahm, B., Walsh, K.B., Edelmann,
whether S1P1-activated endothelial cells degradation of S1P1 are requisite for K.H., McGavern, D., Hatta, Y., Kawaoka, Y.,
produce anti-inflammatory mediators. It immunosuppression by CYM-5442 and Rosen, H., and Oldstone, M.B. (2009). Proc. Natl.
is also possible that the anti-inflammatory other S1P1 agonists remains to be deter- Acad. Sci. USA 106, 15601565.
effect of S1P1 engagement is an indirect mined. In this regard, it is interesting to Rosen, H., Gonzalez-Cabrera, P.J., Sanna, M.G.,
consequence of its effect on vascular speculate whether similar blockade of and Brown, S. (2009). Annu. Rev. Biochem. 78,
inflammation is caused by FTY720 and 743768.
integrity.
Second, why doesnt the endogenous whether the clinical effects of this com- Teijaro, J.R., Walsh, K.B., Cahalan, S., Fremgen,
S1P in the blood or lymph, which is pound may, in part, depend on mecha- D.M., Roberts, E., Scott, F., Martinborough, E.,
maintained at high concentration (100 nisms that extend beyond lymphocyte Peach, R., Oldstone, M.B.A., and Rosen, H.
(2011). Cell 146, this issue, 980991.
1,000 nM in blood and 30300 nM in sequestration.
Following seasonal influenza infection, Tscherne, D.M., and Garca-Sastre, A. (2011). J.
lymph), trigger a similar response in the
endothelial cells? Unlike S1P, the syn- the virus replicates primarily in the lung Clin. Invest. 121, 613.
thetic agonists of S1P1, including epithelium, followed by infection of alveoWalsh, K.B., Teijaro, J.R., Wilker, P.R., Jatzek, A.,
CYM-5442, induce prolonged signaling, lar macrophages, dendritic cells, NK cells, Fremgen, D.M., Das, S.C., Watanabe, T., Hatta,
polyubiquitination of S1P1 followed by and B cells (Manicassamy et al., 2010) M., Shinya, K., Suresh, M., et al. (2011). Proc.
degradation in the lysosomes (Rosen (Figure 1). In contrast, H5N1 avian flu Natl. Acad. Sci. USA 108, 1201812023.
Leading Edge
Previews
Synaptic Plasticity of Feeding Circuits:
Hormones and Hysteresis
Marcelo O. Dietrich1,4 and Tamas L. Horvath1,2,3,*
1Section
of Comparative Medicine
of Obstetrics, Gynecology and Reproductive Sciences
3Department of Neurobiology
Yale University School of Medicine, New Haven, CT 06520, USA
4Programa de Po
s-graduacao em Bioqumica, Department of Biochemistry, Universidade Federal do Rio Grande do Sul,
Porto Alegre RS 90035, Brazil
*Correspondence: tamas.horvath@yale.edu
DOI 10.1016/j.cell.2011.08.031
2Department
The drive to eat is controlled by neuronal circuits in the hypothalamus that respond to hormones
signaling hunger or satiety. In this issue of Cell, Yang et al. (2011) reveal an AMPK-dependent
synaptic pathway that sustains excitatory stimulation of the NPY/AgRP neurons that promote
feeding behavior until satiety signals kick in.
Metabolic hormones such a ghrelin, signaling food deprivation, and leptin, signaling satiety, stimulate synaptic activity
and plasticity within the neuronal circuits
in the hypothalamus that control feeding
behavior (Pinto et al., 2004). This process
was suggested to play an important role
in regulating metabolism (Horvath and
Diano, 2004). However, the mechanisms
that bring about these rapid changes in
synaptic connectivity and activity and
how long they persist remain ill defined.
In this issue of Cell, Yang et al. (2011)
provide a remarkable set of novel findings
pinpointing intracellular and intercellular
substrates of synaptic plasticity on the
orexigenic NPY/AgRP neurons. Their
findings demonstrate why the response
to a pulse of the appetite-stimulating hormone ghrelin can persist for hours, and
how it can be turned off in response to
a pulse of leptin. Using in vitro electrophysiological approaches to record from
neurons marked by the expression of the
neuropeptide NPY, most of which also
synthesize the neuropeptide AgRP, the
authors found that food deprivation increases the frequency of action potentials
in the these neurons (referred to as AgRP
neurons by Yang et al.). The increased
activity occurs through an AMPA-mediated increase in frequency of miniature
excitatory postsynaptic currents (mEPSCs)
onto these cells.
This increase in excitatory inputs is
driven by AMPK- and ryanodine re-
ceptor-mediated calcium release from

internal stores in presynaptic sites. Administration of ghrelin mimicked the
effect of food deprivation on mEPSCs in
in vitro experiments, consistent with
ghrelins previously identified action on
midbrain dopamine neurons (Abizaid
et al., 2006). The authors demonstrate
that calcium/calmodulin-dependent protein kinase kinase (CAMKK) activates
AMPK in response to ghrelin to unleash
synaptic activity that exhibits hysteresis,
i.e., it can persist for hours. The hysteretic nature of this signaling is likely due
to the positive feedback loop between
calcium release triggered by AMPK
activity, which in turn leads to additional
AMPK activity promoted by CAMKK.
The enhanced synaptic transmission
was switched off by the appetite-suppressing hormone leptin, which activates
the pro-opiomelanocortin (POMC) neurons. Intriguingly, the authors show that
opiods derived from these neurons,
most likely b-endorphin, reverse the
AMPK-mediated upregulation of excitatory inputs onto AgRP neurons (Figure 1A). The findings thus provide a putative physiological explanation for why
mammals no longer feel the drive to eat
once satiety is reached.
Changes in synaptic transmission occur within a very short period of time in
response to the changing metabolic environment. One of the ingenious aspects of
the work of Yang et al. is that the authors
recognized that such a physiological

phenomenon is virtually impossible to
thoroughly investigate with any available
transgenic or knockout technology.
Thus, the authors used elegantly applied
pharmacological tools to test and answer,
in a manner not previously attempted,
fundamental questions about synaptic
transmission governed by metabolic
signals.
Conceptually, the work of Yang et al.
(2011) builds on previous reports that
have proposed mechanisms related to
AgRP neuronal plasticity, but the authors
excel in providing a refined and unifying
framework for synaptic regulation, maintenance, and resetting by metabolic alterations. The AgRP neurons have been
considered the primary responders to
gut-derived metabolic hormones such as
ghrelin. This notion is left in the dust by
the study of Yang et al., whose findings
point to an AMPK-dependent presynaptic
mechanism that kicks off adaptation
to ghrelin. This presynaptic mechanism
likely acts in synergy with autonomous
cell adaptations that occur within the
AgRP neurons, allowing them to increase
firing rate in a sustained way in response
to food deprivation (Takahashi and
Cone, 2005) or ghrelin (Kohno et al.,
2008) in the absence of synaptic inputs.
This cell-autonomous mechanism is also
dependent upon AMPK, as well as on
downstream mitochondrial factors and
Sirtuin 1 activation in AgRP neurons
Figure 1. AgRP Neurons Respond to Changing Metabolic States

(A) In response to food deprivation and the ghrelin hormone, a presynaptic pathway involving a positive feedback loop of AMPK-mediated calcium release
stimulates the activity of AgRP neurons. The activity of these neurons is known to drive feeding behavior. The hormone leptin signals satiety and induces the
POMC neurons to release opioids that turn off the AMPK pathway and, consequently, AgRP neuronal activity. POMC neurons are also known to have an inhibitory
synaptic connection with AgRP neurons mediated by GABA, NPY, and/or AgRP.
(B) Under in vivo conditions, a combination of mechanisms likely contributes to the regulation of AgRP neuronal activity. In addition to the presynaptic
signaling pathway uncovered by Yang et al., there may be postsynaptic and cell-autonomous intracellular signaling cascades. In addition, the
AMPK-dependent presynaptic mechanism could also be located in other cell types, such as the POMC neurons and astrocytes (yellow). Ghrsr1 is the ghrelin
receptor.
(Andrews et al., 2008; Kohno et al., 2008;

Dietrich et al., 2010) (Figure 1B).
Yang et al. did not identify the presynaptic cells that undergo AMPK-dependent calcium release to stimulate AgRP
neurons. The presynaptic nature of the
effect was concluded based in part upon
comparing the effects of AMPK inhibitors
when permitted to affect all cells versus
when restricted to AgRP neurons by
direct intracellular delivery. Thus the identities of the cells that are mediating the
responses to food deprivation and ghrelin
remain in question. The data presented
by Yang et al. indicate that presynaptic
boutons in direct contact with the AgRP
neurons might be the site where the intracellular events involving AMPK, calcium
mobilization, and opioid action take place
to affect release probabilities. This is a
very reasonable assumption, and the cells
of origin of these terminals will be of
interest to pursue. However, other mechanisms may also be in place in support
of this process. For example, astrocytes
that surround glutamatergic synapses
could also contain the AMPK-dependent
machinery to modulate synaptic plasticity
and neurotransmitter release probability
by regulating the synaptic bioenergetics
and/or neurotransmitter uptake (Figure 1B). Alternatively, in line with the

conclusions of the current study, the
opioid factor released by the POMC
neurons could diffuse into the arcuate
nucleus, a cluster of hypothalamic neurons that are involved in neuroendocrine
responses, resulting in local modulatory
activity. In addition, although the current
report focused on the role of the excitatory projections onto the AgRP neurons,
it is also possible that inhibitory synapses
may be involved in regulating AgRP neuronal excitability.
Finally, it is worth noting that the experiments performed by Yang et al. were
in vitro recordings. In such settings, the
milieu is different from physiological situations in which AgRP neurons are active.
For example, the media used for recordings contained an 11 mM concentration
of glucose. This level of glucose is at least
a magnitude above that which occurs
during a food-deprived state, and it is
substantially higher than that of the fed
state. Lower, more physiological levels
of glucose in such in vitro conditions
were shown to have differential effects
on AgRP neurons depending on postsynaptic AMPK content (Claret et al., 2007).
Additionally, the effect of the AMPK

activator AICAR (used by Yang and
colleagues as well) on feeding behavior
was blunted in mice lacking mitochondrial
uncoupling protein 2 (Andrews et al.,
2008), suggesting that AMPKs role in
hypothalamic circuit regulation may be
more complex.
In summary, the tour-de-force study by
Yang et al. highlights a fundamentally
novel mechanism that depicts regulatory
principles of hypothalamic AgRP neurons
in a paradigm-shifting manner. The paper
indicates that the current dogma that
AgRP neurons represent the first-order
neurons in brain sensing peripheral metabolic status may need to be revised.
REFERENCES
Abizaid, A., Liu, Z.-W., Andrews, Z.B., Shanabrough, M., Borok, E., Elsworth, J.D., Roth, R.H.,
Sleeman, M.W., Picciotto, M.R., Tschop, M.H.,
et al. (2006). J. Clin. Invest. 116, 32293239.
Andrews, Z.B., Liu, Z.W., Walllingford, N., Erion,
D.M., Borok, E., Friedman, J.M., Tschop, M.H.,
Shanabrough, M., Cline, G., Shulman, G.I., et al.
(2008). Nature 454, 846851.
Claret, M., Smith, M.A., Batterham, R.L., Selman,
C., Choudhury, A.I., Fryer, L.G.D., Clements, M.,
Al-Qassab, H., Heffron, H., Xu, A.W., et al. (2007).

J. Clin. Invest. 117, 23252336.
Horvath, T.L., and Diano, S. (2004). Nat. Rev.

Neurosci. 5, 662667.
Dietrich, M.O., Antunes, C., Geliang, G., Liu, Z.W.,

Borok, E., Nie, Y., Xu, A.W., Souza, D.O., Gao, Q.,
Diano, S., et al. (2010). J. Neurosci. 30, 11815
11825.
Kohno, D., Sone, H., Minokoshi, Y., and Yada, T.

(2008). Biochem. Biophys. Res. Commun. 366,
388392.
Pinto, S., Roseberry, A.G., Liu, H., Diano, S.,

Shanabrough, M., Cai, X., Friedman, J.M., and
Horvath, T.L. (2004). Science 304, 110115.
Takahashi, K.A., and Cone, R.D. (2005). Endocrinology 146, 10431047.
Yang, Y., Atasoy, D., Su, H.H., and Sternson, S.M.
(2011). Cell 146, this issue, 9921003.
Leading Edge
Perspective
DNA Demethylation Dynamics
Nidhi Bhutani,1,2 David M. Burns,1 and Helen M. Blau1,*
1Baxter Laboratory for Stem Cell Biology, Institute for Stem Cell Biology and Regenerative Medicine, Department of Microbiology and
Immunology, Stanford University School of Medicine, Stanford, CA 94305-5175, USA
2Present address: Department of Orthopaedic Surgery, Stanford University, Stanford, CA 94305-5341, USA
*Correspondence: hblau@stanford.edu
DOI 10.1016/j.cell.2011.08.042
The discovery of cytosine hydroxymethylation (5hmC) suggested a simple means of demethylating

DNA and activating genes. Further experiments, however, unearthed an unexpectedly complex
process, entailing both passive and active mechanisms of DNA demethylation by the ten-eleven
translocation (TET) and AID/APOBEC families of enzymes. The consensus emerging from these
studies is that removal of cytosine methylation in mammalian cells can occur by DNA repair. These
reports highlight that in certain contexts, DNA methylation is not fixed but dynamic, requiring
continuous regulation.
Introduction
The significant impact of DNA methylation patterns on cell and
organismal fate is perhaps most graphically exemplified in
honeybees, in which differential DNA methylation determines
whether the bee will be a worker or a queen (Kucharski et al.,
2008). In mammals, DNA methylation has also long been considered integral to fundamental choices, including the long-term
gene silencing that leads to genomic imprinting, X chromosome
inactivation, suppression of transposable elements, and the
establishment and maintenance of stable cellular identities
(Bird, 2002; De Carvalho et al., 2010; Deaton and Bird, 2011;
Goll and Bestor, 2005; Jaenisch and Bird, 2003). Yet, studies
of cellular reprogramming by three approachesnuclear transfer, cell fusion, and induced pluripotency by defined factors
(i.e., iPSCs)all demonstrate that fixed and stable differentiated cellular states can be radically altered (Jullien et al., 2011;
Yamanaka and Blau, 2010). Concurrently, accumulating evidence has suggested that DNA methylation may be reversible
in mammalian cells; however, knowledge of the requisite molecules and mechanisms underlying this process has been lacking.
In this Perspective, we focus on recent reports that now identify
enzymes capable of mediating DNA demethylation in mammalian cells. These findings raise the possibility that regulation by
DNA methylation is at times quite dynamic, providing exciting
insights into why reprogramming of cell fates is possible.
Recent discoveries have generated substantial excitement, as
they show that cytosines in mammalian cells can be hydroxymethylated to 5hmC (5-hydroxymethylcytosine) (Figure 1) (Kriaucionis and Heintz, 2009; Tahiliani et al., 2009). 5hmC is especially
abundant in tissues such as brain and in pluripotent embryonic
stem cells (ESCs), but it is also present at lower levels in blood,
lung, kidney, and muscle (Globisch et al., 2010; Kriaucionis
and Heintz, 2009; Ruzov et al., 2011; Song et al., 2011; Tahiliani
et al., 2009). Initially, hydroxylation seemed like a probable
means of activating genes silenced by methylation (Ito et al.,
2010; Tahiliani et al., 2009), but recent studies rule out this simple
hypothesis (Ficz et al., 2011; Pastor et al., 2011; Williams et al.,
2011; Wu et al., 2011b; Xu et al., 2011). Moreover, although

several groups have investigated the genomic distribution of
DNA hydroxymethylation, the role and functional significance
of this modification are still unclear (Ficz et al., 2011; Pastor
et al., 2011; Williams et al., 2011; Wu et al., 2011b; Xu et al.,
2011).
We postulate here, based on analyses of recent evidence from
our and other laboratories (Bhutani et al., 2010; Cortellino et al.,
2011; Guo et al., 2011; He et al., 2011), that DNA methylation
and demethylation can be a two-way street, characterized by
multiple pathways (Figure 1). Importantly, these findings suggest
that 5hmC may serve as an intermediate for the removal of methylated cytosines either by (1) passive dilution via the presence of
5hmC, which impairs remethylation by DNA methyltransferases
(DNMTs) when cells divide, or (2) active replacement of modified
cytosines via DNA repair in the absence of cell division. The role
of DNA repair in DNA demethylation is well established in plants
(Gehring et al., 2009). However, this pathway was not thought to
operate in mammals, as no mammalian orthologs of the plant
enzymes with similar activities were readily apparent (Gehring
et al., 2009). Moreover, the mammalian enzymes that have
recently been identified as the lead actors in the demethylation
plot are well known for their involvement in other processes,
for example in leukemia (TETs) and in antibody diversification
(AID). Thus, their roles in the saga of DNA demethylation are
entirely new. We postulate that the discovery of these regulators
and their newly identified roles will provide insights into the raison detre for DNA methylation, its modifications, and its role in
gene expression, cell-fate determination, and nuclear reprogramming.
Active Mechanisms for Loss of DNA Methylation:
The Methylation Editors
DNMTs are responsible for the establishment and maintenance
of DNA methylation as well as passive DNA demethylation in
mammalian cells. It has long been thought that an absence or
reduction in the DNMT levels gradually and passively removes
Figure 1. DNA Demethylation Pathways

Passive DNA demethylation has long been known to occur by a reduction in activity or absence of DNA methyltransferases (DNMTs) (black). DNMT3A and 3B are
responsible for de novo DNA methylation, whereas DNMT1 maintains DNA methylation patterns through successive rounds of cell division. Recently, three
enzyme families have been implicated in active DNA demethylation via DNA repair. (1) 5-methylcytosine (5mC) can be hydroxylated by the ten-eleven translocation (TET) family of enzymes (blue) to form 5-hydroxymethylcytosine (5hmC) or further oxidized to 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC). (2)
5mC (or 5hmC) can be deaminated by the AID/APOBEC family members (purple) to form 5-methyluracil (5mU) or 5-hydroxymethyluracil (5hmU). (3) Replacement
of these intermediates (i.e., 5mU, 5hmU, or 5caC) is initiated by the UDG family of base excision repair (BER) glycosylases (green) like TDG or SMUG1, culminating
in cytosine replacement and DNA demethylation.
DNA methylation in early mammalian development (Monk et al.,

1991; Rougier et al., 1998). Specifically, it is well known that de
novo methylation in early development is established by DNA
methyltransferases 3A (DNMT3A) and 3B (DNMT3B) (reviewed
in Law and Jacobsen, 2010). Once established, methylation
patterns are faithfully maintained through cell divisions by
DNMT1 (Law and Jacobsen, 2010). Thus, to date, inhibition of
DNMTs constitutes the primary means of passive DNA demethylation.
A few early studies implicated demethylation of DNA by a rapid
and active mechanism, independent of cell division (Mayer et al.,
2000; Oswald et al., 2000; Paroush et al., 1990; Zhang et al.,
2007), but how this active removal is achieved remained a
mystery. Indeed, for decades an enzyme that could cleave the
methyl group was sought but not found, suggesting that such
a chemical reaction might simply not be possible (Bird, 2002).
In the past year or so, a flurry of studies (Bhutani et al., 2010; Cortellino et al., 2011; Ficz et al., 2011; Guo et al., 2011; He et al.,
2011; Ito et al., 2011; Koh et al., 2011; Pastor et al., 2011; Popp
et al., 2010; Rai et al., 2008; Song et al., 2011; Williams et al.,
2011; Wu et al., 2011a, 2011b; Xu et al., 2011) have identified key
players in this process, which may have been overlooked previously because they act indirectly to mediate active DNA demethylation. These enzymes first modify the methylated cytosine
(by hydroxylation, deamination, oxidation, or a combination of
these modifications), leading to its replacement by DNA repair.
The above studies connect three enzymatic families to active
DNA demethylation: the ten-eleven translocation (TET) family,
which modifies methylated cytosines first by hydroxylation and

then by further oxidation; the AID/APOBEC family, which deaminates the base (5mC or 5hmC); and finally, a family of base
excision repair (BER) glycosylases, which mediate DNA repair
(Figure 1). Here, we synthesize recent data that link these
enzyme families. We suggest a role for them in active DNA demethylation in mammals in response to cell signaling and in early
development, differentiation, and nuclear reprogramming to
new cell states. Furthermore, we provide a speculative scheme
for the circuitry by which DNA can be demethylated and remethylated, and how these states may be rapidly reversed by these
methylation editors.
The TET Family: Mediators of 5mC to 5hmC Conversion
The existence of 5hmC was reported in the 1950s (Wyatt and
Cohen, 1952), but its significance was unknown, and it remained
largely ignored for the next half century. The discovery of the TET
proteins, TET1, 2, and 3, that catalyze the conversion of 5-methylcytosine (5mC) to 5hmC heralded a revival of interest in this
modified base. TET1 was initially identified as a fusion partner
of the MLL protein in acute myeloid leukemia and named
leukemia-associated protein, LCX, although its functional role
in this type of cancer remained unknown (Ono et al., 2002).
More recently, Rao and colleagues rekindled interest in the
TETs when they identified them as potential modifiers of 5mC
(Tahiliani et al., 2009). Based on knowledge that in Trypanosoma,
the J base binding proteins 1 (JBP1) and 2 (JBP2) oxidize
5-methylthymine, Rao and colleagues cloned the mammalian
homolog, TET1, cleverly reasoning that it might serve a similar
function in higher metazoa and mediate 5mC hydroxylation (Tahiliani et al., 2009). Recombinant human TET1 was found to be
capable of converting 5mC to 5hmC in mammalian DNA,
providing evidence for its putative role in mediating DNA demethylation (Tahiliani et al., 2009). Like their human counterparts,
mouse TET1, 2, and 3 catalyze the conversion of 5mC to
5hmC (Ito et al., 2010).
Much of the initial excitement regarding the discovery of 5hmC
was the prediction that it could readily lift the repression of gene
expression imposed by 5mC at many gene promoters (Ito et al.,
2010; Tahiliani et al., 2009). However, like methylation, a high
concentration of 5hmC correlates with transcriptionally nonproductive or altogether inactive gene promoters (Ficz et al., 2011;
Pastor et al., 2011; Williams et al., 2011; Wu et al., 2011a; Xu
et al., 2011). Thus, contrary to expectations, the 5mC to 5hmC
modification is clearly not the functional equivalent of 5mC to
cytosine, which is associated with derepression of certain
gene promoters (Pastor et al., 2011).
The hypothesis that cytosine hydroxylation might play a functional role in maintaining the pluripotent state was first suggested
by Zhang and colleagues (Ito et al., 2010). They reported that
TET1 results in a loss of ESC self-renewal by reducing the
expression of the pluripotency regulator NANOG. This finding,
however, has been challenged by others who suggest that,
although levels of TET1 and TET2 (and therefore 5hmC) are
high in ESCs, these proteins largely mediate regulation of
lineage-specific genes, not the pluripotency regulator NANOG
(Ficz et al., 2011; Koh et al., 2011).
In parallel with studies in ES cells, other experiments have
been performed to determine whether the TET proteins are
involved in active DNA demethylation in early development. In
this case, the third family member, TET3, is most abundant
and plays a role in the rapid and active loss of 5mC in the male
pronucleus upon zygote formation prior to cell division. Recent
experiments have shed light on the mechanism underlying this
process. An increase in 5hmC is concomitant with a decrease
in 5mC in zygotes (as determined by immunohistochemistry),
suggesting that 5mC is converted to 5hmC (Iqbal et al., 2011;
Wossidlo et al., 2011). In addition, knocking down TET3 by
RNA interference (RNAi) led to an increase in 5mC. Thus, TET3
is responsible for 5hmC generation post-fertilization in mouse
zygotes, suggesting a potential role for TET proteins in DNA
demethylation early in development.
An unexpected complication of interpretations of experiments
regarding the effects of TET proteins on gene expression is that
TET1 plays a repressive role, independent of its enzymatic
activity as a hydroxylase (Williams et al., 2011; Wu et al.,
2011b). TET1 has been found to associate with two different
transcriptional repressor complexes containing PRC2 (polycomb repressive complex 2) or SIN3A (Swi-independent 3A)
(Williams et al., 2011). SIN3A and TET1 directly interact with
one another (as shown by coimmunoprecipitation), whereas
PRC2 and TET1 may act indirectly (Williams et al., 2011; Wu
et al., 2011b). Importantly, a high degree of target overlap is
observed between TET1 and PRC2, as well as between TET1
and SIN3A, in global chromatin immunoprecipitation (ChIP) analyses (Wu et al., 2011b; Williams et al., 2011). Furthermore, a
subset of TET1 target genes is upregulated upon loss of SIN3A
function by siRNA knockdown, further substantiating that the

SIN3A corepressor complex is required for TET1-mediated
repression of these genes, independent of its catalytic role in
generating 5hmC (Williams et al., 2011).
The rediscovery of 5hmC and TET proteins has led to a rapid
succession of reports regarding the location and putative function of both 5hmC and TET in regulating DNA demethylation
(Cortellino et al., 2011; Ficz et al., 2011; Guo et al., 2011; He
et al., 2011; Ito et al., 2010, 2011; Koh et al., 2011; Kriaucionis
and Heintz, 2009; Pastor et al., 2011; Tahiliani et al., 2009;
Williams et al., 2011; Wossidlo et al., 2011; Wu et al., 2011a,
2011b; Xu et al., 2011). These studies have answered some
fundamental questions, but they also raise others, underscoring
the need for additional experiments that probe the mechanisms
of DNA methylation and demethylation, how they are regulated,
and how they affect gene expression.
In particular, the discovery of 5hmC raises a new technical
conundrum. Many of the past methylation studies have relied
on two techniques that cannot distinguish between 5mC and
5hmC: bisulfite conversion sequencing and methylation-sensitive restriction digests (Huang et al., 2010; Pastor et al., 2011).
New tools have been developed based on specific modifications of 5hmC coupled with DNA immunoprecipitation and
sequencing, chromatographic separation techniques, and improved immunohistochemical visualization using specific antibodies to distinguish 5mC and 5hmC (He et al., 2011; Pastor
et al., 2011; Song et al., 2011; Wossidlo et al., 2011). Experiments performed using these new methodologies will undoubtedly enhance our understanding of the complex relationship
between cytosine methylation and demethylation, as well as
the new roles of cytosine hydoxymethylation and deamination.
The AID/APOBEC Family: Mediators of 5mC
or 5hmC Deamination
Activation-induced cytidine deaminase (AID) has only recently
been implicated in DNA demethylation (Bhutani et al., 2010; Cortellino et al., 2011; Guo et al., 2011; Popp et al., 2010). However,
AID has been a focus of intense study by numerous groups over
the past 10 years because of its critical role in generating antibody
diversity in lymphocytes (reviewed in Chaudhuri et al., 2007;
Delker et al., 2009). In B lymphocytes, AID participates in somatic
hypermutation and class-switch recombination (Muramatsu
et al., 2000), both of which entail error-prone DNA repair and,
therefore, are mutagenic. AID is a member of a family of proteins,
the APOBECs, which unlike AID were originally identified as
RNA editorshence their name apolipoprotein B mRNA-editing
catalytic polypeptides, or APOBECs (reviewed in Conticello
et al., 2007). AID mediates deamination of cytosine residues to
uracils, which are then repaired by either BER or mismatch repair
(MMR). This repair is error prone, leading to mutations essential
to generating the vast repertoire of diverse antibodies seen in
mammals (Liu and Schatz, 2009; Maul and Gearhart, 2010).
AID was thought to preferentially target the immunoglobulin
(Ig) locus in B lymphocytes by unknown mechanisms, as the
frequency of AID-generated mutations at non-Ig loci is very
low. However, recent studies in B lymphocytes deficient in
BER and MMR repair (i.e., Ung / Msh2 / ) revealed that AID
acts extensively on non-Ig loci as well. These regions are protected from mutations, presumably by high-fidelity error-free
repair mechanisms (Liu et al., 2008). Clearly, an increased understanding of how error-prone and error-free DNA repair pathways
are targeted to Ig versus non-Ig loci warrants further investigation, as AID is key to both antibody generation and DNA demethylation.
A role for AID in global DNA demethylation was first shown in
zebrafish embryos by Cairns and colleagues (Rai et al., 2008).
Upon overexpression of AID or zebrafish APOBEC deaminases
and the glycosylase MBD4, active DNA demethylation was observed in zebrafish embryos injected with a methylated linearized nonreplicating DNA. Reik and colleagues suggested a
similar role for AID in global DNA demethylation at a later stage
of embryogenesis in mice (Popp et al., 2010). Mice completely
lacking AID (AID / ) (Muramatsu et al., 2000) exhibited an
increase in genome-wide hypermethylation in their primordial
germ cells (PGCs), suggesting that AID is involved in DNA demethylation. However, if AID-mediated global DNA demethylation
plays a crucial role in early development, a more profound
phenotype would be expected in AID null mice, which are both
viable and fertile, albeit with somewhat smaller litter sizes
(Popp et al., 2010). These findings raise the possibility that, in
the absence of AID, other family members may play compensatory roles in DNA demethylation.
Studies of nuclear reprogramming provided the first evidence
that AID plays a role in active DNA demethylation in mammals
and in somatic cells (Bhutani, et al., 2010). Upon fusion of an
excess of mouse ESCs with human somatic cells (fibroblasts)
in nondividing heterokaryons, rapid demethylation was detected
at the promoters of the human pluripotency genes OCT4 and
NANOG, accompanied by their transcriptional induction. This
effect on reprogramming was dependent on AID, as a reduction
of AID by four different siRNAs completely blocked pluripotency
gene promoter demethylation and gene expression. In somatic
cells, AID does not appear to act globally as in zebrafish embryos
and mouse PGCs (Popp et al., 2010; Rai et al., 2008), but instead
it is targeted to specific loci (Bhutani et al., 2010). However, if
specific loci are involved, the question arises as to how targeting
of this enzyme is mediated. Unlike TET, which has a DNAbinding motif (Ono et al., 2002), AID does not. Future studies to
decipher the mechanism by which AID is targeted will not only
provide insights into its sites of action but also illuminate its
role in active DNA demethylation in different cell types and in
response to different stimuli.
The BER Glycosylase Family: Mediators of DNA Repair
As described above, the accumulating data from the TET and
AID/APOBEC studies suggest that active demethylation involves
cytosine replacement via DNA repair. In principle, 5hmC or 5mC
can be removed passively in the course of cell division. However,
the rapid loss of methylation that occurs independent of DNA
replication (Bhutani et al., 2010; Frank et al., 1990; Oswald
et al., 2000; Paroush et al., 1990) must be mediated by an active
mechanism. In plants, active DNA demethylation is a well-characterized process in which the accepted mechanism involves
the BER pathway. BER glycosylases mediate the first step in
the repair pathway by removing the methylated cytosine and
creating an abasic site, which is then further acted upon by other
enzymes (Gehring et al., 2009). In mammals, the process is more
complex, as no glycosylases have been identified that act
directly on 5mC or 5hmC. Recently, an intermediate step, deamination, has been suggested to precede BER in mammalian DNA
demethylation (Cortellino et al., 2011; Guo et al., 2011). The
family of glycosylases implicated in the BER pathway are
members of the uracil DNA glycosylase (UDG) family that
includes thymine-DNA glycosylase (TDG) and single-strandselective monofunctional uracil-DNA glycosylase 1 (SMUG1)
(Cortellino et al., 2011; Guo et al., 2011). Therefore, like plants,
mammalian DNA demethylation involves DNA repair pathways.
The DNA glycosylases TDG and SMUG1 are capable of
converting 5hmU to cytosine, suggesting that they act in a partnership with TET and AID/APOBEC (Cortellino et al., 2011; Guo
et al., 2011). Notably, mice lacking TDG (TDG / ) are embryonic
lethal, underscoring the significant role that BER glycosylases
play during development and DNA demethylation. A direct
physical interaction has been demonstrated between AID and
TDG by coimmunoprecipitation experiments (Cortellino et al.,
2011). In addition, recent reports suggest that 5hmC can be
further oxidized by TET proteins to form 5-formylcytosine (5fC)
and 5-carboxylcytosine (5caC) (He et al., 2011; Ito et al., 2011).
These are unstable intermediates that can be detected in
ESCs, but they are 100-fold less abundant than 5hmC. Importantly, 5caC can be repaired by TDG, providing further support
for a DNA repair pathway (He et al., 2011). However, another
study also raises the possibility that a putative decarboxylase
could directly convert 5caC to cytosine independent of DNA
repair (Ito et al., 2011). The extent to which DNA repair pathways
are involved in the removal of 5caC and its relative importance to
DNA demethylation and gene expression remain to be determined.
In summary, the BER glycosylases, along with the TET and
AID/APOBEC families of enzymes, mediate DNA demethylation
via DNA repair. It remains to be tested whether other DNA repair
pathways besides BER participate in DNA demethylation.
Examples of Active DNA Demethylation in Mammals
As described below, a body of evidence is accumulating in this
nascent and rapidly evolving field, which supports the thesis
that active DNA demethylation is more common than previously
anticipated. Examples are presented below that indicate the role
of DNA demethylation in rapid responses to changes in extrinsic
signals, in early stages of development, and in highly specialized
postmitotic cells. This is merely the tip of the iceberg, and more
studies are needed to ascertain the extent to which the methylation editors are involved in the spatial and temporal regulation of
DNA demethylation.
Rapid Active DNA Demethylation in Response to Signals
Perhaps the most striking example of active DNA demethylation in adult cells to date is the activity-dependent DNA demethylation at the promoters of brain-derived neurotrophic factor
(BDNF) and fibroblast growth factor 1 (FGF1) in postmitotic
neurons (Martinowich et al., 2003). Recent studies have elegantly elucidated the molecular mechanisms underlying this
active DNA demethylation process and revealed a partnership
between the TET, AID/APOBEC enzymes, and the BER glycosylases (Guo et al., 2011). Reconstitution in HEK293 cells and
knockdown experiments demonstrate that TET-induced conversion of 5mC to 5hmC is followed by AID/APOBEC-mediated
deamination of 5hmC to 5hmU and its further replacement by an

unmethylated cytosine through the BER pathway. Several laboratories have implicated both SMUG1 and TDG as the BER glycosylases in this process (Cortellino et al., 2011; Guo et al., 2011;
He et al., 2011).
Other examples of active DNA demethylation as a rapid
response to signal transduction include interleukein-2 (IL-2)
stimulation of T lymphocytes and estrogen stimulation of breast
cancer cells (Bruniquel and Schwartz, 2003). IL-2 activates T
lymphocytes to mount an immune response, and this process
is accompanied by a rapid demethylation of 5mCs within
20 min. This signal-dependent DNA demethylation is active
because it is unaffected by the presence of an inhibitor of DNA
synthesis. In breast cancer cells, the promoter of the pS2/trefoil
factor 1 (TFF1) gene undergoes cycles of methylation and
demethylation within 2040 min in response to estrogen, demonstrating a strikingly dynamic interplay of demethylation and de
novo methylation (Kangaspeska et al., 2008; Metivier et al.,
2008). The putative roles of hydroxylation, deamination, and
DNA repair remain to be explored in these scenarios.
Active DNA Demethylation in Early Mammalian
Development
DNA demethylation of paternal and maternal genomes differs in
the zygote, indicating that there is specificity and targeting of
the DNA demethylation machinery even at this early stage of
development. Studies report that, following fertilization, the
paternal pronuclei undergo an extensive loss of 5mC (Oswald
et al., 2000), whereas the maternal pronuclei are resistant to
this loss due to the presence of the protein Stella (Nakamura
et al., 2007). The loss of 5mC in the paternal genome is rapid
and independent of cell division, and it serves as a classic
example of active DNA demethylation. More recent studies
have revealed that the active loss of 5mC is actually a conversion
of 5mC to 5hmC in a TET3-dependent manner (Iqbal et al., 2011;
Wossidlo et al., 2011). The fate of the resulting 5hmC and how it
may be converted back to unmethylated cytosines remain to be
elucidated.
A similar active DNA demethylation process has been reported in mouse PGCs (Popp et al., 2010). It is still unknown,
however, whether 5hmC plays a role in this process. The
genome-wide hypermethylation observed in AID null PGCs has
suggested a role for AID-mediated deamination in active DNA
demethylation (Popp et al., 2010). However, this study used
bisulfite conversion and sequencing, which cannot distinguish
between 5mC and 5hmC (Huang et al., 2010). Thus, it remains
possible that 5hmC generated by TET activity is an intermediate
for AID-mediated deamination and subsequent DNA demethylation. Independently, DNA repair by the BER pathway has been
reported by other investigators to occur in PGCs (Hajkova
et al., 2010). Taken together, it is reasonable to speculate that
the TETAID/APOBECBER pathway plays a role in DNA demethylation in early development both in zygotes and in PGCs.
Active DNA Demethylation in Tissue-Specific
Differentiation
Skeletal muscle constitutes an example of somatic cells in
mammals in which active DNA demethylation has been reported
(Blau et al., 1983; Zhang et al., 2007). In these reprogramming
studies, when human fibroblasts and mouse muscle cells were
fused to form nondividing heterokaryons, active DNA demethylation was observed at the human MyoD promoter, which
accompanied its activation and expression in the fibroblasts.
Remarkably, the Cedar laboratory postulated more than 20
years ago that DNA demethylation occurs by an active mechanism in muscle cells (Paroush et al., 1990; Weiss et al., 1996),
but the factors responsible were unknown. These studies
suggest that a dynamic interplay of methylation and demethylation may also function during differentiation.
Is DNA Methylation-Demethylation Bidirectional?
Although indirect evidence has been accumulating for decades,
recent advances discussed here now support the hypothesis
that DNA demethylation and methylation may be bidirectional
and dynamically regulated throughout early and late development and in certain adult tissues, especially the brain (Guo
et al., 2011; Miller and Sweatt, 2007). Much remains to be
learned, including which loci are targeted for demethylation
and how the process is spatially and temporally regulated in
diverse cell types and stages of development. The long-held
notion that DNA methylation patterns are generally maintained
in stable differentiated states is likely true. Nonetheless, nuclear
reprogramming shows that perturbations are possible (Jullien
et al., 2011; Yamanaka and Blau, 2010). As is the case for the
regulation of gene expression by transcription factors (Blau
and Baltimore, 1991; Jacob and Monod, 1961; Ptashne, 2009;
Blau, 1992), the regulation of DNA methylation may also be
continuous and dictated by a balance of enzymes and targeting
factors.
As shown in Figure 1, our current understanding of the DNA
methylation and demethylation circuitry entails members of the
following enzyme families with roles in either passive or active
DNA demethylation: (1) the DNMT family of three methyltransferases responsible for the de novo generation and maintenance
of 5mC. DNA demethylation can occur passively by a dilution or
inactivation of DNMTs; (2) the TET family of three 5mC hydroxylases, which generate 5hmC (and further oxidized intermediates)
from 5mC; (3) the AID/APOBEC family of deaminases, which
initiate an active process of demethylation by deaminating either
5mC or 5hmC generated by the TET family; (4) the family of BER
glycosylases that initiate DNA repair culminating in the replacement of methylated cytosines with unmethylated cytosines. We
have designated these enzymes as the DNA methylation editors
that are responsible for the regulation of the DNA methylome
associated with a particular cell fate. It remains to be determined
whether active DNA demethylation in different scenarios always
requires a representative member from each of these families. In
other words, does the entire TETAID/APOBECBER pathway
operate broadly, or is only a subset thereof required to achieve
active DNA demethylation in different cell contexts.
The concept of a dynamic interplay of regulators has emerged
in parallel with the demonstration of the remarkable plasticity of
cellular fates illustrated by nuclear reprogramming. When the
balance of transcription factors that recognize DNA sequence
is perturbed by either nuclear transfer, cell fusion, or defined
factors (i.e., in generating iPSCs), it leads to a dramatic shift in
cell fate. A provocative, yet perhaps overly simplistic view of
how cell fate is controlled and maintained is provided by an
analogy with a sailboat. Transcription factors comprise the

rudder that determines the direction of the differentiated state
(i.e., whether it is muscle or liver). Threshold concentrations of
transcription factors, achieved by feedback loops, continuously
regulate the differentiated states. Similarly, the editors of DNA
methylation described in this Review can be regulated actively
and continuously serving as the keel and preventing the cell
from responding to minor changes in wind or current. A blast
of ectopic transcription factors can overwhelm the rudder and
reset it as well as the DNA methylation regulators. This occurs
in cellular reprogramming, either following nuclear transfer into
oocytes, upon cell fusion in heterokaryons, or in induced pluripotency (iPSCs). In the first two cases, the somatic nucleus
encounters an overwhelming abundance of pre-existing
proteins, whereas in iPSCs, this protein abundance is progressive, as it derives from the overexpression of four genes (reviewed in Yamanaka and Blau, 2010).
The recent discoveries that TET and AID/APOBEC enzymes
are active regulators of DNA demethylation support the hypothesis that even apparently stable states are continuously regulated (Blau, 1992; Blau and Baltimore, 1991). Thus, the stable
differentiated state is governed by regulatory pathways that
are surprisingly perturbable. This raises the intriguing question
of how cellular plasticity is kept in check to maintain specific
cell fates. A future goal and major challenge is to understand
how cell plasticity can be first enlisted to reprogram cells and
then regulated to derive stable differentiated cell types. Indeed,
understanding the mechanisms that govern this dichotomy is
critical for successfully applying cellular reprogramming to
regenerative medicine.
ACKNOWLEDGMENTS
We thank Steve Hennikoff, Mary Goll, Mark Ptashne, Peter Jones, Jason
Pomerantz, and members of our laboratory for critical reading of the manuscript and constructive suggestions and Stephane Corbel for expert artwork.
This work was funded by a PHS grant 5T32CA09151 awarded by NCI
to D.M.B. and funding from NIH grants HL096113, HL100397, AG020961,
and AG009521; JDRF grant 34-2008-623; CIRM grants RT1-01001 and
RB1-01292; and the Baxter Foundation to H.M.B.
REFERENCES
Bhutani, N., Brady, J.J., Damian, M., Sacco, A., Corbel, S.Y., and Blau, H.M.
(2010). Nature 463, 10421047.
Bird, A. (2002). Genes Dev. 16, 621.
Blau, H.M. (1992). Annu. Rev. Biochem. 61, 12131230.
Blau, H.M., and Baltimore, D. (1991). J. Cell Biol. 112, 781783.
Blau, H.M., Chiu, C.P., and Webster, C. (1983). Cell 32, 11711180.
Bruniquel, D., and Schwartz, R.H. (2003). Nat. Immunol. 4, 235240.
Chaudhuri, J., Basu, U., Zarrin, A., Yan, C., Franco, S., Perlot, T., Vuong, B.,
Wang, J., Phan, R.T., Datta, A., et al. (2007). Adv. Immunol. 94, 157214.
Conticello, S.G., Langlois, M.A., Yang, Z., and Neuberger, M.S. (2007). Adv.
Immunol. 94, 3773.
Cortellino, S., Xu, J., Sannai, M., Moore, R., Caretti, E., Cigliano, A., Le Coz, M.,
Devarajan, K., Wessels, A., Soprano, D., et al. (2011). Cell 146, 6779.
De Carvalho, D.D., You, J.S., and Jones, P.A. (2010). Trends Cell Biol. 20,
609617.
Deaton, A.M., and Bird, A. (2011). Genes Dev. 25, 10101022.
Delker, R.K., Fugmann, S.D., and Papavasiliou, F.N. (2009). Nat. Immunol. 10,
11471153.
Ficz, G., Branco, M.R., Seisenberger, S., Santos, F., Krueger, F., Hore, T.A.,
Marques, C.J., Andrews, S., and Reik, W. (2011). Nature 473, 398402.
Frank, D., Lichtenstein, M., Paroush, Z., Bergman, Y., Shani, M., Razin, A., and
Cedar, H. (1990). Philos. Trans. R. Soc. Lond. B Biol. Sci. 326, 241251.
Gehring, M., Reik, W., and Henikoff, S. (2009). Trends Genet. 25, 8290.
Globisch, D., Munzel, M., Muller, M., Michalakis, S., Wagner, M., Koch, S.,
Bruckl, T., Biel, M., and Carell, T. (2010). PLoS ONE 5, e15367.
Goll, M.G., and Bestor, T.H. (2005). Annu. Rev. Biochem. 74, 481514.
Guo, J.U., Su, Y., Zhong, C., Ming, G.L., and Song, H. (2011). Cell 145,
423434.
Hajkova, P., Jeffries, S.J., Lee, C., Miller, N., Jackson, S.P., and Surani, M.A.
(2010). Science 329, 7882.
He, Y.F., Li, B.Z., Li, Z., Liu, P., Wang, Y., Tang, Q., Ding, J., Jia, Y., Chen, Z., Li,
L., et al. (2011). Science. Published online August 4, 2011.
Huang, Y., Pastor, W.A., Shen, Y., Tahiliani, M., Liu, D.R., and Rao, A. (2010).
PLoS ONE 5, e8888.
Iqbal, K., Jin, S.G., Pfeifer, G.P., and Szabo, P.E. (2011). Proc. Natl. Acad. Sci.
USA 108, 36423647.
Ito, S., DAlessio, A.C., Taranova, O.V., Hong, K., Sowers, L.C., and Zhang, Y.
(2010). Nature 466, 11291133.
Ito, S., Shen, L., Dai, Q., Wu, S.C., Collins, L.B., Swenberg, J.A., He, C., and
Zhang, Y. (2011). Science. Published online July 21, 2011.
Jacob, F., and Monod, J. (1961). J. Mol. Biol. 3, 318356.
Jaenisch, R., and Bird, A. (2003). Nat. Genet. Suppl. 33, 245254.
Jullien, J., Pasque, V., Halley-Stott, R.P., Miyamoto, K., and Gurdon, J.B.
(2011). Nat. Rev. Mol. Cell Biol. 12, 453459.
Kangaspeska, S., Stride, B., Metivier, R., Polycarpou-Schwarz, M., Ibberson,
D., Carmouche, R.P., Benes, V., Gannon, F., and Reid, G. (2008). Nature 452,
112115.
Koh, K.P., Yabuuchi, A., Rao, S., Huang, Y., Cunniff, K., Nardone, J., Laiho, A.,
Tahiliani, M., Sommer, C.A., Mostoslavsky, G., et al. (2011). Cell Stem Cell 8,
200213.
Kriaucionis, S., and Heintz, N. (2009). Science 324, 929930.
Kucharski, R., Maleszka, J., Foret, S., and Maleszka, R. (2008). Science 319,
18271830.
Law, J.A., and Jacobsen, S.E. (2010). Nat. Rev. Genet. 11, 204220.
Liu, M., and Schatz, D.G. (2009). Trends Immunol. 30, 173181.
Liu, M., Duke, J.L., Richter, D.J., Vinuesa, C.G., Goodnow, C.C., Kleinstein,
S.H., and Schatz, D.G. (2008). Nature 451, 841845.
Martinowich, K., Hattori, D., Wu, H., Fouse, S., He, F., Hu, Y., Fan, G., and Sun,
Y.E. (2003). Science 302, 890893.
Maul, R.W., and Gearhart, P.J. (2010). Adv. Immunol. 105, 159191.
Mayer, W., Niveleau, A., Walter, J., Fundele, R., and Haaf, T. (2000). Nature
403, 501502.
Metivier, R., Gallais, R., Tiffoche, C., Le Peron, C., Jurkowska, R.Z.,
Carmouche, R.P., Ibberson, D., Barath, P., Demay, F., Reid, G., et al. (2008).
Nature 452, 4550.
Miller, C.A., and Sweatt, J.D. (2007). Neuron 53, 857869.
Monk, M., Adams, R.L., and Rinaldi, A. (1991). Development 112, 189192.
Muramatsu, M., Kinoshita, K., Fagarasan, S., Yamada, S., Shinkai, Y., and
Honjo, T. (2000). Cell 102, 553563.
Nakamura, T., Arai, Y., Umehara, H., Masuhara, M., Kimura, T., Taniguchi, H.,
Sekimoto, T., Ikawa, M., Yoneda, Y., Okabe, M., et al. (2007). Nat. Cell Biol. 9,
6471.
Ono, R., Taki, T., Taketani, T., Taniwaki, M., Kobayashi, H., and Hayashi, Y.
(2002). Cancer Res. 62, 40754080.
Oswald, J., Engemann, S., Lane, N., Mayer, W., Olek, A., Fundele, R., Dean,
W., Reik, W., and Walter, J. (2000). Curr. Biol. 10, 475478.
Paroush, Z., Keshet, I., Yisraeli, J., and Cedar, H. (1990). Cell 63, 12291237.
Weiss, A., Keshet, I., Razin, A., and Cedar, H. (1996). Cell 86, 709718.
Pastor, W.A., Pape, U.J., Huang, Y., Henderson, H.R., Lister, R., Ko, M.,
McLoughlin, E.M., Brudno, Y., Mahapatra, S., Kapranov, P., et al. (2011).
Nature 473, 394397.
Williams, K., Christensen, J., Pedersen, M.T., Johansen, J.V., Cloos, P.A.,
Rappsilber, J., and Helin, K. (2011). Nature 473, 343348.
Popp, C., Dean, W., Feng, S., Cokus, S.J., Andrews, S., Pellegrini, M.,
Jacobsen, S.E., and Reik, W. (2010). Nature 463, 11011105.
Wossidlo, M., Nakamura, T., Lepikhov, K., Marques, C.J., Zakhartchenko, V.,
Boiani, M., Arand, J., Nakano, T., Reik, W., and Walter, J. (2011). Nat Commun
2, 241.
Ptashne, M. (2009). Curr. Biol. 19, R234R241.

Rai, K., Huggins, I.J., James, S.R., Karpf, A.R., Jones, D.A., and Cairns, B.R.
(2008). Cell 135, 12011212.
Wu, H., DAlessio, A.C., Ito, S., Wang, Z., Cui, K., Zhao, K., Sun, Y.E., and
Zhang, Y. (2011a). Genes Dev. 25, 679684.
Rougier, N., Bourchis, D., Gomes, D.M., Niveleau, A., Plachot, M., Pa`ldi, A.,
and Viegas-Pequignot, E. (1998). Genes Dev. 12, 21082113.
Wu, H., DAlessio, A.C., Ito, S., Xia, K., Wang, Z., Cui, K., Zhao, K., Sun, Y.E.,
and Zhang, Y. (2011b). Nature 473, 389393.
Ruzov, A., Tsenkina, Y., Serio, A., Dudnakova, T., Fletcher, J., Bai, Y., Chebotareva, T., Pells, S., Hannoun, Z., Sullivan, G., et al. (2011). Cell Res. Published
online July 12, 2011. 10.1038/cr.2011.113.
Song, C.X., Szulwach, K.E., Fu, Y., Dai, Q., Yi, C., Li, X., Li, Y., Chen, C.H.,
Zhang, W., Jian, X., et al. (2011). Nat. Biotechnol. 29, 6872.
Tahiliani, M., Koh, K.P., Shen, Y., Pastor, W.A., Bandukwala, H., Brudno, Y.,
Agarwal, S., Iyer, L.M., Liu, D.R., Aravind, L., and Rao, A. (2009). Science
324, 930935.
Wyatt, G.R., and Cohen, S.S. (1952). Nature 170, 10721073.

Xu, Y., Wu, F., Tan, L., Kong, L., Xiong, L., Deng, J., Barbera, A.J., Zheng, L.,
Zhang, H., Huang, S., et al. (2011). Mol. Cell 42, 451464.
Yamanaka, S., and Blau, H.M. (2010). Nature 465, 704712.
Zhang, F., Pomerantz, J.H., Sen, G., Palermo, A.T., and Blau, H.M. (2007).
Proc. Natl. Acad. Sci. USA 104, 43954400.
Leading Edge
Review
Basic and Therapeutic
Aspects of Angiogenesis
Michael Potente,1,2 Holger Gerhardt,3,4 and Peter Carmeliet5,6,*
1Vascular
Epigenetics Group, Institute for Cardiovascular Regeneration, Center of Molecular Medicine

of Cardiology, Internal Medicine III
Goethe University, D-60590 Frankfurt, Germany
3Vascular Biology Laboratory, London Research Institute, Cancer Research UK, London WC2A 3LY, UK
4Independent consultant for Vascular Patterning Laboratory, Vesalius Research Center, VIB, B-3000 Leuven, Belgium
5Laboratory of Angiogenesis & Neurovascular link, Vesalius Research Center, K.U.Leuven, B-3000 Leuven, Belgium
6Laboratory of Angiogenesis & Neurovascular link, Vesalius Research Center, VIB, B-3000 Leuven, Belgium
*Correspondence: peter.carmeliet@vib-kuleuven.be
DOI 10.1016/j.cell.2011.08.039
2Department
Blood vessels form extensive networks that nurture all tissues in the body. Abnormal vessel growth
and function are hallmarks of cancer and ischemic and inflammatory diseases, and they contribute
to disease progression. Therapeutic approaches to block vascular supply have reached the clinic,
but limited efficacy and resistance pose unresolved challenges. Recent insights establish how
endothelial cells communicate with each other and with their environment to form a branched
vascular network. The emerging principles of vascular growth provide exciting new perspectives,
the translation of which might overcome the current limitations of pro- and antiangiogenic medicine.
Introduction
Blood vessels supply oxygen and nutrients and provide gateways for immune surveillance. Endothelial cells (ECs) line the
inner surface of vessels to support tissue growth and repair. As
this network nourishes all tissues, it is not surprising that structural or functional vessel abnormalities contribute to many
diseases. Inadequate vessel maintenance or growth causes
ischemia in diseases such as myocardial infarction, stroke, and
neurodegenerative or obesity-associated disorders, whereas
excessive vascular growth or abnormal remodeling promotes
many ailments including cancer, inflammatory disorders, and
eye disease (Carmeliet, 2003; Folkman, 2007). Vessels are also
used as routes for tumor cells to metastasize.
Hallmarks of Vessel Growth
In the embryo, new vessels form de novo via the assembly of
mesoderm-derived endothelial precursors (angioblasts) that
differentiate into a primitive vascular labyrinth (vasculogenesis)
(Swift and Weinstein, 2009) (Figure 1A). Subsequent vessel
sprouting (angiogenesis) creates a network that remodels into
arteries and veins (Adams and Alitalo, 2007) (Figure 1A). Recruitment of pericytes and vascular smooth muscle cells that enwrap
nascent EC tubules provides stability and regulates perfusion
(arteriogenesis) (Jain, 2003). In the adult, vessels are quiescent
and rarely form new branches. However, ECs retain high plasticity to sense and respond to angiogenic signals.
The term angiogenesis is commonly used to reference the
process of vessel growth but in the strictest sense denotes
vessel sprouting from pre-existing ones. Recent studies provided tremendous insights into fundamental aspects of angiogenesis that have led to a mechanistic model of vessel branching
(Adams and Alitalo, 2007; Carmeliet and Jain, 2011; Eilken and
Adams, 2010; Phng and Gerhardt, 2009). Attracted by proangiogenic signals, ECs become motile and invasive and protrude
filopodia (Figure 1B). These so-called tip cells spearhead new
sprouts and probe the environment for guidance cues. Following
tip cells, stalk cells extend fewer filopodia but establish a lumen
and proliferate to support sprout elongation. Tip cells anastomose with cells from neighboring sprouts to build vessel loops.
The initiation of blood flow, the establishment of a basement
membrane, and the recruitment of mural cells stabilize new
connections (Figure 1C). The sprouting process iterates until
proangiogenic signals abate, and quiescence is re-established
(Figure 1C).
Although vessels can grow via other mechanisms, such as the
splitting of pre-existing vessels through intussusception or the
stimulation of vessel expansion by circulating precursor cells
(Fang and Salven, 2011; Makanya et al., 2009), we will focus
here on the latest insights on vessel sprouting, which likely
accounts for a substantial fraction of vessel growth.
Therapeutic Expectations and Challenges
The importance of angiogenesis sparked hopes that manipulating this process could offer therapeutic opportunities
(Folkman, 1971). Despite efforts to stimulate angiogenesis therapeutically by proangiogenic factors, most trials failed to meet
these expectations. Alternative strategies, based on proangiogenic cell therapies or targeting of microRNAs, offer new opportunities but are in (pre)clinical development (Bonauer et al.,
2010).
Antiangiogenic approaches aimed at blocking vessel growth in
eye disease and cancer led to the approval of therapeutics
targeting vascular endothelial growth factor (VEGF) (Crawford
and Ferrara, 2009b). Nonetheless, only a fraction of cancer
Figure 1. Hallmarks of Vessel Formation

(A) Angioblasts differentiate into endothelial cells
(ECs), which form cords, acquire a lumen, and are
prespecified to arterial or venous phenotypes.
(B) Steps of vessel sprouting: (1) tip/stalk cell
selection; (2) tip cell navigation and stalk cell
proliferation; (3) branching coordination; (4) stalk
elongation, tip cell fusion, and lumen formation;
and (5) perfusion and vessel maturation.
(C) Sequential steps of vascular remodeling from
a primitive (left box) towards a stabilized and
mature vascular plexus (right box) including
adoption of a quiescent endothelial phalanx
phenotype, basement membrane deposition,
pericyte coverage, and branch regression.
patients show benefit as tumors evolve mechanisms of resistance or are refractory toward VEGF (receptor) inhibitors (Bergers
and Hanahan, 2008; Crawford and Ferrara, 2009a). Conflicting
results about the benefit of VEGF blockade have kick-started a
debate on whether antiangiogenic treatment may trigger more
invasive and metastatic tumors (Ebos and Kerbel, 2011). On the
upside, sustained normalization of abnormal tumor vessels
may offer benefit for combating metastasis (Goel et al., 2011).
For antiangiogenic medicine to have an enduring impact on
cancer patient survival, an integrated understanding of the
molecular principles of vessel growth is needed. Here, we take
a cell biological perspective to explore prototypic principles
and recently discovered regulatory mechanisms, seeking to
develop a framework of the angiogenic process that might
provide the basis for novel pro- and antiangiogenic therapies.
Endothelial Differentiation
Arterial and Venous Specification
Following assembly of primitive vessels in the early embryo (such
as the dorsal aorta and cardinal vein), remodeling transforms
the plexus into a hierarchically organized network of arteries,
capillaries, and veins. Arteries form a high-pressure system,
enabling transportation of blood to capillaries, whereas veins
face low-pressure gradients. The differences in hemodynamic
load are reflected in their structures: arteries are supported by
layers of vascular smooth muscle cells and a specialized matrix,
whereas veins are thinner and surrounded by fewer smooth muscle cells
(Gaengel et al., 2009).
Arterial and venous ECs possess
specific molecular identities (Adams and
Alitalo, 2007; Swift and Weinstein, 2009).
For instance, Notch pathway components are highly expressed in arteries
but are low in veins. Disruption of
Notch signaling causes loss of arterial
markers and re-expression of venous
signature genes, suggesting that Notch
promotes arterial specification by repressing venous identity (Gridley, 2010;
Swift and Weinstein, 2009). Notch also
controls Eph-Ephrin family members,
which configure arterio-venous boundaries. Ephrin-B2 expression in arterial ECs increases in response
to Notch, whereas its receptor EphB4 in venous ECs is repressed by Notch. In zebrafish, Sonic Hedgehog acts upstream
of Notch, where it triggers arterial differentiation by upregulating
VEGF that elevates Notch components. In mice, VEGF secreted
by nerves contributes to arterial differentiation of ECs in cotracking vessels (Carmeliet and Tessier-Lavigne, 2005). Neuropilin-1
(NRP1), a VEGF coreceptor, facilitates transduction of arterial
effects of VEGF. At the level of gene expression, the transcription
factors FOXC1 and FOXC2 drive an arterial gene signature (e.g.,
DLL4, HEY2, CXCR4) by interacting with VEGF and Notch
signaling. Although earlier proposals favored the view that the
venous fate is acquired by default, it has become clear that
venous identity requires repression of Notch signaling by the
vein-specific nuclear receptor COUP-TFII (Swift and Weinstein,
2009). In addition, hemodynamic factors such as blood pressure
and flow codetermine arterio-venous differentiation (Jones et al.,
2006).
Arterio-Venous Segregation
Zebrafish studies indicate that the cardinal vein does not form
by vasculogenesis but instead arises from a common precursor
vessel by segregation (Herbert et al., 2009) (Figure 1A). Venousfated EphB4-positive ECs migrate away from the arterial-fated
ephrinB2-positve ECs in the precursor vessel toward the location of the future cardinal vein. VEGF and Notch both restrain
ventral sprouting, whereas VEGF-C promotes segregation.
Figure 2. Tip Cell Formation

(A) In response to vascular endothelial growth
factor (VEGF) stimulation, endothelial cells (ECs)
degrade the basement membrane and pericytes
detach, allowing ECs to emigrate.
(B) VEGF/Notch signaling selects tip and stalk
cells.
(C) Filopodia guide tip cells by sensing attractive
and repulsive cues. Filopodia formation is regulated by CDC42 and endocytosis of the EphrinB2/
VEGFR2 receptors. ROBO4/UNC5B signaling
promotes stabilization of the endothelial layer
through inhibition of SRC.
However, it needs to be determined whether similar events occur

in mammals.
Sprouting Angiogenesis
Liberating Endothelial Cells
Endothelial and mural cells share a basement membrane
comprised of extracellular matrix proteins that form a sleeve
around endothelial tubules (Eble and Niland, 2009). This basement membrane and the coat of mural cells prevent resident
ECs from leaving their positions. At the onset of sprouting, ECs
therefore must be liberated, a process requiring proteolytic
breakdown of the basement membrane and detachment of
mural cells (Figure 2A). Basement membrane degradation is
mediated by matrix metalloproteases (MMPs) such as MTMMP1, enriched in tip cells. Control of these proteinases is
essential for sprouting, given that excessive degradation of the
extracellular matrix, as occurs in plasminogen activator inhibitor
1 (PAI1) deficiency, leaves too little matrix support for the branch
to sprout (Blasi and Carmeliet, 2002). MMPs also liberate proangiogenic growth factors that are sequestered in the matrix
(Arroyo and Iruela-Arispe, 2010). At the other end, they also
generate antiangiogenic molecules by cleaving plasma proteins,
matrix molecules, or proteases themselves to prevent inappro-
priate sprouting and coordinate branching (Nyberg et al., 2005). Detachment of

mural cells is stimulated by Angiopoietin-2 (ANG2), a proangiogenic growth
factor stored by ECs for rapid release (Augustin et al., 2009; Huang et al., 2010)
(Figure 2A).
Lateral Inhibition Selects
the Tip Cell
The specification of ECs into tip and stalk
cells is controlled by the Notch pathway
(Eilken and Adams, 2010; Phng and Gerhardt, 2009) (Figure 2B). Analysis of
Notch signaling revealed high Notch
activity in stalk cells but low levels of
Notch signaling in tip cells. Conversely,
tip cells express higher levels of the
Notch ligand DLL4. During development
or in tumors, blockade of Notch or DLL4
increases filopodia and sprouting as a
consequence of excessive tip cell formation (Thurston et al., 2007). Although ECs
express several Notch receptors, Notch1 is critical for suppressing tip cell behavior in stalk cells. The hypersprouting phenotype
and excessive number of tip cells following Notch inhibition indicate that the tip cell phenotype is the default endothelial
response to proangiogenic signals. In contrast to DLL4, the
Notch ligand JAGGED1 (JAG1) is expressed primarily by stalk
cells. However, JAG1 poorly activates Notch1, as modification
of Notch by FRINGE glycosyltransferases favors activation by
DLL4 (Eilken and Adams, 2010). Given that some DLL4 protein
is detectable in stalk cells, JAG1 helps to maintain differential
Notch activity by antagonizing DLL4 that signals back to tip cells
(Figure 2B).
VEGF and Dll4/Notch Feedback as a Branching
Pattern Generator
VEGF and Notch co-operate in an integrated intercellular feedback that functions as a branching pattern generator (Figure 2B). VEGF stimulates tip cell induction and filopodia formation via VEGF receptor-2 (VEGFR2), whereas VEGFR2 blockade
causes sprouting defects with blunt-ending channels (Phng and
Gerhardt, 2009). VEGFR3 is expressed in the embryonic vasculature but later becomes confined to lymphatics. However, tip
cells re-express VEGFR3, and its pharmacological inhibition
diminishes sprouting (Tammela et al., 2008). In contrast, loss of
VEGFR1 increases sprouting and vascularization. A soluble

variant or a kinase-dead mutant of VEGFR1 rescues vascular
defects caused by VEGFR1 deficiency, suggesting that this
receptor functions as a VEGF trap. VEGFR1 is predominantly expressed in stalk cells and involved in guidance and limiting tip cell
formation (Chappell and Bautch, 2010; Jakobsson et al., 2010).
The feedback loop between VEGF and Notch involves regulation of all VEGFRs by Notch. VEGF/VEGFR2 enhances DLL4
expression in tip cells (Phng and Gerhardt, 2009). DLL4-mediated activation of Notch in neighboring ECs inhibits tip cell
behavior in these cells by downregulating VEGFR2, VEGFR3,
and NRP1 while upregulating VEGFR1 (Jakobsson et al., 2010;
Phng and Gerhardt, 2009). Computational modeling indicates
that such an integrated negative feedback loop of VEGF and
Notch is sufficient to establish a stable pattern of tip and stalk
cells (Bentley et al., 2009). ECs at the angiogenic front dynamically compete for the tip position through DLL4/Notch signaling
(Jakobsson et al., 2010). Following VEGF exposure, all cells
upregulate DLL4. However, ECs that express DLL4 more quickly
or at higher levels have a competitive advantage to become a tip
cell as they activate Notch signaling in neighboring cells more
effectively. Given the dynamic shuffling of tip-stalk position of
ECs during sprouting and the regular exchange of the leading
tip cell, DLL4 expression must be dynamically regulated. Precise
regulation of DLL4 expression is achieved through a TEL/CtBP
repressor complex at the DLL4 promoter, which is transiently
disassembled upon VEGF stimulation, allowing a temporally
restricted pulse of DLL4 transcription (Roukens et al., 2010). In
line with a central function of DLL4 for vessel patterning
dynamics, several other pathways, such as the Wnt/b-catenin
pathway, converge on the transcriptional control of DLL4
(Corada et al., 2010).
Tip Cell Guidance
Wiring of the nervous system relies on the formation of correct
connections and requires precise guidance of axonal growth
cones. The vasculature must also be correctly patterned for
optimal oxygen delivery. Emerging vessels use tip cells to guide
sprouts properly, and the structure and function of tip cells are
reminiscent of axonal growth cones (Adams and Eichmann,
2010; Carmeliet and Tessier-Lavigne, 2005). Little is known
regarding the molecular mechanisms regulating tip cell filopodia.
Activation of Cdc42 by VEGF triggers filopodia formation,
whereas Rac1 regulates lamellipodia formation (De Smet
et al., 2009) (Figure 2C). Both the axon growth cone and tip
cell use similar attractive and repulsive cues to control guidance.
ECs express guidance receptors including ROBO4, UNC5B,
PLEXIN-D1, NRPs, and EPH family members, which they use
to probe the environment (Figure 2C).
Roundabouts (ROBOs) are guidance receptors. Activation of
ROBO13 by SLIT ligands (SLIT13) provides repulsive signals
for axons. ROBO4 is expressed in ECs and maintains vessel
integrity, and ROBO4 deficiency induces leakiness and hypervascularization (London et al., 2009). At the molecular level,
ROBO4 counteracts the permeability-promoting actions of
VEGF by impeding VEGFR2-mediated activation of the kinase
SRC. The nature of the ROBO4 ligand remains debated, as
ROBO4 lacks SLIT-binding domains. ROBO4 also binds to
UNC5B, another guidance receptor, suggesting that ROBO4/

UNC5B maintains vessel integrity via UNC5B activation (Koch
et al., 2011).
UNC5B is a receptor for Netrins whose expression is enriched
in tip cells. Its inactivation results in enhanced sprouting,
whereas Netrin1 prompts filopodia retraction of ECs, consistent
with a suppressive function of netrins and UNC5B on vessel
growth (Adams and Eichmann, 2010). This function of Netrin1
has not been observed by others, suggesting that Netrin1
signaling might involve other yet unidentified receptors (Adams
and Eichmann, 2010). Alternatively, UNC5B may function as
a dependence receptor that, in the absence of ligand, induces
EC apoptosis (Castets and Mehlen, 2010).
Semaphorins are secreted or membrane-bound guidance
cues that interact with receptor complexes, formed by NRPs
alone or NRP/plexin family proteins (Carmeliet and TessierLavigne, 2005). SEMA3E induces vessel repulsion through interaction with PLEXIN-D1. As ECs express PLEXIN-D1, its loss
causes aberrant sprouting into SEMA3E-expressing tissues in
zebrafish embryos (Adams and Eichmann, 2010). In the mouse
retina, SEMA3E activates PLEXIN-D1 on tip cells to fine-tune
the balance of tip and stalk cells necessary for even-growing
vascular fronts by coordinating VEGFs activity in a negative
feedback (Kim et al., 2011). NRPs bind semaphorins, VEGF,
and other ligands, but the vessel abnormalities in NRP1-deficient
embryos are related to defective VEGF/NRP1 signaling (Fantin
et al., 2009). In fact, most semaphorins suppress angiogenesis
(Serini et al., 2009).
EPH receptors and their ephrin ligands are regulators of cellcontact-dependent signaling (Pitulescu and Adams, 2010).
Eph-ephrin binding leads to bidirectional signaling in cells
expressing the receptor (forward signaling) or ligand (reverse
signaling). Eph-ephrins generate mostly repulsive signals.
Ephrin-B2 is expressed in arterial ECs, whereas EphB4 marks
venous ECs. Both of them regulate vessel morphogenesis, and
loss of ephrin-B2 or EphB4 leads to vascular remodeling defects
(Pitulescu and Adams, 2010). Intriguingly, ephrin-B2-mediated
reverse signaling also controls VEGFR internalization and tip
cell behavior (Figure 2C). ECs lacking ephrin-B2 reverse
signaling are unable to internalize VEGFR2 and VEGFR3 and
cannot transmit VEGF signals properly, together impairing
sprouting (Sawamiphak et al., 2010; Wang et al., 2010).
Endothelial Stalk Cell Formation
Control of Stalk Cell Behavior and Elongation
Stalk cells are equipped with the ability to form tubes and
branches. Compared to tip cells, stalk cells produce fewer filopodia, are more proliferative, and form a vascular lumen (Figures
3A and 3B). They also establish junctions with neighboring cells
and produce basement membrane components to ensure the
integrity of the sprout (Phng and Gerhardt, 2009). ECs with
excess Notch signaling extend less filopodia and are excluded
from the tip position, indicating that Notch activity is dispensable
for tip cell formation but required for stalk cell specification
(Jakobsson et al., 2010). The importance of a balanced tip/
stalk specification by Notch is best illustrated by the paradoxical effects of gene inactivation of DLL4 or Notch1 in the
endothelium: although more vessels are formed, they are poorly
Figure 3. Stalk Cell Formation, Stabilization, and Perfusion

(A) Tip cell fusion and branch anastomosis are
facilitated by macrophages; VE-cadherin promotes cell-cell adhesion between tip cells.
(B) Stalk cell stabilization relies on Notch activity
that is fine-tuned by NRARP and SIRT1. WNT and
Notch intersect via NRARP and LEF1/b-CATENIN
to stabilize connections.
(C) Models of lumen formation: fusion of pinocytotic vesicles (left; C), contraction of the cytoskeleton following exposure of negatively charged
glycoproteins on the lumenal surface of endothelial cells (ECs) (right; C0 ).
perfused and dysfunctional (Phng and Gerhardt, 2009; Thurston

et al., 2007).
Activation of Notch involves the cleavage of Notch receptors
leading to the release of the intracellular domain (NICD), forming a complex with the transcription factor RBPj/CBF1 and
Mastermind-like proteins to drive target gene expression. This
complex not only activates transcription but also promotes its
own turnover to prevent sustained Notch activation. The
Notch-regulated ankyrin repeat protein (NRARP) negatively
regulates Notch responses by dissembling the Notch coactivator complex and promoting NICD degradation. Modulation of
Notch in growing vessels is important, as NRARP allows stalk
cells to proliferate. NRARP also augments Lef1/b-catenin
signaling to maintain stability of nascent vessel connections
(Phng et al., 2009). Control of Notch signaling by reversible
acetylation of NICD is another layer of Notch regulation
(Guarani et al., 2011). Acetylation enhances Notch responses
by interfering with NICD1 turnover, whereas deacetylation by
SIRT1 opposes NICD1 stabilization, thereby limiting Notch
activity.
Negative regulation of Notch signaling in stalk cells might, at
first sight, appear counterintuitive. However, it is important to
note that tip and stalk cells are transient

phenotypes and not stable cell fates. To
expand the vessel network, ECs undergo
iterative cycles of sprouting, branching,
and tubulogenesis, requiring dynamic
transitions between tip and stalk cell
phenotypes (Eilken and Adams, 2010;
Phng and Gerhardt, 2009). Fine-tuning
of the Notch signaling amplitude and
duration by NRARP and SIRT1 could
serve to dynamically adjust the timing of
tip and stalk transitions, thereby adapting
vessel branching frequency.
Lumen Formation
Vessels need to establish a lumen, which
occurs by different mechanisms (IruelaArispe and Davis, 2009; Zeeb et al.,
2010) (Figures 3C and 3C0 ). Observations
in intersomitic vessels indicate that ECs
form a lumen by coalescence of intracellular (pinocytic) vacuoles, which interconnect with vacuoles from neighboring ECs
(cell hollowing) (Figure 3C). Recent studies in large axial vessels
suggest that ECs adjust their shape and rearrange their junctions
to open up a lumen (cord hollowing) (Figure 3C0 ). In this model,
ECs first define apical-basal polarity. Thereafter, the apical
(lumenal) membrane becomes decorated with negatively
charged glycoproteins that confer a repulsive signal, opening
up the lumen. Subsequent changes in EC shape, driven by
VEGF and Rho-associated protein kinase (ROCK), expand the
et al., 2009; Zeeb et al., 2010). Tube morphogenlumen (Strilic
esis also requires Ras-interacting protein 1 (RASIP1), a regulator
of GTPase signaling controlling cytoskeletal rearrangements,
adhesion, and EC polarity (Xu et al., 2011). The mechanisms of
lumen formation likely depend on the vascular bed or type of
vessel formation.
Vessel Branch Fusion and Perfusion
Tip cells contact other tip cells to add new vessel circuits to the
existing network. By accumulating at sites of vessel anastomosis
and interacting with filopodia of neighboring tip cells during
fusion, macrophages can support vessel anastomosis (Fantin
et al., 2010) (Figure 3A). However, anastomosis does not require
macrophages, suggesting that they only facilitate fusion events,
Figure 4. Remodeling and Quiescence

(A) Stalk cells undergo remodeling in response to
flow.
(B) Upregulation of the transcription factor KLF2 in
response to blood flow ensures remodeling of the
vasculature. In consolidated vessels, KLF2 promotes quiescence and the formation of patent
vessels with an antithrombogenic endothelial
lining. Hypoperfused vessels undergo regression.
possibly via cell-to-cell communication. Once the contact

between tip cells is established, VE-cadherin-containing junctions consolidate the connection (Figure 3A).
New vessel connections must become stable to generate an
enduring loop. The deposition of extracellular matrix into the
basement membrane, the recruitment of supporting pericytes,
reduced EC proliferation, and increased formation of cell
junctions all contribute to this process. The onset of blood flow
in the new lumen shapes and remodels vessel connections
and activates the shear stress-responsive transcription factor
Kruppel-like factor 2 (KLF2) (Figures 4A and 4B). In zebrafish,
KLF2 induces vessel remodeling by upregulating the EC-specific
miR-126 that modulates PI3K and MAPK signaling (Nicoli et al.,
2010). Hemodynamic forces also remodel large arteries and are
important for vessel maintenance and collateral vessel expansion. Upon perfusion, oxygen and nutrient delivery reduces
VEGF expression and inactivates endothelial oxygen sensors,
together shifting endothelial behavior toward a quiescent
phenotype.
Vessel Maturation, Stabilization, and Quiescence
For vessels to become functional, they must matureat the level
of the endothelium and vessel wall and as a network. At the
network level, maturation involves remodeling into a hierarchically branched network and adaptation of vascular patterning
to local tissue needs. This involves

recruitment of mural cells and deposition
of extracellular matrix (Jain, 2003). ECs
also acquire tissue-specific differentiation adapted to meet local homeostatic
demands and thus differ in phenotype
(Dyer and Patterson, 2010).
Mural Cell Differentiation
A fundamental feature of vessel maturation is the recruitment of mural cells.
Pericytes establish direct cell-cell contact
with ECs in capillaries and immature
vessels, whereas vascular smooth muscle
cells cover arteries and veins and are
separated from ECs by a matrix (Gaengel
et al., 2009). Vessel maturation relies
partly on transforming growth factor
b (TGF-b) signaling. TGF-b stimulates
mural cell induction, differentiation, proliferation, and migration and promotes
production of extracellular matrix (Pardali
et al., 2010). Loss of function of TGF-b
receptor 2 (TGFBR2), endoglin, or activin
receptor-like kinase 1 (Alk1) in mice causes vessel fragility in
part due to impaired mural cell development (Pardali et al.,
2010). In humans, mutations in ENDOGLIN and ALK1 cause
hereditary hemorrhagic telangiectasia (HHT), a disease characterized by arteriovenous malformations with abnormally remodeled
vessel walls (Pardali et al., 2010). Which of the TGF-b family
members signaling is impaired in HTT and whether smooth
muscle cells are affected directly (or rather indirectly through EC
effects) require further study. For instance, by activating ALK5
(TGFBR1) in ECs, TGF-b signaling contributes to vessel maturation by secretion of PAI1, preventing degradation of the perivascular matrix.
Pericyte Recruitment
Recruitment of mural cells is controlled by platelet-derived
growth factor (PDGF) receptor-b (PDGFR-b) (Gaengel et al.,
2009) (Figure 5A). Endothelial PDGFB signals to PDGFRb expressed by mural cells, stimulating their migration and proliferation. Adequate expression, matrix binding, and spatial
presentation of PDGFB to PDGFR-b are essential for vascular
maturation, and inactivation of either Pdgfb or Pdgfrb induces
pericyte deficiency, vascular dysfunction, micro-aneurysm formation, and bleeding (Gaengel et al., 2009). Pdgfb mouse
mutants with insufficient pericyte coverage display blood brain
barrier defects, causing neuronal damage (Quaegebeur et al.,
2010).
Figure 5. Vessel Maturation, Stabilization,

and Quiescent Phalanx Cell Formation
(A) Vessel stabilization relies on the recruitment of
pericytes involving PDGFR, S1PR1, ephrinB2,
and Notch3 signaling and the formation of Ncadherin junctions. Basement membrane deposition is favored by protease inhibitors (TIMPs).
(B) Perfused vessels become mature through
pericyte coverage and acquisition of an endothelial phalanx phenotype. Right: Inactivation of
PHD2 by low oxygen levels, leading to HIF2amediated upregulation of sVEGFR1 and VE-cadherin, thereby improving perfusion.
Sphingosine-1-phosphate receptor (S1PR) signaling also

controls EC/mural cell interactions. Endothelial-derived S1P
binds to G protein-coupled S1PRs (S1PR15) (Lucke and Levkau, 2010). S1P triggers cytoskeletal, adhesive, and junctional
changes, affecting cell migration, proliferation, and survival.
Disruption of S1PR1 or loss of both S1PR2 and S1PR3 in mice
causes defective coverage of vascular smooth muscle cells
and pericytes, a phenotype reminiscent of Pdgfb and Pdgfrb
mutant mice. However, the primary defect is located in ECs,
where S1P1 controls trafficking of N-cadherin to the ablumenal
side of ECs in order to strengthen EC-pericyte contacts
(Figure 5A).
Angiopoietin-1 (ANG1), produced by mural cells, activates its
endothelial receptor TIE2 (Augustin et al., 2009; Huang et al.,
2010). ANG1 stabilizes vessels, promotes pericyte adhesion,
and makes them leak resistant by tightening endothelial junctions. Contrary to common belief, ANG1 seems less required
for mural cell recruitment than originally thought (Jeansson
et al., 2011). Mural cells also require ephrinB2 for association
around ECs, as mural cell-specific ephrinB2 deficiency causes
mural cell migration and vascular defects (Pitulescu and Adams,
2010) (Figure 5A). Notch signaling also controls maturation and
arterial differentiation of vascular smooth muscle cells (Gridley,
2010). Mice lacking Notch3 lose arterial characteristics and
develop arterial defects, whereas NOTCH3 mutations in humans
cause degeneration of vascular smooth

muscle cells in CADASIL, a human stroke
and dementia syndrome (Figure 5A).
Phalanx ECs Express Oxygen
Sensors to Regulate Vessel
Perfusion
Vessels can adjust their shape and function to meet changing tissue oxygen
demands. Hypoxia-inducible factors
(HIFs) orchestrate adaptive responses of
ECs to changes in oxygen tension by
controlling gene networks that govern
survival, metabolism, and angiogenesis
(Fraisl et al., 2009; Majmundar et al.,
2010). HIF activity is regulated by oxygen-sensing prolyl hydroxylase domain
proteins (PHD13). In normoxia, PHDs
use oxygen to hydroxylate HIFs, thereby
targeting them for proteasomal degradation. Oxygen sensors become inactive
in hypoxic conditions, allowing HIFs to escape degradation.
PHD2 regulates the endothelial phalanx cell phenotype. In
search for a conceptual distinction from angiogenic tip and stalk
cells, the cobblestone-like appearance of quiescent ECs prompted the term phalanx cells given their resemblance to the
ancient Greek military formation (Mazzone et al., 2009).
Haplodeficiency of PHD2 counteracts the abnormal vessel
shape in tumors, promoting a more streamlined phalanx-like
phenotype (Mazzone et al., 2009). Reduced PHD2 levels stabilize HIF2a, thereby enhancing levels of soluble VEGFR1 and
VE-cadherin, counterbalancing endothelial disorganization
(Figure 5B). This oxygen sensor thereby allows ECs to dynamically adapt vessel shape to their primordial function of oxygen
delivery.
Quiescent ECs Have Barrier Properties
Resting ECs form barriers between blood and surrounding
tissues to control the exchange of fluids and solutes and transmigration of immune cells. Essential for this function is the
ability of ECs to regulate cell-cell adhesion between each other
and neighboring cells. This relies on transmembrane-adhesive
proteins, including VE-cadherin and N-cadherin at adherens
junctions, as well as occludins and members of the claudin
and junctional adhesion molecule (JAM) family at tight junctions
(Cavallaro and Dejana, 2011). Tight junction molecules maintain
and regulate paracellular permeability, whereas adherens
junction molecules mediate cell-cell adhesion, cytoskeletal reorganization, and intracellular signaling. VE-cadherin is a key
component of EC junctions. In complex with VEGFR2, VE-cadherin maintains EC quiescence through recruitment of phosphatases that dephosphorylate VEGFR2, thus restraining VEGF
signaling. Distinct types of VE-cadherin-based adherens junctions establish stable or transitory interactions with the cytoskeleton that either solidify EC adhesion and barrier properties
or facilitate EC separation and movement (Falk, 2010). Activation of TIE2 by ANG1 protects vessels from VEGF-induced
leakage by inhibiting VEGFs ability to induce endocytosis of
VE-cadherin.
Vessels Express Survival Signals
As endothelial proliferation decelerates during maturation, ECs
must adopt survival properties to maintain integrity of the vessel
lining. Autocrine and paracrine survival signals from endothelial
and support cells protect the vessel from environmental
stresses. One such survival factor is VEGF, which activates the
PI3K/AKT survival pathway. Interestingly, ECs themselves are
the pivotal source for VEGFs prosurvival activity. Mice lacking
VEGF in ECs suffer bleeding, microinfarcts, and EC rupture
(Warren and Iruela-Arispe, 2010). When produced by ECs as
intracrine factor, VEGF prevents EC apoptosis in nonpathological conditions (Figure 5B). This intracrine activity of VEGF
differs from its paracrine function in stimulating angiogenesis,
as loss of endothelial VEGF does not cause developmental
vascular defects (Warren and Iruela-Arispe, 2010).
Signaling by fibroblast growth factors (FGFs) has also been
implicated in maintaining vascular integrity due to their ability
to anneal adherens junctions (Beenken and Mohammadi,
2009). Inhibition of FGF signaling results in dissociation of
adherens junctions and tight junctions, subsequent loss of
ECs, and vessel disintegration (Murakami et al., 2008). Notch
signaling is critical for generating and maintaining vascular
homeostasis. A consequence of Notch activation is the establishment of mature and patent vessels that promote perfusion
and relieve tissue hypoxia. Conversely, blockade of DLL4 or
Notch1 in the adult causes vascular tumors and hemorrhage
(Liu et al., 2011; Yan et al., 2010). Similarly, endothelial inactivation of RBPj reinitiates vascular growth in adulthood (Figure 5B).
Activation of Notch in mural cells by endothelial DLL4 also
contributes to vessel stability by stimulating deposition of BM
components.
Signaling by TIE2 and ANG1 also controls survival and vessel
quiescence (Augustin et al., 2009). ANG1 clusters TIE2 junctionally at inter-EC junctions in trans to promote survival and EC
quiescence (Figure 5B). Blood flow is another important survival
cue for ECs as fluid shear stress potently inhibits EC apoptosis.
KLF2 is activated by shear stress and evokes quiescence by
upregulating endothelial nitric oxide synthase and the anticoagulant factor thrombomodulin, keeping vessels dilated, perfused,
and free of clots, and by downregulating VEGFR2, which prevents tip cell formation (Figure 4B). Other EC quiescence factors
include bone morphogenic protein 9 (BMP9) and cerebral
cavernous malformation proteins (CCM13), whose defective
signaling causes vascular malformations (Leblanc et al., 2009).
ECs in nonperfused vessels regress from their locations or
undergo apoptosis (Figure 4B).
Other Signaling Pathways and Limitations of the Model

Although the described model offers a framework to explain the
activity of numerous pro- and antiangiogenic molecules, there
are other angiogenic pathways, with documented effects on
vessel growth in vivo, whose roles in vessel branching have
not or have only incompletely been characterized. Examples
include chemokines, integrins (Desgrosellier and Cheresh,
2010), several transcriptional regulators, Wnt ligands and their
frizzled receptors (Franco et al., 2009), other members of the
FGF, PDGF, and TGF- superfamilies, or the VEGF homolog
PlGF that transmits angiogenic signals through VEGFR1 (Fischer
et al., 2008). Identifying their role in vessel branching or the other
types of vessel growth will generate a unifying model that can
serve as a source for future drug development.
The Vascular-Metabolic Interface
Blood vessels transport nutrients to energy-utilizing tissues, and
hence, vessels as well as proangiogenic signals can affect
metabolism (Fraisl et al., 2009) (Figures 6A and 6C). In metabolically active tissues, the uptake of nutrients is linked to energy
demand to maintain tissue homeostasis. Interestingly, high
levels of VEGF-B, a VEGF member with poor angiogenic activity,
are found in metabolically active tissues, where it is coexpressed
with genes like VEGF, stimulating mitochondrial biogenesis, and
controls trans-endothelial uptake of fatty acids into other tissues
(Hagberg et al., 2010). Through this mechanism, VEGF-B
prepares tissues for fatty acid consumption. Notably, besides
their role in supplying nutrients, ECs themselves can also
promote growth and repair of metabolically active tissues independent of perfusion by secreting angiocrine factors (Butler
et al., 2010; Ding et al., 2010). How vascular growth signals coordinate metabolism is only beginning to become understood.
The converse crosstalk is also true, with metabolism affecting
vascular growth (Fraisl et al., 2009) (Figures 6B and 6C). Metabolic sensors and regulators control vessel growth, often stimulating angiogenesis in nutrient-deprived conditions in order to
prepare the tissue for oxidative metabolism upon repletion of
oxygen and nutrients. Examples include PGC1a, LKB1, AMPK,
FOXOs, and SIRT1 (Fraisl et al., 2009). In conditions of oxygen
and nutrient scarcity, PGC1a stimulates angiogenesis by upregulating VEGF through interaction with ERRa; this angiogenic
burst, coupled to mitochondrial biogenesis, prepares the
ischemic tissue for oxidative metabolism upon revascularization
(Fraisl et al., 2009). Also, an increase in cellular levels of AMP
(reflecting energy deprivation) induces VEGF-driven angiogenesis through activation of AMPK. Vascular growth is similarly
controlled by LKB1, an activating kinase of AMPK and regulator
of metabolism. The vascular-metabolic interface is further regulated by FOXO transcription factors, which are upregulated
during fasting and restrict angiogenic behavior (Fraisl et al.,
2009). Interestingly, FOXO1 and Notch1 are controlled by
SIRT1, a deacetylase activated by NAD+ in conditions of energy
distress and nutrient deprivation.
Vessel Growth in Disease
Insufficient vessel growth and regression contribute to numerous disorders, ranging from myocardial infarction and stroke
to neurodegeneration. Conversely, uncontrolled vessel growth
Figure 6. AngiogenesisMetabolism Crosstalk

(A) Endothelial cells (ECs) promote growth and repair of metabolically active tissues by releasing angiocrine signals, whereas angiogenic molecules stimulate
trans-endothelial transport of fuel to surrounding tissues.
(B) Metabolic sensors and regulators stimulate angiogenesis and mitochondrial biogenesis in order to prepare the ischemic tissue for oxidative metabolism upon
repletion of oxygen and nutrients following revascularization.
(C) Schematic models of the molecular basis of angiogenesismetabolism crosstalk.
promotes tumorigenesis and ocular disorders such as agerelated macular degeneration. Historically, this has led to
concepts of pro- and antiangiogenic therapy, aiming to restore
adequate vessel densities. However, sprouting angiogenesis
alone might be insufficient to fully revascularize ischemic tissues,
as also collateral vessels have to enlarge to supply bulk flow
(Schaper, 2009). It has become clear that vessel densities can
no longer be considered separately from vessel function when
designing angiogenic therapeutics. We anticipate that insights
into pathological angiogenesis, guiding future diagnostic and
therapeutic approaches, will increasingly focus on the functional
quality of vessels and their effects on local metabolism rather
than on vessel quantity alone.
Tumor Vessels Are Abnormal
Tumor vessels display abnormal structure and function (Goel
et al., 2011; Jain, 2005) with seemingly chaotic organization
(Figure 7A). Highly dense regions neighbor vessel-poor areas,
and vessels vary from abnormally wide, irregular, and tortuous
serpentine-like shape to thin channels with small or compressed
lumens. Every layer of the tumor vessel wall is abnormal. ECs
lack a cobblestone appearance, are poorly interconnected,
and are occasionally multilayered. Also, arterio-venous identity

is ill defined, and shunting compromises flow. The basement
membrane is irregular in thickness and composition, and fewer,
more loosely attached hypocontractile mural cells cover tumor
vessels, though tumor-type-specific differences exist.
The resulting irregular perfusion impairs oxygen, nutrient, and
drug delivery (Goel et al., 2011; Jain, 2005). Vessel leakiness
together with growing tumor mass increases the interstitial
pressure and thereby impedes nutrient and drug distribution.
The loosely assembled vessel wall also facilitates tumor cell
intravasation and dissemination. As a consequence of poor
oxygen, nutrient, and growth factor supply, tumor cells further
stimulate angiogenesis in an effort to compensate for the
poor functioning of the existing ones. However, this excess of
proangiogenic molecules only leads to additional disorganization as the angiogenic burst is nonproductive, further aggravating tumor hypoperfusion in a vicious cycle. The hypoxic and
acidic tumor milieu constitutes a hostile microenvironment
that is believed to drive selection of more malignant tumor cell
clones and further promotes tumor cell dissemination. The
uneven delivery of chemotherapeutics together with a reduced
Figure 7. Antiangiogenesis versus Vessel

Normalization
(A) Antiangiogenic agents that destroy abnormal
tumor vessels and prune the tumor microvasculature can aggravate intratumor hypoxia, which
can activate a prometastatic switch; the question
mark reflects ongoing debate as to whether this
metastatic switch exists in patients treated with
VEGF (receptor) inhibitors.
(B) Antivascular targeting strategies that normalize
abnormal tumor vessels are believed not to
aggravate tumor hypoxia or even to improve
oxygen supply, thereby impeding the hypoxiadriven prometastatic switch. Their effect on
stabilizing and tightening of the tumor vessel wall
makes the vessels less penetrable for disseminating tumor cells. When improving drug delivery
and tumor oxygenation, vessel normalization can
also enhance conventional chemotherapy and
irradiation.
efficacy of radiotherapy, owing to the lower intratumoral oxygen

levels, limit the success of conventional anticancer treatment.
Modes of Tumor Vascularization
Besides sprouting, tumors utilize other modes of vessel growth.
For example, tumor cells can co-opt pre-existing vasculature
without a need to stimulate vessel branching initially. Once the
tumor outgrows this supply, hypoxia evokes a secondary angiogenic response. Bone marrow-derived progenitors can also
promote tumor vascularization or control the angiogenic switch
during metastasis, but their importance is debated and context
dependent (Fang and Salven, 2011). If tumors would be able to
switch mechanisms of vascular growth and some of these mechanisms rely less on VEGF, they would possess the means to
escape from treatment with VEGF (receptor) inhibitors. Identifying the molecular basis of these alternative modes of vessel
growth will thus be critical to improve

the efficacy of antiangiogenic treatment.
Role of Myeloid Cells in Tumor
Vessel Vascularization
Various hematopoietic lineages influence
tumor angiogenesis (Kerbel, 2008).
VEGFR1+ hematopoietic precursors or
TIE2-expressing monocytes (TEMs) are
located close to growing tumor vessels
and release angiogenic molecules (De
Palma and Naldini, 2009). Expression of
ANG2 by tumor ECs activates TEMs to
stimulate angiogenesis (Mazzieri et al.,
2011). Tumor-associated macrophages,
especially those polarized to a proangiogenic M2-like phenotype, stimulate
angiogenesis by releasing PlGF that also
contributes to vessel disorganization (Grivennikov et al., 2010; Qian and Pollard,
2010; Rolny et al., 2011). Mast cells
promote tumor angiogenesis by secretion of proteases that liberate proangiogenic factors from the extracellular
matrix. Additionally, CD11B+Gr1+ neutrophils release the proangiogenic factor Bv8, particularly in tumors
that are resistant against VEGF blockade (Ferrara, 2010b).
Recruitment of other bone marrow-derived cells (BMDCs) can
also contribute to tumor vascularization. For instance, CXCR4+
BMDCs are retained inside the cancer via production of
SDF1a, the ligand of CXCR4, and boost tumor vascularization
by releasing angiogenic factors. An increasing body of evidence
implicates myeloid cells in the resistance of tumors against
treatment with VEGF (receptor) inhibitors (Ferrara, 2010b).
Role of Cancer-Associated Fibroblasts
in Tumor Vessel Vascularization
Another stromal cell type gaining increasing attention is the
cancer-associated fibroblast (Crawford and Ferrara, 2009a;
Nyberg et al., 2008; Pietras and Ostman, 2010). These cells originate from local mesenchyme in organs where tumors grow or
become recruited from the bone marrow (Wels et al., 2008).

Cancer-associated fibroblasts promote tumor vascularization
by recruiting endothelial progenitor cells (EPCs) or releasing
proangiogenic factors (Crawford and Ferrara, 2009a; Erez et al.,
2010). In chronic myeloid leukemia, malignant cells upregulate
PlGF in bone marrow stromal cells to create a vascularized soil
for leukemia cells (Schmidt et al., 2011).
Clinically Approved Antiangiogenic Therapies
VEGF has become the prime antiangiogenic drug target with
approval by the US Food and Drug Administration of several
VEGF (receptor)-based inhibitors for clinical use (Crawford and
Ferrara, 2009b). The anti-VEGF antibody (bevacizumab [Avastin])
is approved in combination with chemotherapy or cytokine
therapy for several advanced metastatic cancers, including
non-squamous non-small cell lung cancer, colorectal cancer,
renal cell cancer, and metastatic breast cancer. Based on
a randomized phase II trial, bevacizumab monotherapy has
been approved for recurrent glioblastoma. Additionally, four multitargeted pan-VEGF receptor tyrosine kinase inhibitors (RTKIs)
have been approved: Sunitinib [Sutent] and Pazopanib [Votrient]
for metastatic RCC, Sorafenib [Nexavar] for metastatic RCC and
unresectable hepatocellular carcinoma, and Vandetanib [Zactima] for medullary thyroid cancer. Sunitinib has also been recommended for treatment of advanced pancreatic neuroendocrine
tumors. Clinical agents for wet age-related macular degeneration,
characterized by neovascularization of leaky vessels, include an
anti-VEGF Fab (ranibizumab [Lucentis]) and a VEGF aptamer (pegaptanib [Macugen]), with Avastin being used off-label. VEGF
blockade prolongs progression-free survival or overall survival
of cancer patients in the range of weeks to months and improves
visual acuity in patients with age-related macular degeneration.
The clinical benefit of treatment with VEGF (receptor) inhibitors
is attributable to several mechanisms. First, these blockers
inhibit tumor vessel expansion by blocking vascular branching
or inhibiting homing of BMDCs (Figure 7A). Additionally, these
drugs induce regression of pre-existing tumor vessels and sensitize ECs to effects of chemotherapy and irradiation by depriving
them of VEGFs survival activity. Normalization of abnormal
tumor vessels by pruning immature pericyte-devoid vessels
and by promoting maturation into more functional vessels is
another mechanism (Goel et al., 2011) (Figure 7B). The resulting
sensitization to cytotoxic or radiation therapies relying on
conversion of oxygen to radicals in combination with improved
chemotherapeutic delivery may explain partly why combination
delivery of bevacizumab/cytotoxic agents is often superior
(Jain, 2005). However, the importance of vessel normalization
versus pruning for the overall anticancer effect of VEGF
(receptor) inhibitor treatment requires future study. Furthermore,
vessel normalization observed with treatment is transient, as
these drugs induce excessive vessel regression, or tumor vascularization escapes VEGF blockade. In conditions where vascular
leakage causes life-threatening intracranial edema (e.g., in glioblastoma) or blindness (e.g., in wet age-related macular degeneration), restoration of normal barrier properties by VEGF
(receptor) blockade may be a relevant mechanism (Goel et al.,
2011). Besides targeting tumor vessels, these inhibitors also
target tumor cells expressing VEGF (receptor), whose growth
is stimulated by VEGF.
Challenges and Concerns of VEGF (Receptor) Inhibitor

Treatment
Contrary to preclinical experiments, where long-term benefit of
VEGF (receptor) inhibition can be achieved, the clinical benefit
in prolonging cancer patient survival with advanced disease is
limited, and a fraction of patients are intrinsically refractory or
acquire resistance (Bergers and Hanahan, 2008; Ebos and Kerbel, 2011; Ferrara, 2010a). Recent trials using VEGF (receptor)
blockers showed that the benefit, initially reported for progression-free survival, was no longer detected when analyzing overall survival (Ebos and Kerbel, 2011). The first phase III trial evaluating the adjuvant effect of anti-VEGF therapy following surgical
tumor resection did not prolong disease-free survival (Van Cutsem et al., 2011). It is also curious why monotherapy with
VEGF receptor kinase inhibitors induces benefit in some tumors
but is ineffective in others or evokes side effects when combined
with chemotherapy. Validated genetic or molecular biomarkers
for anti-VEGF (receptor) responsiveness are much needed to
identify responsive patients and tailor antiangiogenic therapy
but are not yet available (Jain et al., 2009). Mechanism-based
side effects of anti-VEGF (receptor) treatment (hypertension)
show predictive value for antitumor efficacy.
The relative inefficacy of VEGF (receptor) inhibitors in oncological practice calls for more suitable preclinical cancer models
(Bagri et al., 2010; Francia et al., 2011) and has spurred research
into mechanisms underlying resistance (Box 1) (Bergers and Hanahan, 2008; Ebos and Kerbel, 2011; Ferrara, 2010a). Certain
tumors produce proangiogenic factors besides VEGF, even prior
to treatment, and are thus relatively insensitive to VEGF
(receptor) inhibition. Others become unresponsive during treatment, when hypoxia upregulates rescue angiogenic molecules (e.g., PlGF, FGFs, IL-8). Second, vessel co-option or lining
of tumor channels by ECs with cytogenetic abnormalities may
not be as sensitive to VEGF (receptor) inhibitors. Also, the
precise modes of vascular supply in the pre- and micrometastatic niches remain insufficiently characterized (Figure 7B).
Poor vascularization, as in pancreatic cancer, or mature tumor
capillaries, as in hepatocellular carcinoma, may reduce sensitivity to VEGF (receptor) inhibitor treatment. Finally, depriving
the tumor of its vascular supply may select hypoxia-resistant
tumor clones (Ebos and Kerbel, 2011).
Recent preclinical data also raised concerns that VEGF
(receptor) inhibitors might fuel cancer invasiveness and metastasis by aggravating intratumoral hypoxia and creating a proinflammatory environment (Ebos and Kerbel, 2011) (Figure 7A).
These findings are debated, as other preclinical studies have
not observed an increase in malignancy (Padera et al., 2008),
and large meta-analyses have not shown a worse clinical
outcome (Ebos and Kerbel, 2011; Miles et al., 2011). One exception is glioblastoma that exhibits a more invasive phenotype after
VEGF (receptor) blockade in preclinical models and patients,
possibly as a consequence of a hypoxic cancer stem cell niche
that drives recurrence of a more aggressive tumor (Norden
et al., 2009). Conflicting reports on whether discontinuation
of VEGF (receptor) blockade boosts a tumor (angiogenesis)
rebound call for further clarification. Moreover, the most effective
dosing and duration of VEGF (receptor) inhibitor treatment
remain to be determined.
Box 1. Mechansisms of Resistance against VEGF (receptor)

Blockade
VEGF-independent vessel growth: Tumors produce additional
proangiogenic molecules besides VEGF, before or after treatment
with VEGF (receptor) blockers.
Sprouting-independent vessel growth: Tumors possess/switch to
modes of vessel growth (vessel co-option, vascular mimicry, intussusception, etc.) that can be less sensitive to VEGF (receptor)
blockade.
Stromal cells: Both myeloid cells and cancer-associated fibroblasts
produce other proangiogenic factors besides VEGF or recruit proangiogenic bone marrow-derived cells.
Endothelial cell (EC) instability: Endothelial cells with cytogenetic
abnormalities or tumor ECs, which differentiate from cancer stem
cell-like cells (as in glioblastoma), may not be as sensitive to VEGF
(receptor) blockade as sprouting ECs.
Vascular independence: Mutant tumor clones or inflammatory cells
are able to survive in hypoxic tumors; their reduced vascular dependence impairs the antiangiogenic response. Certain tumors have
a hypovascular stroma. Tumors can also metastasize via lymphatics;
their growth may not be blocked by antiangiogenic therapy.
Mature vessels: Mature supply vessels are covered by vascular
smooth muscle cells and not easily pruned by EC-targeted treatment.
EC radioresistance: Hypoxic activation of HIF1a renders ECs resistant to irradiation.
Organ-specific differences: Tumors show opposite invasive
behaviors depending on the organ of inoculation.
Gene variations: Gene variations in VEGF receptors determine the
responsiveness to VEGF (receptor) blockade.
Vessel normalization: Transient vessel normalization can reduce
antiangiogenic drug delivery and efficacy; alternatively, barrier tightening could impede drug penetration.
Primary tumor versus metastasis: Distinct signals regulate angiogenesis in primary versus metatstatic tumors.
Alternative Therapeutic Antitumor

Vascularization Strategies
All approved antiangiogenic therapies have been developed to
starve tumors by destroying their vascular supply. Approaches
with a similar mechanism of action but different targets are under
development. However, alternative strategies that are not solely
based on vessel destruction are being considered as well. We
will highlight a few prototypic examples.
Given that VEGF (receptor) inhibitors are more efficient at destroying capillaries devoid of pericytes, simultaneous targeting
of ECs and pericytes might enhance their antiangiogenic efficacy.
Preclinical treatment with PDGFR inhibitors reduces tumor
progression by facilitating pericyte detachment, thereby rendering vessels more immature and vulnerable to regression. Also,
multitargeted tyrosine kinase inhibitors blocking both PDGFR
and VEGF receptors (besides many other targets) were more efficient than inhibitors of VEGF signaling alone. However, combining
selective PDGFR and VEGF receptor blockers did not meet
expectations (Nisancioglu et al., 2010). PDGFR blocking studies
also highlighted the importance of considering not only effects on
the primary tumor alone but also on metastasis, as poor pericyte
attachment promotes metastasis (Gerhardt and Semb, 2008).
The sustained vascular normalization concept proposes not
to destroy tumor vessels but to restore their structure and
function, so that improved perfusion and oxygenation counteract the hypoxia-driven expression of genes controlling epithelial-mesenchymal transition, invasion, and intravasation, which
prompt the metastatic switch (Goel et al., 2011; Mazzone
et al., 2009; Rolny et al., 2011) (Figure 7B). The normalized vessel
wall also restricts tumor cell intravasation (Mazzone et al., 2009),
while responses to chemo- or immunotherapy can be improved
(Goel et al., 2011; Rolny et al., 2011).
Conclusions and Perspectives
Despite progress in understanding the molecular basis of angiogenesis, and successful translation of VEGF blockade for the
treatment of age-related macular degeneration and some cancer
patients, challenges must be overcome to improve the overall
efficacy of antivascular strategies to combat cancer more efficiently. A question of high priority is whether the approved
antiangiogenic regimes are optimally used in terms of dosing,
duration, and combination therapy. The role of VEGF (receptor)
inhibitors in micrometastatic disease in adjuvant settings (e.g.,
upon resection of the primary tumor) will require further research
given the paucity of available preclinical data and suitable animal
models. Another priority is to identify predictive biomarkers,
tailored for particular tumors, stages, and treatment. Third, development of additional antiangiogenic drugs, independent of VEGF
signaling, and evaluation of their potential in clinical trials, in
particular as combination therapy with current VEGF (receptor)
inhibitors, is likely to expand the antiangiogenic armamentarium.
Fourth, the therapeutic potential of sustained vessel normalization to suppress metastasis and enhance chemotherapy will
need to be evaluated clinically, and additional studies are
required to establish how it could be combined best with available vessel pruning therapies. Also, antivascular approaches
could be beneficial for the treatment of nonsolid malignancies
(e.g., leukemias) or for the treatment of children or pregnant
women with cancer or individuals with inflammatory disorders
(e.g., arthritis) who have not been considered eligible for VEGF
blockade because of side effects. Finally, the recent molecular
breakthroughs in our understanding of vessel growth should
kindle renewed interest in developing strategies to revascularize
ischemic tissues.
ACKNOWLEDGMENTS
We acknowledge L. Notebaert and A. Truyens for help with the illustrations. We
apologize to authors whose work we could not cite because of the limit on the
number of references and therefore mostly cited overview articles. The work of
P.C. is supported by a Federal Government Belgium grant (IUAP06/30), longterm structural Methusalem funding by the Flemish Government, a Concerted
Research Activities Belgium grant (GOA2006/11), a grant from The Research
FoundationFlanders (FWO G.0673.08), and the Foundation Leducq Transatlantic Network ARTEMIS. The research of M.P. is supported by grants from the
Deutsche Forschungsgemeinschaft (DFG - SFB 834/A6 and Exc 147/1). The
research of H.G. is supported by Cancer Research UK, the Lister Institute of
Preventive Medicine, the EMBO Young Investigator Programme, and the
Foundation Leducq Transatlantic Network ARTEMIS.
REFERENCES
Adams, R.H., and Alitalo, K. (2007). Molecular regulation of angiogenesis and
lymphangiogenesis. Nat. Rev. Mol. Cell Biol. 8, 464478.
Adams, R.H., and Eichmann, A. (2010). Axon guidance molecules in vascular

patterning. Cold Spring Harb. Perspect. Biol. 2, a001875.
Ebos, J.M.L., and Kerbel, R.S. (2011). Antiangiogenic therapy: impact on invasion, disease progression, and metastasis. Nat. Rev. Clin. Oncol. 8, 210221.
Arroyo, A.G., and Iruela-Arispe, M.L. (2010). Extracellular matrix, inflammation,

and the angiogenic response. Cardiovasc. Res. 86, 226235.
Eilken, H.M., and Adams, R.H. (2010). Dynamics of endothelial cell behavior in
sprouting angiogenesis. Curr. Opin. Cell Biol. 22, 617625.
Augustin, H.G., Koh, G.Y., Thurston, G., and Alitalo, K. (2009). Control of
vascular morphogenesis and homeostasis through the angiopoietin-Tie
system. Nat. Rev. Mol. Cell Biol. 10, 165177.
Erez, N., Truitt, M., Olson, P., Arron, S.T., and Hanahan, D. (2010). Cancerassociated fibroblasts are activated in incipient neoplasia to orchestrate
tumor-promoting inflammation in an NF-kappaB-dependent manner. Cancer
Cell 17, 135147.
Bagri, A., Kouros-Mehr, H., Leong, K.G., and Plowman, G.D. (2010). Use of
anti-VEGF adjuvant therapy in cancer: challenges and rationale. Trends Mol.
Med. 16, 122132.
Beenken, A., and Mohammadi, M. (2009). The FGF family: biology, pathophysiology and therapy. Nat. Rev. Drug Discov. 8, 235253.
Bentley, K., Mariggi, G., Gerhardt, H., and Bates, P.A. (2009). Tipping the
balance: robustness of tip cell selection, migration and fusion in angiogenesis.
PLoS Comput. Biol. 5, e1000549.
Falk, M.M. (2010). Adherens junctions remain dynamic. BMC Biol. 8, 34.
Fang, S., and Salven, P. (2011). Stem cells in tumor angiogenesis. J. Mol. Cell.
Cardiol. 50, 290295.
Fantin, A., Maden, C.H., and Ruhrberg, C. (2009). Neuropilin ligands in
vascular and neuronal patterning. Biochem. Soc. Trans. 37, 12281232.
Bergers, G., and Hanahan, D. (2008). Modes of resistance to anti-angiogenic

therapy. Nat. Rev. Cancer 8, 592603.
Fantin, A., Vieira, J.M., Gestri, G., Denti, L., Schwarz, Q., Prykhozhij, S., Peri,
F., Wilson, S.W., and Ruhrberg, C. (2010). Tissue macrophages act as cellular
chaperones for vascular anastomosis downstream of VEGF-mediated endothelial tip cell induction. Blood 116, 829840.
Blasi, F., and Carmeliet, P. (2002). uPAR: a versatile signalling orchestrator.

Nat. Rev. Mol. Cell Biol. 3, 932943.
Ferrara, N. (2010a). Pathways mediating VEGF-independent tumor angiogenesis. Cytokine Growth Factor Rev. 21, 2126.
Bonauer, A., Boon, R.A., and Dimmeler, S. (2010). Vascular microRNAs. Curr.
Drug Targets 11, 943949.
Ferrara, N. (2010b). Role of myeloid cells in vascular endothelial growth factorindependent tumor angiogenesis. Curr. Opin. Hematol. 17, 219224.
Butler, J.M., Kobayashi, H., and Rafii, S. (2010). Instructive role of the vascular
niche in promoting tumour growth and tissue repair by angiocrine factors. Nat.
Rev. Cancer 10, 138146.
Fischer, C., Mazzone, M., Jonckx, B., and Carmeliet, P. (2008). FLT1 and its
ligands VEGFB and PlGF: drug targets for anti-angiogenic therapy? Nat.
Rev. Cancer 8, 942956.
Carmeliet, P. (2003). Angiogenesis in health and disease. Nat. Med. 9,

653660.
Folkman, J. (1971). Tumor angiogenesis: therapeutic implications. N. Engl.

J. Med. 285, 11821186.
Carmeliet, P., and Jain, R.K. (2011). Molecular mechanisms and clinical applications of angiogenesis. Nature 473, 298307.
Folkman, J. (2007). Angiogenesis: an organizing principle for drug discovery?

Nat. Rev. Drug Discov. 6, 273286.
Carmeliet, P., and Tessier-Lavigne, M. (2005). Common mechanisms of nerve

and blood vessel wiring. Nature 436, 193200.
Fraisl, P., Mazzone, M., Schmidt, T., and Carmeliet, P. (2009). Regulation of
angiogenesis by oxygen and metabolism. Dev. Cell 16, 167179.
Castets, M., and Mehlen, P. (2010). Netrin-1 role in angiogenesis: to be or not

to be a pro-angiogenic factor? Cell Cycle 9, 14661471.
Francia, G., Cruz-Munoz, W., Man, S., Xu, P., and Kerbel, R.S. (2011). Mouse
models of advanced spontaneous metastasis for experimental therapeutics.
Nat. Rev. Cancer 11, 135141.
Cavallaro, U., and Dejana, E. (2011). Adhesion molecule signalling: not always
a sticky business. Nat. Rev. Mol. Cell Biol. 12, 189197.
Chappell, J.C., and Bautch, V.L. (2010). Vascular development: genetic mechanisms and links to vascular disease. Curr. Top. Dev. Biol. 90, 4372.
Corada, M., Nyqvist, D., Orsenigo, F., Caprini, A., Giampietro, C., Taketo,
M.M., Iruela-Arispe, M.L., Adams, R.H., and Dejana, E. (2010). The Wnt/
beta-catenin pathway modulates vascular remodeling and specification by
upregulating Dll4/Notch signaling. Dev. Cell 18, 938949.
Franco, C.A., Liebner, S., and Gerhardt, H. (2009). Vascular morphogenesis:

a Wnt for every vessel? Curr. Opin. Genet. Dev. 19, 476483.
Gaengel, K., Genove, G., Armulik, A., and Betsholtz, C. (2009). Endothelialmural cell signaling in vascular development and angiogenesis. Arterioscler.
Thromb. Vasc. Biol. 29, 630638.
Gerhardt, H., and Semb, H. (2008). Pericytes: gatekeepers in tumour cell
metastasis? J. Mol. Med. 86, 135144.
Crawford, Y., and Ferrara, N. (2009a). Tumor and stromal pathways mediating
refractoriness/resistance to anti-angiogenic therapies. Trends Pharmacol. Sci.
30, 624630.
Goel, S., Duda, D.G., Xu, L., Munn, L.L., Boucher, Y., Fukumura, D., and Jain,
R.K. (2011). Normalization of the vasculature for treatment of cancer and other
diseases. Physiol. Rev. 91, 10711121.
Crawford, Y., and Ferrara, N. (2009b). VEGF inhibition: insights from preclinical
and clinical studies. Cell Tissue Res. 335, 261269.
Gridley, T. (2010). Notch signaling in the vasculature. Curr. Top. Dev. Biol. 92,
277309.
De Palma, M., and Naldini, L. (2009). Tie2-expressing monocytes (TEMs):

novel targets and vehicles of anticancer therapy? Biochim. Biophys. Acta
1796, 510.
Grivennikov, S.I., Greten, F.R., and Karin, M. (2010). Immunity, inflammation,

and cancer. Cell 140, 883899.
De Smet, F., Segura, I., De Bock, K., Hohensinner, P.J., and Carmeliet, P.
(2009). Mechanisms of vessel branching: filopodia on endothelial tip cells
lead the way. Arterioscler. Thromb. Vasc. Biol. 29, 639649.
Desgrosellier, J.S., and Cheresh, D.A. (2010). Integrins in cancer: biological
implications and therapeutic opportunities. Nat. Rev. Cancer 10, 922.
Ding, B.S., Nolan, D.J., Butler, J.M., James, D., Babazadeh, A.O., Rosenwaks,
Z., Mittal, V., Kobayashi, H., Shido, K., Lyden, D., et al. (2010). Inductive angiocrine signals from sinusoidal endothelium are required for liver regeneration.
Nature 468, 310315.
Guarani, V., Deflorian, G., Franco, C.A., Kruger, M., Phng, L.K., Bentley, K.,
Toussaint, L., Dequiedt, F., Mostoslavsky, R., Schmidt, M.H., et al. (2011).
Acetylation-dependent regulation of endothelial Notch signalling by the
SIRT1 deacetylase. Nature 473, 234238.
Hagberg, C.E., Falkevall, A., Wang, X., Larsson, E., Huusko, J., Nilsson, I., van
Meeteren, L.A., Samen, E., Lu, L., Vanwildemeersch, M., et al. (2010). Vascular
endothelial growth factor B controls endothelial fatty acid uptake. Nature 464,
917921.
Dyer, L.A., and Patterson, C. (2010). Development of the endothelium: an

emphasis on heterogeneity. Semin. Thromb. Hemost. 36, 227235.
Herbert, S.P., Huisken, J., Kim, T.N., Feldman, M.E., Houseman, B.T., Wang,
R.A., Shokat, K.M., and Stainier, D.Y. (2009). Arterial-venous segregation by
selective cell sprouting: an alternative mode of blood vessel formation.
Science 326, 294298.
Eble, J.A., and Niland, S. (2009). The extracellular matrix of blood vessels.
Curr. Pharm. Des. 15, 13851400.
Huang, H., Bhat, A., Woodnutt, G., and Lappe, R. (2010). Targeting the
ANGPT-TIE2 pathway in malignancy. Nat. Rev. Cancer 10, 575585.
Iruela-Arispe, M.L., and Davis, G.E. (2009). Cellular and molecular mechanisms of vascular lumen formation. Dev. Cell 16, 222231.
Jain, R.K. (2003). Molecular regulation of vessel maturation. Nat. Med. 9,
685693.
Jain, R.K. (2005). Normalization of tumor vasculature: an emerging concept in
antiangiogenic therapy. Science 307, 5862.
Jain, R.K., Duda, D.G., Willett, C.G., Sahani, D.V., Zhu, A.X., Loeffler, J.S.,
Batchelor, T.T., and Sorensen, A.G. (2009). Biomarkers of response and resistance to antiangiogenic therapy. Nat. Rev. Clin. Oncol. 6, 327338.
Nisancioglu, M.H., Betsholtz, C., and Genove, G. (2010). The absence of pericytes does not increase the sensitivity of tumor vasculature to vascular endothelial growth factor-A blockade. Cancer Res. 70, 51095115.
Norden, A.D., Drappatz, J., and Wen, P.Y. (2009). Antiangiogenic therapies for
high-grade glioma. Nat. Rev. Neurol. 5, 610620.
Nyberg, P., Xie, L., and Kalluri, R. (2005). Endogenous inhibitors of angiogenesis. Cancer Res. 65, 39673979.
Nyberg, P., Salo, T., and Kalluri, R. (2008). Tumor microenvironment and
angiogenesis. Front. Biosci. 13, 65376553.
Jakobsson, L., Franco, C.A., Bentley, K., Collins, R.T., Ponsioen, B., Aspalter,
I.M., Rosewell, I., Busse, M., Thurston, G., Medvinsky, A., et al. (2010). Endothelial cells dynamically compete for the tip cell position during angiogenic
sprouting. Nat. Cell Biol. 12, 943953.
Padera, T.P., Kuo, A.H., Hoshida, T., Liao, S., Lobo, J., Kozak, K.R., Fukumura,
D., and Jain, R.K. (2008). Differential response of primary tumor versus
lymphatic metastasis to VEGFR-2 and VEGFR-3 kinase inhibitors cediranib
and vandetanib. Mol. Cancer Ther. 7, 22722279.
Jeansson, M., Gawlik, A., Anderson, G., Li, C., Kerjaschki, D., Henkelman, M.,
and Quaggin, S.E. (2011). Angiopoietin-1 is essential in mouse vasculature
during development and in response to injury. J. Clin. Invest. 121, 22782289.
Pardali, E., Goumans, M.J., and ten Dijke, P. (2010). Signaling by members of
the TGF-beta family in vascular morphogenesis and disease. Trends Cell Biol.
20, 556567.
Jones, E.A., le Noble, F., and Eichmann, A. (2006). What determines blood
vessel structure? Genetic prespecification vs. hemodynamics. Physiology
(Bethesda) 21, 388395.
Phng, L.K., and Gerhardt, H. (2009). Angiogenesis: a team effort coordinated

by notch. Dev. Cell 16, 196208.
Kerbel, R.S. (2008). Tumor angiogenesis. N. Engl. J. Med. 358, 20392049.

Kim, J., Oh, W.J., Gaiano, N., Yoshida, Y., and Gu, C. (2011). Semaphorin 3EPlexin-D1 signaling regulates VEGF function in developmental angiogenesis
via a feedback mechanism. Genes Dev. 25, 13991411.
Koch, A.W., Mathivet, T., Larrivee, B., Tong, R.K., Kowalski, J., PibouinFragner, L., Bouvree, K., Stawicki, S., Nicholes, K., Rathore, N., et al. (2011).
Robo4 maintains vessel integrity and inhibits angiogenesis by interacting
with UNC5B. Dev. Cell 20, 3346.
Leblanc, G.G., Golanov, E., Awad, I.A., and Young, W.L.; Biology of Vascular
Malformations of the Brain NINDS Workshop Collaborators. (2009). Biology of
vascular malformations of the brain. Stroke 40, e694e702.
Liu, Z., Turkoz, A., Jackson, E.N., Corbo, J.C., Engelbach, J.A., Garbow, J.R.,
Piwnica-Worms, D.R., and Kopan, R. (2011). Notch1 loss of heterozygosity
causes vascular tumors and lethal hemorrhage in mice. J. Clin. Invest. 121,
800808.
London, N.R., Smith, M.C., and Li, D.Y. (2009). Emerging mechanisms of
vascular stabilization. J. Thromb. Haemost. 7 (Suppl 1), 5760.
Lucke, S., and Levkau, B. (2010). Endothelial functions of sphingosine-1-phosphate. Cell. Physiol. Biochem. 26, 8796.
Phng, L.K., Potente, M., Leslie, J.D., Babbage, J., Nyqvist, D., Lobov, I., Ondr,
J.K., Rao, S., Lang, R.A., Thurston, G., and Gerhardt, H. (2009). Nrarp coordinates endothelial Notch and Wnt signaling to control vessel density in angiogenesis. Dev. Cell 16, 7082.
Pietras, K., and Ostman, A. (2010). Hallmarks of cancer: interactions with the
tumor stroma. Exp. Cell Res. 316, 13241331.
Pitulescu, M.E., and Adams, R.H. (2010). Eph/ephrin moleculesa hub for
signaling and endocytosis. Genes Dev. 24, 24802492.
Qian, B.Z., and Pollard, J.W. (2010). Macrophage diversity enhances tumor
progression and metastasis. Cell 141, 3951.
Quaegebeur, A., Segura, I., and Carmeliet, P. (2010). Pericytes: blood-brain
barrier safeguards against neurodegeneration? Neuron 68, 321323.
Rolny, C., Mazzone, M., Tugues, S., Laoui, D., Johansson, I., Coulon, C., Squadrito, M.L., Segura, I., Li, X., Knevels, E., et al. (2011). HRG inhibits tumor
growth and metastasis by inducing macrophage polarization and vessel
normalization through downregulation of PlGF. Cancer Cell 19, 3144.
Roukens, M.G., Alloul-Ramdhani, M., Baan, B., Kobayashi, K., PetersonMaduro, J., van Dam, H., Schulte-Merker, S., and Baker, D.A. (2010). Control
of endothelial sprouting by a Tel-CtBP complex. Nat. Cell Biol. 12, 933942.
Majmundar, A.J., Wong, W.J., and Simon, M.C. (2010). Hypoxia-inducible

factors and the response to hypoxic stress. Mol. Cell 40, 294309.
Sawamiphak, S., Seidel, S., Essmann, C.L., Wilkinson, G.A., Pitulescu, M.E.,
Acker, T., and Acker-Palmer, A. (2010). Ephrin-B2 regulates VEGFR2 function
in developmental and tumour angiogenesis. Nature 465, 487491.
Makanya, A.N., Hlushchuk, R., and Djonov, V.G. (2009). Intussusceptive

angiogenesis and its role in vascular morphogenesis, patterning, and remodeling. Angiogenesis 12, 113123.
Schaper, W. (2009). Collateral circulation: past and present. Basic Res.

Cardiol. 104, 521.
Mazzieri, R., Pucci, F., Moi, D., Zonari, E., Ranghetti, A., Berti, A., Politi, L.S.,
Gentner, B., Brown, J.L., Naldini, L., and De Palma, M. (2011). Targeting the
ANG2/TIE2 axis inhibits tumor growth and metastasis by impairing angiogenesis and disabling rebounds of proangiogenic myeloid cells. Cancer Cell 19,
512526.
Mazzone, M., Dettori, D., Leite de Oliveira, R., Loges, S., Schmidt, T., Jonckx,
B., Tian, Y.M., Lanahan, A.A., Pollard, P., Ruiz de Almodovar, C., et al. (2009).
Heterozygous deficiency of PHD2 restores tumor oxygenation and inhibits
metastasis via endothelial normalization. Cell 136, 839851.
Miles, D., Harbeck, N., Escudier, B., Hurwitz, H., Saltz, L., Van Cutsem, E.,
Cassidy, J., Mueller, B., and Sirzen, F. (2011). Disease course patterns after
discontinuation of bevacizumab: pooled analysis of randomized phase III
trials. J. Clin. Oncol. 29, 8388.
Schmidt, T., Kharabi Masouleh, B., Loges, S., Cauwenberghs, S., Fraisl, P.,
Maes, C., Jonckx, B., De Keersmaecker, K., Kleppe, M., Tjwa, M., et al.
(2011). Loss or inhibition of stromal-derived PlGF prolongs survival of mice
with imatinib-resistant Bcr-Abl1(+) leukemia. Cancer Cell 19, 740753.
Serini, G., Maione, F., Giraudo, E., and Bussolino, F. (2009). Semaphorins and
tumor angiogenesis. Angiogenesis 12, 187193.
, B., Kucera, T., Eglinger, J., Hughes, M.R., McNagny, K.M., Tsukita, S.,
Strilic
Dejana, E., Ferrara, N., and Lammert, E. (2009). The molecular basis of
vascular lumen formation in the developing mouse aorta. Dev. Cell 17,
505515.
Swift, M.R., and Weinstein, B.M. (2009). Arterial-venous specification during
development. Circ. Res. 104, 576588.
Murakami, M., Nguyen, L.T., Zhuang, Z.W., Moodie, K.L., Carmeliet, P., Stan,
R.V., and Simons, M. (2008). The FGF system has a key role in regulating
vascular integrity. J. Clin. Invest. 118, 33553366.
Tammela, T., Zarkada, G., Wallgard, E., Murtomaki, A., Suchting, S., Wirzenius, M., Waltari, M., Hellstrom, M., Schomber, T., Peltonen, R., et al. (2008).
Blocking VEGFR-3 suppresses angiogenic sprouting and vascular network
formation. Nature 454, 656660.
Nicoli, S., Standley, C., Walker, P., Hurlstone, A., Fogarty, K.E., and Lawson,
N.D. (2010). MicroRNA-mediated integration of haemodynamics and Vegf signalling during angiogenesis. Nature 464, 11961200.
Thurston, G., Noguera-Troise, I., and Yancopoulos, G.D. (2007). The Delta
paradox: DLL4 blockade leads to more tumour vessels but less tumour
growth. Nat. Rev. Cancer 7, 327331.
Van Cutsem, E., Lambrechts, D., Prenen, H., Jain, R.K., and Carmeliet, P.
(2011). Lessons from the adjuvant bevacizumab trial on colon cancer: what
next? J. Clin. Oncol. 29, 14.
Wang, Y., Nakayama, M., Pitulescu, M.E., Schmidt, T.S., Bochenek, M.L.,
Sakakibara, A., Adams, S., Davy, A., Deutsch, U., Luthi, U., et al. (2010).
Ephrin-B2 controls VEGF-induced angiogenesis and lymphangiogenesis.
Nature 465, 483486.
Warren, C.M., and Iruela-Arispe, M.L. (2010). Signaling circuitry in vascular
morphogenesis. Curr. Opin. Hematol. 17, 213218.
Wels, J., Kaplan, R.N., Rafii, S., and Lyden, D. (2008). Migratory neighbors and
distant invaders: tumor-associated niche cells. Genes Dev. 22, 559574.
Xu, K., Sacharidou, A., Fu, S., Chong, D.C., Skaug, B., Chen, Z.J., Davis, G.E.,
and Cleaver, O. (2011). Blood vessel tubulogenesis requires Rasip1 regulation
of GTPase signaling. Dev. Cell 20, 526539.
Yan, M., Callahan, C.A., Beyer, J.C., Allamneni, K.P., Zhang, G., Ridgway, J.B.,
Niessen, K., and Plowman, G.D. (2010). Chronic DLL4 blockade induces
vascular neoplasms. Nature 463, E6E7.
Zeeb, M., Strilic, B., and Lammert, E. (2010). Resolving cell-cell junctions:
lumen formation in blood vessels. Curr. Opin. Cell Biol. 22, 626632.
Chromosome Catastrophes Involve

Replication Mechanisms Generating
Complex Genomic Rearrangements
Pengfei Liu,1,16 Ayelet Erez,1,4,16 Sandesh C. Sreenath Nagamani,1,4 Shweta U. Dhar,1,4 Katarzyna E. Ko1odziejska,1
Avinash V. Dharmadhikari,1 M. Lance Cooper,1 Joanna Wiszniewska,1 Feng Zhang,1,14 Marjorie A. Withers,1
Carlos A. Bacino,1,4 Luis Daniel Campos-Acevedo,5 Mauricio R. Delgado,6 Debra Freedenberg,7,15 Adolfo Garnica,8
Theresa A. Grebe,9 Dolores Hernandez-Almaguer,5 LaDonna Immken,10 Seema R. Lalani,1,2,4 Scott D. McLean,11
Hope Northrup,12 Fernando Scaglia,1,4 Lane Strathearn,2,3,4 Pamela Trapane,13 Sung-Hae L. Kang,1 Ankita Patel,1
Sau Wai Cheung,1 P.J. Hastings,1 Pawe1 Stankiewicz,1 James R. Lupski,1,2,4,* and Weimin Bi1
1Department
of Molecular and Human Genetics

of Pediatrics
3Department of Psychiatry and Behavioral Sciences
Baylor College of Medicine, Houston, TX 77030, USA
4Texas Childrens Hospital, Houston, TX 77030, USA
5Departamento de Gene
tica, Hospital Universitario, Universidad Autonoma de Nuevo Leon, CP 64460 Monterrey, NL, Mexico
6Texas Scottish Rite Hospital, Dallas, TX 75219, USA
7Division of Medical Genetics, Vanderbilt University Medical Center, Nashville, TN 37232, USA
8St. Francis Hospital, Tulsa, OK 74136, USA
9Division of Genetics and Metabolism, Phoenix Childrens Hospital, Phoenix, AZ 85016, USA
100 Specially for Children, Austin, TX 78723, USA
11Department of Pediatrics, San Antonio Military Medical Center, San Antonio, TX 78234, USA
12Division of Medical Genetics, Department of Pediatrics, The University of Texas Medical School at Houston, Houston, TX 77030, USA
13Childrens Hospital of Wisconsin, Milwaukee, WI 53201, USA
14Present address: School of Life Sciences, Fudan University, Shanghai 200433, China
15Present address: Texas Department of State Health Services, Austin, TX 78756, USA
16These authors contributed equally to this work
*Correspondence: jlupski@bcm.edu
DOI 10.1016/j.cell.2011.07.042
2Department
SUMMARY
Complex genomic rearrangements (CGRs) consisting of two or more breakpoint junctions have been
observed in genomic disorders. Recently, a chromosome catastrophe phenomenon termed chromothripsis, in which numerous genomic rearrangements
are apparently acquired in one single catastrophic
event, was described in multiple cancers. Here, we
show that constitutionally acquired CGRs share similarities with cancer chromothripsis. In the 17 CGR
cases investigated, we observed localization and
multiple copy number changes including deletions,
duplications, and/or triplications, as well as extensive
translocations and inversions. Genomic rearrangements involved varied in size and complexities; in
one case, array comparative genomic hybridization
revealed 18 copy number changes. Breakpoint sequencing identified characteristic features, including
small templated insertions at breakpoints and microhomology at breakpoint junctions, which have been
attributed to replicative processes. The resemblance
between CGR and chromothripsis suggests similar
mechanistic underpinnings. Such chromosome catastrophic events appear to reflect basic DNA metabolism operative throughout an organisms life cycle.
INTRODUCTION
Human genomic rearrangements with two or more breakpoint
junctions are referred to as complex genomic rearrangements
(CGRs) (Zhang et al., 2009a). CGRs have been identified
frequently during characterization of nonrecurrent microduplications associated with genomic disorders. Based on the microhomologies identified at breakpoint junctions, the apparent template driven insertional complexities at breakpoints, and the
fusions of distantly distributed sequences in complex genomic
rearrangements, a replication based mechanism, the fork stalling and template switching (FoSTeS) model, has been proposed
to explain the formation of such rearrangement complexities
in the human genome (Lee et al., 2007; Slack et al., 2006).
Other similar replication based models such as microhomology
mediated break-induced replication (MMBIR) (Hastings et al.,
2009a, 2009b), microhomology/microsatellite induced replication (MMIR) (Payen et al., 2008), and microhomology-mediated
replication-dependent recombination (MMRDR) (Chen et al.,
2010) have also been proposed. Recent studies on genomic
Cell 146, 889903, September 16, 2011 2011 Elsevier Inc. 889
Table 1. Complex Genomic Rearrangements Assessed by Gains and Losses of Specific Human Genome Intervals
Subjects
(BAB#)
Sex Location
Array CGH-Inferred
Rearrangement
Sizes of Copy
Pattern
Number Changes
Size (Mb)
Clinical Indication
Parental Study Total CNVs Gain Loss
Patient 1 F
1q32.3-q43
del-nml-dup-nmldup-nml-del
del 4.7 Mb-nml

11.5 Mb-dup 3.1 Mb-nml
8.1 Mbdup 4.4 Mb-nml
0.3 Mb-del 1.8 Mb
Severe DD/ID
de novo
34.0 14.0
7.5
6.5
2920
1q43-q44
dup-del
dup 0.7 Mb-del 4 Mb
Microcephaly,
epilepsy
mother nml
4.7
0.7
4.0
mat
4.7
3047
4q33-q34.1
trp
trp 1.2 Mb
DD
1.2
1.2
1.2
2760
4q35.1-q35.2
dup-nml-dup
dup 115 kb-nml 2.6

Mb-dup 298 kb
Autistic spectrum, NA
DD/ID, prenatal
teratogen exposure
3.0
0.4
0.4
3012
5q31.1-q31.2
dup-trp
dup 28 kb-trp 1.9 Mb
Mild DD, DF
mat (47%)
1.9
1.9
1.9
2778
6q27
dup-nml-terminal
dup, r(6)
dup 2.2 Mb-nml

666 kb-dup 1.0 Mb
Chromosomal
abnormality,
microcephaly,
speech delay
father nml
3.9
3.2
3.2
3103
7q33-ter
dup-nml-del
dup 25.6 Mb-nml

7.9 kb-del 269 kb
DF, VSD
de novo
25.9 25.9
25.6 0.3
3105
9q31.1-q33.1
dup-nml-dup-nmldup-nml-dup-nmldup-nml-dup-nmldup-nml-dup-nmldup-trp-dup-nmldup-nml-trp-dupnml-dup-nml-dupnml-dup-nml-dup
dup 2.1 Mb-nml

0.8 Mb-dup 37 kb-nml
0.2 Mb-dup 0.2 Mb- nml
8.1 Mb-dup 0.1 Mb-trp
0.1 Mbdup 44 kb-nml
33 kbtrp 5.5 Mb-dup
0.4 Mb-nml 0.2 Mb-dup
20 kbnml 35 kb-dup
0.8 Mb-nml 0.2 Mb-dup
1.4 Mbnml 0.4 Mb-dup
29 kb
DF
de novo
51.0 11.6
11.6 0
3015
12q21.1
dup-trp
dup 9.4 kb-trp 2.2 Mb
DD, DF
mat
2.3
2780
12q24
dup-nml-dup
dup 250 kb-nml

162 kb-dup 10.7 Mb
DD
de novo
11.1 10.9
10.9 0
2785
13q34
dup-nml-dup
dup 508 kb-nml

127 kb-dup 500 kb
Moderate DD/ID,
DF
mat
1.1
1.0
3011
14q12-q21.3
dup-nml-del
dup 2.8 Mb-nml

1.7 Mb-del 15.4 Mb
DD, DF
de novo
19.9 18.2
2.8
15.4
3050
15q26.3
dup-trp-duptrp-dup
dup 1.3 Mb-trp

0.9 Mb-dup 40 kb-trp
0.3 Mbdup 41 kb
DD, DF, MCA,

immunodeficiency
de novo
2.5
2.5
2.5
2783
18p11.32
dup-nml-dup
dup 590 kb-nml

1.4 Mb-dup 443 kb
ID
mother nml
2.4
1.0
1.0
0.0
del 2.0 Mb-nml

0.3 Mbdup 1.7 Mb- nml
2.5 Mbdup 2.2 Mb-nml
8.8 Mb-del 0.4 Mb-nml
3.1 Mbdup 1.6 Mb
Chromosomal
abnormality,
epilepsy
NA
35.1 14.5
Patient 2 F
22q11.1-q13.33 del-nml-dup-nmldup-nml-delnml-dup
890 Cell 146, 889903, September 16, 2011 2011 Elsevier Inc.
2.2
1.0
2.2
12.1 2.4
Table 1. Continued
Subjects
(BAB#)
Sex Location
Array CGH-Inferred
Rearrangement
Sizes of Copy
Pattern
Number Changes
Size (Mb)
Clinical Indication
Parental Study Total CNVs Gain Loss
3104
del 0.4 Mb-nml 295 kb-dup Hypotonia,

22q12.2-q13.31 del-nml-dup-nmlSeizure
dup-nml-dup-nml- 4.3 Mb-nml 175 kb-dup
del-nml-del-nml-del 0.5 Mb-nml 1.1 Mb-dup
0.3 Mb-nml 10 Mbdel
76 kb-nml 23 kb-del
2.8 Mb-nml 738 kb-del
0.9 Mb
NA
21.8 9.4
5.1
4.3
3032
Xq27.1
mother nml
0.7
0.7
0.0
dup-trp-duptrp-dup
dup 194 kb-trp 191 kb-dup Moderate

18 kb-trp 28 kb-dup 200 kb DD/ID
0.7
Abbreviations: dup-, duplication; trp-, triplication; del-, deletion; nml-, normal; DD-, developmental delay; ID-, intellectual disability; DF-, dysmorphic
features; VSD-, ventricular septal defect; MCA-, multiple congenital anomalies; NA-, not available; mat-, maternal. See also Table S1 for detailed clinical features and Figure S5 for descriptions of low-copy repeats at breakpoints.
disorder associated nonrecurrent rearrangements identified

CGRs on chromosome X (Carvalho et al., 2009; Liu et al.,
2011) and at multiple genomic loci on autosomes such as
17p13.3 (Bi et al., 2009), 17p12 (Zhang et al., 2009b), 17p11.2
(Zhang et al., 2010), 9q34.3 (Yatsenko et al., 2009), and 1p36
(Gajecka et al., 2010).
Chromosome rearrangements are also frequently observed
in cancers. At an organismal level, the rearrangements acquired
in cancers differ from the ones in genomic disorders in the time
they arise during the life cycle (Lupski, 2010). Genomic disorders
frequently result from constitutional germline rearrangements
that occur during gametogenesis or early postzygotic development, whereas rearrangements acquired in cancers involve
somatic differentiated cells. Thus, genomic rearrangement
may be less complex in genomic disorders than in cancers reflecting selective forces because an organism cannot endure/
survive excessive toxicities from massive genomic changes
early in development. However, on a cellular level, the mechanisms underlying these DNA rearrangements occurring in cells
at different stages of the human life cycle (i.e., germline, postzygotic development, somatic differentiated cells) are likely to be
the same.
Recently, the phenomenon of chromothripsis (Stephens et al.,
2011), an apparent chromosome catastrophe with several copy
number changes and multiple breakpoints concentrated on a
single chromosome, was described in cancer cells. Remarkably,
2%3% of all cancers, and up to 25% of bone cancers, demonstrated chromothripsis. In contrast to the generally accepted
concept for cancer biogenesis, in which a mutational accumulation model appears operative, the profound level and complexity
of rearrangements observed in chromothripsis are generated on
a much shorter time scale, probably in a single mutational event.
The mechanisms behind these cataclysmic genome disruptions
are unknown.
We identified apparent CGR in subjects referred with developmental delay and cognitive anomalies. Using diverse highresolution genome analysis techniques, we show that such
constitutional CGRs share many structural and breakpoint
features with cancer chromothripsis. Constitutional CGR can
involve multiple copy number changes, translocations, and
inversions in a genomic region-focused manner; frequently,

two or more genomic segments from separate loci are assembled into one larger rearranged piece. These characteristic
features are consistent with formation of the highly complex
pattern of chromosome catastrophe by a replicative mechanism
in a single event. We propose that both the constitutional CGR
and the chromothripsis processes reflect basic DNA metabolism
and share a cellular DNA replication/repair mechanism.
RESULTS
Complex Genomic Rearrangements Identified
by Clinical CMA
We investigated 17 casesten males and seven females,
referred to the Medical Genetics Laboratories (MGL; http://
www.bcm.edu/geneticlabs/) for various developmental problems. Clinical chromosomal microarray analysis (CMA) by array
comparative genomic hybridization (aCGH) in each subject
showed apparent multiple copy number changes involving a
single chromosome potentially representing a CGR (Table 1
and Figure 1A). A subset of copy number changes, with a genomic size large enough to be resolved, were further confirmed
by fluorescence in situ hybridization (FISH) or G-banded chromosome analyses (Figures 1B1E). Multiple rearrangement
patterns were observed: a duplication (dup) followed by a normal
copy (nml) sequence and then by a duplication (i.e., dup-nmldup), a duplication followed by a deletion (del) (i.e., dup-del),
a duplication followed by a normal copy sequence followed by
a deletion (dup-nml-del), and a duplication followed by a triplication (trp) (i.e., dup-trp). Triplications were identified in six cases.
Rearrangements with complex patterns including three or more
apparent copy number changes were also identified. Additional
chromosomal structural aberrations were found in case
BAB2778, which has a terminal dup-nml-dup rearrangement
within a ring chromosome 6 (Figure 1B).
All CGRs show localization to a single chromosome. The copy
number changes in most cases (15 out of 17) were confined
within the distal half of the involved chromosomal arms. Three
cases (BAB2778, 3103 and patient 2) contained a terminal rearrangement. BAB3103 had a terminal deletion whereas BAB2778
Figure 1. Cases with Complex Rearrangements Identified by Clinical CMA

(A) Array CGH data for subjects on the left with the interpreted rearrangement patterns depicted on the right. The piled rectangles depict copy number status for
the region (one for deletion, two for normal diploid copy, three for duplication, and four for triplication). The V-shaped lines connecting different segments indicate
regions of unknown copy number state due to lack of interrogating oligonucleotide coverage in the array.
(BE) Representative FISH and chromosome analyses independently confirm the CMA findings and provide positional information. The chromosomes with
rearrangements are indicated by arrows.
(B) Metaphase FISH analysis using two probes located within the duplications show that the rearrangement was within a ring chromosome 6 in BAB2778.
had a terminal duplication. Patient 2 had a terminal duplication

that was part of its very complex rearrangement. Additional
CMA cases with complex rearrangements involving terminal
changes frequently consist of a terminal deletion followed by
an inverted duplication (Figure S1 available online). The rearrangements were apparently distributed on all chromosomes.
Of the 17 patients, parental studies when available showed
that five patients had a de novo rearrangement whereas four
patients had a maternally inherited rearrangement, demonstrating transmission of the CGR. The mother of individual
BAB3012 with a triplication was mosaic for the triplication in
47% of the cells examined by FISH analysis and the mosaicism
was independently confirmed by aCGH (data not shown). These
data suggest a postzygotic, mitotic origin for this CGR.
Characterization of Complex Genomic Rearrangements
by High-Density Arrays
More extensive genome resolution of these complex rearrangements was achieved by aCGH with custom-designed highdensity Agilent 60K or 180K arrays, or commercially available
Nimblegen or Agilent arrays (Figure 2). These studies confirmed
the CMA findings, refined the genomic intervals involved, and
mapped most breakpoint junctions to a genomic region of a
few kilobases or less in size (Table S2). The genomic sizes of
the intervals involved in these CGR ranged from 0.7 Mb to 51 Mb,
whereas the combined individual intervals with copy number
changes within the CGR ranged from 0.4 Mb to 25.9 Mb in
size; the latter reduced amounts reflecting normal copy intervals
within the CGRs. For a single CNV event, the sizes ranged
between 9.4 kb and 25.6 Mb in size. Two copy number changes
were either adjacent to each other or separated by a normal copy
sequence as small as 7.9 kb or as large as 11.7 Mb in size.
Higher-resolution aCGH analysis revealed additional complexity in eight cases, BAB3012, 3015, 3103, 3032, 3104,
3105, patient 1, and patient 2. Duplications were identified proximal and adjacent to triplications in two of the three cases with a
single triplication detected initially by clinical CMA. The sizes of
the duplications in these two cases, BAB3012 and BAB3015,
were 28 kb and 9 kb, respectively. BAB3103 had a duplication
of 25.6 Mb in size apparently juxtaposed to a terminal deletion
of 269 kb identified by CMA. High-density aCGH revealed that
the duplication and deletion are separated by a 7.9 kb normal
copy sequence. BAB3032 had a trp-dup in Xq revealed by
CMA. High-density array detected a 194 kb duplication adjacent
and proximal to the triplication and an 18 kb duplication within
the triplicated segment.
Some Genomic Rearrangements Are Very Complex
Four cases, BAB3104 and 3105, patient 1, and patient 2, had
highly complex rearrangements. Using a multitude of techniques, we found that these cases all had a combination of interspersed multiple copy number changes, including duplications,
deletions, or triplications together with additional structural

changes, such as translocations or inversions in one single
chromosome.
Patient BAB3105, a one and a half-year-old boy with dysmorphic features, had 18 copy number change events, with the
rearrangement pattern nml-dup-nml-dup-nml-dup-nml-dupnml-dup-nml-dup-nml-dup-nml-dup-nml-dup-trp-dup-nml-dupnml-trp-dup-nml-dup-nml-dup-nml-dup-nml-dup-nml (Figures
3A and 3B and Figures S2AS2F). Partial chromosome analysis
showed that this highly complex rearrangement was constrained
within one chromosome with the abnormal chromosome 9
having additional material inserted proximal to 9q21; the chromosome homolog was cytogenetically normal (Figure S2H).
Multiple FISH analyses were performed to determine whether
the gained material in the duplications and triplications were
inserted in tandem with their original copies or translocated
into different loci. Interestingly, for all the intervals tested (four
duplications and one triplication), FISH results demonstrated
breakpoint clustering with all the additional copies translocated
to a proximal region at 9q21, adjacent to the pericentric heterochromatin (Figure 3A and Figures S2IS2L). In addition, FISH
also revealed an inversion event with one copy of the 5.5 Mb triplicated segment inversely oriented in comparison to the other
two copies (Figure 3A). Breakpoint sequencing analyses successfully resolved six breakpoint junctions (Figures 3A and 4A).
A striking breakpoint-clustering pattern was evident, whereby
all junctions presented fusion of dispersed duplicated or triplicated segments. The FISH and breakpoint junction data are
consistent with a general impression that most, if not all, of the
additional copies of duplications and triplications in BAB3105
are randomly conjoined and assembled into a large breakpoint
junction cluster.
Breakpoint junction sequence also revealed that novel sequence insertions were frequently observed at junctions (five
out of six), with four of them being relatively long sequences
(541542 bp) (Figure 4A). These insertions are all nonrandom
sequences because their sequences match to either a single
interval or a joining of two intervals from chromosome 9 in the
reference genome. Microhomology and inversion are also
frequently observed in this subject. For example, breakpoint #5
in Figure 4A was a head-to-tail junction of gained genomic
sequences corresponding to the purple and the gray segments
(Figure 4A), whose mapping in the reference genome were
located 34 Mb apart. A 1542 bp sequence was inserted at
the junction. The 1542 bp insertion consists of two segments:
one 1524 bp segment that matches to a region on the opposite
DNA strand 1500 bp distal to the reference purple arrow
sequence and one 18 bp segment that matches to a region on
9p13.3. Both of the two segments were inserted into the junction
with flanking microhomology. The constellation of the above
features strongly suggests a replicative mechanism for formation
of the highly complex chromosome catastrophe event in
(C and D) G-banded chromosome (C) and FISH (D) analyses using two probes within the two duplications show that neither of the two individual duplicated
segments appears to be simple tandem events; subsequent sequence analysis demonstrated that one duplication in BAB2780 is inserted in between the other
duplication.
(E) Interphase FISH analysis revealed three red signals, which confirmed the triplication in BAB3050. The BAC probes used were indicated.
See also Figure S1 for additional rearrangements with CMA results indicating a duplication next to a terminal deletion.
A
BAB 3012
BAB 3032
dup
dup -
trp
BAB 2783
BAB 3015
dup
trp -dup-trp - dup
trp
dup
nml
dup
BAB 2785
BAB 3103
dup
dup
nml-del
nml
dup
BAB 2778
nml-dup-trp
BAB 3012
nml-dup-trp
BAB 3015
dup-nml-del
BAB 3103
dup
nml
dup
B
BAB 2780
dup-nml
BAB 3011
dup
dup-nml
BAB 2920
dup
del
BAB 3050
del
dup
trp
Figure 2. Characterization of the Rearrangements with High-Density Arrays

(A) Array CGH with customized Agilent arrays. Copy number changes (indicated by black circles) in addition to the ones detected by clinical CMA were seen for
patients BAB3012, 3015, 3103, and 3032. The boxed regions were enlarged in the left bottom panels. For both BAB3012 and 3015, a small duplication was
detected proximal and next to the triplicated regions. For BAB3103, a 7.9 kb normal copy sequence was detected between the terminal deletion and the
duplicated segment. BAB3032 had a trp-dup rearrangement revealed by CMA. High-density array showed a duplication proximal to the triplication and a 18 kb
duplication within the triplicated segment.
(B) Array CGH using Nimblegen 4.2 M arrays. BAB3050 had a dup-trp rearrangement identified by CMA. High density array showed a 40 kb duplication within the
triplicated segment which was highlighted in a blue circle. This duplication is within a CNV in the database of genomic variants, and most likely is not part of the
complex event, but rather reflects relative signal intensities from a benign CNV.
See also Table S2.
BAB3105. A potential replication fork collapse at 9q21 could

explain the breakpoint clustering therein.
The parental aCGH analysis (Figure 3B) and chromosome
analyses (data not shown) independently confirmed that all the
highly complicated rearrangements on chromosome 9 were de
novo events in the patient. Further experiments were performed
to narrow the time frame when these catastrophic rearrangements occurred. High-density informative single-nucleotide
polymorphism (SNP) array analysis in the trio suggested that
the additional copies of the duplicated and triplicated genomic

segments were all (eight out of eight) derived from the paternal
allele that was not transmitted (Figures 3C and 3D and Figure S2G), indicating that the rearrangements arose in the father,
either in early development as a postzygotic event, or in germline
during spermatogenesis. PCR analysis with patient-breakpointjunction-specific primers and paternal lymphoblast DNA as
template failed to yield the junction product, suggesting that
the father, at least in the blood tissue, was unlikely to be
somatically mosaic for the rearrangements (data not shown). In

aggregate, these data suggest that the CGR in BAB3105 likely
occurred during spermatogenesis in the father.
Patient BAB3104 was referred for CMA due to hypotonia and
epilepsy. In total, there were seven copy number changes and
14 breakpoints over a 21.8 Mb interval (Figures 5A and 5B
and Figure S3A). Multiple FISH experiments demonstrated
extensive structural changes occurring in chromosome 22
(Figures S3BS3G). The proposed structure and derivation resulting in the CGR are illustrated in Figure 5C.
The rearrangements studied from two other patients also
presented localized but extensive chromosomal copy number
and structural changes. Patient 1, an 11-year-old girl with
severe developmental delay and intellectual disability, had a de
novo complex rearrangement including four copy number
changestwo deletions and two duplications between bands
1q32.2 and 1q43, spanning the distal half of the long arm of
chromosome 1 (Figure 6A and Figures S4A and S4B). All four
copy number changes were separated by normal copy number
segments in their original positions. Data from partial chromosome and metaphase FISH analyses suggested additional inversion and revealed that the two duplicated segments were
brought into proximity after rearrangement (Figure 6A and
Figures S4CS4G). Patient 2, a 3-year-old girl with an epilepsy
disorder who was referred due to abnormal chromosome findings, had two losses and five gains interspersed along the entire
long arm of chromosome 22 (Figure 6B and Figure S4H). Retrospective chromosome analysis (Figure S4I) and FISH results
(data not shown) showed that the rearranged chromosome 22
was nearly metacentric in appearance without stalk and satellites and that the duplicated segments on the long arm were
translocated into the short arm.
The common gestalt of a region-focused complex rearrangement and breakpoint clustering, with multiple CNVs and other
structural changes that is observed in these subjects bears
a striking resemblance with the phenomenon of chromosome
shattering (chromothripsis) recently reported in 2%3% of all
cancers (Stephens et al., 2011).
Low-Copy Repeats Are Enriched in the Breakpoints
with Highly Complicated Rearrangements
Genome architecture incites genomic instability. Low-copy
repeats (LCR) represent paralogous segments of the human
genome usually greater than 10 kb in length with >97%
sequence identity (Stankiewicz and Lupski, 2002). LCR have
been shown to mediate recurrent rearrangements by NAHR
and stimulate nonrecurrent rearrangements (Carvalho et al.,
2009; Stankiewicz et al., 2003). To examine whether the LCRs
might potentially incite instability of a region during CGR formation, we analyzed genomic DNA sequences of breakpoint intervals, as delineated by high resolution arrays, for the presence
of segmental duplications (repeat sequences of >1 kb and
>90% identity) (Bailey et al., 2002) or other short repeats. Of
the 116 segments encompassing 658,119 bp of reference
genome sequence that were analyzed, six segmental duplications (91,782 bp) and seven short repeats from the self-chain
tracks in the UCSC genome browser, i.e., representing both
direct and inverted repeats that can be shorter than the 1 kb,
were identified in 13 breakpoints of seven cases (Figure S5

and Table S2). The percentage of segmental duplication in these
breakpoint regions (13.95%) is significantly higher than the
genome-wide average (5.53%) (Chi square test, p < 0.0001).
Most of the repeat-associated breakpoints (eight out of 13)
were found in the highly complicated rearrangements with four
or more copy number changes. Noteworthy, paralogous LCR
pairs, i.e., homologous sequences that were probably derived
from a common ancestral segment (Bailey et al., 2002), were
identified flanking the duplication proximal and adjacent to a triplication in BAB3032 and were inversely orientated. In addition,
two inversely orientated LCRs were found flanking a normal
copy sequence between two duplicated regions in BAB3105.
Breakpoint Junction Sequencing Analyses
in Eight CGRs
Long range PCR and DNA sequencing of the breakpoint junctions
were attempted for all novel joints with the hypothesis that examination of rearrangement products could allow inferences regarding potential mechanisms for their formation. Besides the
six junctions sequenced from BAB3105, 14 breakpoint sequences were obtained from eight other patients with CGRs,
many of which revealed the presence of microhomologies, inversions (the fusion of two reference sequences belonging to opposite DNA strands), short DNA sequence inserts, or additional
small copy number gains at the breakpoint junctions.
CGRs in four individuals (BAB2760, 2780, 2783, and 2785) had
an interstitial nml-dup-nml-dup-nml pattern. Breakpoint sequencing revealed that the CGRs in these four patients share
the same configuration in their final structures: one copy of the
duplicated segment is inserted in an inverted orientation in
between the two copies of the other duplicated segments (Figure 4B and Figures S6A and S6B) (Carvalho et al., 2011a). For
the CGRs in individuals BAB2760 and 2783, sequences of one
breakpoint showed that the rearrangement occurred in two paralogous Alu repeats in an inverted orientation (Figure S6A). The
lengths of perfect homology at the breakpoints in these two Alu
elements are below the length of minimum efficient processing
segment required for homologous recombination, making
NAHR unlikely to be the possible mechanism. The sequence
homology between Alu elements can serve as microhomology
and may facilitate a template switch during a replicative process.
Intriguingly, additional complexities were identified in the breakpoints in two CGRs. In BAB2780, two segments flanking the
proximal duplicated interval, 90 bp and 406 bp in size, were found
adjacent to each other in one breakpoint. In addition, an 8 bp
novel sequence was inserted within the 406 bp segment (Figure 4B). In BAB2785, an 89 bp segment is inserted within one
of the breakpoints, resulting in triplication of this segment
(Figure S6B). Next and proximal to this segment is a 13 bp fragment, which could be a sequence synthesized during the repair
of this breakpoint, or a result of complex joining of two shorter
sequences that could be copied from multiple genomic intervals
located within the rearranged chromosome region. These additional small-scale complexities may reflect that a low-processivity polymerase, perhaps allowing iterative template switches,
could have been used to initiate the replicative repair when the
replication fork collapsed (Hastings et al., 2009a).
Figure 3. Highly Complex Rearrangements in BAB3105

(A) Schematic drawing of copy number and rearrangement structure profiles on chromosome 9q summarized from aCGH, FISH and G-banded chromosome
analysis results. Bars (red for duplication and blue for triplication) above the chromosome ideogram and piled rectangular boxes below the ideogram represent
Individual BAB2778 had a duplication followed by a normal

copy sequence, then a terminal duplication in a ring chromosome 6. DNA sequence was obtained for one breakpoint
showing that one copy of the two duplicated segments was
adjacent to another consistent with an inverted segment
(Figure S6C).
Two types of structures were revealed in three patients with
triplications. In BAB3032 and 3015, the triplicated segments
were inserted in an inverted orientation between the two copies
of the duplicated segmentsi.e., dup-trp/inv-dup (Figure S6D)
(Carvalho et al., 2011b)whereas in BAB3050, the triplicated
segment was apparently in the same orientation as the duplicated segments (Figure S6E). One pair of inverted LCRs flanks
the first duplicated segment in the CGR in BAB3032.
Although the cases with breakpoint sequences presented
above do not have as many copy number changes as the four
chromothripsis-like CGR cases, most of them present evidence
of breakpoint clustering, i.e., dispersed genomic fragments
involving breakpoints are jumbled together forming a junction
that apparently contains a medley of multiple genomic fragments, which is also a key feature of chromothripsis (Stephens
et al., 2011). This may potentially be a vestige of replicative
repair. It also suggests that the complex joining is a result of
one single catastrophic event, as opposed to progressive rearrangement, in which the expectation would be that the individual
duplications or deletions arise independently at their original loci
as simple deletions or tandem duplications.
with more severe phenotypes (Table 1). Indeed, of the ten cases
with combined copy number changes larger than 2.5 Mb, six
rearrangements were de novo events and the other four had
either no or incomplete parental studies. These de novo cases
were associated with severe DD/intellectual disability, epilepsy,
and dysmorphic features. In contrast, in the seven cases with
relatively smaller copy number changes (0.4 Mb2.2 Mb), four
rearrangements were inherited and the other three had incomplete parental information. Interestingly, in each of these
CGRs, only copy number gains were present, and most of these
patients showed a comparatively milder phenotype such as
moderate DD. It is unclear at the present time whether these relatively smaller complex rearrangements are clinically significant.
The fact that four rearrangements were maternally inherited,
but none were paternally inherited is consistent with the previous
observations in complex chromosomal rearrangements that
familial transmission is mainly observed through female carriers
(Batista et al., 1994).
As noted for cancer chromothripsis, such complex genomic
rearrangement can have multigenic consequences wherein
many genes have an altered dosage or copy number, others
are potentially disrupted, and there may be novel gene fusion
formed at the multiple breakpoints. Perhaps the observed
common clinical phenotype of neurodevelopmental delay
reflects multiple potential gene disruptions or dosage alterations
by CGR and the large number of genes in the human genome
that contribute to neurological function.
Clinical Phenotypes of Patients with Complex

Rearrangements
The complex rearrangements were identified among patients
referred by clinical geneticists for genomic studies using chromosomal microarray analysis, i.e., CMA. Most were referred
because of developmental delay (DD) in attaining milestones
in motor and language development (ten out of 12). Clinical information from 14 patients is summarized in Table S1. The combination of intellectual disability, failure to thrive, behavioral
problems, dysmorphic features and congenital anomalies was
present in more than four out of 12 subjects. However, DD was
observed in 12 out of 12 patients (Table S1).
We hypothesized that the rearrangements with large combined copy number changes may be de novo and associated
DISCUSSION
Do Somatic Chromothripsis Events Occur
in a Mechanistically Similar Manner
to Constitutional CGR?
Recently, the phenomenon of chromosome shattering termed
chromothripsis, i.e., a massively complex genomic rearrangement which occurs in a single catastrophic event involving local
apparent shattering of a chromosome and subsequent reassembly proposed to be potentially mediated by nonhomologous end
joining, was described in 2%3% of all cancers analyzed (Stephens et al., 2011). This phenomenon was also reported in the
germline (Borg et al., 2005; Kloosterman et al., 2011). The attractively simple idea that chromothripsis stems from the shattering
the copy number states for each genomic segment. The sizes of the rearrangements and normal copy number intervals are listed in megabases. The arrows
indicate translocations (upward facing) and insertions (downward facing) as indicated by FISH analysis with locus-specific BAC clones. All these additional
segments of the 5.1 Mb duplication, 2.1 Mb duplication, 0.78 Mb duplication, 1.4 Mb duplication, and 5.5 Mb triplication were translocated close to the pericentric
region, proximal to the original location of the 2.1 Mb duplication region. The orders of these additional segments are deduced from FISH results. It is unknown
whether the additional 1.4 Mb duplicated segment is located proximal to or between the two copies of the 5.5 Mb triplicate segments. Two representative FISH
images are shown. The FISH image on the left shows that the duplicated (RP11-35N6) and triplicated (RP11-18B16) segments in 9q31-q33 were translocated
close to the 9q21 region. The FISH image on the right shows that the two additional copies of the 5.5 Mb triplication marked by an arrow are in an inverted
orientation with each other. RP11-203L12 is mapped at distal end and RP11-104M22 is mapped at the proximal end of the triplicated segment. Note that the
additional complexities of the rearranged chromosome are represented as curvilinear connections of breakpoint regions. The purple curves indicate junctions
between two segments with copy number gain revealed by breakpoint sequencing.
(B) Nimblegen 4.2M aCGH plots for the trio. The displayed regions correspond to the regions in (A). Parental aCGH analyses demonstrated that all the copy
number changes were de novo. Triplication was indicated by blue and duplication was indicated by red. The pedigrees are on the bottom.
(C) Representative SNP transmission patterns that suggest a paternal interchromosome origin of the rearrangements.
(D) Zoomed-in view of the SNP array data for the 5.5 Mb triplication and the 0.78 Mb duplication in BAB3105. Five different genotypes are observed across the
entire triplicated region.
See also Figure S2.
Figure 4. Representative Breakpoint Sequences

In each panel, the top graph shows schematic representation of the aCGH result. Regions of copy number gains are highlighted in red. The sizes of different
segments are not in proportion to the actual rearrangement size. Below the array result is the schematic diagram illustrating the structure of a reference genomic
region corresponding to the region in the array result. Below the reference structure is the proposed rearranged structure in the patient, which is surmised from
of chromosomes or regions of chromosomes, and their re-ligation with scant regard for their site of origin, is based on data
obtained by next generation paired-end sequencing. This technique provides excellent data on the relative positions of
sequences, but cannot directly reveal copy number dosages.
Whereas next generation sequencing can infer copy number
dosage based on depth of read coverage, in the analysis the
raw read length data are first processed through matching to a
reference haploid human genome build; a filtering process notoriously challenged by low copy repeats, repetitive sequences,
and other sequence complexities.
By use of a combination of molecular techniques, we have obtained descriptions of chromosomes that have experienced a
chromothripsis-like process that includes full detail of copy
number, position and orientation in noncancerous patients. This
has revealed that, in addition to inversion and translocation, there
is extensive duplication and triplication of sequence. We also
show that some novel junctions have microhomology and smallscale complexity in the form of apparent templated insertion of
fragments of nearby sequence at the junctions. These features
are characteristic of events that we have previously attributed to
postulated replicative processes of chromosomal structural
changes, long-distance template-switching by FoSTeS/MMBIR;
the latter apparent insertional complexity near the junctions can
be attributed to a synthesis product resultant from a new lowprocessivity fork as proposed for the MMBIR model (Hastings
et al., 2009a). Interestingly, DNA polymerase(s) involved in
break-induced replication (BIR) may also have poor fidelity
(Deem et al., 2011; Hicks et al., 2010).
Although a duplication might potentially result from chromosome shattering and re-ligation, either because the chromosomes have replicated, or by involvement of both homologs,
explanation of triplication in this way would become complicated. Furthermore, such a shattering and re-ligation mechanism predicts the presence of a deletion reciprocal to any
duplication, and this is not observed. A more parsimonious
explanation for the origin of the extra copies is that they were
formed by replication. Although re-replication, the inappropriate
firing of replication origins, might explain some cases of overreplication (Doksani et al., 2009; Green et al., 2010), BIR could
also lead to over-replication, either by use of ectopic homology
or non-homologous processes using microhomology to anneal
single-strands that act as primers for DNA replication. Thus,
MMBIR provides an explanation for duplication and triplication
as well as deletion, inversion, and translocation. In addition,
MMBIR can explain both the observed microhomology at
selected breakpoints and the insertion of short segments flanked
by microhomology around the junctions. Can MMBIR also
account for the multiple breaks over extended regions described

in chromothripsis?
We have previously pointed out that replication forks formed
by BIR are believed to differ from those formed at an origin in
that they might involve a Holliday junction (Hastings et al.,
2009a). The Holliday junction might be resolved at the time of
formation of the replication fork, or it might follow the fork. If a
replication fork stalls, a following Holliday junction might process
through the fork leading to fork collapse. This was offered as the
origin of discontinuities on the scale of hundreds of kilobases or
megabases seen in many duplications. Another way in which
BIR differs from origin-dependent replication is that it appears,
at least in some cases, to replicate to the telomere, thereby
ignoring intervening replicon signals. This was shown in yeast
in the break copy duplication model, wherein a foreign chromosome fragment can invade into an endogenous target
chromosome and initiate a unidirectional replication fork that
proceeds to duplicate all sequences distal to the site of initiation
(Morrow et al., 1997). Smith and Symington later showed that
BIR can go all the way from near the centromere to the end of
the chromosome arm in yeast (Smith et al., 2007). Multiple
template-switches using homologies or microhomologies is
also a signature of BIR (Hicks et al., 2010; Smith et al., 2007).
From these considerations, one may argue that, once a replication fork has collapsed, and been restarted by BIR, chromosomal structural changes of all kinds can form over distances
of tens of megabases of sequence on the same chromosome.
In this context, it is also of interest to note both the observed
long arm and the distal chromosome preference of the constitutional CGR events we described here. Viewed from this
mechanistic perspective, chromothripsis is less likely to represent a blowing apart of a chromosome and putting the pieces
of the puzzle together that is peculiar to a cancer cell (Stephens
et al., 2011), but rather may reflect an inherent cellular DNA
replicative/repair process for maintaining genome stability. Perhaps the phenomenon termed chromothripsis might be better
referred to as chromoanasynthesis (chromosome reconstitution or chromosome reassortment).
Time Frame of Formation of the Chromosome
Catastrophe CGR Event
As indicated by Stephens et al., chromothripsis occurs in a single
catastrophic event, rather than in a step-wise cumulative way
(Stephens et al., 2011). Furthermore, recent studies of multiple
myeloma identified large numbers of genomic rearrangements
with the hallmarks of chromothripsis in 1.3% of samples (Magrangeas et al., 2011). This catastrophic event confers a poor
outcome as indicated by shortened time to relapse and death
breakpoint sequencing data. Specific breakpoint sequences and alignments to the reference sequences are shown below. Reference sequences are named as
PLUS, MINUS, PLU1 (plus1), and MIN1 (minus1), etc., in order to indicate the orientation of DNA strand. A transition between the plus and minus strands within
a breakpoint indicates that such a rearrangement causes an inversion. Microhomologies found at the breakpoints are boxed in red, whereas short sequences
insertions at the breakpoints with no microhomology are boxed in black.
(A) BAB3105. Six breakpoint junctions were obtained, providing information about their relative positions in the rearranged chromosome. In breakpoints #3#6,
novel sequence insertions, ranging from 541542 bp, were identified at the fusion point.
(B) BAB2870. In breakpoint junction #2, a 406 bp and a 90 bp segments copied from nearby genomic sequences are inserted. There is one 8 bp novel sequence
inserted within the 90 bp segment (green).
See also Figure S6.
Figure 5. Highly Complex Rearrangements in BAB3104

(A) Copy number profiles in BAB3104 revealed by aCGH. Deletions are shown in green and duplications in red.
(B) Nimblegen 2.1M aCGH plot in BAB3104. Note that a 23 kb copy number normal region is indicated by an arrow.
(C) The rearrangement structure in BAB3104 proposed based on multiple FISH results. To the left is an ideogram of a normal chromosome 22. Shown in
middle is the proposed replicative process that produced the genomic imbalances. The segments involving rearrangement are indicated by colored brackets.
The figure to the right shows the derivative chromosome 22 with the segments involved in the final order based on FISH analyses. The mapping positions of
FISH BAC RP11- probes used are indicated. The orientation of these segments is not certain.
See also Figure S3.
Figure 6. Highly Complex Rearrangements in Patient 1 and Patient 2

Copy number profiles and schematic drawings of the genomic rearrangements are shown on chromosome 1q for patient 1 (A) and chromosome 22q for patient 2
(B). Structural changes are revealed by FISH or G-banded chromosome analyses. Insertional translocations are indicated by arrows. Inversion in patient 1 is
indicated by a curly bracket. See also Figure S4 for aCGH, FISH, and chromosome analyses data.
within two years. Several lines of evidence support the contention that in the constitutional CGR cases in our cohort, in
particular the four chromothripsis-like cases, the majority of
rearranged genomic intervals also arose as one singular event.
First, the multiple rearrangements are localized to a focused
region on one homolog of a chromosome, most frequently in
the long arms. In a mutation accumulation model, the rearrangements would be expected to disperse randomly over the
genome. Second, there is no evidence of differential level of
mosaicism among individual rearrangements, which would be
the expectation when new rearrangements accumulate as cells
proliferate. Third, clustering and juxtaposition of breakpoints
indicate that these breakpoints were interrelated when they
were formed. Notably, inversion of genomic segments gives
proximity of what appears to be distant breakpoints (Branzei
and Foiani, 2010; Carvalho et al., 2011b; Futcher, 1986).
If replicative mechanisms are used, CGR can occur during
DNA replication in gametogenesis and, in some cases, as postzygotic events (Zhang et al., 2009b). Here, we provide evidence
for examples of both types (BAB3105 and 3012). Given the
resemblance of key features between constitutional CGR and
cancer chromothripsis, we propose that such chromosome
catastrophe events can occur by essentially similar mechanisms
in a variety of stages in an organisms life cycle.
In summary, genomic studies of rearrangements associated
with cancer and genomic disorders reveal unanticipated
complexities. Such observations have important ramifications,

such as inducing multigenic changes in a singular event, with
implications for both cancer genesis and species evolution.
Elucidating the mechanism underlying such apparent one-off
events is essential to understanding mutational processes. Our
findings in CGR suggest replicative processes may be operative
in both genomic disorders and cancers.
EXPERIMENTAL PROCEDURES
Patients
Patients with complex rearrangements were identified by CMA clinical testing
at Baylor College of Medicine (BCM) MGL. Patient samples chosen for further
higher-resolution genome analysis studies had evidence for either a triplication
or at least two separate copy number changes in one chromosome, with no
additional clinically significant genomic findings in other chromosomes. In
addition, rearrangements involving a terminal deletion in one arm and a
terminal duplication in the other arm of the same chromosome were not
included because they tend to be products of unbalanced translocation. In
all, 17 cases were included in this study out of over 23,000 CMA cases studied
in the MGL. Clinical information was collected after procuring informed consents approved by the Institutional Review Board for Human Subject Research
at BCM.
CMA with Clinical BCM Arrays
Clinical CMA was performed on custom designed Agilent arrays as described
(Boone et al., 2010; Cheung et al., 2005; Lu et al., 2007; Ou et al., 2008).
Patients were identified either by the BAC emulation oligo-based arrays with
at least one BAC clone for one chromosome band at a resolution of 550 band
level, or by the oligo-based arrays with one probe for every 30 kb interval.
High-Density aCGH
To characterize further the complex rearrangements identified by CMA, two
customized Agilent arrays were designed including an 8X 60K array format
for investigation of the rearrangements in seven patients, BAB2760, 2778,
2785, 3012, 3015, 3047, and 3050, and a 4X 180K array format for four patients
BAB2780, 2783, 3032, and 3103. The designed probe density was two to four
oligonucleotides per kilobase for intervals with copy number changes and
breakpoint regions as assessed by transitions between two separate copy
number states. The high-density arrays also interrogate the flanking genomic
regions of up to 5 Mb in size with probe density of one to three oligos per
kilobase.
Patients BAB2780, 2920, 3011, 3050, 3104, 3105, and patients 1 and 2 were
analyzed by Nimblegen 2.1M or 4.2M arrays. Patient BAB3104 was also
analyzed by a whole-genome 244K array (Agilent Technologies); assays
were performed according to the manufacturers instructions.
Fluorescence In Situ Hybridization
After identification of a complex rearrangement by clinical CMA, confirmatory
FISH analyses were performed on Phytohemagglutinin-stimulated cultured
blood cells with standard procedures (Cheung et al., 2005). For the two-color
FISH analyses, the test probes were directly labeled by nick-translation with
rhodamine (red), while the control probes were directly labeled with FITC
(green). To confirm a deletion detected by CMA, ten metaphase cells were
examined using the centromere probe as a control. To confirm a copy number
gain detected by CMA, 50 interphase nuclei were examined. A region was
considered to be duplicated when two red signals or triplicated when three
signals were observed in more than 70% of the cells examined. Metaphase cells
were also examined to determine chromosome/genome position and investigate whether the duplicated segments were translocated to a different location.
SNP Array Analysis
SNP array analysis was performed on Illumina Infinium HD assay platform with
HumanOmni1-Quad BeadChip (Illumina) DNA whole-genome amplification,
fragmentation, hybridization, enzymatic single base extension, slides staining,
and washing were performed according to the manufacturers instructions
(Illumina). The Illumina iScan System was used for image registration, image
extraction, and data output. The GenomeStudio software (Illumina) was used
for genotyping data normalization, genotype calling, clustering, data intensity
analysis, and calculation of Log R ratio and B-allele frequency. Illumina CNV
partition statistical algorithm was used for copy number variation and copy
number neutral absence of heterozygosity analyses. The reference model file
provided by Illumina was used as a reference for data analysis. Parent of origin
was determined by comparison of probands and parental genotypes and
B-allele frequencies for SNPs in the rearranged regions on chromosome 9.
Breakpoint Analysis
Genomic coordinates for potential breakpoints were estimated from aCGH
interrogating oligonucleotides revealing gain/no gain or loss/no loss copy
number transitions. Rearrangement breakpoints located in subtelomeric or
pericentromeric regions were not included in this analysis. The presence of
LCRs was revealed by examination of the segmental_dups track in the UCSC
genome browser (assembly hg19), while the presence of both direct and inverted short repeats was revealed by examination of the self_chain track.
The breakpoint junctions were amplified by long-range PCR with the
TAKARA LA Taq kit (TAKARA Bio) as previously described (Zhang et al.,
2009b). PCR products were sequenced by the Sanger dideoxy method and
DNA sequences were compared to the human genome reference assembly
(hg19).
ACKNOWLEDGMENTS
We thank the patients and their families for participation in this research, the
Molecular Cytogenetics Laboratories in the MGL at BCM (https://www.bcm.
edu/geneticlabs/) for providing technical support, Dr. Katarzyna Derwinska
for helpful discussion, and Roche NimbleGen for supplying NimbleGen
aCGH materials. This work was supported in part by National Institute of
Neurological Disorders and Stroke, National Institutes of Health grant R01
NS058529 to J.R.L., National Institutes of General Medical Science grant
R01 GM064022 to P.J.H., and the BCM IDDRC P30HD024064 funded from
the Eunice Kennedy Shriver National Institute of Child Health and Human
Development. A.E. is supported by DK081735-02 and S.C.S.N. is supported
by a fellowship grants by the LCRC from the Osteogenesis Imperfecta Foundation and the National Urea Cycle Disorders Foundation. J.R.L. is a consultant
for Athena Diagnostics, has stock ownership in 23andMe and Ion Torrent
Systems and is a coinventor on multiple United States and European patents
for DNA diagnostics. The Department of Molecular and Human Genetics
derives revenue from clinical testing by high-resolution human genome
analysis.
Received: May 17, 2011
Revised: June 6, 2011
Accepted: July 25, 2011
Published: September 15, 2011
REFERENCES
Bailey, J.A., Gu, Z., Clark, R.A., Reinert, K., Samonte, R.V., Schwartz, S.,
Adams, M.D., Myers, E.W., Li, P.W., and Eichler, E.E. (2002). Recent
segmental duplications in the human genome. Science 297, 10031007.
Batista, D.A., Pai, G.S., and Stetten, G. (1994). Molecular analysis of a complex
chromosomal rearrangement and a review of familial cases. Am. J. Med.
Genet. 53, 255263.
Bi, W., Sapir, T., Shchelochkov, O.A., Zhang, F., Withers, M.A., Hunter, J.V.,
Levy, T., Shinder, V., Peiffer, D.A., Gunderson, K.L., et al. (2009). Increased
LIS1 expression affects human and mouse brain development. Nat. Genet.
41, 168177.
Boone, P.M., Bacino, C.A., Shaw, C.A., Eng, P.A., Hixson, P.M., Pursley, A.N.,
Kang, S.H., Yang, Y., Wiszniewska, J., Nowakowska, B.A., et al. (2010).
Detection of clinically relevant exonic copy-number changes by array CGH.
Hum. Mutat. 31, 13261342.
Borg, K., Stankiewicz, P., Bocian, E., Kruczek, A., Obersztyn, E., Lupski, J.R.,
and Mazurczak, T. (2005). Molecular analysis of a constitutional complex
genome rearrangement with 11 breakpoints involving chromosomes 3, 11,
12, and 21 and a approximately 0.5-Mb submicroscopic deletion in a patient
with mild mental retardation. Hum. Genet. 118, 267275.
Branzei, D., and Foiani, M. (2010). Leaping forks at inverted repeats. Genes
Dev. 24, 59.
Carvalho, C.M., Zhang, F., Liu, P., Patel, A., Sahoo, T., Bacino, C.A., Shaw, C.,
Peacock, S., Pursley, A., Tavyev, Y.J., et al. (2009). Complex rearrangements
in patients with duplications of MECP2 can occur by fork stalling and template
switching. Hum. Mol. Genet. 18, 21882203.
Carvalho, C., Bartnik, M., Pehlivan, D., Fang, P., Shen, J., and Lupski, J.
(2011a). Evidence for disease penetrance relating to CNV size: PelizaeusMerzbacher disease and manifesting carriers with a familial 11 Mb duplication
at Xq22. Clin. Genet. Published online May 30, 2011. 10.1111/j.1399-0004.
2011.01716.x.
SUPPLEMENTAL INFORMATION
Carvalho, C.M., Ramocki, M.B., Pehlivan, D., Franco, L.M., GonzagaJauregui, C., Fang, P., McCall, A., Pivnick, E.K., Hines-Dowell, S., Seaver,
L., et al. (2011b). Inverted genomic segments and complex triplication
rearrangements are mediated by inverted repeats in the human genome.
Nat. Genet., in press.
Supplemental Information includes six figures and two tables and can be
found with this article online at doi:10.1016/j.cell.2011.07.042.
Chen, J.M., Cooper, D.N., Ferec, C., Kehrer-Sawatzki, H., and Patrinos, G.P.
(2010). Genomic rearrangements in inherited disease and cancer. Semin.
Cancer Biol. 20, 222233.
Cheung, S.W., Shaw, C.A., Yu, W., Li, J., Ou, Z., Patel, A., Yatsenko, S.A.,
Cooper, M.L., Furman, P., Stankiewicz, P., et al. (2005). Development and
validation of a CGH microarray for clinical cytogenetic diagnosis. Genet.
Med. 7, 422432.
Deem, A., Keszthelyi, A., Blackgrove, T., Vayl, A., Coffey, B., Mathur, R.,
Chabes, A., and Malkova, A. (2011). Break-induced replication is highly
inaccurate. PLoS Biol. 9, e1000594.
Doksani, Y., Bermejo, R., Fiorani, S., Haber, J.E., and Foiani, M. (2009).
Replicon dynamics, dormant origin firing, and terminal fork integrity after
double-strand break formation. Cell 137, 247258.
Futcher, A.B. (1986). Copy number amplification of the 2 micron circle plasmid
of Saccharomyces cerevisiae. J. Theor. Biol. 119, 197204.
Gajecka, M., Saitta, S.C., Gentles, A.J., Campbell, L., Ciprero, K., Geiger, E.,
Catherwood, A., Rosenfeld, J.A., Shaikh, T., and Shaffer, L.G. (2010).
Recurrent interstitial 1p36 deletions: Evidence for germline mosaicism and
complex rearrangement breakpoints. Am. J. Med. Genet. A. 152A, 30743083.
Green, B.M., Finn, K.J., and Li, J.J. (2010). Loss of DNA replication control is
a potent inducer of gene amplification. Science 329, 943946.
Hastings, P.J., Ira, G., and Lupski, J.R. (2009a). A microhomology-mediated
break-induced replication model for the origin of human copy number variation. PLoS Genet. 5, e1000327.
Hastings, P.J., Lupski, J.R., Rosenberg, S.M., and Ira, G. (2009b). Mechanisms of change in gene copy number. Nat. Rev. Genet. 10, 551564.
Hicks, W.M., Kim, M., and Haber, J.E. (2010). Increased mutagenesis and
unique mutation signature associated with mitotic gene conversion. Science
329, 8285.
Kloosterman, W.P., Guryev, V., van Roosmalen, M., Duran, K.J., de Bruijn, E.,
Bakker, S.C., Letteboer, T., van Nesselrooij, B., Hochstenbach, R., Poot, M.,
and Cuppen, E. (2011). Chromothripsis as a mechanism driving complex
de novo structural rearrangements in the germline. Hum. Mol. Genet. 20,
19161924.
Lee, J.A., Carvalho, C.M., and Lupski, J.R. (2007). A DNA replication mechanism for generating nonrecurrent rearrangements associated with genomic
disorders. Cell 131, 12351247.
Liu, P., Erez, A., Nagamani, S.C., Bi, W., Carvalho, C.M., Simmons, A.D.,
Wiszniewska, J., Fang, P., Eng, P.A., Cooper, M.L., et al. (2011). Copy number
gain at Xp22.31 includes complex duplication rearrangements and recurrent
triplications. Hum. Mol. Genet. 20, 19751988.
Lu, X., Shaw, C.A., Patel, A., Li, J., Cooper, M.L., Wells, W.R., Sullivan, C.M.,
Sahoo, T., Yatsenko, S.A., Bacino, C.A., et al. (2007). Clinical implementation
of chromosomal microarray analysis: summary of 2513 postnatal cases. PLoS
ONE 2, e327.
Lupski, J.R. (2010). New mutations and intellectual function. Nat. Genet. 42,
10361038.
Magrangeas, F., Avet-Loiseau, H., Munshi, N.C., and Minvielle, S. (2011).

Chromothripsis identifies a rare and aggressive entity among newly diagnosed
multiple myeloma patients. Blood 118, 675678.
Morrow, D.M., Connelly, C., and Hieter, P. (1997). Break copy duplication:
a model for chromosome fragment formation in Saccharomyces cerevisiae.
Genetics 147, 371382.
Ou, Z., Kang, S.H., Shaw, C.A., Carmack, C.E., White, L.D., Patel, A., Beaudet,
A.L., Cheung, S.W., and Chinault, A.C. (2008). Bacterial artificial chromosomeemulation oligonucleotide arrays for targeted clinical array-comparative
genomic hybridization analyses. Genet. Med. 10, 278289.
Payen, C., Koszul, R., Dujon, B., and Fischer, G. (2008). Segmental duplications arise from Pol32-dependent repair of broken forks through two alternative replication-based mechanisms. PLoS Genet. 4, e1000175.
Slack, A., Thornton, P.C., Magner, D.B., Rosenberg, S.M., and Hastings, P.J.
(2006). On the mechanism of gene amplification induced under stress in
Escherichia coli. PLoS Genet. 2, e48.
Smith, C.E., Llorente, B., and Symington, L.S. (2007). Template switching
during break-induced replication. Nature 447, 102105.
Stankiewicz, P., and Lupski, J.R. (2002). Genome architecture, rearrangements and genomic disorders. Trends Genet. 18, 7482.
Stankiewicz, P., Shaw, C.J., Dapper, J.D., Wakui, K., Shaffer, L.G., Withers,
M., Elizondo, L., Park, S.S., and Lupski, J.R. (2003). Genome architecture
catalyzes nonrecurrent chromosomal rearrangements. Am. J. Hum. Genet.
72, 11011116.
Stephens, P.J., Greenman, C.D., Fu, B., Yang, F., Bignell, G.R., Mudie, L.J.,
Pleasance, E.D., Lau, K.W., Beare, D., Stebbings, L.A., et al. (2011). Massive
genomic rearrangement acquired in a single catastrophic event during cancer
development. Cell 144, 2740.
Yatsenko, S.A., Brundage, E.K., Roney, E.K., Cheung, S.W., Chinault, A.C.,
and Lupski, J.R. (2009). Molecular mechanisms for subtelomeric rearrangements associated with the 9q34.3 microdeletion syndrome. Hum. Mol. Genet.
18, 19241936.
Zhang, F., Carvalho, C.M., and Lupski, J.R. (2009a). Complex human chromosomal and genomic rearrangements. Trends Genet. 25, 298307.
Zhang, F., Khajavi, M., Connolly, A.M., Towne, C.F., Batish, S.D., and Lupski,
J.R. (2009b). The DNA replication FoSTeS/MMBIR mechanism can generate
genomic, genic and exonic complex rearrangements in humans. Nat. Genet.
41, 849853.
Zhang, F., Seeman, P., Liu, P., Weterman, M.A., Gonzaga-Jauregui, C.,
Towne, C.F., Batish, S.D., De Vriendt, E., De Jonghe, P., Rautenstrauss, B.,
et al. (2010). Mechanisms for nonrecurrent genomic rearrangements associated with CMT1A or HNPP: rare CNVs as a cause for missing heritability.
Am. J. Hum. Genet. 86, 892903.
BET Bromodomain Inhibition as a

Therapeutic Strategy to Target c-Myc
Jake E. Delmore,1,9 Ghayas C. Issa,1,9 Madeleine E. Lemieux,2 Peter B. Rahl,3 Junwei Shi,4 Hannah M. Jacobs,1
Efstathios Kastritis,1 Timothy Gilpatrick,1 Ronald M. Paranal,1 Jun Qi,1 Marta Chesi,5 Anna C. Schinzel,1
Michael R. McKeown,1 Timothy P. Heffernan,1 Christopher R. Vakoc,4 P. Leif Bergsagel,5 Irene M. Ghobrial,1,6
Paul G. Richardson,1,6 Richard A. Young,3,7 William C. Hahn,1,8 Kenneth C. Anderson,1,6 Andrew L. Kung,2
James E. Bradner,1,6,* and Constantine S. Mitsiades1,6,*
1Department
of Medical Oncology, Dana-Farber Cancer Institute, 450 Brookline Avenue, Boston, MA 02215, USA
of Pediatric Oncology, Dana-Farber Cancer Institute and Childrens Hospital Boston, 450 Brookline Avenue, Boston,
MA 02215, USA
3Whitehead Institute for Biomedical Research, 9 Cambridge Center, Cambridge, MA 02142, USA
4Cold Spring Harbor Laboratory, 1 Bungtown Road, Cold Spring Harbor, NY 11724, USA
5Comprehensive Cancer Center, Mayo Clinic Arizona, Scottsdale, AZ 85259, USA
6Department of Medicine, Harvard Medical School, 25 Shattuck Street, Boston, MA 02115, USA
7Department of Biology, Massachusetts Institute of Technology, Cambridge, MA 02142, USA
8Broad Institute of Harvard and MIT, 7 Cambridge Center, Cambridge, MA 02142, USA
*Correspondence: james_bradner@dfci.harvard.edu (J.E.B.), constantine_mitsiades@dfci.harvard.edu (C.S.M.)
DOI 10.1016/j.cell.2011.08.017
2Department
SUMMARY
MYC contributes to the pathogenesis of a majority of

human cancers, yet strategies to modulate the function of the c-Myc oncoprotein do not exist. Toward
this objective, we have targeted MYC transcription
by interfering with chromatin-dependent signal transduction to RNA polymerase, specifically by inhibiting
the acetyl-lysine recognition domains (bromodomains) of putative coactivator proteins implicated in
transcriptional initiation and elongation. Using a selective small-molecule bromodomain inhibitor, JQ1,
we identify BET bromodomain proteins as regulatory
factors for c-Myc. BET inhibition by JQ1 downregulates MYC transcription, followed by genome-wide
downregulation of Myc-dependent target genes. In
experimental models of multiple myeloma, a Mycdependent hematologic malignancy, JQ1 produces
a potent antiproliferative effect associated with cellcycle arrest and cellular senescence. Efficacy of
JQ1 in three murine models of multiple myeloma
establishes the therapeutic rationale for BET bromodomain inhibition in this disease and other malignancies characterized by pathologic activation of c-Myc.
INTRODUCTION
c-Myc is a master regulatory factor of cell proliferation (Dang
et al., 2009). In cancer, pathologic activation of c-Myc plays
a central role in disease pathogenesis by the coordinated upregulation of a transcriptional program influencing cell division,
metabolic adaptation, and survival (Dang, 2009; Kim et al.,
2008). Amplification of MYC is among the most common genetic

alterations observed in cancer genomes (Beroukhim et al.,
2010). Validation of c-Myc as a therapeutic target is supported
by numerous lines of experimental evidence. Murine models of
diverse malignancies have been devised by introducing genetic
constructs overexpressing MYC (Harris et al., 1988; Leder et al.,
1986; Stewart et al., 1984). In addition, conditional transgenic
models featuring tunable transcriptional suppression have
shown that even transient inactivation of MYC is capable of promoting tumor regression (Jain et al., 2002; Soucek et al., 1998;
Soucek et al., 2002). Elegant studies of systemic induction of
a dominant-negative MYC allele within an aggressive, KRASdependent murine model of lung adenocarcinoma have further
suggested the putative therapeutic benefit of c-Myc inhibition
(Fukazawa et al., 2010). Importantly, these studies establish
the feasibility of c-Myc inhibition within an acceptable therapeutic window of tolerability.
Nevertheless, a therapeutic approach to target c-Myc has remained elusive. The absence of a clear ligand-binding domain
establishes a formidable obstacle toward direct inhibition, which
is a challenging feature shared among many compelling transcriptional targets in cancer (Darnell, 2002). c-Myc functions as
a DNA-binding transcriptional activator upon heterodimerization
with another basic-helix-loop-helix leucine zipper (bHLH-LZ)
transcription factor, Max (Amati et al., 1993; Blackwood and
Eisenman, 1991). High-resolution structures of the complex fail
to identify a hydrophobic involution compatible with the positioning of an organic small molecule (Nair and Burley, 2003).
Therefore, we have targeted c-Myc transcriptional function by
another means, namely the disruption of chromatin-dependent
signal transduction (Schreiber and Bernstein, 2002). c-Myc transcription is associated locally and globally with increases in
histone lysine side-chain acetylation, a covalent modification of
chromatin that is regionally associated with transcriptional
activation (Frank et al., 2003; Vervoorts et al., 2003). Histone

acetylation templates the assembly of higher-ordered transcriptional complexes by recruiting proteins with one or more acetyllysine-binding modules or bromodomains (Dhalluin et al., 1999;
Haynes et al., 1992). Members of the bromodomain and extraterminal (BET) subfamily of human bromodomain proteins (BRD2,
BRD3, and BRD4) associate with acetylated chromatin and facilitate transcriptional activation by increasing the effective molarity of recruited transcriptional activators (Rahman et al.,
2011). Notably, BRD4 has been shown to mark select M/G1
genes in mitotic chromatin as transcriptional memory and direct
postmitotic transcription (Dey et al., 2009) via direct interaction
with the positive transcription elongation factor complex
b (P-TEFb) (Bisgrove et al., 2007). The discovery that c-Myc
regulates promoter-proximal pause release of Pol II, also
through the recruitment of P-TEFb (Rahl et al., 2010), established
a rationale for targeting BET bromodomains to inhibit c-Mycdependent transcription.
Recently, we reported the development and biochemical characterization of a potent, selective small-molecule inhibitor of BET
bromodomains, JQ1 (Figure 1A) (Filippakopoulos et al., 2010).
JQ1 is a thieno-triazolo-1,4-diazepine that displaces BET
bromodomains from chromatin by competitively binding to the
acetyl-lysine recognition pocket. In the present study, we
leverage the properties of JQ1 as a chemical probe (Frye,
2010) to interrogate the role of BET bromodomains in Mycdependent transcription and to explore the role of BET bromodomains as cancer dependencies.
Multiple myeloma (MM) represents an ideal model system for
these mechanistic and translational questions, given the known
role of MYC in disease pathophysiology. MM is an incurable
hematologic malignancy that is typified by the accumulation of
malignant plasma cells harboring diverse genetic lesions
(Chapman et al., 2011). Dysregulation of transcription factors
features prominently in the biology of MM, including NF-kB
(Keats et al., 2007), c-Maf (Hurt et al., 2004), XBP1 (Claudio
et al., 2002), HSF1 (Mitsiades et al., 2002), GR (Gomi et al.,
1990), IRF4 (Shaffer et al., 2008), Myb (Palumbo et al., 1989),
and, notably, c-Myc (Dean et al., 1983). Rearrangement or translocation of MYC are among the most common somatic events in
early- and late-stage MM (Shou et al., 2000), and transcriptional
profiling identifies Myc pathway activation in more than 60% of
patient-derived MM cells (Chng et al., 2011). Experimental
support for the central role of c-Myc in the pathogenesis of
MM is contributed by an informative, genetically engineered
murine model of MM. Lineage-specific and stochastic activation-induced deaminase (AID)-dependent activation of a conditional MYC transgene in the late stages of B cell differentiation
establishes genetically engineered mice with a plasma cell
malignancy that shares clinically relevant features of MM (Chesi
et al., 2008). Thus, MYC dysregulation represents a largely unifying molecular feature observed across the otherwise complex
genetic landscape of MM.
In this study, we report that c-Myc transcriptional function can
be modulated pharmacologically by BET bromodomain inhibition. Unexpectedly, we have discovered that MYC itself is transcriptionally regulated by BET bromodomains. Chromatin immunoprecipitation studies show that BRD4 is strongly enriched
at immunoglobulin heavy-chain (IgH) enhancers in MM cells

bearing IgH rearrangement at the MYC locus. BET inhibition
with JQ1 depletes enhancer-bound BRD4 and promptly inhibits
MYC transcription in a dose- and time-dependent manner. In
translational models of MM, JQ1 leads to depletion of the
c-Myc oncoprotein and selective downregulation of the coordinated c-Myc transcriptional program, prompting cell-cycle
arrest and cellular senescence. These results indicate that
targeting protein-protein interactions within the c-Myc transcriptional signaling network can modulate the function of c-Myc in
cancer.
RESULTS
BET Bromodomains as Therapeutic Targets in MM
We first evaluated the expression of BRD2, BRD3, BRD4, and
BRDT transcripts in MM by integrating publicly available compendia of gene expression data sets. Among asymptomatic
patients with premalignant disease (Zhan et al., 2007), we
observed increasing expression of BRD4 in monoclonal gammopathy of undetermined significance (MGUS) and smoldering
MM (SMM) compared to normal bone marrow plasma cells (Figure 1B). In a second, independent data set (Mattioli et al., 2005),
we observed significantly higher expression of BRD4 in plasma
cell leukemia (PCL) compared to MM or MGUS samples (Figure 1C). Thus, BRD4 expression correlates positively with
disease progression. BRD2 and BRD3 are also expressed in
MM, but expression does not clearly correlate with stage of
disease (data not shown). BRDT, a testis-specific bromodomain-containing protein, is not expressed in MM.
Analysis of copy number polymorphism (CNP) data collected
on 254 MM patients by the Multiple Myeloma Research Consortium (MMRC) revealed that the BRD4 locus is frequently amplified in MM patient samples (Figure 1D). The majority of patient
samples exhibit broad amplification of chromosome 19p, but
focal amplification at the BRD4 locus is observed (Figure S1
available online). Among 45 established MM cell lines, expression of BRD4 was pronounced and did not correlate with amplification status (Figure 1E).
Human MM cells are highly osteotropic in vivo, and interaction
with bone marrow stromal cells (BMSCs) induces proliferation
and contributes to drug resistance (McMillin et al., 2010). Analysis of BET bromodomain expression, as influenced by MM
cell binding to BMSCs (McMillin et al., 2010), revealed marked
upregulation of BRD4 in the INA-6 human MM cell line upon
interaction with HS5 stromal cells (Figure 1F), suggesting a plausible role for BRD4 function in MM cells within the bone marrow
microenvironment.
To explore the function of BET bromodomains in MM, we
examined the effect on proliferation of small hairpin RNAs
(shRNAs) targeting each of the four BET proteins in comparison
to shRNAs targeting 1011 kinases, phosphatases, and oncogenes in a lentivirally delivered, arrayed shRNA screen in INA-6
cells. As illustrated in Figures 1G and 1H, shRNA constructs
targeting each of the expressed BET bromodomains are identified as reducing INA-6 proliferation as shown by normalized
B scores (Malo et al., 2006). Together, these data establish
B
N O
O
O
Cl
N O
N
H
Cl
(+)-JQ1
C
BRD4
iBET
4
2
0
1.5
1.0
0.5
0.0
NPCs
BRD4
2.0
Expression (log2)
Expression (log2)
MGUS
SMM
MGUS
MM
PCL
E
Chr19
0.06
p11
q13.2
BRD4
q13.42
Multiple Myeloma Samples
Normalized Expression
BRD4
0.04
*
Neutral
* *
0.02
0.00
Deleted
BRDT
Amplified
* * * *
* * *
INA-6
BRD2
BRD4
BRD2
BRD4
BRD3
BRD2
-2
-4
226054_at
226052_at
BRD3
BRD4
-2
-4
BRD3
-6
-6
202102_s_at
B-Score (Mean)
BRD3
100
300
B-Score
BRD4 Expression
INA-6
INA-6 + HS5
200
H
BRD2
400
OCI-MY5
KHM-1B
KMS-12BM
OCI-MY7
KARPAS620
XG-1
FLAM-76
LP-1
ANBL-6
SACHI
L363
CAG
OPM-2
PE1
KMS-12PE
NCI-H929
MM.1.144
KMS-18
PE2
RPMI-8226
EJM
FR4
KMS-28PE
JK-6L
OCI-MY1
INA-6
H1112
KMS-34
KMS-26
OPM-1
KMS-28BM
JJN3
ARP-1
XG-7
KHM-11
SKMM-2
KMS-11
UTMC-2
JIM3
DELTA-47
MM-M1
KMM-1
U266
XG-2
SKMM-1
p13.1
1000 2000 3000 4000 5000

Order
Figure 1. Integrated Genomic Rationale for BET Bromodomains as Therapeutic Targets in MM

(A) Structures of the BET bromodomain inhibitors JQ1 and iBET.
(B and C) Expression levels (log2 transformed, median-centered values) for BRD4 transcripts were evaluated in oligonucleotide microarray data from normal
plasma cells (NPCs) from healthy donors, individuals with MGUS, or SMM patients (B, data set GSE5900; Zhan et al., 2007) and in plasma cells from MGUS, MM,
and PCL patients (C, data set GSE2113; Mattioli et al., 2005). Increased BRD4 expression is observed in SMM (or MGUS) compared to NPCs (B) and in PCL
compared to MM (C) (nonparametric Kruskal-Wallis one-way analysis of variance; p < 0.001 and p = 0.0123, respectively; Dunns Multiple Comparison post-hoc
tests; p < 0.05 in both cases). For each box plot, the whiskers represent minimum and maximum values, the lower and upper boundaries denote the 25th and 75th
percentile, respectively, and the horizontal line represents the median value for each group.
(D) Copy number analysis of the BRD4 locus at human chromosome 19p13.1 in primary samples from 254 MM patients. Chromosome 19p amplifications are
common in MM. See also Figure S1.
a rationale for the study of BET bromodomains, and BRD4 in

particular, as tumor dependencies in MM.
BET Inhibition with JQ1 Arrests c-Myc Transcriptional
Programs
To test the hypothesis that BET inhibition will specifically abrogate Myc-dependent transcription, we utilized global transcriptional profiling and unbiased gene set enrichment analysis
(GSEA). We first characterized the transcriptional consequences
of BET inhibition in three MM cell lines with genetically distinct
activating lesions at the MYC locus (KMS11, MM.1S, and
OPM1) (Dib et al., 2008). Unsupervised hierarchical clustering
segregated samples based on treatment assignment, suggesting a common transcriptional consequence in response to JQ1
(Figure 2A). Acute JQ1 treatment did not prompt global, nonspecific transcriptional silencing but instead produced significant
changes in a finite number of genes (88 down- and 25 upregulated genes by 2-fold or greater in all three MM lines).
To examine higher-order influences on biological networks
regulated by c-Myc, we evaluated four canonical transcriptional
signatures of MYC-dependent genes (Kim et al., 2006; Schlosser
et al., 2005; Schuhmacher et al., 2001; Zeller et al., 2003). All four
signatures were strongly correlated with downregulation of expression by JQ1 (Figure 2B). As a measure of the specificity of
this effect, an open-ended enrichment analysis was performed
on the entire set of transcription factor target gene signatures
available from the Molecular Signatures Database (MSigDB).
Gene sets defined by adjacency to Myc-binding motifs were in
almost all cases significantly enriched in JQ1-suppressed genes
(Figures 2C and 2D). In marked contrast, JQ1 treatment did not
exert significant suppression of gene sets for other transcription
factors linked to pathophysiology of MM, including NF-kB, AP-1,
STAT3, GR, and XBP-1 (Figure 2E and Figure S2). Notably, 27 of
the 28 significantly correlated gene sets are annotated as predicted targets of MYC or E2F (Figure 2C and Figure S2). Consistent with Myc-specific inhibition, biological modules associated
with Myc (e.g., ribosomal biogenesis and assembly and glycolysis) were also anticorrelated with JQ1 treatment (Figure 2E).
BET bromodomain inhibition by JQ1 confers a selective repression of transcriptional networks induced by c-Myc.
Regulation of MYC Transcription by BET Bromodomains
An unexpected finding was the observed, robust inhibition of
MYC expression following treatment with JQ1 (Figure 2A). As
MYC is commonly activated by upstream oncogenic signaling
pathways, we studied the consequence of JQ1 treatment on
the expression of 230 cancer-related genes in a human MM
cell line (MM.1S) using a multiplexed transcript detection assay
(Figure 3A). Excellent concordance was observed between replicate measurements of expressed genes (Figure S3A). Unsupervised hierarchical clustering segregated replicate data correctly
into early- and late-treatment time points. Surprisingly, we observed immediate, progressive, and profound downregulation
of MYC transcription itself, a unique finding among all transcripts
studied (p < 0.05).
Downregulation of MYC was further confirmed by RT-PCR
and immunoblot (Figures 3B and 3C). This effect was BET bromodomain specific, supported by the nearly comparable activity
of an analogous BET inhibitor subsequently published by Glaxo
SmithKline (iBET) (Figure 1A) (Nicodeme et al., 2010) and the lack
of activity of the inactive (-)- JQ1 enantiomer, which we previously characterized as structurally incapable of inhibiting BET
bromodomains (Filippakopoulos et al., 2010) (Figure 3C). Inhibition of MYC transcription by JQ1 was observed to be dose and
time dependent, with peak inhibition at submicromolar concentrations (Figures 3D and 3E). Rapid depletion of chromatinbound c-Myc was confirmed by nuclear ELISA transcription
factor-binding assays (Figure 3F). In contrast, NF-kB and AP-1
chromatin-binding assays failed to reveal any decrease in DNA
binding within 8 hr of JQ1 treatment (Figure 3G and Figure S4A).
To assess the breadth of these findings in MM, we expanded
gene expression studies to three MM cells with distinct lesions at
the MYC locus. MM.1S cells have a complex MYC rearrangement involving an IgH insertion at the breakpoint of a derivative
chromosome der3t(3;8); KMS-11 cells have both MYC duplication and inversion; and OPM1 cells feature a der(8)t(1;8)
(Dib et al., 2008). Among 230 genes studied, MYC was one of
only four genes downregulated by treatment with JQ1, along
with MYB, TYRO3, and TERT (Figure 3H and Figure S4B).
Immunoblotting analyses confirmed the JQ1 suppression of
c-Myc protein expression in a further expanded panel of Mycdependent MM cell lines (Figure 3I). Despite the intriguing potential effect on E2F transcriptional function and MYB gene expression, JQ1 did not influence E2F or MYB protein abundance
through 24 hr of drug exposure (Figures S4C and S4D). Together,
these data support the general observation that BET inhibition
specifically suppresses MYC transcription across MM cells
with different genetic lesions affecting the MYC locus and with
striking selectivity in comparison to other oncogenic transcriptions factors with established roles in MM pathophysiology.
BRD4 Binds IgH Enhancers, Regulating MYC Expression
and Function
Based on the integrated, functional genomic analysis of BET bromodomains in MM (Figure 1), we pursued further mechanistic
studies of BRD4. Silencing of BRD4 using directed shRNAs
(E) Expression levels of BRD4 (compared to BRDT) in human MM cell lines. Asterisks denote cell lines with amplification of the BRD4 locus (19p13.1).
(F) BRD4 expression (depicted on a linear scale for three different oligonucleotide microarray probes) in INA-6 MM cells cultured in vitro in the presence or
absence of HS-5 bone marrow stromal cells.
(G) Silencing of BET bromodomains impairs proliferation in MM cells. Results of an arrayed lentiviral screen using a diverse shRNA library in INA-6 MM cells are
presented in rank order of ascending B scores. The effect of shRNAs targeting BET bromodomains on INA-6 cell viability is highlighted by red circles and
annotated by gene. Gray dots represent results for non-BET bromodomain shRNAs.
(H) Silencing of BET bromodomain family members in MM cells. Viability of INA-6 MM cells exposed to shRNAs directed against BRD2, BRD3, and BRD4 are
reported as mean B scores ( SD of the two normalized replicates).
VEH JQ1
0.0
-0.2
-0.4
-0.6
-0.8 q < 0.0001
Enrichment Score
Enrichment Score
Schlosser and colleagues

0.0
-0.2
-0.4
-0.6
-0.8
q < 0.0001
0.0
-0.2
-0.4
-0.6
q < 0.0001
Kim and colleagues

0.0
-0.2
-0.4
-0.6
q < 0.0001
D
V$MYCMAX_01
MYC
E2F
Enrichment Score
1
0.8
FDR
SLC38A5
ACSL5
NME1
SORD
MAP1D
PTPN22
CCR1
KCNQ5
MYB
AMPD1
SLC7A2
MORC1
MYC
ADAT2
HBD
ALDH1B1
ZNF485
POLR3G
UNQ3104
NAV1
SRM
KCNA3
MGC29506
GTF3C6
ABCC4
RNF125
MTMR2
RRS1
KAT2A
SFXN4
GALNT14
SLC16A9
MAP4K1
CDC25A
MMACHC
FKBP11
RAI14
ABLIM1
TYRO3
MANEAL
XTP3TPA
BTN3A2
MTHFD1L
ACSM3
DERL3
BDH1
FADS1
TTC27
POLE2
SLC19A1
MAP2
CALCOCO1
CNTN5
PAG1
SYT11
YPEL1
NFKBIZ
ZSWIM6
TMEM2
HEXIM1
APOLD1
STAT2
SAT1
KLHL24
PNPLA8
JARID1B
BMPR2
SCN9A
SLC12A6
ZFYVE1
SESN3
C9ORF95
SERPINI1
KIAA0913
WDR47
BNIP3L
HIST2H2BE
HHLA3
C1ORF63
C13ORF31
KIAA0825
OR2B6
USP11
DOPEY2
RNF19B
DNM3
JHDM1D
YPEL5
ZFP36
ITFG3
LYST
SATB1
C1ORF26
DNAJC28
HIST2H4A
CLDN12
LMNA
SEPP1
LGALS1
SQSTM1
Dang and colleagues

Enrichment Score
Enrichment Score
Schuhmacher and colleagues
KMS11
MM.1S
OPM1
KMS11
MM.1S
OPM1
0.6
0.4
0.0
-0.2
-0.4
-0.6
q < 0.0001
0.2
0
-2 -1.5 -1 -0.5
0.5
1.5
NES
Gene Set
n
SCHUHMACHER_MYC_TARGETS_UP
67
DANG_MYC_TARGETS_UP
1 27
SCHLOSSER_MYC_TARGETS_AND_SERUM_RESPONSE_UP 47
KIM_MYC_AMPLIFICATION_TARGETS_UP
153
MORI_EMU_MYC_LYMPHOMA_BY_ONSET_TIME_UP
96
RIBOSOME_BIOGENESIS_AND_ASSEMBLY
14
YU_MYC_TARGETS_UP
37
MOOTHA_GLYCOLYSIS
21
NES
-2.61
-2.29
-2.29
-2.27
-1.953
-1.78
-1.65
-1.42
FDR q-val
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.0001
0.0006
0.004
0.045
V$MYCMAX_01
V$MYCMAX_02
V$NFKAPPAB65
V$AP1_Q4
V$STAT3_02
V$MYB_Q3
V$HSF1_01
V$GR_01
V$XBP1_01
-2.078
- 1 .7 2
-1.13
-1.11
-1.04
-0.92
-0.92
-0.88
-0.89
< 0.0001
0 .0 0 1 8
NS
NS
NS
NS
NS
NS
NS
192
200
190
214
111
176
197
155
107
Figure 2. Inhibition of Myc-Dependent Transcription by the JQ1 BET Bromodomain Inhibitor

(A) Heatmap representation of the top 50 down- and upregulated genes (p < 0.001) following JQ1 treatment in MM cell lines. Data are presented row normalized
(range from 3- to 3 standard deviations from median in expression). MYC (arrow) is downregulated by JQ1 treatment.
(B) GSEA of four Myc-dependent gene sets (Kim et al., 2006; Schlosser et al., 2005; Schuhmacher et al., 2001; Zeller et al., 2003) in transcriptional profiles of MM
cells treated (left) or untreated (right) with JQ1.
validated by RT-PCR analysis (Figure 4A) elicited a marked

decrease in MYC transcription (Figure 4B) accompanied by G1
cell-cycle arrest in JQ1-sensitive MM cells (OPM-1) (Figure 4B
and Figure S5A). We reasoned that early and sustained JQ1induced suppression of MYC transcription may be mechanistically explained by physical interaction of BRD4 with regulatory
elements influencing MYC expression. Indeed, avid binding of
BRD4 to established IgH enhancers was observed by chromatin
immunoprecipitation (ChIP) in MM.1S cells (Figure 4C and Figure S6), which harbor an IgH insertion proximal to the MYC transcriptional start site (TSS). BRD4 binding was not observed at
five characterized enhancer regions adjacent to the MYC gene
(Pomerantz et al., 2009a, 2009b). JQ1 treatment (500 nM) for
24 hr significantly depleted BRD4 binding to IgH enhancers
and the TSS, supporting direct regulation of MYC transcription
by BET bromodomains and a model whereby BRD4 acts as a
coactivator of MYC transcription potentially through long-range
interactions with distal enhancer elements. Forced overexpression of c-Myc in MM cells (OPM1) by retroviral infection rescues,
in part, the cell-cycle arrest observed with JQ1 treatment (Figures 4D and 4E), arguing that MYC downregulation by JQ1
contributes functionally to cell physiology in MM.
Therapeutic Implications of BET Inhibition in MM
Based on this mechanistic rationale, we evaluated the therapeutic opportunity of MYC transcriptional inhibition using established translational models of MM. Antiproliferative activity of
JQ1 was assessed using a panel of 25 MM cell lines or isogenic
derivative lines (Figure 5A). MM cell proliferation was uniformly
inhibited by JQ1 (Figure 5A), including several MM cell lines
selected for resistance to FDA-approved agents (dexamethasone-resistant MM.1R and melaphalan-resistant LR5). As expected, MM cells possessing diverse genetic lesions involving
MYC (Dib et al., 2008) were comparably sensitive to JQ1
(Figure 5B).
As interaction of MM cells with BMSCs is widely recognized to
confer resistance to numerous therapeutic agents (Hideshima
et al., 2007; McMillin et al., 2010), we sought to characterize
the effect of BMSCs on MM cell sensitivity to BET inhibition.
Using compartment-specific bioluminescence imaging assays
(CS-BLI), we observed that the sensitivity of MM cell lines to
JQ1 is largely unchanged by the presence of HS-5 bone marrow
stroma cells (Figure 5C). This pattern of broad activity in MM
without evident stroma-mediated chemoresistance has been
associated with efficacy of FDA-approved agents bortezomib
and lenalidomide.
MM cells were then further phenotyped for Myc-specific biological effects of BET inhibition. Flow cytometry of JQ1-treated
MM.1S cells revealed a pronounced decrease in the proportion
of cells in S phase, with a concomitant increase in cells arrested
in G0/G1 (Figure 6A). Only a modest induction of apoptosis was

observed after 48 hr of JQ1 treatment (Figure 6B), in contrast to
the nonselective cytotoxic kinase inhibitor staurosporine (Figure S5B). Transcripts previously associated with induction of
cellular senescence were enriched following treatment with
JQ1, by GSEA (Figure 6C). Experimentally, treatment with JQ1
resulted in pronounced cellular senescence by b-galactosidase
staining (Figure 6D). Overall, these phenotypes of arrested proliferation, G1 cell-cycle arrest, and cellular senescence are highly
specific to anticipated effects of inhibiting cellular c-Myc function (Wu et al., 2007).
We next extended the study of JQ1 in MM cells to primary MM
samples. JQ1 exposure led to a significant reduction in cell
viability among the majority of CD138+ patient-derived MM
samples tested (Figure 7A). In primary cells isolated from
a patient with relapsed/refractory MM, JQ1 treatment ex vivo
conferred a time-dependent suppression of c-Myc expression
(Figure 7B). In contrast, JQ1 treatment of phytohemaglutinin
(PHA)-stimulated peripheral blood mononuclear cells (PBMCs)
suppressed PHA-induced proliferation but did not adversely
influence cell viability, indicating that the anti-MM effect of JQ1
is not accompanied by a nonspecific, toxic effect on all hematopoietic cells (Figure S7A).
To model the therapeutic effect of JQ1 in vivo, we evaluated
anti-MM efficacy in multiple orthotopic models of advanced
disease. First, JQ1 was studied using an established, bioluminescent MM model (MM.1S-luc), which recapitulates the clinical
sequelae, anatomic distribution of MM lesions, and hallmark
bone pathophysiology observed in MM patients (Mitsiades
et al., 2004). Tumor-bearing mice were treated with JQ1 administered by intraperitoneal injection (50 mg/kg daily) or vehicle
control. JQ1 treatment significantly decreased the burden of
disease measured by serial, whole-body, noninvasive bioluminescence imaging (Figures 7C and 7D). Importantly, treatment
with JQ1 resulted in a significant prolongation in overall survival
compared to vehicle-treated animals (Figure 7E). In a second
plasmacytoma xenograft that more accurately models extramedullary disease, JQ1 also exhibited a significant diseasemodifying response (Figure S7B). Finally, the effect of JQ1 was
explored in the genetically engineered model of MYC-dependent
MM (Chesi et al., 2008). To date, two Vk*MYC mice with established disease and measurable M-protein have completed
14 days of JQ1 treatment (25 mg/kg daily, adjusted to tolerability). Both animals reveal objective evidence of response assessed by reduction of serum immunoglobulins, including a
complete response (CR) in the second animal (Figure 7F and Figure S7C). In this faithful orthotopic and nonproliferative model,
the only FDA-approved agents, bortezomib, melphalan and
cyclophosphamide, have previously prompted a CR (Chesi
et al., 2008). These results establish in vivo proof of concept
(C) Quantitative comparison of all transcription factor target gene sets available from the MSigDB by GSEA for reduced expression in JQ1-treated MM cell lines.
Data are presented as scatterplot of false discovery rate (FDR) versus normalized enrichment score (NES) for each evaluated gene set. Colored dots indicate gene
sets for MYC (red), E2F (black), or other (gray) transcription factors.
(D) GSEA showing downregulation in JQ1-treated MM cells of a representative set of genes with proximal promoter regions containing Myc-Max-binding sites.
(E) Table of gene sets enriched among genes downregulated by JQ1 in MM cells (top group), highlighting the number of genes in each set (n), the normalized
enrichment score (NES), and test of statistical significance (FDR q value). The bottom group represents comparisons of top-ranking transcription factor target
gene sets of MM master regulatory proteins, enriched among genes downregulated by JQ1 in MM cells. See also Figure S2.
Figure 3. BET Inhibition Suppresses MYC Transcription in MM

(A) Heatmap of cancer-related genes expressed in MM cells (MM.1S), treated with JQ1 (500 nM over 1, 4, and 8 hr). MYC (red) is downregulated by JQ1 in a timedependent manner, uniquely among all oncogenes studied (230 total). MYC was identified as the only statistically significant decrease in transcription at all four
time points analyzed (p < 0.05). See also Figure S3.
(B) Quantitative RT-PCR analysis for MYC levels in JQ1-treated MM.1S cells (500 nM, 08 hr). Data are presented as ratio of MYC expression at each time point
compared to baseline MYC expression (mean SD). Asterisks denote the level of statistical significance (*p < 0.01, **p < 0.0002, ***p < 0.006; paired Students t
test each relative to t = 0 hr).
(C) The active JQ1 enantiomer and the structurally analogous BET inhibitor iBET (Nicodeme et al., 2010), but not the inactive (-)-JQ1 enantiomer, downregulate
c-Myc expression, as determined by immunoblotting of MM.1S cells treated with compounds (500 nM) or vehicle control for 24 hr.
(D and E) Immunoblotting analyses of the (D) dose- and (E) time-dependent effects of JQ1 treatment on c-Myc expression in MM.1S cells.
(F and G) Selective depletion of nuclear c-Myc following JQ1 treatment (500 nM) as measured by ELISA-based DNA-binding assays for the activity of (F) c-Myc
(depleted after 12 hr) and (G) NF-kB family members (unaffected). Data represent mean SEM. See also Figure S4.
Relative mRNA
1.0
shRluc
shBRD4.533
shBRD4.602
shBRD4.1838
0.8
54%
RLuc
BRD4
533
17.9%
0.6
0.4
Count
0.2
0.0
BRD4
MYC
APC
C
Enrichment over
background region
350
Empty
VEH
JQ1
300
JQ1
Myc
250
200
Myc
150
100
-actin
50
0
NR2 NR3
E1 E2 E3 E4 E5
TS1 TS2
E1 E2 E3 E4
Negative
Region
MYC Enhancers
MYC TSS
IgH Enhancers
E
OPM1
Empty
Vehicle
OPM1
Empty
JQ1
OPM1
MYC
Vehicle
62 %
OPM1
MYC
JQ1
47 %
63 %
Count
31 %
APC
Figure 4. Regulation of MYC Transcription by BET Bromodomains

(A) BRD4 and MYC expression in OPM1 cells transduced with either shBRD4 (three different hairpins) or a Renilla control hairpin (shRluc). shRNAs were induced
for 5 days before analysis by qRT-PCR. Data normalized to GAPDH are presented as mean SEM of three biological replicates.
(B) Cell-cycle analysis of OPM1 cells transduced with the indicated shRNA for 6 days. BrdU staining (APC) identifies the fraction of cells in S phase. See also
Figure S5.
(C) ChIP studies of BRD4 (anti-BRD4; Bethyl) binding to MYC TSS or proximal enhancers in MM.1S cells. Competitive displacement of BRD4 from IgH enhancers
is observed upon treatment with JQ1 (500 nM for 24 hr, red bars) compared to vehicle control (black bars). Data represent mean SEM of three replicates.
See also Figure S6.
(D) Immunoblotting of whole-cell lysates from empty MSCV vector- or Myc overexpression vector-transduced OPM1 cells after treating with JQ1 (500 nM, 24 hr)
or DMSO control.
(E) Cell-cycle analysis of either empty or Myc-overexpressing OPM1 cells treated with JQ1 (500 nM, 24 hr). BrdU staining (APC) identifies the fraction of cells
in S phase.
for the investigational study of BET bromodomain inhibitors in

the treatment of MM.
DISCUSSION
Despite the centrality of Myc in the pathogenesis of cancer,
conventional approaches toward direct Myc inhibition have not
proven successful. To date, efforts to target c-Myc have identified only a small number of molecules with low biochemical
potency and limited biological characterization (Bidwell et al.,
2009; Hammoudeh et al., 2009; Jeong et al., 2010), underscoring
both the challenge of targeting c-Myc as well as the enduring
need for chemical probes of c-Myc transcriptional function.
Considering chromatin as a platform for signal transduction
(H) Heatmap of clustered gene expression data from multiplexed measurement (Nanostring) of cancer-associated genes in three human MM cell lines treated
with JQ1 or vehicle control. Among 230 genes studied (Figure S4), four genes (MYC, TERT, TYRO3, and MYB) exhibited statistically significant (p < 0.05)
downregulation. Replicate expression measurements exhibited high concordance among low and highly expressed genes (Figure S3B).
(I) Immunoblotting study of four MM lines (KMS11, LR5, OPM1, and INA-6) identifies a JQ1-induced decrease in c-Myc expression (500 nM, 24 hr).
B
0
25
50
100
200
400
KMS-34
LR5
MOLP-8
Dox 40
INA-6
MM.1S
KMS-5
KMS-12-PE
AMO-1
MR20
OPM-1
KMS-11
KMS-12-BM
KMS-26
L363
EJM
K MM1
MM.1S-myrAkt
KMS-20
MM. 1S-Bc l-2
RPMI-8226/S
OCI-MY5
KMS-28-BM
KMS-18
MM. 1R
100
50
IC-50
800 (nM)
68
98
136
122
164
152
116
108
124
167
155
163
230
235
241
285
317
365
341
326
395
432
502
1150
1490
MYC Annotation
der8t(8;?)
der16t(16;8;22)
der16t(16;8;22)
IgH insertion at breakpoint of der3t(3;8)
IgH insertion near MYC
der16t(16;8;22)
der(8)t(1;8) with insertion of IgH
MYC insertion, duplication and inversion
IgH insertion near MYC
der(8)t(8;22)
t(5;8); der8x2+der5
Negative
MYC insertion at IgH locus
der16t(16;8;22)
der(der14)t(8;14); duplication and inversion
t(8;14)
der(6&21)t(6&21;8)
10
50
0
0.0
0.2
150
0.4 0.6 0.8

JQ1 (uM)
1.0
L-363
100
50
0
0.0
0.2
150
0.4 0.6 0.8

JQ1 (uM)
1.0
Dox40
50
0
0.2
0.4 0.6 0.8

JQ1 (uM)
1.0
150
KMS-20
100
50
0
0.0
0.2
200
0.4 0.6 0.8

JQ1 (uM)
1.0
AMO-1
150
100
100
0.0
RPMI-8226
Normalized Proliferation
100
150
MM.1S
150
100
50
0
0.0
0.2
0.4 0.6 0.8

JQ1 (uM)
1.0
50
0
0.0
0.2
0.4 0.6 0.8

JQ1 (uM)
1.0
Figure 5. Antimyeloma Activity of JQ1 In Vitro

(A) A panel of MM cell lines was tested for in vitro sensitivity to JQ1 (12.5800 nM, 72 hr) by measurement of ATP levels (Cell TiterGlo; Promega).
(B) MYC genomic status of selected MM cell lines from (A), as annotated (Dib et al., 2008).
(C) Activity of JQ1 against MM cell lines cultured in the presence (red lines) or absence (black lines) of the HS-5 stromal cell line, assessed by CS-BLI (McMillin
et al., 2010). Data represent mean SD for four biological replicates.
(Schreiber and Bernstein, 2002), we have undertaken to inhibit

Myc transcription and function through displacement of chromatin-bound, coactivator proteins using competitive small molecules. Using a first-in-class, small-molecule bromodomain inhibitor developed by our laboratories, JQ1, we validate BET
bromodomains as determinants of c-Myc transcription and as
therapeutic targets in MM, an ideal model system for the mechanistic and translational study of Myc pathway inhibitors.
Most importantly, we illustrate the feasibility of selectively
downregulating transcription of MYC itself via the molecular
action of a selective, small molecule. The ensuing suppression
of c-Myc protein levels, depletion of chromatin-bound c-Myc,
and concomitant downregulation of the Myc-dependent transcriptional network lead to growth-inhibitory effects sharing the
specificity of phenotypes associated with prior genetic models
of Myc inhibition. These are notable observations that distinguish
the transcriptional consequences of BET inhibition from other
nonselective transcriptional inhibitors, such as actinomycin D,
a-amanitin, and flavopiridol.
A compelling finding is the observed, direct interaction of
BRD4 with IgH enhancers in MM cells possessing IgH rearrangement into the MYC locus and the depletion of BRD4 binding by
JQ1. This suggests BET inhibition as a strategy for targeting
other structural rearrangements in cancer involving IgH or other
A
VEH
Count
S 40%
G1 42%
JQ1 (24h)
JQ1 (48h)
S 5%
G1 80%
S 5%
G1 81%
PI
B
VEH
12%
12%
JQ1 (24h)
3%
23%
5%
PI
3%
JQ1 (48h)
AV
D
Enrichment Score
C
FRIDMAN_SENESCENCE_UP
0.4
0.3
0.2
0.1
0.0
Vehicle (40x)
JQ1 (500 nM; 40x)
FDR q < 0.05
Figure 6. JQ1 Induces Cell-Cycle Arrest and Cellular Senescence in MM Cells

(A and B) Flow cytometric evaluation of propidium iodide (PI) staining for cell-cycle analysis (A) and detection of Annexin V-positive apoptotic cells (B) in
JQ1-treated MM.1S cells (048 hr, 500 nM).
(C) Enrichment of senescence-associated genes among JQ1-suppressed genes in MM.1S cells.
(D) Induction of cellular senescence in JQ1-treated MM.1S cells (500 nM, 48 hr), as detected by b-galactosidase staining.
strong enhancers and has potential implications for the modulation of immunoglobulin gene expression in autoimmune
diseases.
An unexpected finding was the pronounced and concordant
suppression of multiple E2F-dependent transcriptional signatures. In this instance, E2F1 protein and transcript levels were
not affected by BET inhibition, suggesting either an unrecognized
function of BET bromodomains in E2F transcriptional complexes
or a dominant effect of Myc downregulation causing cell-cycle
arrest in G1 leading to silencing of E2F. These observations are
also compatible with the known role of Myc and E2F1 as transcriptional collaborators in cell-cycle progression and tumor
cell survival (Matsumura et al., 2003; Trimarchi and Lees, 2002).
Insights provided by our study identify rational strategies for
combination therapeutic approaches warranting exploration in
MM. MYC activation is commonly accompanied by antiapoptotic signaling in human cancer. In MM, constitutive or microenvironment-inducible activation of antiapoptotic Bcl-2 proteins
has been reported (Harada et al., 1998; Legartova et al., 2009).
Thus, Myc pathway inhibition by JQ1 may demonstrate synergism with targeted proapoptotic agents (e.g., ABT-737) (Oltersdorf et al., 2005; Trudel et al., 2007). Additionally, the selective
effect of JQ1 on Myc and E2F1 transcriptional programs provides an opportunity to combine BET inhibitors with pathwaydirected antagonists of the NF-kB, STAT3, XBP1, or HSF1 transcriptional programs.
Direct inhibition of c-Myc remains a central challenge in the
discipline of ligand discovery. Inhibition of MYC expression and
function, demonstrated herein, presents an immediate opportunity to study and translate the concept of c-Myc inhibition more
B
100
% Survival
Duration of JQ1 Treatement (h)

0
1
4
8
24
Pt#1
Pt#2
Pt#3
Pt#4
Pt#5
50
c-Myc
GAPDH
0
0
200
400
JQ1 (nM)
600
800
D
Bioluminescence
(ph/s/cm 2/sr)
2.5 x 1010
VEH
JQ1
JQ1
Vehicle
1.5 x 1010
5.0 x 10 9
p = 0.0002
0
0
Luminescence (x106)
10
15
20
25
Day of Treatment
0.5
1.0
1.5
2.0
2.5
3.0
Percent survival
100
Mouse 1
Day
JQ1
80
Mouse 2
14
14
Vehicle
60
40
20
p < 0.0001
0
0
10
M Protein
20
30
40
Day of Treatment
Figure 7. Translational Implications of BET Bromodomain Inhibition in MM

(A) JQ1 arrests the proliferation of primary, patient-derived CD138+ MM cells (Cell TiterGlo; Promega). Data represent the mean SD of four replicates per
condition.
(B) c-Myc immunoblot shows JQ1-induced downregulation in short-term culture of primary, patient-derived MM cells (500 nM, duration as indicated).
(C) Representative whole-body bioluminescence images of SCID-beige mice orthotopically xenografted after intravenous injection with MM.1S-luc+ cells and
treated with JQ1 (50 mg/kg IP daily) or vehicle control.
(D) Tumor burden of SCID-beige mice orthotopically xenografted after intravenous injection with MM.1S-luc+ cells. Upon detection of MM lesions diffusely
engrafted in the skeleton, mice were randomly assigned to receive JQ1 (50 mg/kg IP daily) or vehicle control. Data are presented as mean SEM (n = 10/group).
(E) Survival curves (Kaplan-Meier) of mice with orthotopic diffuse MM lesions show prolongation of overall survival with JQ1 treatment compared to vehicle
control (log-rank test, p < 0.0001).
(F) Serum protein electrophoresis to detect monoclonal, tumor-derived immunoglobulin (M-protein) in two MM-bearing Vk*myc mice before or after 7 and 14 days
of JQ1 treatment. JQ1 treatment induced partial and complete responses, respectively, in mouse 1 and mouse 2.
See also Figure S7.
broadly in human cancer. During the course of this research,

a collaborative effort with the laboratories of Christopher Vakoc
and Scott Lowe revealed BRD4 as a tumor dependency in acute
myeloid leukemia. Consistent with our observations described
here in MM, leukemia cells similarly require BRD4 to sustain
MYC expression to enforce aberrant self-renewal (Zuber et al.,

2011). Collectively, these findings highlight a broad role for
BRD4 in maintaining MYC expression in diverse hematopoietic
malignancies and suggest the utility of drug-like BET bromodomain inhibitors as therapeutic agents in these diseases.
Gene Expression Analysis
MM cells treated with JQ1 (500 nM, 24 h) were processed for transcriptional
profiling using Affymetrix Human Gene 1.0 ST microarrays. Expression of individual genes was assessed in the context of dose- and time-ranging experiments by real-time quantitative polymerase chain reaction, multiplexed direct
detection (Nanostring), and immunoblotting using antibodies as described in
the Extended Experimental Procedures.
Chromatin Immunoprecipitation
ChIP was performed on MM.1S cells cultured in the presence or absence of
JQ1 (500 nM, 24 hr). Specific antibodies, detailed methods, and primer sequences for MYC and IgH enhancers, as well as the MYC TSS, are described
in the Extended Experimental Procedures.
In Vitro and In Vivo MM Studies
The impact of JQ1 on cell viability, proliferation, and cell cycle was assessed in
human MM cells as documented in the Extended Experimental Procedures.
In vivo efficacy studies were performed with protocols approved by Institutional Animal Care and Use Committees at the DFCI or Mayo Clinic Arizona.
JQ1 was administered by intraperitoneal injection into SCID-beige mice with
MM lesions established after subcutaneous or intravenous injections and in
nonimmunocompromised tumor-bearing Vk*myc mice. Tumor burden in these
models was quantified by caliper measurement, whole-body bioluminescence
imaging, and serum protein electrophoresis, respectively, as detailed in the
Extended Experimental Procedures.
ACCESSION NUMBERS
Oligonucleotide microarray data have been deposited in the Gene Expression
Omnibus under the accession number GSE31365.
Supplemental Information includes Extended Experimental Procedures,
seven figures, and one table and can be found with this article online at
doi:10.1016/j.cell.2011.08.017.
ACKNOWLEDGMENTS
We are grateful to S. Lowe for sharing unpublished information; A. Azab, D.
McMillin, C. Ott, and A. Roccaro for technical support; E. Fox for microarray
data; J. Daley and S. Lazo-Kallanian for flow cytometry; the MMRF, MMRC,
and Broad Institute for establishing the MM Genomics Portal (http://www.
broadinstitute.org/mmgp/). This research was supported by NIHK08CA128972 (J.E.B.), NIH-R01CA050947 (C.S.M.), NIH-R01HG002668
(R.A.Y.), and NIH-R01CA46455 (R.A.Y.); the Chambers Medical Foundation
(P.G.R., C.S.M.); the Stepanian Fund for Myeloma Research (P.G.R.,
C.S.M.); and the Richard J. Corman Foundation (P.G.R., C.S.M.); an American
Cancer Society Postdoctoral Fellowship, 120272-PF-11-042-01-DMC
(P.B.R.); the Burroughs-Wellcome Fund, the Smith Family Award, the
Damon-Runyon Cancer Research Foundation, and the MMRF (to J.E.B.).
J.E.B. and C.S.M. designed the study, analyzed data, and prepared the
manuscript. J.E.D., H.M.J., and E.K. assayed MM drug sensitivity. G.C.I.
and J.E.D. assessed the effects of JQ1 on Myc expression. J.Q. performed
scaling synthesis and purification of JQ1. P.G.R. and K.C.A. provided primary
MM samples. P.B.R. and T.G. conducted ChIP experiments, and P.B.R. and
R.A.Y. contributed to their interpretation. R.M.P., T.P.H., and M.R.M. performed RNA expression analysis. I.M.G. and K.C.A. provided support and interpreted cellular data. A.C.S. and W.C.H. designed and performed shRNA
screens. M.E.L. analyzed expression array data. J.S. and C.R.V. performed
Myc rescue experiments. A.L.K. supervised in vivo efficacy and biostatistical
studies. M.C. and P.L.B. performed in vivo GEMM studies. J.E.B. and
C.S.M. supervised the research. All authors edited the manuscript.
Received: July 19, 2011

Revised: August 13, 2011
Accepted: August 15, 2011
Published online: September 1, 2011
REFERENCES
Amati, B., Brooks, M.W., Levy, N., Littlewood, T.D., Evan, G.I., and Land, H.
(1993). Oncogenic activity of the c-Myc protein requires dimerization with
Max. Cell 72, 233245.
Beroukhim, R., Mermel, C.H., Porter, D., Wei, G., Raychaudhuri, S., Donovan,
J., Barretina, J., Boehm, J.S., Dobson, J., Urashima, M., et al. (2010). The landscape of somatic copy-number alteration across human cancers. Nature 463,
899905.
Bidwell, G.L., 3rd, Davis, A.N., and Raucher, D. (2009). Targeting a c-Myc
inhibitory polypeptide to specific intracellular compartments using cell penetrating peptides. J. Control. Release 135, 210.
Bisgrove, D.A., Mahmoudi, T., Henklein, P., and Verdin, E. (2007). Conserved
P-TEFb-interacting domain of BRD4 inhibits HIV transcription. Proc. Natl.
Acad. Sci. USA 104, 1369013695.
Blackwood, E.M., and Eisenman, R.N. (1991). Max: a helix-loop-helix zipper
protein that forms a sequence-specific DNA-binding complex with Myc.
Science 251, 12111217.
Chapman, M.A., Lawrence, M.S., Keats, J.J., Cibulskis, K., Sougnez, C.,
Schinzel, A.C., Harview, C.L., Brunet, J.P., Ahmann, G.J., Adli, M., et al.
(2011). Initial genome sequencing and analysis of multiple myeloma. Nature
471, 467472.
Chesi, M., Robbiani, D.F., Sebag, M., Chng, W.J., Affer, M., Tiedemann, R.,
Valdez, R., Palmer, S.E., Haas, S.S., Stewart, A.K., et al. (2008). AID-dependent activation of a MYC transgene induces multiple myeloma in a conditional
mouse model of post-germinal center malignancies. Cancer Cell 13, 167180.
Chng, W.J., Huang, G.F., Chung, T.H., Ng, S.B., Gonzalez-Paz, N., TroskaPrice, T., Mulligan, G., Chesi, M., Bergsagel, P.L., and Fonseca, R. (2011).
Clinical and biological implications of MYC activation: a common difference
between MGUS and newly diagnosed multiple myeloma. Leukemia 25,
10261035.
Claudio, J.O., Masih-Khan, E., Tang, H., Goncalves, J., Voralia, M., Li, Z.H.,
Nadeem, V., Cukerman, E., Francisco-Pabalan, O., Liew, C.C., et al. (2002).
A molecular compendium of genes expressed in multiple myeloma. Blood
100, 21752186.
Dang, C.V. (2009). MYC, microRNAs and glutamine addiction in cancers. Cell
Cycle 8, 32433245.
Dang, C.V., Le, A., and Gao, P. (2009). MYC-induced cancer cell energy
metabolism and therapeutic opportunities. Clin. Cancer Res. 15, 64796483.
Darnell, J.E., Jr. (2002). Transcription factors as targets for cancer therapy.
Nat. Rev. Cancer 2, 740749.
Dean, M., Kent, R.B., and Sonenshein, G.E. (1983). Transcriptional activation
of immunoglobulin alpha heavy-chain genes by translocation of the c-myc
oncogene. Nature 305, 443446.
Dey, A., Nishiyama, A., Karpova, T., McNally, J., and Ozato, K. (2009). Brd4
marks select genes on mitotic chromatin and directs postmitotic transcription.
Mol. Biol. Cell 20, 48994909.
Dhalluin, C., Carlson, J.E., Zeng, L., He, C., Aggarwal, A.K., and Zhou, M.M.
(1999). Structure and ligand of a histone acetyltransferase bromodomain.
Nature 399, 491496.
Dib, A., Gabrea, A., Glebov, O.K., Bergsagel, P.L., and Kuehl, W.M. (2008).
Characterization of MYC translocations in multiple myeloma cell lines. J.
Natl. Cancer Inst. Monogr., 2531.
Filippakopoulos, P., Qi, J., Picaud, S., Shen, Y., Smith, W.B., Fedorov, O.,
Morse, E.M., Keates, T., Hickman, T.T., Felletar, I., et al. (2010). Selective inhibition of BET bromodomains. Nature 468, 10671073.
Frank, S.R., Parisi, T., Taubert, S., Fernandez, P., Fuchs, M., Chan, H.M.,
Livingston, D.M., and Amati, B. (2003). MYC recruits the TIP60 histone acetyltransferase complex to chromatin. EMBO Rep. 4, 575580.
Frye, S.V. (2010). The art of the chemical probe. Nat. Chem. Biol. 6, 159161.
Mattioli, M., Agnelli, L., Fabris, S., Baldini, L., Morabito, F., Bicciato, S.,
Verdelli, D., Intini, D., Nobili, L., Cro, L., et al. (2005). Gene expression profiling
of plasma cell dyscrasias reveals molecular patterns associated with distinct
IGH translocations in multiple myeloma. Oncogene 24, 24612473.
Fukazawa, T., Maeda, Y., Matsuoka, J., Yamatsuji, T., Shigemitsu, K., Morita,
I., Faiola, F., Durbin, M.L., Soucek, L., and Naomoto, Y. (2010). Inhibition of
Myc effectively targets KRAS mutation-positive lung cancer expressing high
levels of Myc. Anticancer Res. 30, 41934200.
McMillin, D.W., Delmore, J., Weisberg, E., Negri, J.M., Geer, D.C., Klippel, S.,
Mitsiades, N., Schlossman, R.L., Munshi, N.C., Kung, A.L., et al. (2010). Tumor
cell-specific bioluminescence platform to identify stroma-induced changes to
anticancer drug activity. Nat. Med. 16, 483489.
Gomi, M., Moriwaki, K., Katagiri, S., Kurata, Y., and Thompson, E.B. (1990).
Glucocorticoid effects on myeloma cells in culture: correlation of growth inhibition with induction of glucocorticoid receptor messenger RNA. Cancer Res.
50, 18731878.
Mitsiades, C.S., Mitsiades, N.S., McMullan, C.J., Poulaki, V., Shringarpure, R.,
Akiyama, M., Hideshima, T., Chauhan, D., Joseph, M., Libermann, T.A., et al.
(2004). Inhibition of the insulin-like growth factor receptor-1 tyrosine kinase
activity as a therapeutic strategy for multiple myeloma, other hematologic
malignancies, and solid tumors. Cancer Cell 5, 221230.
Hammoudeh, D.I., Follis, A.V., Prochownik, E.V., and Metallo, S.J. (2009).
Multiple independent binding sites for small-molecule inhibitors on the oncoprotein c-Myc. J. Am. Chem. Soc. 131, 73907401.
Harada, N., Hata, H., Yoshida, M., Soniki, T., Nagasaki, A., Kuribayashi, N.,
Kimura, T., Matsuzaki, H., and Mitsuya, H. (1998). Expression of Bcl-2 family
of proteins in fresh myeloma cells. Leukemia 12, 18171820.
Harris, A.W., Pinkert, C.A., Crawford, M., Langdon, W.Y., Brinster, R.L., and
Adams, J.M. (1988). The E mu-myc transgenic mouse. A model for highincidence spontaneous lymphoma and leukemia of early B cells. J. Exp.
Med. 167, 353371.
Haynes, S.R., Dollard, C., Winston, F., Beck, S., Trowsdale, J., and Dawid, I.B.
(1992). The bromodomain: a conserved sequence found in human, Drosophila
and yeast proteins. Nucleic Acids Res. 20, 2603.
Hideshima, T., Mitsiades, C., Tonon, G., Richardson, P.G., and Anderson, K.C.
(2007). Understanding multiple myeloma pathogenesis in the bone marrow to
identify new therapeutic targets. Nat. Rev. Cancer 7, 585598.
Hurt, E.M., Wiestner, A., Rosenwald, A., Shaffer, A.L., Campo, E., Grogan, T.,
Bergsagel, P.L., Kuehl, W.M., and Staudt, L.M. (2004). Overexpression of
c-maf is a frequent oncogenic event in multiple myeloma that promotes proliferation and pathological interactions with bone marrow stroma. Cancer Cell 5,
191199.
Jain, M., Arvanitis, C., Chu, K., Dewey, W., Leonhardt, E., Trinh, M., Sundberg,
C.D., Bishop, J.M., and Felsher, D.W. (2002). Sustained loss of a neoplastic
phenotype by brief inactivation of MYC. Science 297, 102104.
Jeong, K.C., Ahn, K.O., and Yang, C.H. (2010). Small-molecule inhibitors of
c-Myc transcriptional factor suppress proliferation and induce apoptosis of
promyelocytic leukemia cell via cell cycle arrest. Mol. Biosyst. 6, 15031509.
Keats, J.J., Fonseca, R., Chesi, M., Schop, R., Baker, A., Chng, W.J., Van Wier,
S., Tiedemann, R., Shi, C.X., Sebag, M., et al. (2007). Promiscuous mutations
activate the noncanonical NF-kappaB pathway in multiple myeloma. Cancer
Cell 12, 131144.
Kim, J., Chu, J., Shen, X., Wang, J., and Orkin, S.H. (2008). An extended transcriptional network for pluripotency of embryonic stem cells. Cell 132, 1049
1061.
Kim, Y.H., Girard, L., Giacomini, C.P., Wang, P., Hernandez-Boussard, T.,
Tibshirani, R., Minna, J.D., and Pollack, J.R. (2006). Combined microarray
analysis of small cell lung cancer reveals altered apoptotic balance and distinct
expression signatures of MYC family gene amplification. Oncogene 25,
130138.
Leder, A., Pattengale, P.K., Kuo, A., Stewart, T.A., and Leder, P. (1986).
Consequences of widespread deregulation of the c-myc gene in transgenic
mice: multiple neoplasms and normal development. Cell 45, 485495.
Legartova, S., Krejci, J., Harnicarova, A., Hajek, R., Kozubek, S., and Bartova,
E. (2009). Nuclear topography of the 1q21 genomic region and Mcl-1 protein
levels associated with pathophysiology of multiple myeloma. Neoplasma 56,
404413.
Malo, N., Hanley, J.A., Cerquozzi, S., Pelletier, J., and Nadon, R. (2006). Statistical practice in high-throughput screening data analysis. Nat. Biotechnol. 24,
167175.
Matsumura, I., Tanaka, H., and Kanakura, Y. (2003). E2F1 and c-Myc in cell
growth and death. Cell Cycle 2, 333338.
Mitsiades, N., Mitsiades, C.S., Poulaki, V., Chauhan, D., Fanourakis, G., Gu, X.,
Bailey, C., Joseph, M., Libermann, T.A., Treon, S.P., et al. (2002). Molecular
sequelae of proteasome inhibition in human multiple myeloma cells. Proc.
Natl. Acad. Sci. USA 99, 1437414379.
Nair, S.K., and Burley, S.K. (2003). X-ray structures of Myc-Max and Mad-Max
recognizing DNA. Molecular bases of regulation by proto-oncogenic transcription factors. Cell 112, 193205.
Nicodeme, E., Jeffrey, K.L., Schaefer, U., Beinke, S., Dewell, S., Chung, C.W.,
Chandwani, R., Marazzi, I., Wilson, P., Coste, H., et al. (2010). Suppression of
inflammation by a synthetic histone mimic. Nature 468, 11191123.
Oltersdorf, T., Elmore, S.W., Shoemaker, A.R., Armstrong, R.C., Augeri, D.J.,
Belli, B.A., Bruncko, M., Deckwerth, T.L., Dinges, J., Hajduk, P.J., et al.
(2005). An inhibitor of Bcl-2 family proteins induces regression of solid
tumours. Nature 435, 677681.
Palumbo, A.P., Pileri, A., Dianzani, U., Massaia, M., Boccadoro, M., and
Calabretta, B. (1989). Altered expression of growth-regulated protooncogenes
in human malignant plasma cells. Cancer Res. 49, 47014704.
Pomerantz, M.M., Ahmadiyeh, N., Jia, L., Herman, P., Verzi, M.P., Doddapaneni, H., Beckwith, C.A., Chan, J.A., Hills, A., Davis, M., et al. (2009a). The
8q24 cancer risk variant rs6983267 shows long-range interaction with MYC
in colorectal cancer. Nat. Genet. 41, 882884.
Pomerantz, M.M., Beckwith, C.A., Regan, M.M., Wyman, S.K., Petrovics, G.,
Chen, Y., Hawksworth, D.J., Schumacher, F.R., Mucci, L., Penney, K.L.,
et al. (2009b). Evaluation of the 8q24 prostate cancer risk locus and MYC
expression. Cancer Res. 69, 55685574.
Rahl, P.B., Lin, C.Y., Seila, A.C., Flynn, R.A., McCuine, S., Burge, C.B., Sharp,
P.A., and Young, R.A. (2010). c-Myc regulates transcriptional pause release.
Cell 141, 432445.
Rahman, S., Sowa, M.E., Ottinger, M., Smith, J.A., Shi, Y., Harper, J.W., and
Howley, P.M. (2011). The Brd4 extraterminal domain confers transcription
activation independent of pTEFb by recruiting multiple proteins, including
NSD3. Mol. Cell. Biol. 31, 26412652.
Schlosser, I., Holzel, M., Hoffmann, R., Burtscher, H., Kohlhuber, F., Schuhmacher, M., Chapman, R., Weidle, U.H., and Eick, D. (2005). Dissection of
transcriptional programmes in response to serum and c-Myc in a human
B-cell line. Oncogene 24, 520524.
Schreiber, S.L., and Bernstein, B.E. (2002). Signaling network model of chromatin. Cell 111, 771778.
Schuhmacher, M., Kohlhuber, F., Holzel, M., Kaiser, C., Burtscher, H., Jarsch,
M., Bornkamm, G.W., Laux, G., Polack, A., Weidle, U.H., and Eick, D. (2001).
The transcriptional program of a human B cell line in response to Myc. Nucleic
Acids Res. 29, 397406.
Shaffer, A.L., Emre, N.C., Lamy, L., Ngo, V.N., Wright, G., Xiao, W., Powell, J.,
Dave, S., Yu, X., Zhao, H., et al. (2008). IRF4 addiction in multiple myeloma.
Nature 454, 226231.
Shou, Y., Martelli, M.L., Gabrea, A., Qi, Y., Brents, L.A., Roschke, A., Dewald,
G., Kirsch, I.R., Bergsagel, P.L., and Kuehl, W.M. (2000). Diverse karyotypic
abnormalities of the c-myc locus associated with c-myc dysregulation and
tumor progression in multiple myeloma. Proc. Natl. Acad. Sci. USA 97,
228233.
Soucek, L., Helmer-Citterich, M., Sacco, A., Jucker, R., Cesareni, G., and Nasi,
S. (1998). Design and properties of a Myc derivative that efficiently homodimerizes. Oncogene 17, 24632472.
Soucek, L., Jucker, R., Panacchia, L., Ricordy, R., Tato`, F., and Nasi, S. (2002).
Omomyc, a potential Myc dominant negative, enhances Myc-induced
apoptosis. Cancer Res. 62, 35073510.
Stewart, T.A., Pattengale, P.K., and Leder, P. (1984). Spontaneous mammary
adenocarcinomas in transgenic mice that carry and express MTV/myc fusion
genes. Cell 38, 627637.
Trimarchi, J.M., and Lees, J.A. (2002). Sibling rivalry in the E2F family. Nat. Rev.
Mol. Cell Biol. 3, 1120.
Trudel, S., Stewart, A.K., Li, Z., Shu, Y., Liang, S.B., Trieu, Y., Reece, D.,
Paterson, J., Wang, D., and Wen, X.Y. (2007). The Bcl-2 family protein inhibitor,
ABT-737, has substantial antimyeloma activity and shows synergistic effect
with dexamethasone and melphalan. Clin. Cancer Res. 13, 621629.
Vervoorts, J., Luscher-Firzlaff, J.M., Rottmann, S., Lilischkis, R., Walsemann,
G., Dohmann, K., Austen, M., and Luscher, B. (2003). Stimulation of c-MYC
transcriptional activity and acetylation by recruitment of the cofactor CBP.

EMBO Rep. 4, 484490.
Wu, C.H., van Riggelen, J., Yetil, A., Fan, A.C., Bachireddy, P., and Felsher,
D.W. (2007). Cellular senescence is an important mechanism of tumor regression upon c-Myc inactivation. Proc. Natl. Acad. Sci. USA 104, 1302813033.
Zeller, K.I., Jegga, A.G., Aronow, B.J., ODonnell, K.A., and Dang, C.V. (2003).
An integrated database of genes responsive to the Myc oncogenic transcription factor: identification of direct genomic targets. Genome Biol. 4, R69.
Zhan, F., Barlogie, B., Arzoumanian, V., Huang, Y., Williams, D.R., Hollmig, K.,
Pineda-Roman, M., Tricot, G., van Rhee, F., Zangari, M., et al. (2007). Geneexpression signature of benign monoclonal gammopathy evident in multiple
myeloma is linked to good prognosis. Blood 109, 16921700.
Zuber, J., Shi, J., Wang, E., Rappaport, A.R., Herrmann, H., Sison, E.A.,
Magoon, D., Qi, J., Blatt, K., Wunderlich, M., et al. (2011). RNAi screen identifies Brd4 as a therapeutic target in acute myeloid leukaemia. Nature.
Published online August 3, 2011. 10.1038/nature10334.
USP1 Deubiquitinates ID Proteins

to Preserve a Mesenchymal
Stem Cell Program in Osteosarcoma
Samuel A. Williams,1 Heather L. Maecker,2 Dorothy M. French,3 Jinfeng Liu,4 Andrew Gregg,1 Leah B. Silverstein,1
Tim C. Cao,5 Richard A.D. Carano,5 and Vishva M. Dixit1,*
1Department
of Physiological Chemistry
of Translational Oncology
3Department of Pathology
4Department of Bioinformatics
5Department of Biomedical Imaging
Genentech, Inc., 1 DNA Way, South San Francisco, CA 94080, USA
*Correspondence: dixit@gene.com
DOI 10.1016/j.cell.2011.07.040
2Department
SUMMARY
Inhibitors of DNA binding (IDs) antagonize basichelix-loop-helix (bHLH) transcription factors to inhibit
differentiation and maintain stem cell fate. ID ubiquitination and proteasomal degradation occur in differentiated tissues, but IDs in many neoplasms appear
to escape degradation. We show that the deubiquitinating enzyme USP1 promotes ID protein stability
and stem cell-like characteristics in osteosarcoma.
USP1 bound, deubiquitinated, and thereby stabilized
ID1, ID2, and ID3. A subset of primary human osteosarcomas coordinately overexpressed USP1 and ID
proteins. USP1 knockdown in osteosarcoma cells
precipitated ID protein destabilization, cell-cycle
arrest, and osteogenic differentiation. Conversely,
ectopic USP1 expression in mesenchymal stem cells
stabilized ID proteins, inhibited osteoblastic differentiation, and enhanced proliferation. Consistent with
USP1 functioning in normal mesenchymal stem cells,
USP1-deficient mice were osteopenic. Our observations implicate USP1 in preservation of the stem
cell state that characterizes osteosarcoma and identify USP1 as a target for differentiation therapy.
INTRODUCTION
Basic-helix-loop-helix (bHLH) transcription factors comprise the
third-largest family of recognized transcription factors in the
human genome (Tupler et al., 2001) and are essential regulators
of development and differentiation through binding DNA elements termed E boxes (Massari and Murre, 2000). Class I
bHLH homodimers are expressed broadly and promote expression of antiproliferative genes such as CDKN1A, CDKN2A, and
CDKN2B (Yokota and Mori, 2002). Class II bHLH proteins
show more restricted expression and form heterodimers with
class I proteins to drive tissue-specific genes such as IGH@

and SP7/OSTERIX (Lassar et al., 1991; Weintraub et al., 1994).
Through the combined induction of tissue-specific and antiproliferative genes, bHLH transcription factors serve as integrators
of lineage commitment.
DNA binding of bHLH proteins is limited by heterodimerization
with inhibitor of DNA-binding proteins, or IDs. The ID family
consists of four members, ID1, ID2, ID3, and ID4 (Lasorella
et al., 2001), with overlapping spatial and temporal expression
profiles. All four IDs bind the various bHLH proteins with similar
affinities to regulate gene expression (Prabhu et al., 1997). IDs
are essential for mammalian development; disruption of two or
more ID genes results in embryonic lethality (Lyden et al.,
1999). In contrast, overexpression of ID proteins in transgenic
mice produces fatal malignancies (Kim et al., 1999). Similarly,
elevated ID protein levels are observed in a broad range of
dedifferentiated primary human malignancies ranging from
pancreatic carcinoma to neuroblastoma (Perk et al., 2005).
Although ID proteins are scarce in normal adult differentiated
tissues, they are abundant in proliferating tissues, including
embryonic and adult stem cell populations, which suggests
that IDs might maintain stemness (Yokota and Mori, 2002).
The stem cell expression and tumor-promoting qualities of ID
proteins suggest that ID genes may be pivotal in cancer stem
cell biology. An engineered ID-suppressing HLH protein was
reported to differentiate neuroblastoma tumors (Ciarapica
et al., 2009).
IDs are induced transcriptionally by myriad growth factors
including bone morphogenic proteins, platelet-derived growth
factor, epidermal growth factor, as well as by T cell receptor ligation (Yokota and Mori, 2002). ID1, ID2, and ID3, but not ID4, are
subject to K48-linked polyubiquitination and subsequent degradation by the 26S proteasome. Consequently, IDs are short lived
in most tissues (Bounpheng et al., 1999). The ubiquitously expressed APC/Cdh1 complex is an E3 ubiquitin ligase that
governs ID stability and abundance (Lasorella et al., 2006), but
ID proteins are stable in some contexts. Therefore, we sought
a deubiquitinating enzyme that counters ID ubiquitination.
We show that USP1 deubiquitinates and stabilizes ID1, ID2,

and ID3, resulting in their increased abundance. Significantly,
elevated USP1 protein and mRNA in a subset of primary osteosarcoma tumors correlated with increased ID protein levels.
USP1 knockdown in osteosarcoma cells caused ID protein
destabilization, p53-independent induction of CDKN1A encoding cyclin-dependent kinase inhibitor (CDKI) p21, and cellcycle arrest. In addition, expression of mesenchymal stem cell
markers was decreased and osteogenic differentiation resumed.
These data suggest that osteosarcomas, like acute promyelocytic leukemia, may be amenable to differentiation therapies
(Soignet et al., 1998). In contrast to USP1 knockdown, USP1
overexpression in primary human mesenchymal stem cells
(hMSCs) caused ID protein accumulation and interfered with
normal differentiation. Indeed, USP1 promoted transformation
in a mesenchymal cell line. Finally, loss of Usp1 in gene-targeted
mice caused severe osteopenia, which is consistent with a role
for USP1 in the mesenchymal lineage. Our results strongly suggest that USP1 has oncogenic potential and promotes tumorigenesis through disruption of normal mesenchymal stem cell
commitment and differentiation.
RESULTS
USP1 Deubiquitinates and Stabilizes ID Proteins
To identify deubiquitinases (DUBs) that stabilize ID proteins, we
overexpressed 94 human DUBs with C-terminal Flag epitopes
in 293T cells and assessed endogenous ID2 abundance by
western blot (see Table S1 available online). 293T cells degrade
ID2 in a proteasome-dependent manner because ID2 accumulates after treatment with the proteasome inhibitor MG-132
(Figure S1A). DUBs that increased endogenous ID2 were
USP36, JOSD2, USP33, SENP3, SENP5, USP37, OTUD5,
USP9Y, USP45, and USP1 (Figure S1A). To exclude indirect
mechanisms increasing ID2 expression, we focused on ID2-interacting DUBs USP1 and USP33 (Figure S1B). However, unlike
USP1, USP33 lacked deubiquitinating activity against ID2 (data
not shown).
To determine whether USP1 extended the half-life of ID2, we
transfected 293T cells with USP1 and monitored ID2 abundance
after treatment with the translational inhibitor cycloheximide. In
the absence of new protein synthesis, ID2 was cleared rapidly
from cells transfected with a control vector, with a half-life of
approximately 2 min (Figure 1A). Overexpressed USP1 extended
the half-life of ID2 to over 80 min. The proteolytic activity of
USP1 was necessary for ID2 accumulation because the catalytically inactive point mutant USP1 C90S (Nijman et al., 2005)
neither increased the half-life of ID2 (Figure 1A) nor altered ID2
steady-state abundance (Figure 1B). Similar results were obtained with ID1 and ID3. USP1 appeared to target IDs specifically
because it did not enhance expression of labile IkBa (Palombella
et al., 1994).
Next, we assessed the effect of USP1 on ID2 ubiquitination.
Wild-type USP1, but not USP1 C90S, reduced the amount of
ID2 modified with HA-tagged ubiquitin (Figure 1C). Both basal
and USP1-induced ID2 deubiquitination was enhanced by coexpression of USP1 cofactor WDR48 (Cohn et al., 2007) (Figure 1C).
To address whether USP1 deubiquitinated ID2 directly, ubiquiti-
nated ID2 purified from 293T cells was incubated in vitro with
either wild-type USP1 or USP1 C90S purified separately from
293T cells. Ubiquitinated ID2 was decreased by wild-type
USP1 but not USP1 C90S (Figure 1D), indicating that deubiquitination was unlikely a consequence of a coeluted protease. The
decrease in ID2 ubiquitination also was sensitive to N-ethylmaleimide, confirming the involvement of a cysteine protease
(Figure 1D). Consistent with ubiquitinated ID2 being a USP1
substrate, deletion mutant USP1D260300 interacted poorly
with ID2 (Figure S1C) and did not enhance ID2 abundance
(Figure S1D).
USP1 and ID2 Are Coordinately Overexpressed in
a Subset of Primary Osteosarcoma Tumors
To identify a biological context in which USP1 deubiquitinates ID
proteins, we examined USP1 expression patterns. Microarray
analyses of healthy and diseased human tissues revealed that
osteosarcoma tumors expressed more USP1 mRNA than
healthy or osteoarthritic bone biopsies (Figure 2A). Western blotting of a separate set of primary human osteosarcoma biopsies
found that USP1 was elevated in 7 of 14 osteosarcomas when
compared to 3 normal primary human osteoblast samples (Figure 2B). Strikingly, ID2 protein abundance in these primary
human tumor samples correlated well with USP1 abundance.
One anomalous sample contained abundant USP1 but little
ID2 (Figure 2B, lane 6), perhaps due to poor expression of the
USP1 cofactor WDR48. Another sample contained abundant
ID2 and little USP1 (Figure 2B, lane 16), which may reflect
reduced ID2 ubiquitination or that other DUBs are active.
The amount of USP1 protein in the primary osteosarcomas
correlated largely with USP1 mRNA abundance (Figure 2C),
suggesting that elevated USP1 in osteosarcoma is due to transcriptional upregulation. In contrast, ID2 protein and mRNA
levels correlated poorly (Figure 2D). The coincident overexpression of USP1 and ID2 in primary osteosarcoma was confirmed by
immunohistochemistry (Figures 2E2G). These results strongly
suggest that USP1 modifies ID proteins posttranslationally in
osteosarcoma.
USP1 Stabilizes ID Proteins in Osteosarcoma
We also assessed USP1 abundance and ID2 stability in human
osteosarcoma cell lines and in primary osteoblasts (Figure S2A).
In U2-OS osteosarcoma cells, USP1 was elevated, and the normally labile ID2 was stable (Figures S2A and S2B). Knockdown of
USP1 with two distinct USP1 shRNAs caused a reduction in ID1,
ID2, and ID3 but had no effect on ID4 (Figure 3A). ID1, ID2, and
ID3 mRNAs were not reduced, excluding decreased transcription as the reason for the drop in ID protein abundance (Figure S3I). USP1 knockdown specificity was confirmed with
shRNA-resistant USP1, which restored ID1, ID2, and ID3 to basal
levels. USP1 catalytic activity was essential for ID stability
because shRNA-resistant USP1 C90S did not restore ID protein
levels. Similar results were observed in osteosarcoma cell lines
HOS, SAOS, and SJSA (Figure S2C). USP1 knockdown did not
impact ID2 abundance in MG-63 osteosarcoma cells, likely
because these cells express very little WDR48 (Figure S2C).
Consistent with WDR48 deficiency limiting USP1 activity in
MG-63 cells, ectopic WDR48 increased ID2 (Figure S2D).
40
80
120
Time (min)
D
WDR48
ID2-FLAG
USP1
T
90
S
USP1
USP1
T
90
S
USP1
IB-FLAG
WT
C90S
USP1
ID3-FLAG
CTL
ID2-FLAG
17 18 19 20 21 22 23 24
25%
9 10 11 12 13 14 15 16
MG-132
WT
C90S
CTL
WT
C90S
CTL
USP1
MG-132
WT
C90S
CTL
ID1-FLAG
MG-132
MG-132
50%
- - - + + +
+ + + + + +
188
WB: Flag
(ID/IB)
WDR48
ID2-Ub
NEM
T
90
S
W
T
WB: Actin
75%
- + + +
+ + + +
- - - +
C
WB: USP1
CTL
USP1 WT
USP1 C90S
100%
WB: ID2
ID2 Levels (% Untreated)
USP1 C90S
0
7
15
30
60
12
0
24
0
48
0
0
7
15
30
60
12
0
24
0
48
0
CHX (min)
USP1 WT
0
7
15
30
60
12
0
24
0
48
0
CTL
98
IP: FLAG/
62
WB:
HA-Ub
WB: USP1
Poly-Ub
49
38
WB: Tub
1
10 11 12
13
14
15
28
16
WB:
HA-Ub
5
4
3
17
IP: FLAG/
WB:
ID2-FLAG
WB: USP1
WB: USP1
WB: WDR48
WB: WDR48
WB: ID2-FLAG
WB: ID2
1
WB: Tub
1
Figure 1. USP1 Deubiquitinates and Stabilizes ID Proteins

(A) Western blot (WB) analysis of 293T cells transfected with vector only (CTL), wild-type USP1 (WT), or catalytically inactive USP1 C90S. Cells were treated with
25 mg/ml cycloheximide (CHX) for the times indicated (left panel). ID2 was quantified by densitometry (right panel).
(B) 293T cells were cotransfected with Flag-tagged ID1, ID2, ID3, or IkBa, and empty vector (CTL), wild-type USP1, or USP1 C90S. Where indicated, cells were
treated with 10 mM MG-132 for 4 hr.
(C) Deubiquitination of ID2-Flag by USP1 or USP1 C90S and WDR48 in 293T cells cotransfected with HA-tagged ubiquitin.
(D) USP1-Flag, USP1 C90S-Flag, WDR48-Flag, and ubiquitinated ID2-Flag were affinity purified separately from 293T extracts and then combined together for
6 hr in an in vitro deubiquitination assay. NEM, N-ethylmaleimide.
See also Figure S1.
To determine if USP1 knockdown and decreased IDs 13

modulated bHLH transcriptional activity, U2-OS cells were
transfected with an E box-driven luciferase reporter gene.
USP1 shRNAs enhanced expression of this reporter 7- to
10-fold over a control shRNA, consistent with activation of
bHLH proteins when ID proteins are decreased (Figure 3B).
shRNA-resistant wild-type USP1, but not USP1 C90S, suppressed E box-driven reporter activity caused by USP1 knockdown, confirming that USP1 catalytic activity is required for
bHLH-dependent transcription as well as ID protein stabilization.
The acute loss of IDs 13 following knockdown of endogenous
USP1 suggested ID protein destabilization via proteasomemediated degradation. As anticipated, the proteasome inhibitor
MG-132 did not by itself alter ID protein abundance in U2-OS
cells, suggesting that the IDs are intrinsically stable in cells that
express USP1 highly (Figure 3C). However, MG-132 treatment

did restore ID expression after USP1 knockdown, indicating
that ID proteins are subject to proteasome-mediated clearance
upon USP1 depletion. In keeping with this scenario, USP1
knockdown in MG-132-treated U2-OS cells increased the
amount of ubiquitinated ID2 (Figure 3D).
Next, we confirmed that endogenous USP1 associates with an
endogenous ID in osteosarcoma cells. ID2 coimmunoprecipitated with endogenous USP1 from U2-OS cells (Figure 3E), albeit
not with 1:1 stoichiometry, but this is to be expected for a
transient enzyme-substrate interaction. Similar results were
obtained in HOS cells (Figure S2E). USP1 also coimmunoprecipitated with ID2 (Figure 3F). Collectively, our results suggest
that USP1 is a potent DUB and stabilizing factor for ID1, ID2,
and ID3 in osteosarcoma.
6000
A B C
5000
Primary Osteosarcomas
28271
28272
28273
28274
28275
28276
28277
7000
4000
WB: USP1
3000
WB: ID2
2000
WB: WDR48
1000
Osteoblastic
Osteosarcoma
Osteoarthritis
10
11 12 13 14 15 16 17
8
7
6
5
4
3
2
1
0
A B C
A B C
293T-ID2shRNA
293T-ID2
Normal
Bone
ID2 mRNA (Arbitrary Units)
USP1 mRNA (Arbitrary Units)
WB: Actin
1
2915
4384
4898
4931
10047
10049
10050
28271
28272
28273
28274
28275
28276
28277
2915
4384
4898
4931
10047
10049
10050
28271
28272
28273
28274
28275
28276
28277
USP1 mRNA (microarray MAS5

signal intensity)
8000
2915
4384
4898
4931
10047
10049
10050
9000
Primary
Osteoblasts
Control
USP1
ID2
Figure 2. USP1 Is Overexpressed in Osteosarcoma and Correlates with ID2 Protein Expression
(A) Box and whisker plots of USP1 mRNA expression in primary human bone biopsies from normal and diseased tissue.
(B) Western blot (WB) analysis of USP1 and ID2 protein expression in primary human osteoblasts and osteosarcoma tumor samples.
(C and D) RT-PCR quantification of USP1 (C) and ID2 (D) expression in the samples in (B). Bars represent the mean SD of triplicate observations.
(E and F) Immunohistochemical detection of ID2 in 293T cells transfected with an ID2 expression vector (top panel) or an ID2 shRNA (bottom panel) (E) or in
a primary human osteosarcoma biopsy (F).
(G) Immunohistochemical staining of USP1 and ID2 in serial sections from primary osteosarcoma tissue. Control staining was with an isotype-control antibody.
ID2 stabilization by USP1 was not limited to the setting of

osteosarcoma. USP1 / DT40 chicken B cells (Oestergaard
et al., 2007) expressed less ID2 protein than their wild-type counterparts (Figure S2F) despite expressing similar levels of ID2
mRNA (Figure S2G). Consistent with USP1 deubiquitinating
and stabilizing ID2, proteasome inhibition with MG-132 increased ID2 in USP1 / , but not wild-type, DT40 cells (Figure S2H). In addition, USP1 / DT40 cells reconstituted with
wild-type USP1, but not USP1 C90S, contained equivalent ID2
to wild-type DT40 cells (Figure S2I).
USP1 Suppresses p21-Mediated Cell-Cycle Arrest

in Osteosarcoma
One potential consequence of USP1 deficiency and increased
bHLH transcriptional activity in osteosarcoma cells is induction
of bHLH-regulated CDKI p21. Indeed, p21 was increased in
U2-OS cells transfected with USP1 shRNAs relative to cells
transfected with a control shRNA (Figure 4A). shRNA-resistant
wild-type USP1, but not USP1 C90S, reduced p21 to levels
observed in control cells, confirming knockdown specificity in
this setting. The tumor suppressor p53, a well-known inducer
shUSP1-B
shUSP1-A
shCTL
shRes
USP1
CTL
WT
- - + - - +
- + - - + + - - + - -
shRes
USP1
C90S
- - +
- + + - -
C
14
WB: ID1
WB: ID2
WB: ID3
WB: ID4
WB: E47
WB: USP1
CTL
USP1-WT
USP1-C90S
12
E-Box Luciferase Activity

(Fold Induction, Renilla Norm.)
MG-132
10
WB: ID1
WB: ID2
WB: ID3
WB: USP1
WB: Tub
WB: Tub
MG-132 -
IP
IP: ID2-FLAG
IB: ID2-FLAG
- + +
IP: ID2-FLAG
IB: HA-Ub
shCTL shUSP1 shUSP1

A
B
WB: ID2
WB: USP1
WB: USP1
WB: ID2
IP
-
IgG
ID2
Input
IgG
USP1
Input
shCTL
shUSP1
shCTL
shUSP1
sh
sh
sh USP1 USP1
B
CTL
A
- + - + - +
WB: GAPDH
WB: Tub
1
1
IB: USP1
IB: ID2-FLAG
IB: GAPDH
1
Figure 3. USP1 Physically Engages and Stabilizes ID Proteins in Osteosarcoma

(A) Western blot (WB) analysis of U2-OS cells cotransfected with USP1 or control (CTL) shRNAs, plus either empty vector (CTL) or shRNA-resistant USP1
(wild-type [WT] or USP1 mutant C90S).
(B) Luciferase activity of U2-OS cells treated as in (A) and cotransfected with an E box-driven luciferase reporter. Bars represent the mean SD of triplicate
observations.
(C) U2-OS cells were transfected with shRNAs and, where indicated, treated with 10mM MG-132 for 4 hr.
(D) U2-OS cells were cotransfected with ID2-Flag, HA-ubiquitin, and either CTL or USP1 shRNAs. Where indicated, cells were treated with 10mM MG-132 for 4 hr.
ID2-Flag was immunoprecipitated from SDS/heat-denatured cell lysates.
(E and F) USP1 (E) or ID2 (F) was immunoprecipitated from U2-OS cells. Control immunoprecipitations were with nonspecific IgG. Asterisk (*) denotes a band of
unknown identity recognized by the anti-ID2 antibody.
See also Figure S3.
of CDKN1A, was not increased by USP1 knockdown, suggesting that increased p21 was p53 independent.
p21 is a potent inhibitor of cell cycle progression (Polyak et al.,
1996), so we assessed the proliferative capacity of U2-OS cells
following USP1 knockdown. Consistent with increased p21,
USP1 knockdown reduced U2-OS cell proliferation (Figure 4B
and Figure S3A). shRNA-resistant wild-type USP1, but neither
USP1 C90S nor USP1D260300, restored cell proliferation
(Figures S3B and S3C), indicating that both USP1 catalytic activity
and ID substrate recognition are required to maintain U2-OS cell
proliferation. USP1 knockdown similarly reduced proliferation in
HOS, SAOS, and SJSA, but not MG-63 osteosarcoma cells (Figure S3D). Flow cytometric analysis of the DNA content in U2-OS
cells after USP1 knockdown revealed a moderate increase in cells
in G1 and G2 phases of the cell cycle with a pronounced reduction

of cells in S phase (Figure 4C and Figure S3E). Apoptosis induction
following USP1 knockdown was not prominent; few cells with
a subdiploid DNA content were observed, there was no increase
in cells stained with annexin V, and increased processing of
caspase-3 was not detected (Figure S3E; data not shown).
Significantly, CDKN1A siRNAs restored S phase entry in USP1deficient U2-OS cells (Figure S3F and S3G), indicating that p21
is essential for the cell-cycle arrest induced by USP1 knockdown.
USP1 Regulates p21 Expression and Cell-Cycle Arrest
in Osteosarcoma via ID Proteins
If ID degradation in the absence of USP1 caused p21 induction,
then knockdown of the ID proteins should phenocopy USP1
CTL
+
+
-
+
-
+
-
+
-
+
-
+
-
100
Total Viable Cells (x103)
shUSP1-B
shUSP1-A
shCtl
shRes
shRes
USP1 WT USP1 C90S
WB: p21
WB: p53
WB: USP1
WB: Tub
80
60
40
Ctl
USP1
WT
siCTL
sip21
50
si-CTL + - + - + si-p21 - + - + - +
40
% Cells in S-phase
WB: p53
WB: ID1
WB: ID2
WB: ID3
G1
G2/M
shCTL shUSP1
ID1, ID2, ID3 - - + - - +

shRes USP1 - + - - + CTL + - - + - -
WB: ID2
20
WB: ID3
10
WB: USP1
shCTL
shUSP1
shID1,
ID2, ID3
WB: Actin
1
CTL
USP1
ID1,2,3
30
% Cells in S-phase
USP1
C90S
WB: ID1
30
WB: Actin
2
20
WB: p21
WB: USP1
30
10
WB: p21
40
sh
sh
shID1,
CTL USP1 ID2, ID3
shCTL
shUSP1-A
shUSP1-B
60
50
20
1
C
shCTL
shUSP1-A
shUSP1-B
% Cell Cycle Distribution
20
10
shCTL
shUSP1
Figure 4. USP1 Regulates Cell Cycling via ID Proteins in Osteosarcoma

(A) Western blot (WB) analysis of U2-OS cells treated as in Figure 3A.
(B) Outgrowth of U2-OS cells treated as in (A) was enumerated after 5 days of culture.
(C) Cell cycle status of propidium iodide-stained U2-OS cells treated as in (A).
(D) U2-OS cells transfected with indicated shRNAs and control or CDKN1A/p21 siRNAs.
(E) Quantification of cells in S phase in cells treated as in (D).
(F) U2-OS cells transfected with indicated shRNAs and shRNA-resistant USP1 (shRes USP1), ID1, ID2, and ID3, or control expression vectors.
(G) Quantification of cells in S phase in U2-OS cells treated as in (F).
Bars represent the mean SD of triplicate observations.
See also Figure S4.
knockdown. shRNA knockdown of IDs 13 individually did not

alter p21 levels, but combined knockdown of ID1, ID2, and ID3
increased p21 similar to USP1 knockdown (Figure S3H). IDand USP1-deficient cells also expressed comparable levels of
CDKN1A mRNA (Figure S3I). Consistent with these observations, ID deficiency caused cell-cycle arrest similar to USP1 deficiency (Figures S3J and S3K), and this was rescued by p21
knockdown (Figures 4D and 4E).
CDKN1A is regulated by many transcription factors, including
p53, which is activated in response to DNA damage (Kastan
et al., 1991). p53 knockdown inhibited etoposide-induced p21
in U2-OS cells but did not block the increase in p21 protein
seen after USP1 knockdown (Figure S3L), supporting a p53independent mechanism of p21 induction. Because USP1 is reported to target PCNA and FANCD2 during DNA repair (Nijman
et al., 2005; Huang et al., 2006), we also checked if USP1 knockdown produced DNA damage. H2AX phosphorylation that is
associated with DNA damage (Rogakou et al., 1999) increased
after etoposide treatment but not USP1 knockdown (data not
shown). These observations, combined with the ability of USP1
shRNAs to arrest p53-deficient SAOS cells (Figure S3D), exclude
general DNA damage, and p53 in particular, as intermediaries in
p21 induction following USP1 knockdown.
We confirmed that USP1 regulates p21 expression and cell
cycling via the IDs by rescuing the effects of USP1 knockdown
in U2-OS cells with ectopic expression of ID1, ID2, and ID3. ID
expression in USP1-depleted cells inhibited p21 expression
(Figure 4F) and blocked cell-cycle arrest (Figure 4G). Taken
together, our results demonstrate that USP1 suppresses p21
via ID protein stabilization and inhibition of bHLH transcriptional
activity in osteosarcoma.
USP1 and ID Proteins Restrict Osteogenic Commitment
in Osteosarcoma
Osteosarcomas are heterogeneous tumors comprised of disorganized masses of osteoblasts, chondrocytes, and adipocytes.
These tumors are thought to develop from a mesenchymal stem
cell population that can give rise to all three lineages (Tang et al.,
2008). Accordingly, osteosarcoma cell lines fail to express classical osteoblast markers such as RUNX2, OSTERIX, SPARC/
OSTEONECTIN, and alkaline phosphatase (ALP) (Luo et al.,
2008). Osteosarcoma cell lines also express surface markers characteristic of mesenchymal stem cells, including CD90, CD105, and
CD106 (Di Fiore et al., 2009). In light of the role that IDs play in stem
cell maintenance and regulation of differentiation, we investigated
whether either USP1 or ID knockdown in osteosarcoma would
trigger osteoblastic differentiation. U2-OS cells transfected with
USP1 or ID shRNAs expressed less CD105, CD106, and CD90
relative to control cells, whereas all cells expressed equivalent
amounts of the unrelated surface marker CD144 (Figure S4A).
Similar results were observed with HOS, SJSA, and SAOS cell
lines. USP1 or ID knockdown also increased expression of osteoblastic RUNX2, OSTERIX, and OSTEONECTIN (Figure S4B), and
increased ALP activity (Figure S4C). Increased E-cadherin expression and reduced N-cadherin and fibronectin following USP1
knockdown in U2-OS cells indicated a reversal of the epithelial to
mesenchymal transition that accompanies the malignant state of
osteosarcoma (Figures 5A and 5B) (Thiery et al., 2009). Collectively,
these data suggest that ID protein stabilization by USP1 in osteosarcoma blocks a normal osteogenic differentiation program.
The potential of USP1 inhibition as a tumor differentiation
strategy was investigated in the 143B osteosarcoma xenograft
model. A doxycycline-induced USP1 shRNA suppressed USP1
expression and reduced ID1 and ID2 in the xenografts (Figure 5C
and Figure S4D). ID3 was not detectable in this setting (data
not shown). USP1 knockdown also reduced 143B tumor
growth (Figure 5D), promoted OSTEONECTIN, RUNX2, SPP1/
OSTEOPONTIN, OSTERIX, and BGLAP/OSTEOCALCIN expression (Figure 5E and Figure S4E), and enhanced ALP activity
(Figure 5F). Remarkably, four of ten USP1-deficient xenograft
tumors achieved stasis and differentiation in situ, displaying
markedly altered cellular morphology and accumulation of
acellular collagenous masses consistent with proto-ossification
(Figure 5G). The tumors that continued to proliferate showed
evidence of escape from knockdown, presumably due to loss
or silencing of the shRNA (Figure S4F). These data indicate
that reducing USP1 is sufficient to initiate an osteogenic differentiation program in osteosarcoma.
Dysregulated USP1 Expression Inhibits hMSC
Differentiation
Next, we determined if USP1 stabilization of the IDs contributes
to normal mesenchymal stem cell maintenance. USP1 was expressed in primary hMSCs but declined steadily as the cells
were cultured in conditions favoring osteoblastic differentiation
(Figure 6A). Consistent with a previous study (Peng et al., 2003),
ID1 and ID2 were induced transiently and then declined as well.
ID3 was not detected (data not shown). These data, together
with a study showing that misregulated ID expression inhibits
osteogenic differentiation (Peng et al., 2004), prompted us
to investigate whether USP1 overexpression disrupts hMSC
differentiation. hMSCs overexpressing USP1 and cultured in
osteogenic differentiation medium expressed abnormally high
levels of ID1 and ID2 (Figure 6B), exhibited low ALP activity (Figure 6C), showed minimal induction of RUNX2, OSTERIX, and
OSTEONECTIN (Figure 6D), and stained poorly with alizarin
red, which reveals mineral deposition that is a classic marker of
osteoblast activity (Figure 6E). These data imply that the hMSCs
overexpressing USP1 failed to differentiate. A similar differentiation defect was observed in hMSCs overexpressing ID2, whereas
hMSCs overexpressing USP1 C90S differentiated similarly to
control cells. Thus, the catalytic activity of USP1 was necessary
and ID stabilization sufficient to inhibit osteogenic differentiation.
Coincident with their apparent failure to differentiate, hMSCs
overexpressing USP1 or ID2 proliferated significantly in the presence of excess osteogenic differentiation factors (Figure 6F). In
contrast, proliferation of control hMSCs, or those expressing
USP1 C90S, slowed as they differentiated in culture. Collectively, our observations suggest that overexpression of USP1
or ID2 is sufficient to block osteoblastic differentiation, promote
retention of stem-like features, and render cells resistant to
differentiation cues.
USP1 Promotes Transformation and Tumor Formation
The ability of USP1 to inhibit mesenchymal stem cell differentiation and sustain proliferation of osteosarcoma cell lines
B
shCTL
WB: E-cadherin
C
-DOX
shUSP1
Fibronectin
+DOX
IHC: USP1
WB: N-cadherin
N-cadherin
WB: Fibronectin
IHC: ID2
WB: USP1
1400
Tumor Volume mm3
shUSP1
shCTL
E-cadherin
WB: ID2
-DOX
600
400
200
0
12 15 18
Days
16
2
1.5
12
0.5
- + - +
USP1 ID2
- + - + - + - + DOX
ON RX2 OSX OP
+DOX
2
ALP Activity
(nmol pNPP/g protein/hr)
mRNA levels (arbitrary units)
800
+DOX
1000
WB: GAPDH
-DOX
1200
H&E
1.5
0.5
Trichrome
shCTL shUSP1
Figure 5. USP1 Promotes Retention of Stem Cell Identity in Osteosarcoma

(A) Western blot (WB) analysis of U2-OS cells transfected with CTL or USP1 shRNAs.
(B) Cells in (A) were stained and analyzed by fluorescence microscopy.
(C) Immunohistochemical staining for USP1 or ID2 in xenografts of 143B cells with doxycycline (DOX)-inducible shUSP1.
(D) Quantification of tumor volume of 143B xenografts as described in (C). Bars represent the mean SD of ten xenografts.
(E and F) RT-PCR quantification of USP1, ID2, OSTEONECTIN (ON), RUNX2 (RX2), OSTERIX (OSX), and OSTEOPONTIN (OP) mRNA levels (E) and ALP activity (F)
from 143B xenografts in (C). Bars represent the mean SD of triplicate observations.
(G) Representative xenograft tumors from (C) were stained with hematoxylin and eosin (H&E) or trichrome stain.
Scale bars, 100 mm.
See also Figure S5.
suggested that USP1 might promote cell transformation. We

explored this possibility with NIH 3T3 cells stably transduced
with empty vector, ID2, USP1, or USP1 C90S. Wild-type USP1
increased expression of IDs 13 (Figure S5A) and caused
anchorage-independent cell proliferation in soft agar (Figure S5B), which is a classic hallmark of oncogenic transformation
(Hanahan and Weinberg, 2000). In contrast, cells transduced
with empty vector or USP1 C90S did not grow well in soft agar
(Figures S5B and S5C). Interestingly, USP1 produced larger
and more numerous colonies than ID2 (Figure S5C), suggesting
that stabilization of multiple ID proteins may be more transforming than ID2 overexpression alone.
Our in vitro observations were recapitulated in vivo when NIH
3T3 cells were implanted subcutaneously into C.B-17 SCID.bg
mice. Control cells and cells expressing USP1 C90S failed to
produce measurable tumors, whereas cells overexpressing
USP1 or ID2 produced measurable tumors as early as 7 days
postimplantation (Figure S5D). Gross visual inspection of tumors
at the study endpoint confirmed that cells overexpressing USP1

or ID2 produced aggressive malignancies (Figure S5E). Similar
results were observed in NCr nude mice.
We assessed the contribution of the IDs to NIH 3T3 cell
transformation by USP1 with Id1, Id2, and Id3 shRNAs. Suppression of IDs 13 (Figure S5F) blocked colony formation in soft
agar (Figure S5G), indicating that the IDs are essential for
USP1 transformation of NIH 3T3 cells.
USP1 Regulates Bone Development
Because USP1 overexpression impaired osteoblastic differentiation of mesenchymal precursors, whereas USP1 loss caused
osteoblastic differentiation of osteosarcoma cells, we investigated whether USP1 regulates normal bone development with
Usp1 gene-targeted mice. P12 Usp1 / mice were osteopenic
with defects in ossification of the cranial and long bones (Figure 7A). Underdeveloped sternal ribs likely contribute to the
lethal cyanotic respiratory failure in Usp1 / pups (Kim et al.,
0 1 3 6 9 Un
WB: USP1
WB: ID2
WB: GAPDH
WB: GAPDH
1
hMSC
hMSC +ODM
hMSC USP1 WT +ODM
hMSC USP1 C90S +ODM
hMSC ID2 +ODM
10
Runx2
Osterix
20
0
CTL
50
CTL
40
USP1 WT
USP1 C90S
CTL
USP1 USP1
WT
C90S
+ODM
ID2
hMSC
hMSC +ODM
hMSC USP1 WT +ODM
hMSC USP1 C90S +ODM
hMSC ID2 +ODM
30
20
10
1
0
40
60
F
ODM
15
20
+ + + +
WB: ID2
WB: ID1
25
ODM
WB: ID1
WB: USP1
CTL ID2 WT C90S
Cell Number (x103)
+ODM (days)
mRNA levels (arbitrary units)
USP1
ALP Activity
(nmoles pNPP/g protein/hr)
80
ID2
0
Osteonectin
3
6
Culture Period (Days)
Figure 6. USP1 and IDs Regulate Mesenchymal Stem Cell Differentiation

(A) Western blot (WB) analysis of hMSCs grown in osteogenic differentiation medium (ODM), or in nondifferentiating medium (Un).
(B) hMSCs were transduced with ID2, USP1 wild-type (WT), USP1 C90S, or empty vector (CTL) and cultured in ODM for 9 days.
(C and D) hMSCs in (B) were assessed for ALP activity (C) and OSTEONECTIN, RUNX2, and OSTERIX mRNA (D). Bars represent the mean SD of triplicate
observations.
(E) hMSCs in (B) stained with alizarin red to visualize calcium deposition. Scale bars, 100 mm.
(F) Enumeration of hMSCs in (B) after the indicated number of days of culture. Bars represent the mean SD of triplicate observations.
See also Figure S6.
2009). Bone mineral density and volume in Usp1 / neonates

and E18.5 embryos were much less than in wild-type littermates
(Figures 7B and 7C and Figures S6B and S6C). Neither FANCD2nor PCNA-deficient mice exhibit perinatal lethality (Parmar et al.,
2010; Roa et al., 2008), excluding destabilization of these USP1
substrates as the primary cause of the perinatal lethality associated with USP1 deficiency.
Usp1 / and Usp1+/+ femurs contained similar numbers of
resting, transitional, proliferating, and hypertrophic chondrocytes, but deposition of osteoid on emergent bone spicules
was diminished, suggesting reduced activity of osteoid-depositing osteoblasts (Figures S6DS6F). Consistent with a defect
in osteoblast function, serum levels of bone alkaline phosphatase (BALP), a marker of systemic osteoblast activity, were
reduced in Usp1 / E18.5 embryos (Figure 7E). USP1 deficiency
did not alter osteoclast abundance or activity (Figures S6GS6I),
excluding increased bone resorption in the Usp1 / mice. Significantly, and in keeping with observations made in osteosarcoma
and mesenchymal stem cell cultures, Usp1 / femoral metaphyses contained less ID1 and ID2 than their wild-type counterparts
(Figure 7D and Figure S6J). These data indicate that the USP1-ID
axis regulating differentiation in osteosarcoma is recapitulated
in normal skeletal development.
DISCUSSION
In this study we show that ID stabilization by USP1 sustains a
significant fraction of human osteosarcomas. USP1 was overexpressed frequently in primary osteosarcomas and osteosarcoma
cell lines (Figure 2), and by deubiquitinating the ID proteins
(Figures 1 and 3), inhibited bHLH-dependent expression of
CDKI p21 (Figure 4) resulting in unchecked cell proliferation
(Figure 5). USP1 overexpression not only was necessary for
the proliferation of several osteosarcoma cell lines, it also was
sufficient to prevent normal mesenchymal cell differentiation,
capturing the cells in a stem-like state (Figure 6). By contrast,
USP1 knockdown in osteosarcoma cell lines reduced expression of mesenchymal stem cell markers and initiated an osteogenic development program (Figure 5). USP1 deficiency in
mice impaired normal osteogenesis and resulted in pronounced
350
340
Minz. Vol. (mm3)
Usp1-/-
WT
BMD (mg/cm3)
330
320
310
300
290
280
10 mm
Usp1-/-
WT
E
BALP (U/L)
WB: ID2
1 mm
1
0.5
WT
Usp1-/-
12
10
WB: ID1
1 mm
1.5
Usp1-/-
WT
10 mm
2.5
WB: USP1
WB: WDR48
8
6
4
2
WB: GAPDH
1
WT
Usp1-/-
Figure 7. USP1 Is Required for Normal Skeletogenesis

(A) Microcomputed tomography of 12-day-old Usp1+/+ (WT) and Usp1 / mice (top) and femurs (bottom).
(B and C) Mean bone mineralized density (BMD) (B) and mineralized bone volume (Minz. Vol.) (C) of mice in (A). Bars represent the mean SD of four femurs of
each genotype.
(D) Western blot (WB) analysis of femoral metaphyses from E18.5 Usp1+/+ (WT) and Usp1 / mice.
(E) BALP in the sera of E18.5 Usp1+/+ (WT) and Usp1 / embryos. Bars represent the mean SD of four embryos of each genotype.
See also Figure S6.
osteopenia (Figure 7). Therefore, we posit that overexpressed

USP1 interferes with mesenchymal stem cell differentiation
and thereby fosters the development of malignant mesenchymal
cell populations.
USP1 Is an ID DUB Overexpressed in Osteosarcoma
Our screen for DUBs capable of stabilizing ID2 (Figure S1)
identified both USP1 and USP33, although USP33 was unable
to deubiquitinate ID2 (data not shown). USP33 binding ID2
may have precluded ID2 recognition by the proteasome and
prevented its degradation. Other DUBs that enhanced ID2
expression in our screen did not appear to interact with ID2
and must influence ID2 abundance indirectly. These DUBs may
upregulate ID gene expression, interfere with the ubiquitinconjugation machinery, or otherwise impair proteasome function. For example USP9X may upregulate ID2 gene expression
by deubiquitinating and stabilizing the transcription factor
SMAD4 (Dupont et al., 2009).
The mechanism responsible for USP1 overexpression in a
subset of osteosarcomas (Figure 2) is unclear. USP1 mRNA
and protein levels correlated strongly implying transcriptional
upregulation. Notably, recent CGH analyses found that the
USP1 locus 1p31.3 was amplified in 26%57% of osteosarcoma
tumors (Ozaki et al., 2003; Stock et al., 2000).
Figure S3). Thus, USP1 overexpression perturbs normal osteoblast differentiation, which is characterized by p53-independent
upregulation of multiple CDKIs (Funato et al., 2001; Kenner et al.,
2004; Matsumoto et al., 1998; Yan et al., 1997; Zhang et al.,
1997). CDKI function often is compromised in osteosarcomas;
CDKN2A/p16INK4a and CDKN2B/p15INK4b gene deletions are
common (Miller et al., 1996; Nielsen et al., 1998), as is gene
inactivation due to promoter methylation (Oh et al., 2006). In
contrast, CDK4, a target of CDKIs, is frequently overexpressed
in osteosarcoma due to gene amplification (Ozaki et al., 2003).
ID-mediated transcriptional repression of p21 represents an
additional oncogenic mechanism in osteosarcoma.
ID protein overexpression has been observed in various
human cancers but has been attributed largely to increased ID
transcription (Perk et al., 2005). For example ID2 is transcriptionally upregulated by the EWS-Ets translocation in Ewings
sarcoma (Nishimori et al., 2002), which is an osteoid tumor
bearing strong resemblance to osteosarcoma. Patients with a
disrupted copy of the RB1 gene are strongly sensitized to development of osteosarcoma (Friend et al., 1986), RB being able to
sequester and inactivate ID2 (Iavarone et al., 1994; Lasorella
et al., 2000). Our study reveals an additional mechanism by
which ID proteins and, in turn, CDKIs can be dysregulated in
osteosarcoma.
USP1 Promotes Proliferation via ID-Mediated

Repression of CDKI p21
We show that ID protein stabilization by USP1 disrupts bHLHdependent p21 expression in osteosarcoma (Figure 4 and
ID Proteins Modulate Osteogenic Development

of Mesenchymal Precursors
Our data also implicate ID proteins in normal osteogenic development. ID2 or USP1 overexpression in mesenchymal stem cells
inhibited osteogenic differentiation and promoted retention of

mesenchymal stem cell features (Figure 6). These findings
support a recent study describing a role for ID proteins in mesenchymal differentiation (Peng et al., 2004). Intriguingly, Id1/Id3
compound heterozygous mutant mice display calvarial defects
and reduced osteoblast outgrowth (Maeda et al., 2004), suggestive of a mesenchymal proliferation defect. It is unknown if
additional Id2 deficiency would exacerbate this phenotype due
to early lethality. Restricting Id gene deletion to the mesenchymal lineage may prove informative.
The osteopenia that occurs in Usp1 / mice is consistent with
the phenotype predicted by USP1 knockdown in osteosarcoma
and USP1 overexpression in primary hMSCs (Figure 7 and Figure S6). In each setting our data suggest the USP1-ID axis
inhibits lineage commitment. An independent USP1-deficient
mouse strain also demonstrated runting and perinatal lethality
(Kim et al., 2009). Mice lacking multiple Id genes die early in
embryogenesis (Lyden et al., 1999), which could indicate that
an additional DUB regulates ID protein stability in early development, or that other DUBs can compensate for the absence of
USP1.
Recent studies suggest that the bHLH proteins inhibited by
IDs 13 during osteogenesis may belong to the Hey/Hes family.
Hey1 overexpression promoted osteoblastic differentiation,
whereas Hey1 knockdown inhibited it (Sharff et al., 2009). Similarly, Hes1 overexpression promoted osteocommitment (Suh
et al., 2008). It is possible that multiple bHLH transcription
factors act in parallel to promote osteoblast development.
USP1 and ID proteins would be positioned to broadly restrain
bHLH-driven commitment signals engaged during differentiation
of mesenchymal stem cells. We propose that USP1 belongs to
an emerging set of caulo-oncogenes that promotes tumorigenesis through subversion of normal stem (Latin caulo) cell
biology.
A consequence of our findings, one that has significant
therapeutic ramifications, is that inhibition of USP1 protease
activity should institute a differentiation program in malignant
osteosarcoma leading to a precipitous decline in proliferative
capacity and potential reversal of the transformed phenotype.
Targeting USP1 would be expected to impact all USP1
substrates including FANCD2, but this may be beneficial
because defective DNA repair in tumor cells lacking a normal
p53 checkpoint is predicted to sensitize them to crosslinking
chemotherapeutic agents or PARP inhibitors (DAndrea, 2010).
Differentiation treatments for cancer, as evidenced by the spectacular success of arsenic as a differentiation therapy for acute
promyelocytic leukemia, provide an exciting option for the effective treatment of previously lethal cancers. Targeting USP1 may
provide such an opportunity for osteogenic sarcoma.
Cell Lines
Human 143B, 293T, HOS, MG-63, SAOS-2, SJSA, and U2-OS cells, murine
NIH 3T3 cells (ATCC), and primary human osteoblasts (PromoCell) were grown
in DMEM with 10% FBS (Sigma-Aldrich), 10 U/ml penicillin, and 10 mg/ml
streptomycin (GIBCO). Primary hMSCs derived from normal bone marrow
were cultured in hMSC medium (Lonza) or osteogenic differentiation medium
(Lonza) with 100 ng/ml BMP-9 (R&D Systems). Primary human osteosarcomas
were from CytoMix, LLC, and the CHTN. DT-40 cells from K. Patel (MRC-LMB)
were cultured in RPMI with 7% FBS and 3% chicken serum (GIBCO). Where
indicated, cells were treated with 10 mM MG-132 (Calbiochem) or 25 mg/ml
cycloheximide (Sigma-Aldrich). Complete experimental procedures are
available in the Extended Experimental Procedures.
Vectors
USP1 variants, ID1, ID2, and ID3 were in pRK2001 with or without a C-terminal
Flag epitope. A WDR48 expression vector was from OriGene. pMACS was
from Miltenyi Biotec. ID2 and USP1 variants also were subcloned into pQCXIP
(Clontech) or pHUSH.Lenti.Puro (Genentech). Knockdown experiments used
pRS-based shRNA vectors (OriGene) or the pTRIPZ shRNA system (Open
Biosystems).
Transfections
Osteosarcoma cells were transfected with the plasmids indicated in combination with the marker plasmid pMACS using FuGENE 6 (Sigma-Aldrich).
Transfected cells were sorted with MACS-select H-2Kk microbeads (Miltenyi
Biotec). Proteins were extracted in lysis buffer (1% NP-40, 120 mM NaCl,
50 mM Tris-HCl [pH 7.4], and 1 mM EDTA) containing protease inhibitor
cocktail 1 and phosphatase inhibitor cocktails 1 and 2 (Calbiochem). For
dual shRNA/siRNA experiments, cells transfected with plasmids in FuGENE
6 then underwent nucleofection. DT-40 cells were transfected by
nucleofection.
Antibodies
Rat 5E10 anti-USP1 and 9F10 anti-WDR48 antibodies (Genentech) were
raised against the C-terminal 100 amino acids of the human proteins. Other
antibodies recognized ID1, ID2, ID3, E47, and p53 (Santa Cruz Biotechnology),
GAPDH (Assay Designs), Flag, HA, tubulin, and actin (Sigma-Aldrich), and p21
(Cell Signaling Technologies). Immunoprecipitations included 10 mM MG-132
and used protein A/G agarose beads (Pierce). Abcam AB52093 anti-ID2
antibody was used for immunohistochemistry in Figures 2E and 2F.
Gene Expression
RNA was extracted with RNeasy kits (QIAGEN) and amplified with the QuantiTect SYBR Green RT-PCR system (QIAGEN). Primary tumor RNA data were
obtained from GeneLogic microarray analyses (Ocimum Biosolutions) with
HGU133P Affymetrix chips.
Flow Cytometry
Cells staining with FITC-conjugated anti-H-2Kk antibodies (Miltenyi Biotec)
were stained with propidium iodide (Sigma-Aldrich) in the presence of RNase
A (Sigma-Aldrich), and their DNA content was measured in a FACSCalibur
flow cytometer (BD Biosciences). Other antibodies used recognized CD90
(Chemicon), CD105 (R&D Systems), CD106 (SouthernBiotech), and CD144
(eBioscience).
Immunohistochemistry
Formalin-fixed, paraffin-embedded tissue sections were incubated in target
retrieval solution (Dako) at 99 C for 20 min and cooled to 74 C for 20 min.
Sections were blocked at room temperature in avidin/biotin blocking buffer
(Vector Labs) and then 3% BSA for 30 min. Staining with primary antibody
was at room temperature for 60 min. Sections were rinsed twice in Dako
wash buffer and then incubated in VECTASTAIN reagents (Vector Labs) for
30 min. Staining was revealed with peroxidase substrate buffer (Pierce), then
sections were counterstained with Mayers Hematoxylin. TRAP staining
(Sigma-Aldrich) was on 5 mm sections from formalin-fixed, paraffin-embedded
P12 mouse femurs. TRAP-positive osteoclasts were counted in ten fields.
Alizarin red staining was performed according to manufacturer instructions
(Ricca Chemical).
Immunofluorescence Staining
Stably transduced U2-OS cells treated with 3 mg/ml doxycycline (Clontech) for
14 days were fixed in 1% PFA in PBS and stained with antibodies recognizing
E-cadherin, N-cadherin (BD), or fibronectin (Calbiochem).
In Vivo Deubiquitination
Transfected 293T cells were cultured with 10 mM MG-132 for 30 min prior to
extraction in lysis buffer containing 10 mM MG-132 and 10 mM N-ethylmaleimide (Sigma-Aldrich). Soluble lysates were denatured with 1% SDS at 95 C
for 5 min, then diluted 20-fold in lysis buffer prior to immunoprecipitation
with M2 anti-Flag agarose (Sigma-Aldrich).
In Vitro Deubiquitination
293T cells were transfected with USP1-Flag, USP1-C90S-Flag, or ID2-Flag
and HA-ubiquitin. ID2-Flag was immunoprecipitated from SDS-denatured
cell lysate. USP1 was immunoprecipitated from regular cell lysate. Elutions
were with 500 mg/ml 3X Flag peptide (Sigma-Aldrich). Samples were combined
in DUB buffer (20 mM HEPES [pH 8.3], 20 mM NaCl, 100 mg/ml BSA, 500 mM
EDTA, 1 mM DTT) at room temperature.
Osteosarcoma Differentiation
U2-OS, HOS, or SAOS cells transfected serially with pRS.shUSP1 (or shCTL)
were selected based on H-2Kk expression and cultured for 12 days. 143B cells
transduced with pTRIPZ inducible USP1 shRNA were selected in puromycin.
ALP activity was determined in cell lysates lacking phosphatase inhibitors after
normalizing for protein content. p-nitrophenyl phosphatase cleavage was
analyzed by colorimetric assay (Sigma-Aldrich).
3T3 Transformation
Transduced 3T3 cells were plated in DMEM/0.5% low-melting agar on 1%
agar. Colonies of eight or more cells were scored visually after 21 days. For
ID knockdown, USP1-expressing 3T3 cells were transduced with pTRIPZ
inducible shRNAs and cultured with 3 mg/ml doxycycline for 72 hr prior to
embedding in agar containing 3 mg/ml doxycycline.
Mice
Eight-week-old female NCr nude mice (Taconic) or C.B-17 SCID.bg mice
(CRL) were injected subcutaneously in the right hind flank with 1 3 106 3T3
cells or 2.5 3 106 143B cells in 100 ml HBSS and provided 1 mg/ml doxycycline
in 5% sucrose water. Usp1 gene-targeted C57BL/6 ES cells were from the
Knockout Mouse Project Repository (Davis, CA). Microcomputed tomography
was a mCT 40 (SCANCO Medical). Amniotic fluid from E18.5 embryos was
assayed for deoxypyridinoline (TSZ ELISA), BALP (EIAab & USCNLIFE), and
creatinine (R&D Systems). The Genentech Institutional Animal Care and Use
Committee approved all animal studies.
ACCESSION NUMBERS
mRNA expression data of USP1 were extracted from the Gene Logic expression database of Affymetrix Human Genome U133 Plus 2.0 data. The probe set
202412_s_at was chosen to represent the expression of USP1.
REFERENCES
Bounpheng, M.A., Dimas, J.J., Dodds, S.G., and Christy, B.A. (1999).
Degradation of Id proteins by the ubiquitin-proteasome pathway. FASEB J.
13, 22572264.
Ciarapica, R., Annibali, D., Raimondi, L., Savino, M., Nasi, S., and Rota, R.
(2009). Targeting Id protein interactions by an engineered HLH domain induces
human neuroblastoma cell differentiation. Oncogene 28, 18811891.
Cohn, M.A., Kowal, P., Yang, K., Haas, W., Huang, T.T., Gygi, S.P., and
DAndrea, A.D. (2007). A UAF1-containing multisubunit protein complex
regulates the Fanconi anemia pathway. Mol. Cell 28, 786797.
DAndrea, A.D. (2010). Susceptibility pathways in Fanconis anemia and breast
cancer. N. Engl. J. Med. 362, 19091919.
Di Fiore, R., Santulli, A., Ferrante, R.D., Giuliano, M., De Blasio, A., Messina, C.,
Pirozzi, G., Tirino, V., Tesoriere, G., and Vento, R. (2009). Identification and
expansion of human osteosarcoma-cancer-stem cells by long-term 3-aminobenzamide treatment. J. Cell. Physiol. 219, 301313.
Dupont, S., Mamidi, A., Cordenonsi, M., Montagner, M., Zacchigna, L.,
Adorno, M., Martello, G., Stinchfield, M.J., Soligo, S., Morsut, L., et al.
(2009). FAM/USP9x, a deubiquitinating enzyme essential for TGFbeta
signaling, controls Smad4 monoubiquitination. Cell 136, 123135.
Friend, S.H., Bernards, R., Rogelj, S., Weinberg, R.A., Rapaport, J.M., Albert,
D.M., and Dryja, T.P. (1986). A human DNA segment with properties of the
gene that predisposes to retinoblastoma and osteosarcoma. Nature 323,
643646.
Funato, N., Ohtani, K., Ohyama, K., Kuroda, T., and Nakamura, M. (2001).
Common regulation of growth arrest and differentiation of osteoblasts by
helix-loop-helix factors. Mol. Cell. Biol. 21, 74167428.
Hanahan, D., and Weinberg, R.A. (2000). The hallmarks of cancer. Cell 100,
5770.
Huang, T.T., Nijman, S.M., Mirchandani, K.D., Galardy, P.J., Cohn, M.A., Haas,
W., Gygi, S.P., Ploegh, H.L., Bernards, R., and DAndrea, A.D. (2006).
Regulation of monoubiquitinated PCNA by DUB autocleavage. Nat. Cell
Biol. 8, 339347.
Iavarone, A., Garg, P., Lasorella, A., Hsu, J., and Israel, M.A. (1994). The
helix-loop-helix protein Id-2 enhances cell proliferation and binds to the
retinoblastoma protein. Genes Dev. 8, 12701284.
Kastan, M.B., Onyekwere, O., Sidransky, D., Vogelstein, B., and Craig, R.W.
(1991). Participation of p53 protein in the cellular response to DNA damage.
Cancer Res. 51, 63046311.
Kenner, L., Hoebertz, A., Beil, T., Keon, N., Karreth, F., Eferl, R., Scheuch, H.,
Szremska, A., Amling, M., Schorpp-Kistner, M., et al. (2004). Mice lacking
JunB are osteopenic due to cell-autonomous osteoblast and osteoclast
defects. J. Cell Biol. 164, 613623.
Kim, D., Peng, X.C., and Sun, X.H. (1999). Massive apoptosis of thymocytes in
T-cell-deficient Id1 transgenic mice. Mol. Cell. Biol. 19, 82408253.
Kim, J.M., Parmar, K., Huang, M., Weinstock, D.M., Ruit, C.A., Kutok, J.L., and
DAndrea, A.D. (2009). Inactivation of murine Usp1 results in genomic instability and a Fanconi anemia phenotype. Dev. Cell 16, 314320.

six figures, and one table and can be found with this article online at
doi:10.1016/j.cell.2011.07.040.
Lasorella, A., Noseda, M., Beyna, M., Yokota, Y., and Iavarone, A. (2000). Id2 is
a retinoblastoma protein target and mediates signalling by Myc oncoproteins.
Nature 407, 592598.
ACKNOWLEDGMENTS
Lasorella, A., Uo, T., and Iavarone, A. (2001). Id proteins at the cross-road of
development and cancer. Oncogene 20, 83268333.
We thank Kim Newton for editorial assistance. All authors were employees of
Genentech, Inc.
Lasorella, A., Stegmuller, J., Guardavaccaro, D., Liu, G., Carro, M.S.,
Rothschild, G., de la Torre-Ubieta, L., Pagano, M., Bonni, A., and Iavarone,
A. (2006). Degradation of Id2 by the anaphase-promoting complex couples
cell cycle exit and axonal growth. Nature 442, 471474.
Received: December 7, 2010

Revised: May 10, 2011
Lassar, A.B., Davis, R.L., Wright, W.E., Kadesch, T., Murre, C., Voronova, A.,
Baltimore, D., and Weintraub, H. (1991). Functional activity of myogenic HLH
proteins requires hetero-oligomerization with E12/E47-like proteins in vivo.
Cell 66, 305315.
Luo, X., Chen, J., Song, W.X., Tang, N., Luo, J., Deng, Z.L., Sharff, K.A., He, G.,
Bi, Y., He, B.C., et al. (2008). Osteogenic BMPs promote tumor growth of
human osteosarcomas that harbor differentiation defects. Lab. Invest. 88,
12641277.
Peng, Y., Kang, Q., Luo, Q., Jiang, W., Si, W., Liu, B.A., Luu, H.H., Park, J.K., Li,
X., Luo, J., et al. (2004). Inhibitor of DNA binding/differentiation helix-loop-helix
proteins mediate bone morphogenetic protein-induced osteoblast differentiation of mesenchymal stem cells. J. Biol. Chem. 279, 3294132949.
Lyden, D., Young, A.Z., Zagzag, D., Yan, W., Gerald, W., OReilly, R., Bader,
B.L., Hynes, R.O., Zhuang, Y., Manova, K., and Benezra, R. (1999). Id1 and
Id3 are required for neurogenesis, angiogenesis and vascularization of tumour
xenografts. Nature 401, 670677.
Perk, J., Iavarone, A., and Benezra, R. (2005). Id family of helix-loop-helix

proteins in cancer. Nat. Rev. Cancer 5, 603614.
Maeda, Y., Tsuji, K., Nifuji, A., and Noda, M. (2004). Inhibitory helix-loop-helix
transcription factors Id1/Id3 promote bone formation in vivo. J. Cell. Biochem.
93, 337344.
Massari, M.E., and Murre, C. (2000). Helix-loop-helix proteins: regulators of
transcription in eucaryotic organisms. Mol. Cell. Biol. 20, 429440.
Polyak, K., Waldman, T., He, T.C., Kinzler, K.W., and Vogelstein, B. (1996).
Genetic determinants of p53-induced apoptosis and growth arrest. Genes
Dev. 10, 19451952.
Prabhu, S., Ignatova, A., Park, S.T., and Sun, X.H. (1997). Regulation of the
expression of cyclin-dependent kinase inhibitor p21 by E2A and Id proteins.
Mol. Cell. Biol. 17, 58885896.
Matsumoto, T., Sowa, Y., Ohtani-Fujita, N., Tamaki, T., Takenaka, T.,
Kuribayashi, K., and Sakai, T. (1998). p53-independent induction of WAF1/
Cip1 is correlated with osteoblastic differentiation by vitamin D3. Cancer
Lett. 129, 6168.
Roa, S., Avdievich, E., Peled, J.U., Maccarthy, T., Werling, U., Kuang, F.L.,
Kan, R., Zhao, C., Bergman, A., Cohen, P.E., et al. (2008). Ubiquitylated
PCNA plays a role in somatic hypermutation and class-switch recombination
and is required for meiotic progression. Proc. Natl. Acad. Sci. USA 105,
1624816253.
Miller, C.W., Aslo, A., Campbell, M.J., Kawamata, N., Lampkin, B.C., and
Koeffler, H.P. (1996). Alterations of the p15, p16,and p18 genes in osteosarcoma. Cancer Genet. Cytogenet. 86, 136142.
Rogakou, E.P., Boon, C., Redon, C., and Bonner, W.M. (1999). Megabase
chromatin domains involved in DNA double-strand breaks in vivo. J. Cell
Biol. 146, 905916.
Nielsen, G.P., Burns, K.L., Rosenberg, A.E., and Louis, D.N. (1998). CDKN2A
gene deletions and loss of p16 expression occur in osteosarcomas that lack
RB alterations. Am. J. Pathol. 153, 159163.
Sharff, K.A., Song, W.X., Luo, X., Tang, N., Luo, J., Chen, J., Bi, Y., He, B.C.,
Huang, J., Li, X., et al. (2009). Hey1 basic helix-loop-helix protein plays an
important role in mediating BMP9-induced osteogenic differentiation of
mesenchymal progenitor cells. J. Biol. Chem. 284, 649659.
Nijman, S.M., Huang, T.T., Dirac, A.M., Brummelkamp, T.R., Kerkhoven, R.M.,
DAndrea, A.D., and Bernards, R. (2005). The deubiquitinating enzyme USP1
regulates the Fanconi anemia pathway. Mol. Cell 17, 331339.
Nishimori, H., Sasaki, Y., Yoshida, K., Irifune, H., Zembutsu, H., Tanaka, T.,
Aoyama, T., Hosaka, T., Kawaguchi, S., Wada, T., et al. (2002). The Id2 gene
is a novel target of transcriptional activation by EWS-ETS fusion proteins in
Ewing family tumors. Oncogene 21, 83028309.
Oestergaard, V.H., Langevin, F., Kuiken, H.J., Pace, P., Niedzwiedz, W.,
Simpson, L.J., Ohzeki, M., Takata, M., Sale, J.E., and Patel, K.J. (2007).
Deubiquitination of FANCD2 is required for DNA crosslink repair. Mol. Cell
28, 798809.
Oh, J.H., Kim, H.S., Kim, H.H., Kim, W.H., and Lee, S.H. (2006). Aberrant
methylation of p14ARF gene correlates with poor survival in osteosarcoma.
Clin. Orthop. Relat. Res. 442, 216222.
Ozaki, T., Neumann, T., Wai, D., Schafer, K.L., van Valen, F., Lindner, N.,
Scheel, C., Bocker, W., Winkelmann, W., Dockhorn-Dworniczak, B., et al.
(2003). Chromosomal alterations in osteosarcoma cell lines revealed by
comparative genomic hybridization and multicolor karyotyping. Cancer Genet.
Cytogenet. 140, 145152.
Palombella, V.J., Rando, O.J., Goldberg, A.L., and Maniatis, T. (1994). The
ubiquitin-proteasome pathway is required for processing the NF-kappa B1
precursor protein and the activation of NF-kappa B. Cell 78, 773785.
Parmar, K., Kim, J., Sykes, S.M., Shimamura, A., Stuckert, P., Zhu, K.,
Hamilton, A., Deloach, M.K., Kutok, J.L., Akashi, K., et al. (2010). Hematopoietic stem cell defects in mice with deficiency of Fancd2 or Usp1. Stem Cells 28,
11861195.
Peng, Y., Kang, Q., Cheng, H., Li, X., Sun, M.H., Jiang, W., Luu, H.H., Park,
J.Y., Haydon, R.C., and He, T.C. (2003). Transcriptional characterization of
bone morphogenetic proteins (BMPs)-mediated osteogenic signaling. J.
Cell. Biochem. 90, 11491165.
Soignet, S.L., Maslak, P., Wang, Z.G., Jhanwar, S., Calleja, E., Dardashti, L.J.,
Corso, D., DeBlasio, A., Gabrilove, J., Scheinberg, D.A., et al. (1998). Complete
remission after treatment of acute promyelocytic leukemia with arsenic
trioxide. N. Engl. J. Med. 339, 13411348.
Stock, C., Kager, L., Fink, F.M., Gadner, H., and Ambros, P.F. (2000). Chromosomal regions involved in the pathogenesis of osteosarcomas. Genes
Chromosomes Cancer 28, 329336.
Suh, J.H., Lee, H.W., Lee, J.W., and Kim, J.B. (2008). Hes1 stimulates
transcriptional activity of Runx2 by increasing protein stabilization during
osteoblast differentiation. Biochem. Biophys. Res. Commun. 367, 97102.
Tang, N., Song, W.X., Luo, J., Haydon, R.C., and He, T.C. (2008). Osteosarcoma development and stem cell differentiation. Clin. Orthop. Relat. Res.
466, 21142130.
Thiery, J.P., Acloque, H., Huang, R.Y., and Nieto, M.A. (2009). Epithelialmesenchymal transitions in development and disease. Cell 139, 871890.
Tupler, R., Perini, G., and Green, M.R. (2001). Expressing the human genome.
Nature 409, 832833.
Weintraub, H., Genetta, T., and Kadesch, T. (1994). Tissue-specific gene activation by MyoD: determination of specificity by cis-acting repression
elements. Genes Dev. 8, 22032211.
Yan, Y., Frisen, J., Lee, M.H., Massague, J., and Barbacid, M. (1997). Ablation
of the CDK inhibitor p57Kip2 results in increased apoptosis and delayed
differentiation during mouse development. Genes Dev. 11, 973983.
Yokota, Y., and Mori, S. (2002). Role of Id family proteins in growth control. J.
Cell. Physiol. 190, 2128.
Zhang, P., Liegeois, N.J., Wong, C., Finegold, M., Hou, H., Thompson, J.C.,
Silverman, A., Harper, J.W., DePinho, R.A., and Elledge, S.J. (1997). Altered
cell differentiation and proliferation in mice lacking p57KIP2 indicates a role
in Beckwith-Wiedemann syndrome. Nature 387, 151158.
Selective Bypass of a Lagging Strand

Roadblock by the Eukaryotic
Replicative DNA Helicase
Yu V. Fu,1 Hasan Yardimci,1 David T. Long,1,7 The Vinh Ho2,4,7Angelo Guainazzi,2,5,7 Vladimir P. Bermudez,3
Jerard Hurwitz,3 Antoine van Oijen,1,6 Orlando D. Scharer,2 and Johannes C. Walter1,*
1Department
of Biological Chemistry and Molecular Pharmacology, Harvard Medical School, Boston, MA 02115, USA
of Pharmacological Sciences and Chemistry, Stony Brook University, Stony Brook, NY 11794, USA
3Program of Molecular Biology, Memorial Sloan Kettering Cancer Center, 1275 York Avenue, New York, NY 10021
4Present address: Tecan Group Ltd., 8708 Ma
nnedorf, Switzerland
5Present address: Helsinn Therapeutics (U.S.) Inc., Bridgewater, NJ 08807, USA
6Present address: The Zernike Institute for Advanced Materials, University of Groningen, 9747 AG Groningen, The Netherlands
*Correspondence: johannes_walter@hms.harvard.edu
DOI 10.1016/j.cell.2011.07.045
2Department
SUMMARY
The eukaryotic replicative DNA helicase, CMG, unwinds DNA by an unknown mechanism. In some
models, CMG encircles and translocates along one
strand of DNA while excluding the other strand. In
others, CMG encircles and translocates along duplex
DNA. To distinguish between these models, replisomes were confronted with strand-specific DNA
roadblocks in Xenopus egg extracts. An ssDNA translocase should stall at an obstruction on the translocation strand but not the excluded strand, whereas
a dsDNA translocase should stall at obstructions on
either strand. We found that replisomes bypass large
roadblocks on the lagging strand template much
more readily than on the leading strand template.
Our results indicate that CMG is a 30 to 50 ssDNA translocase, consistent with unwinding via steric exclusion. Given that MCM2-7 encircles dsDNA in G1, the
data imply that formation of CMG in S phase involves
remodeling of MCM2-7 from a dsDNA to a ssDNA
binding mode.
INTRODUCTION
In eukaryotic cells, DNA replication initiates at many chromosomal locations called origins. At each origin, two sister replisomes are assembled that move away from the origin in opposite
directions. An essential component of the replisome is the replicative DNA helicase, which unwinds parental DNA, generating
substrates for leading and lagging strand DNA polymerases.
Current evidence indicates that the eukaryotic replicative DNA
helicase contains at least three components, a heterohexameric
ATPase called MCM2-7 and two cofactors, Cdc45 and GINS
(Bochman and Schwacha, 2009; Ilves et al., 2010; Moyer et al.,
2006; Pacek et al., 2006; Takahashi et al., 2005). The complex

of Cdc45, MCM2-7, and GINS is called CMG (Moyer et al., 2006).
A striking feature of eukaryotic DNA replication is the intricate,
bi-phasic assembly of CMG, which underlies the cell cycle regulation of DNA replication (reviewed in Arias and Walter, 2007; Labib, 2010). CMG assembly is best understood in yeast. In the G1
phase, when cyclin-dependent kinase (CDK) activity is low, ORC,
Cdc6, and Cdt1 recruit MCM2-7 onto origins of DNA replication,
forming a prereplication complex (pre-RC). Within the pre-RC,
MCM2-7 complexes bind to double-stranded DNA (dsDNA) as
inactive double hexamers (Evrin et al., 2009; Gambus et al.,
2011; Remus et al., 2009) At the G1/S transition, when CDK
activity rises, numerous additional factors cooperate to convert
the MCM2-7 double hexamer into two CMG complexes. In
particular, Cdc7-Dbf4 protein kinase (DDK) phosphorylates
MCM2-7 (Randell et al., 2010 and references therein). CDK phosphorylates Sld2 and Sld3 (Treslin/Ticrr in metazoans), promoting
their interaction with Dpb11 (TopBP1 in metazoans). The Sld3Sld2-Dpb11 complex enables the stable binding of Cdc45 and
GINS to phosphorylated MCM2-7. Once formed, CMG unwinds
the origin, allowing replisome assembly. Importantly, the high
CDK activity present in the S and G2 phases inhibits the functions
of ORC, Cdc6, and Cdt1, such that de novo MCM2-7 recruitment
is blocked. As a result, CMG complexes that leave the origin
during the first initiation event cannot be replaced, and each
origin fires only once. Although the regulation of CMG assembly
in metazoans is very similar, it appears to be even more complicated, depending on several additional factors, including
Mcm9, DUE-B, GEMC1, Geminin, and CRL4Cdt2.
A crucial question in the field is why so many proteins are
required to convert bound MCM2-7 complexes into the active
CMG helicase. This question is inextricably connected to another open question, which is how CMG unwinds DNA in S
phase. Only by describing the final disposition of CMG on DNA
will it be possible to understand how it is assembled and activated. MCM2-7, the core molecular motor of CMG, is composed
of six related AAA+ ATPase subunits, Mcm2-Mcm7 (Tye, 1999).
Thus, GINS and Cdc45 have been proposed to bind to the

outside of the MCM2-7 complex and to regulate its ATPase
activity (Ilves et al., 2010). Crystallography and electron microscopy reveal that eukaryotic and archaeal MCM complexes form
ring-shaped hexamers whose central channel can accommodate single-stranded DNA (ssDNA) or dsDNA (Evrin et al.,
2009; Fletcher et al., 2003; Pape et al., 2003; Remus et al.,
2009). Thus, DNA unwinding likely involves one or both strands
of the DNA passing through the MCM central channel.
Most well-studied replicative DNA helicases, such as E. coli
DnaB and Bacteriophage T7 gene product 4 (T7 gp4), function
as single hexamers that encircle and translocate along one
strand of DNA while excluding the other strand (steric exclusion). In biochemical experiments using purified proteins,
archaeal and eukaryotic MCM proteins including the CMG translocated along ssDNA in the 30 to 50 direction (Bochman and
Schwacha, 2008; Ilves et al., 2010; Kelman et al., 1999; Moyer
et al., 2006; Shechter et al., 2000). Using elegant manipulation
of DNA templates, steric exclusion by a purified Mcm4/6/7 subcomplex was directly demonstrated (Kaplan et al., 2003). However, the physiological relevance of these studies is not clear
since the reactions bypassed the endogenous helicase loading
and activation pathway.
More recently, several considerations led to speculation that
CMG might unwind DNA by translocating along dsDNA (Laskey
and Madine, 2003; Mendez and Stillman, 2003; Takahashi et al.,
2005). First, Mcm4/6/7 can translocate along dsDNA with
considerable force (Kaplan et al., 2003). Second, MCM2-7 interacts with dsDNA as part of the pre-RC (Evrin et al., 2009; Remus
et al., 2009). It is therefore attractive to propose that MCM2-7
does not fundamentally change its interaction with DNA upon
assembly into the CMG. Several models propose how CMG
might unwind DNA while translocating along dsDNA. One idea
is that CMG functions as an obligate dimer that pumps dsDNA
toward its dimer interface, and that single-stranded DNA is
extruded through lateral channels, analogous to a mechanism
proposed for SV40 large T-antigen (Li et al., 2003; Wessel
et al., 1992). However, there is no need for helicase dimerization
during eukaryotic DNA replication, arguing against this model
(Yardimci et al., 2010). Alternatively, a single CMG complex
motors along dsDNA. As DNA emerges from the rear of CMGs
central channel, a proteinaceous pin or plougshare bisects
the duplex, leading to strand separation (Takahashi et al.,
2005). In a variation of this idea, ssDNA exits through side channels midway along the longitudinal axis of MCM (Brewster et al.,
2008). In all of these models, DNA enters the MCM central
channel as a duplex.
To understand how the replicative helicase interacts with
DNA in S phase, we examined the collision of replisomes with
various DNA roadblocks in Xenopus egg extracts. We previously
showed that when two replisomes converge on a DNA interstrand crosslink (ICL), the 30 ends of the leading strands initially arrest 20-40 nucleotides (nt) from the ICL (Raschle et al.,
2008). Here, we provide evidence that the footprint of CMG on
DNA underlies these distal arrest points. We had also speculated
(Raschle et al., 2008) that the wide distribution of arrest points
might suggest that CMG translocates along dsDNA, since the
first replisome to arrive at the lesion could engulf the ICL and
impose a more distal arrest on the second replisome. However,

we show here that blocking the arrival of one replisome using
biotin-streptavidin (biotin-SA) roadblocks has no effect on the
leading strand arrest points of the other converging replisome.
To directly examine the CMG translocation mode, we employed
strand-specific roadblocks. We reasoned that if CMG translocates along dsDNA, it should be arrested by a bulky roadblock
on either the leading or lagging strand templates. In contrast, if
it moves along ssDNA in the 30 to 50 direction, it should be arrested by a roadblock on the leading strand template but not
on the excluded, lagging strand template. We found that the
nascent leading strand stalls 30 nucleotides from a biotin-SA
complex on the leading strand template, consistent with CMG
stalling, whereas a biotin-SA complex on the lagging strand
template induced little arrest. Similar results were obtained in
a single-molecule assay using strand-specific quantum dot
(QDot) roadblocks. Together with previous biochemical analyses, our data strongly suggest that CMG unwinds DNA by
translocating along ssDNA in the 30 to 50 direction, consistent
with DNA unwinding by steric exclusion. The data imply a series
of discrete molecular events that underlie the conversion of
MCM2-7 to an active CMG in S phase.
RESULTS
To understand how the eukaryotic replicative DNA helicase is
configured on DNA in S phase, we examined how the replisome interacts with specific DNA lesions. To this end, we employed nucleus-free Xenopus egg extracts (Walter et al., 1998),
which support efficient DNA replication of plasmids or l DNA.
In this system, DNA is first incubated with a high speed supernatant of egg cytoplasm (HSS), which chromatinizes the
template and also promotes sequence nonspecific MCM2-7
recruitment to the DNA by ORC, Cdc6, and Cdt1 (Arias and
Walter, 2004). Subsequently, a concentrated nucleoplasmic
extract (NPE) is added, which supports CMG assembly dependent on Cdc7-Drf1, Cdk2-Cyclin E, Mcm10, and Cdc45 (Takahashi and Walter, 2005; Walter and Newport, 2000; Wohlschlegel et al., 2002). The Xenopus CMG travels with the replisome
and unwinds DNA throughout S phase (Pacek et al., 2006;
Pacek and Walter, 2004). These observations indicate that
nucleus-free Xenopus egg extracts promote activation of the
replicative DNA helicase by the same events that occur in
cells.
CMG Binding Correlates with Leading Strand Arrest 20
Nucleotides from a DNA Interstrand Crosslink
We first sought to detect the CMG footprint on DNA in S phase.
To this end, we examined the collision of the replisome with two
different lesions: a DNA inter-strand crosslink, which should
arrest CMG, and a DNA intra-strand crosslink, which should stall
the DNA polymerase but not CMG. Plasmids containing a sitespecific cisplatin inter-strand crosslink (pICLInter; Figure 1A, red
line) or 1,2 cisplatin intra-strand crosslink (pICLIntra; Figure 1B,
red bracket) were incubated sequentially in HSS and NPE containing [a32P]dATP. At different times after NPE addition,
DNA was extracted and digested with StuI, which cuts the
plasmid once 268 nt to the right of both lesions (Figure 1A and
pICLInter
Stu I (-268)
0
10
20
30
5-CCCTCTTCCGCTCTTCTTTCGTGCGCGGCCGCGATCCGCTGCATT
GAGAAGAAAGCACGCGCCGGCGCTAGGCGACGTAA
AGGCCT-3
TCCGGA
Leftward Leading Strand
3-GGGAGAAGGCGAGAAGAAAGCACGCGCCGGCGCTAGGCGACGTAA
pICLIntra
10
20
TCCGGA-5
Stu I (-268)
30
5-CCCTCTCCTTGGTTCTTCTCGTGCGCGGCCGCGATCCGCTGCATT
CAAGAAGAGCACGCGCCGGCGCTAGGCGACGTAA
AGGCCT-3
TCCGGA
Leftward Leading Strand
3-GGGAGAGGAACCAAGAAGAGCACGCGCCGGCGCTAGGCGACGTAA
pICLInter
pICLIntra
Primer M
Time (min): 0 10 15 20 30 60 90 120 0 10 15 20 30 60 90 120 G A T
StuI Digest - Leftward Fork
TCCGGA-5
0
-1
10
20
-20
30
40
-40
50
6 7
9 10 11 12 13 14 15 16
17 18 19 20
Mcm7 ChIP Control Region
Relative amount
Mcm7 ChIP ICL Region

120
-20 Arrest
100
-1 Arrest
Figure 1. CMG Causes Leading Strand

Stalling 20 Nucleotides from a DNA Interstrand Crosslink
(A) Cartoon depicting the DNA sequence surrounding the ICL in pICLInter. Red line, inter-strand
crosslink. Blue arrow, StuI cleavage site, which is
used to map leftward leading strands in (C).
The sequence of the longest leading strand detected at the 10 min time point (see [C]) is shown in
red letters, and the product generated after the
leading strand advances toward the ICL after
15 min (see [C]) is shown in green letters. Blue
letters, sequence differences between pICLInter
and pICLIntra.
(B) Same as (A), except for pICLIntra. Red bracket,
1,2 intra-strand crosslink. The sequence of the
longest leading strand seen in (C) is shown in red
letters.
(C) Mapping leading strands near DNA inter- and
intrastrand crosslinks. pICLInter (lanes 1-8) or
pICLIntra (lanes 916) was incubated sequentially
in HSS and NPE containing [a32P]dATP. At the
indicated times after NPE addition, replication
intermediates were digested with StuI, and separated on a DNA sequencing gel alongside a
sequencing ladder generated with primer M (see
Figure S1A). The distance of sequencing products
from the bold G in panel (A) and T in panel (B) is
indicated on the right.
(D) Kinetics of Mcm7 binding to an ICL. pICLInter
was replicated as in (C) but lacking radioactivity,
and samples were withdrawn for Mcm7 ChIP
using ICL proximal (pink) and control (purple)
primer pairs (see plasmid cartoon). The relative
ChIP signal adjacent to the ICL (pink circles) and
distal to the ICL (purple triangles) was plotted. In
parallel reactions containing [a32P]dATP, replication intermediates were digested with AflIII, and
separated on a DNA sequencing gel alongside
a sequencing ladder generated with primer S (see
Figure S1A). The leading strands stalled between
20 and 40 and at the 1 position (see Figure S1B) were quantified and plotted (blue diamonds and gray squares). Error bars represent the
standard deviation of three experiments.
strands initially stalled 20-40 nucleotides

from the ICL (Figure 1C, lane 2, and FigICL
ICL
ure 1A, red strand)(Raschle et al., 2008).
60
Region
Subsequently, DNA synthesis resumed
40
and the leading strand stalled again 1 nt
from the ICL (Figure 1C, green arrow
20
and Figure 1A, green strand) (Raschle
0
et al., 2008). In contrast, during replicaControl
0
20
40
60
80
100
120
tion of pICLIntra, the 30 end of the leading
Region
strand immediately advanced all the
Time (min)
way to the lesion, where it stalled (Figure 1C, lanes 1016; Figure 1B, red
strand). These data imply that leading
Figure S1A available online). Leftward leading strands were strands arrest 20-40 nt from an inter-strand crosslink due to
visualized on a sequencing gel alongside an appropriate the footprint of the replicative DNA helicase, which we denote
sequencing ladder. On pICLInter, the 30 ends of the leading as CMG.
80
To test this idea, we performed chromatin immunoprecipitation (ChIP) for Mcm7, a CMG subunit. pICLInter was replicated,
and at different times, the reaction was crosslinked with formaldehyde, sonicated, immunoprecipitated with Mcm7 antibody,
and the recovered DNA amplified with ICL-proximal and control
primer pairs (Figure 1D). ChIP revealed that soon after replication
began, Mcm7 was depleted from the control region, as expected if CMG travels toward the ICL while displacing any latent
MCM2-7 complexes (Figure 1D, purple triangles). In contrast,
Mcm7 initially accumulated at the ICL, concurrent with the arrival
of leading strands at the 20 position (Figure 1D, compare pink
circles and blue diamonds). The subsequent disappearance of
Mcm7 from the ICL after 10 min correlated with the disappearance of the 20 to 40 leading strand cluster (Figure 1D), and
the advance of the leading strand to the 1 position (Figure 1D,
gray squares). These data indicate that arrest of the leading
strand 20-40 nt from a cisplatin ICL is caused by the footprint
of CMG on DNA (and any dead volume of the DNA polymerase)
and that dissociation of CMG facilitates resumption of leading
strand synthesis toward the ICL.
Converging Replisomes Do Not Interfere with Each
Other at an ICL
We postulated that the wide range of leading strand arrest points
near a DNA inter-strand crosslink (Figure 1C, red bracket) could
reflect heterogeneity in the CMG footprint and/or interference by
converging CMG complexes. In the latter view, the first CMG to
arrive at the ICL would impose a more distal stoppage point on
the second replisome, giving rise to the distribution of leading
strand products. Such interference might be particularly pronounced if CMG travels along dsDNA, since an ICL might
enter deep into the central channel of CMG, allowing the first
replisome to prevent approach of the second replisome
(Figure S2A, top).
To test whether replisome interference occurs, we designed
a plasmid, pICLLead/Lag, in which the arrival of replication forks
on one side of an ICL can be blocked. pICLLead/Lag contains
a nitrogen mustard (NM)-like ICL [which causes a 24 position
arrest (Raschle et al., 2008)], as well as four biotinylated thymidine nucleotides placed 34-40 nt to the right of the ICL, two on
the leading strand template and two on the lagging strand
template of the leftward replication fork (Figure 2A). In the presence of SA, the leftward DNA replication fork should not be able
to reach the ICL. To test whether this is the case, pICLLead/Lag
was preincubated with and without SA and then replicated in
the presence of [a32P]dATP. At different times, DNA was digested with StuI and nascent strands of the leftward replication
fork were visualized on sequencing gels (Figure 2B). In the
absence of SA, the leftward leading strand initially advanced to
within 24 nt of the ICL (Figure 2B, lane 1, red bracket), after which
it crept forward, nearly reaching the ICL (Figure 2B, lanes 5, 7, 9,
11, 13, green arrow), as expected based on our previous results
(Raschle et al., 2008). In the presence of SA, the 24 cluster was
largely absent, and we observed a new set of products starting
at the 70 position (Figure 2B, lanes 2, 4, 6, 8, 10, 12, orange
bracket), which is 30 nt from the outermost biotin-SA complex
at the 40 position (Figure 2B, bottom white arrow on DNA
sequencing ladder). The arrest of the leading strand 30 nt from
the biotin-SA complex is similar to the arrest observed at

cisplatin (20) and NM ICLs (24). After the leading strand
paused for 10-15 min at 70, it was further extended to within
one nucleotide of the outermost biotin-SA complex located at
the 41 position (Figure 2B, lanes 4, 6, 8, 10, 12, 14, black
arrow). We infer that the leading strand arrest at 70 is due to
the footprint of CMG, which has stalled at the outermost
biotin-SA complex, and that the advance of the leading strand
to the 41 position reflects CMG dissociation and advance of
DNA polymerase to the lesion. This interpretation is supported
by the fact that purified DNA polymerase advanced to within
a single nucleotide of a biotin-SA complex in primer extension
reactions (Figure S2B). In summary, biotin-SA complexes placed
to the right of the ICL efficiently arrest the leftward CMG complex
for 10-30 min.
To look for interference of the rightward fork by the leftward
fork, pICLLead/Lag was replicated in the presence and absence
of SA and then digested with AflIII, which cuts 151 nt to the
left and 581 nt to the right of the ICL (Figure 2A). Strikingly,
the 24 arrest pattern of the rightward fork was unchanged
by the addition of SA (Figure 2C, blue bracket, compare SA
lanes), even though the leftward fork was efficiently arrested
by the roadblock (Figure S2C, even lanes, orange arrow). We
conclude that the presence of a replisome on one side of the
ICL does not affect the position of the second replisome on
the other side of the lesion. This result can be interpreted in
two ways. First, CMG is a dsDNA translocase, but the ICL
cannot enter its central channel, preventing the first replisome
from engulfing the ICL. However, this explanation is unlikely
since the NM-like ICL is nondistorting (A.G., Z. T., S.C., and
O.D.S., unpublished data) and does not expand the diameter
of the duplex (Angelov et al., 2009). The second interpretation
is that MCM2-7 translocates along ssDNA, such that both replisomes arrest just before the ICL, avoiding interference (Figure S2A, bottom).
Preferential Bypass of a Lagging Strand Roadblock
by the Replicative DNA Helicase
To interrogate directly the translocation mode of CMG on DNA,
we confronted the replisome with a roadblock on only the
leading or lagging strand templates. To this end, we prepared
two derivatives of pICLLead/Lag called pICLLead and pICLLag,
which contain biotins on the leading or lagging strand templates,
respectively, for the leftward replication fork (Figures 3B and 3C).
Using these DNA templates, we could address how the biotin-SA
roadblocks placed on the leading or lagging strand templates
influence the leftward moving CMG without interference from
the rightward moving CMG, whose arrival is prevented by the
ICL (Figure 2C). If CMG translocates along ssDNA in the 30 to
50 direction, it will stall at a biotin-SA complex located on the
leading strand template, yielding the same 70 leading strand
arrest as seen on pICLLead/Lag (Figures 3A and 3B). In contrast,
a biotin-SA complex located on the lagging strand template
might not affect CMG progression since this strand would be
excluded from the central channel of the helicase (Figure 3C).
In this case, the leftward CMG should only stall once it hits the
ICL, manifesting as a 24 arrest (Figure 3C). On the other
hand, if CMG translocates along dsDNA, it is predicted to stall
AflIII (151)
Stu I (-309)
ICL
-35 -40
-70
Primer M
rS
Rightward
Fork
AflIII (-581)
Prime
Leftward
Fork
-34 -39
Biotin-SA
Primer M
G A T C
10
15
20
25
30
60
: Time (min)
: SA
-1
ICL 0
-10
-20
-24
-30
-40
-41
-50
-50
-60
-70
-70
-80
-80
-90
1 2
3 4
5 6
7 8
9 10
11 12
13 14
10
15
20
25
30
60
Primer S
AflIII Digest - Rightward Fork
G A T C
: Time (min)
+ : SA
-1
ICL 0
10
20
-24
30
40
50
-50
1
3 4
5 6
9 10
11 12
13 14
Figure 2. Replisomes Converging on an ICL Do Not Interfere with Each Other

(A) Locations of restriction sites, the nitrogen-mustard-like ICL, biotins, and sequencing primers on pICLLead/Lag. Figure S2A presents two alternative scenarios for
how replication forks might interact at an ICL.
(B) pICLLead/Lag was preincubated with buffer or streptavidin, as indicated, and replicated in egg extracts in the presence of [a32P]dATP. At the indicated times
after NPE addition, replication intermediates were digested with Stu I and separated on a DNA sequencing gel alongside a sequencing ladder generated with
primer M. The distance of products from the ICL is indicated on the left of the gel. White arrows on the DNA sequencing ladder indicate the location of biotins.
Red bracket, leading strand arrest 2450 nt from the ICL in the absence of SA. Orange bracket, leading strand arrest 70-80 nt from the ICL in the presence of SA
(3040 nt from the outermost biotin). Green arrow (1 postion), leading strands that have advanced to the ICL. Black arrow (41 position), leading strands that
have advanced to the outermost biotin-SA complex. Figure S2B shows that DNA polymerase can advance to within one nt of a biotin-SA complex on the leading
strand template.
(C) pICLLead/Lag was preincubated with buffer or streptavidin, as indicated, and replicated in egg extracts in the presence of [a32P]dATP. At the indicated times
after NPE addition, replication intermediates were digested with AflIII (see Figure 2A) and separated on a DNA sequencing gel alongside a sequencing ladder
generated with primer S. Products of the rightward fork are shown. The leftward fork was efficiently arrested by the biotin-SA (see Figure S2C).
3 to 5 ssDNA translocation
Position (nt): 0
-24
-40
dsDNA translocation
Bio-SA
A
pICLLead/Lag
-24
-70
pICLLead/Lag
-70
Bio-SA
ICL
-40
ICL
pICLLead
pICLLead
ICL
ICL
pICLLag
pICLLag
ICL
10
15
ICL
20
30
: Time (min)
: Plasmid
Primer M
G A T
-1
ICL 0
: SA
-10
-20
- 24
-30
-40
- 41
-50
- 50
-60
-70
- 70
-80
- 80
-90
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Figure 3. Biotin-SA Complexes Located on the Leading Strand Template But Not on the Lagging Strand Template Arrest the Replisome
(AF) The 30 to 50 ssDNA translocation (AC) and dsDNA translocation models (DF) for CMG make different predictions regarding how the leftward
moving replisome (CMG, green; DNA polymerase, gray) will interact with SA molecules bound to pICLLead/Lag, pICL Lead, or pICLLag (see main text). On all
plasmids, the rightward replisome (not depicted) will be prevented from approaching the biotin-SA complexes by the ICL. The yellow line in (DF) represents
the ploughshare postulated to split the duplex as it emerges from the central channel. (G) pICLLead/Lag, pICL Lead, or pICLLag was preincubated with buffer
or streptavidin, as indicated, and replicated in the presence of [a32P]dATP as in Figure 2B. SA was not displaced from pICLLag during replication
(Figure S3).
at the SA whether the obstruction is located on the leading or

lagging strand templates since both strands pass through the
central channel of the helicase (Figures 3E and 3F; 70 arrest).
In short, only if CMG translocates along ssDNA is the leading
strand predicted to reach the 24 position on pICLLag bound
to SA (Figure 3C).
pICLLead/Lag, pICLLead, and pICLLag were replicated separately

in the presence or absence of SA, and nascent strands were
analyzed after digestion with StuI, as in Figure 2B. When SA
was bound to pICLLead, leading strands arrested at the 70 position, almost exactly as seen for pICLLead/Lag (Figure 3G, compare
lanes 2 and 4, 8 and 10, 14 and 16, 20 and 22). In contrast, when
biotin-SA was located on the lagging strand template (pICLLag),

the result was very different. While a small fraction (maximally
28% of total signal) of leading strands transiently arrested near
the 80 position (Figure 3G, lane 6), the majority advanced
directly to the 24 position (Figure 3G, lanes 6, 12, 18, and 24,
red bracket). Therefore, CMG bypasses a lagging strand roadblock much more readily than a leading strand roadblock.
Using gel shift analysis, we verified that pICLLead and pICLLag
bound equally to SA (Figure S3A). To assess the retention of
biotin-SA complexes on DNA after replication in egg extracts,
we immunoprecipitated SA and quantified the associated
plasmid by real-time PCR. Importantly, pICLLag was recovered
at least as efficiently as pICLLead, indicating that the absence of
replisome arrest at the 70 position on pICLLag was not due to
reduced binding by SA to this DNA template in extracts (Figure S3B). To rule out that SA dissociated from biotin and
then reassociated, we added excess free biotin to the replication extract to trap dissociated SA. This manipulation had no
significant effect on the retention of SA on either template
(compare Figure S3B and Figure S3C; Figure S3D), arguing
there was no transient dissociation of the roadblock. The arrest
of the CMG by a leading strand but not a lagging strandspecific roadblock clearly favors the 30 to 50 ssDNA translocation model.
Single-Molecule Analysis of Replisome Collisions
with Strand-Specific Roadblocks
To examine the interaction of replisomes with strand-specific
DNA roadblocks by an independent approach where replisomes
do not converge, we employed single-molecule analysis. l DNA
Figure 4. Single-Molecule Analysis of Replisome Collision with Leading and Lagging

Strand-Specific Roadblocks
(A) Reaction scheme for the replication of immobilized l DNA in a microfluidic flow cell using a
single pair of diverging replisomes. SYTOX Orange
and dig-dUTP detection of replicated DNA are
indicated schematically. (B and C) l DNA containing a QDot on the bottom strand (19 kb from
the end) or top strand (15 kb from the end), as
indicated, was immobilized within a microfluidic
flow cell and replicated as depicted in (A). After
protein removal, total DNA (SYTOX Orange), digdUTP (fluorescein-labeled anti-dig Antibody), and
the QDot, were visualized and presented individually or as a merged image. Each dig tract was
classified as a rightward or leftward moving fork
depending on the location of the origin. If the tract
ended within 2 pixels (0.3 mm) of the QDot, it was
considered arrested. Cartoons depicting each
type of collision, the expected outcome based on
the 30 to 50 ssDNA translocation model (checkmarks), representative examples of the raw data,
and the frequency of each outcome are included.
A hypothetical model in which a dsDNA translocase bypasses a lagging strand roadblock is
presented in Figure S4.
(48.5 kb) biotinylated at one or both 30 ends was attached to the

streptavidinated surface of a microfluidic flow cell. We recently
showed that these substrates are extensively replicated by
HSS/NPE in an MCM2-7 dependent manner (Yardimci et al.,
2010). DNA replication can be limited to a single pair of diverging
replisomes when replication initiation is restricted by adding
the Cdk2 inhibitor p27Kip shortly after NPE addition (Figure 4A)(Yardimci et al., 2010). Replicated DNA is detected using
SYTOX Orange (SYTOX), a nonspecific DNA dye, which gives
rise to a fluorescence signal that is twice as strong in regions
containing two daughter duplexes as in regions containing
a single parental duplex (Figure 4A; red molecule). We also
added dig-dUTP for the last 25 min of the reaction and then
stained the DNA with fluorescent anti-dig antibody (Figure 4A,
blue tracts). This allowed us to determine the approximate position of the replication origin, which is located roughly halfway
between the stained dig-dUTP tracts, as well as the directionality
of each replication fork (Figure 4A) (Yardimci et al., 2010).
For the present study, we attached a quantum dot (QDot;
20 nm diameter) to the bottom strand of l DNA (see Materials
and Methods), 19 kb from the end (Figure 4Bi; green dot). Since
DNA replication initiates randomly on DNA in Xenopus egg
extracts, forks hit the QDot from both directions. The QDot
resides on the leading strand template for rightward moving
replication forks (Figure 4Bi and 4Bii) and on the lagging strand
template for leftward moving forks (Figures 4Biii and 4Biv). If
CMG translocates on ssDNA, as indicated by the biotin-SA
experiments presented above, leftward forks should bypass
the QDot (Figure 4Biv), whereas rightward forks should be arrested (Figure 4Bi). In contrast, if CMG is a dsDNA translocase,
neither rightward nor leftward forks should be able to bypass the

QDot (Figures 4Bi and 4Biii).
In agreement with the 30 to 50 ssDNA translocation model, we
found that all rightward DNA replication forks stalled at the QDot
(Figure 4Bi), whereas 80% of leftward forks bypassed the QDot
(Figure 4Biv). We also attached the QDot to the top strand,
15 kb from the end of l DNA (Figure 4Ci, green dot). On this
DNA template, all polarities are reversed. Thus, if CMG is a 30
to 50 ssDNA translocase, leftward forks should stall at the QDot
(Figure 4Ciii) and rightward forks should bypass it (Figure 4Cii).
In agreement with this prediction, 93% of leftward forks stalled
at the QDot (Figure 4Ciii), whereas 74% of rightward forks bypassed it (Figure 4Cii). The fact that 20%25% of forks stalled
at a QDot on the lagging strand template suggests that while
CMG is a 30 to 50 ssDNA translocase, it can be transiently
arrested by a lesion on the excluded strand (see Discussion).
DISCUSSION
This study addresses whether the eukaryotic replicative DNA
helicase (CMG) travels along ssDNA or dsDNA in S phase,
a fundamental, unanswered question in cell biology. Our strategy
was based on the supposition that both 30 to 50 ssDNA and
dsDNA translocases should arrest at an obstruction on the
leading strand template (Figures 3B and 3E), whereas only a 30
to 50 ssDNA should be able to bypass a roadblock on the lagging
strand template (Figure 3C versus Figure 3F). Using the arrest
point of the leading strand as a marker for CMG stalling, we
observed that the helicase was arrested much more efficiently
by a leading strand biotin-SA roadblock than a lagging strand
roadblock. Analogous results were obtained using singlemolecule assays, in which replisomes were collided with QDots.
These data support the idea that CMG translocates along ssDNA
in the 30 to 50 direction. We cannot formally rule out an alternative
interpretation of the data in which MCM2-7 translocates along
dsDNA and transient opening of the helicase central channel
allows bypass of a lagging strand roadblock. However, this
model is improbable, principally because it does not explain
how the breached helicase could bypass the lesion (see Figure S4 and legend thereof).
Importantly, other aspects of our data are also consistent with
CMG being a ssDNA translocase. First, we observed no interference between converging replisomes at an ICL, as might have
been expected in the dsDNA translocation model (see Results).
Second, the apparent size of the CMG footprint is consistent
with a ssDNA translocation model. Thus, upon initial collision
of the replisome with a cisplatin ICL, the 30 end of the leading
strand can advance to within 20 nucleotides of the lesion. Given
that the longitudinal axis of the eukaryotic MCM2-7 complex
comprises 115 A (Remus et al., 2009), MCM2-7 should protect
34 basepairs of dsDNA (115 A O 3.4 A per basepair = 34 basepairs), which likely occurs in the context of prereplication complexes. Importantly, ssDNA can be extended to 1.7 times the
length of B-form DNA (Smith et al., 1996; van Oijen, 2007). Therefore, 20 nt (34 O 1.7) of fully extended ssDNA would suffice to
traverse the entire MCM2-7 channel. Thus, a ssDNA translocation model is compatible with the 20 nt arrest point seen for
the leading strand if the active site of DNA polymerase is located
immediately behind the rear exit channel of CMG. Although

stretching ssDNA to this extent is energetically unfavorable,
this might occur as a result of interactions between DNA and
the CMG central channel and/or the pulling force exerted by
DNA polymerase. It is also worth noting that most of the leading
strands arrest further than 20 nt from the ICL, indicating that only
a small fraction of stalled CMG complexes encircle fully extended ssDNA. The reason why leading strands arrest 4 nucleotides further away from a nitrogen-mustard like ICL versus
a cisplatin ICL (Raschle et al., 2008) is likely that the former lesion
stabilizes the duplex, preventing full approach to the lesion by
CMG. In the case of SA roadblocks, a steric clash between
CMG and SA might explain the more distal arrest point
(30 nt) observed for this lesion.
Finally, our conclusion that the replicative helicase moves
along ssDNA is supported by extensive biochemical analysis
of purified MCM complexes. Archaeal MCMs (Kelman et al.,
1999; Shechter et al., 2000), Mcm4/6/7 (Ishimi, 1997), and
CMG (Moyer et al., 2006) exhibit 30 to 50 ssDNA translocation
activity on model DNA templates. In light of our findings, the
observation that MCM complexes can slide (Evrin et al., 2009;
Remus et al., 2009) or actively translocate along dsDNA (Kaplan
et al., 2003) could have other implications. As previously proposed (Remus et al., 2009), the collision of a replisome with
latent MCM2-7 complexes could induce the latter to slide or
translocate along dsDNA ahead of the fork. These mobilized
MCMs might represent a cadre of reserve helicases that are
deployed to rescue stalled replication forks (Remus et al.,
2009), or they might serve to increase the local concentration
of pre-RCs in unreplicated DNA to potentiate an increased initiation frequency late in S phase (Lucas et al., 2000).
In summary, the data are most consistent with CMG translocation along ssDNA in the 30 to 50 direction. The evidence thus indicates that CMG unwinds DNA via steric exclusion, and it strongly
disfavors models in which dsDNA passes through any part of the
CMG central channel. Future experiments will be required to
address the precise molecular mechanism by which CMG translocates along ssDNA and how this leads to strand separation.
Implications for CMG and Replisome Architecture
Our results have implications for the overall architecture of the
DNA replication fork. As discussed above, the 20 nt leading
strand arrest point matches precisely what is expected if ssDNA
is fully extended within the central channel of MCM2-7. Barring
rapid dissociation of replication proteins upon collision with the
ICL or a major difference in the dimensions of yeast and Xenopus
MCM2-7, this observation implies that the C-terminal ATPase
domains of the MCM2-7 complex comprise the leading edge
of the replisome and that Cdc45, GINS, and any other helicaseassociated factors do not substantially alter the footprint of
MCM2-7 on DNA. Moreover, the data imply that the active site
of DNA polymerase epsilon resides immediately behind the rear
exit channel of CMG, which is consistent with our finding that
purified pol can advance to within 1 nucleotide of a biotin-SA
complex on the leading strand template. This configuration is
advantageous, as emerging ssDNA will be immediately replicated, minimizing the possibility of strand reannealing or
cleavage.
We noted that when a biotin-SA complex was placed on the

lagging strand template, there was a low level of transient stalling
(Figure 3G, lane 6). Similarly, in single-molecule analysis, QDots
placed on the lagging strand template induced 20% stalling.
These results suggest that the excluded strand might interact
intimately with the outer face of the CMG, leading to steric
clashes with the bulky lesions that cause occasional or transient
arrest. Recent structural analysis of the Drosophila CMG complex supports this conclusion since it reveals a potential
groove/channel on the outer surface of CMG that might interact
with the excluded strand (Costa et al., 2011).
Interactions of the Replisome with DNA Damage
Our data have implications for the early steps of DNA interstrand
crosslink repair, which is initiated by the collision of two DNA
replication forks with an ICL in Xenopus egg extracts (Knipscheer
et al., 2009; Raschle et al., 2008). It was previously unknown why
leading strands pause at the 20 to 40 positions and then
undergo further elongation toward the ICL, followed by lesion
bypass. Our experiments indicate that the delay involves the
dissociation of CMG from the site of the lesion. Another implication is that when the replisome first arrives at an ICL, the DNA
immediately adjacent to the lesion is single stranded since one
strand is sequestered within the CMG central channel while
the other strand is excluded.
Implications for the Mechanism of CMG Assembly
In G1, latent MCM2-7 double hexamers appear to interact with
dsDNA, as evidenced by the absence of ssDNA within pre-RCs
(our unpublished results; Bowers et al., 2004; Geraghty et al.,
2000) and the ability of latent MCM2-7 complexes to slide along
DNA (Evrin et al., 2009; Remus et al., 2009). Our data indicate
that in S phase, CMG encircles ssDNA with no duplex DNA remaining inside the central channel of the helicase. Furthermore,
CMG likely functions as a monomer (Gambus et al., 2006; Ilves
et al., 2010; Moyer et al., 2006; Yardimci et al., 2010). Assuming
these starting and end points, a number of discrete stages can be
envisioned that convert MCM2-7 to CMG, not necessarily in the
following order (Figure 5). (1) The MCM2-7 double hexamer is
split. (2) The latent MCM2-7 rings straddling dsDNA are opened,
perhaps via regulation of the Mcm2-Mcm5 gate (Bochman and
Schwacha, 2008). Both events might depend on MCM2-7 phosphorylation by DDK. (3) One strand of the DNA duplex is extruded
from the central channel, perhaps by the binding of Mcm10 or
Sld2 to ssDNA (Kanter and Kaplan, 2010; Warren et al., 2008).
(4) The gate is reclosed and the helicase motor is jump-started
by Cdc45 and GINS (Ilves et al., 2010). Whatever the precise
sequence of events, the reconfiguration of the MCM2-7 complex
from a dsDNA binding mode in G1 to a ssDNA binding mode
in S phase helps explain why so many initiation factors are
required to assemble CMG. This view of initiation establishes
a roadmap for future investigations into the functions of replication initiation factors. Notably, the DNA tumor virus initiator
proteins E1 and possibly Large T Antigen initially bind to dsDNA
within the viral origin of replication but unwind DNA via steric
exclusion (Enemark and Joshua-Tor, 2006). Thus, the transition
from dsDNA to ssDNA binding appears to be widely conserved
among eukaryotic replicative DNA helicases.
Side View
Latent MCM2-7
double hexamer
End-on view
(left hexamer)
G1 Phase
(1)
Split Dimer
(2)
Open 2/5 gate
(3)
Extrude strand
S Phase
(4)
Activate ATPase
with GINS/Cdc45
DNA unwinding
Figure 5. Model for Replication Initiation

Model for helicase activation in which MCM2-7 encircles dsDNA in G1 and
ssDNA in S phase. See text for details.
Preparation of Plasmids
To make pICLLead/Lag, pICLLead, and pICLLag, various ICL-biotin oligonucleotides (see Supplementary Methods) were purified by polyacrylamide gel
electrophoresis, and ligated into pSVRLuc to form the three plasmids (see Figure 3) (Guainazzi et al., 2010; Raschle et al., 2008). pICLControl contains the
identical sequence as pICLLead/Lag, except for the four biotinylated thymidine
nucleotides. pICLInter and pICLIntra were constructed as previously described
(Raschle et al., 2008; Tremeau-Bravard et al., 2004). Compared to pICLInter
and pICLIntra, the plasmids pICLControl, pICLLead/Lag, pICLLead, and pICLLag
contain a 41 nt insert near the ICL.
Xenopus Egg Extracts and Replication
The preparation of Xenopus egg extracts (NPE and HSS) was as described
(Walter et al., 1998). For DNA replication, the plasmids (75ng/ml) were first
incubated with an equal volume of Streptavidin (5 mg/ml) (SouthernBiotech,
Birmingham, AL, USA) or buffer for 1 hr at room temperature (RT), after which
this mixture was added to HSS for 5 min at 22 C (1215 ng/ml final plasmid
concentration), followed by addition of two volumes of NPE containing
[a32P]dATP. Replication was stopped with Stop Solution (0.5% SDS,
25mM EDTA, 50 mM Tris, [pH 7.5]) at different time points. The purification
of DNA replication products was as described (Raschle et al., 2008).
Nascent Strand Analysis
The nascent strand analysis was carried out as described (Raschle et al.,
2008). Briefly, purified DNA replication products were isolated and digested
with the indicated enzymes. Restriction fragments were separated on 5% or
7% polyacrylamide sequencing gels. Gels were transferred to filter paper,
dried, and nascent strands visualized with a phosphorimager. Sequencing
ladders using primers S and M (see Figure 2A) were generated using the
Cycle Sequencing kit from USB (USB Corporation, Cleveland, OH, USA).
Quantification was performed by Image Gauge V4.22 (Fuji Photo Film Corporation, Tokyo, Japan).
Chromatin Immunoprecipitation
Chromatin immunoprecipitation (ChIP) was modified from our existing procedure (Pacek et al., 2006). After the addition of NPE, aliquots of the reaction
were crosslinked through the addition of 1% formaldehyde in ELB. After
10 min incubation at RT, glycine was added to stop the crosslinking reaction.
The crosslinked material was spun through Micro Bio-Spin 6 Chromatography
Columns (BIO-RAD, Hercules, CA, USA) to remove formaldehyde. The flowthrough was diluted with sonication buffer and subjected to sonication. Immonoprecipitation with the indicated antibodies and quantitative real-time PCR
were performed as described (Pacek et al., 2006). Anti-Mcm7 antibody was
previously described (Walter and Newport, 2000); anti-Streptavidin antibody
was purchased from Spring Bioscience (Spring Bioscience, Pleasanton,
CA, USA).
Single-Molecule Assays
A QDot was attached to the top strand 15 kb from the right end of l DNA
or to the bottom strand 19 kb away from the end of l DNA through modification of a previous procedure (Kochaniak et al., 2009; Kuhn and FrankKamenetskii, 2008). Anti-digoxigenin antibody (Roche Applied Sciences,
Indianapolis, IN, USA) was attached to QDot605 using Invitrogen QDot
Antibody Conjugation Kit. After immobilizing l DNA in the flow cell, 1 nM
of anti-digoxigenin conjugated QDot was injected. After 10-20 min incubation, the flow cell was washed with buffer and extracts were introduced.
Replication of surface immobilized l DNA from single initiations in Xenopus
egg extracts and visualization of replicated products was performed as
described (Yardimci et al., 2010). SYTOX, fluorescein labeled anti-dig, and
the QDot were imaged using 568 nm, 488 nm, and 405 nm laser light,
respectively. To assess replication fork arrest near QDots, replication forks
were labeled with dig-dUTP for 25 min. Since the average fork rate in
these experiments was 268 bp/minute (Yardimci et al., 2010), uninterrupted
replication would yield average dig-dUTP tracts of 6.7 kb (25 min 3
0.268 kb/minute). Therefore, only dig-dUTP tracts that were significantly
shorter than expected (%4.5 kb), which ended at a QDot, were considered
to represent stalled forks.
Supplemental Information contains Extended Experimental Procedures and
four figures and can be found with this article online at doi:10.1016/j.cell.
2011.07.045.
ACKNOWLEDGMENTS
We thank Milica Enoiu for the gift of pICLIntra, Anna Loveland, Puck Knipscheer,
and Tatsuro Takahashi for helpful discussions, and the members of our laboratory for valuable feedback on the manuscript. J.C.W was supported by
National Institutes of Health (NIH) grant GM62267, by a Leukemia and
Lymphoma Scholar Award, and by the American Cancer Society grant RSG08-234-01-GMC. Y.V.F. was supported by a Human Frontier Science Program
Long-Term Fellowship (LT000307/2009). D.T.L was supported by a postdoctoral fellowship (PF-10-146-01-DMC) from the American Cancer Society.
O.D.S. was supported by grants from the New York State Office of Science
and Technology and Academic Research NYSTAR (C040069) and the NIH
(GM08454 and CA092584). A.M.v.O. acknowledges support from the American Cancer Society grant RSG-08-234-01 and Searle Scholarship 05-L-104.
J.H. was supported by NIH grant GM034559 and American Cancer Society
grant RSG-08-234-01-GMC.
REFERENCES
Angelov, T., Guainazzi, A., and Scharer, O.D. (2009). Generation of DNA
interstrand cross-links by post-synthetic reductive amination. Org. Lett. 11,
661664.
Arias, E.E., and Walter, J.C. (2004). Initiation of DNA replication in Xenopus egg
extracts. Front. Biosci. 9, 30293045.
Arias, E.E., and Walter, J.C. (2007). Strength in numbers: preventing rereplication via multiple mechanisms in eukaryotic cells. Genes Dev. 21, 497518.
Bochman, M.L., and Schwacha, A. (2008). The Mcm2-7 complex has in vitro
helicase activity. Mol. Cell 31, 287293.
Bochman, M.L., and Schwacha, A. (2009). The Mcm complex: unwinding the
mechanism of a replicative helicase. Microbiol. Mol. Biol. Rev. 73, 652683.
Bowers, J.L., Randell, J.C., Chen, S., and Bell, S.P. (2004). ATP hydrolysis by
ORC catalyzes reiterative Mcm2-7 assembly at a defined origin of replication.
Mol. Cell 16, 967978.
Brewster, A.S., Wang, G., Yu, X., Greenleaf, W.B., Carazo, J.M., Tjajadia, M.,
Klein, M.G., and Chen, X.S. (2008). Crystal structure of a near-full-length
archaeal MCM: functional insights for an AAA+ hexameric helicase. Proc.
Natl. Acad. Sci. USA 105, 2019120196.
Costa, A., Ilves, I., Tamberg, N., Petojevic, T., Nogales, E., Botchan, M.R., and
Berger, J.M. (2011). The structural basis for MCM2-7 helicase activation by
GINS and Cdc45. Nat. Struct. Mol. Biol. 18, 471477.
Enemark, E.J., and Joshua-Tor, L. (2006). Mechanism of DNA translocation in
a replicative hexameric helicase. Nature 442, 270275.
Evrin, C., Clarke, P., Zech, J., Lurz, R., Sun, J., Uhle, S., Li, H., Stillman, B., and
Speck, C. (2009). A double-hexameric MCM2-7 complex is loaded onto origin
DNA during licensing of eukaryotic DNA replication. Proc. Natl. Acad. Sci. USA
106, 2024020245.
Fletcher, R.J., Bishop, B.E., Leon, R.P., Sclafani, R.A., Ogata, C.M., and Chen,
X.S. (2003). The structure and function of MCM from archaeal M. Thermoautotrophicum. Nat. Struct. Biol. 10, 160167.
Gambus, A., Jones, R.C., Sanchez-Diaz, A., Kanemaki, M., van Deursen, F.,
Edmondson, R.D., and Labib, K. (2006). GINS maintains association of
Cdc45 with MCM in replisome progression complexes at eukaryotic DNA
replication forks. Nat. Cell Biol. 8, 358366.
Gambus, A., Khoudoli, G.A., Jones, R.C., and Blow, J.J. (2011). MCM2-7 form
double hexamers at licensed origins in Xenopus egg extract. J. Biol. Chem.
286, 1185511864.
Geraghty, D.S., Ding, M., Heintz, N.H., and Pederson, D.S. (2000). Premature
structural changes at replication origins in a yeast minichromosome maintenance (MCM) mutant. J. Biol. Chem. 275, 1801118021.
Guainazzi, A., Campbell, A.J., Angelov, T., Simmerling, C., and Scharer, O.D.
(2010). Synthesis and molecular modeling of a nitrogen mustard DNA interstrand crosslink. Chemistry (Easton) 16, 1210012103.
Ilves, I., Petojevic, T., Pesavento, J.J., and Botchan, M.R. (2010). Activation of
the MCM2-7 helicase by association with Cdc45 and GINS proteins. Mol. Cell
37, 247258.
Ishimi, Y. (1997). A DNA helicase activity is associated with an MCM4, 6,
and 7 protein complex. J. Biol. Chem. 272, 2450824513.
Kanter, D.M., and Kaplan, D.L. (2010). Sld2 binds to origin single-stranded
DNA and stimulates DNA annealing. Nucleic Acids Res. 39, 25802592.
Kaplan, D.L., Davey, M.J., and ODonnell, M. (2003). Mcm4,6,7 uses a pump
in ring mechanism to unwind DNA by steric exclusion and actively translocate
along a duplex. J. Biol. Chem. 278, 4917149182.
Kelman, Z., Lee, J.K., and Hurwitz, J. (1999). The single minichromosome
maintenance protein of Methanobacterium thermoautotrophicum DeltaH
contains DNA helicase activity. Proc. Natl. Acad. Sci. USA 96, 1478314788.
Knipscheer, P., Raschle, M., Smogorzewska, A., Enoiu, M., Ho, T.V., Scharer,
O.D., Elledge, S.J., and Walter, J.C. (2009). The Fanconi anemia pathway
promotes replication-dependent DNA interstrand cross-link repair. Science
326, 16981701.
Kochaniak, A.B., Habuchi, S., Loparo, J.J., Chang, D.J., Cimprich, K.A.,
Walter, J.C., and van Oijen, A.M. (2009). Proliferating cell nuclear antigen
uses two distinct modes to move along DNA. J. Biol. Chem. 284, 17700
17710.
Kuhn, H., and Frank-Kamenetskii, M.D. (2008). Labeling of unique sequences
in double-stranded DNA at sites of vicinal nicks generated by nicking endonucleases. Nucleic Acids Res. 36, e40.
Labib, K. (2010). How do Cdc7 and cyclin-dependent kinases trigger the initiation of chromosome replication in eukaryotic cells? Genes Dev. 24, 1208
1219.
Laskey, R.A., and Madine, M.A. (2003). A rotary pumping model for helicase
function of MCM proteins at a distance from replication forks. EMBO Rep. 4,
2630.
Li, D., Zhao, R., Lilyestrom, W., Gai, D., Zhang, R., DeCaprio, J.A., Fanning, E.,
Jochimiak, A., Szakonyi, G., and Chen, X.S. (2003). Structure of the replicative
helicase of the oncoprotein SV40 large tumour antigen. Nature 423, 512518.
Lucas, I., Chevrier-Miller, M., Sogo, J.M., and Hyrien, O. (2000). Mechanisms
ensuring rapid and complete DNA replication despite random initiation in
Xenopus early embryos. J. Mol. Biol. 296, 769786.
Mendez, J., and Stillman, B. (2003). Perpetuating the double helix: molecular
machines at eukaryotic DNA replication origins. Bioessays 25, 11581167.
Moyer, S.E., Lewis, P.W., and Botchan, M.R. (2006). Isolation of the Cdc45/
Mcm2-7/GINS (CMG) complex, a candidate for the eukaryotic DNA replication
fork helicase. Proc. Natl. Acad. Sci. USA 103, 1023610241.
Pacek, M., Tutter, A.V., Kubota, Y., Takisawa, H., and Walter, J.C. (2006).
Localization of MCM2-7, Cdc45, and GINS to the site of DNA unwinding during
eukaryotic DNA replication. Mol. Cell 21, 581587.
Remus, D., Beuron, F., Tolun, G., Griffith, J.D., Morris, E.P., and Diffley, J.F.
(2009). Concerted loading of Mcm2-7 double hexamers around DNA during
DNA replication origin licensing. Cell 139, 719730.
Shechter, D.F., Ying, C.Y., and Gautier, J. (2000). The intrinsic DNA helicase
activity of Methanobacterium thermoautotrophicum delta H minichromosome
maintenance protein. J. Biol. Chem. 275, 1504915059.
Smith, S.B., Cui, Y., and Bustamante, C. (1996). Overstretching B-DNA: the
elastic response of individual double-stranded and single-stranded DNA
molecules. Science 271, 795799.
Takahashi, T.S., and Walter, J.C. (2005). Cdc7-Drf1 is a developmentally regulated protein kinase required for the initiation of vertebrate DNA replication.
Genes Dev. 19, 22952300.
Takahashi, T.S., Wigley, D.B., and Walter, J.C. (2005). Pumps, paradoxes and
ploughshares: mechanism of the MCM2-7 DNA helicase. Trends Biochem.
Sci. 30, 437444.
Tremeau-Bravard, A., Riedl, T., Egly, J.M., and Dahmus, M.E. (2004). Fate of
RNA polymerase II stalled at a cisplatin lesion. J. Biol. Chem. 279, 77517759.
Tye, B.K. (1999). MCM proteins in DNA replication. Annu. Rev. Biochem. 68,
649686.
van Oijen, A.M. (2007). Honey, I shrunk the DNA: DNA length as a probe for
nucleic-acid enzyme activity. Biopolymers 85, 144153.
Walter, J., and Newport, J. (2000). Initiation of eukaryotic DNA replication:
origin unwinding and sequential chromatin association of Cdc45, RPA, and
DNA polymerase alpha. Mol. Cell 5, 617627.
Walter, J., Sun, L., and Newport, J. (1998). Regulated chromosomal DNA
replication in the absence of a nucleus. Mol. Cell 1, 519529.
Pacek, M., and Walter, J.C. (2004). A requirement for MCM7 and Cdc45 in
chromosome unwinding during eukaryotic DNA replication. EMBO J. 23,
36673676.
Warren, E.M., Vaithiyalingam, S., Haworth, J., Greer, B., Bielinsky, A.K.,
Chazin, W.J., and Eichman, B.F. (2008). Structural basis for DNA binding by
replication initiator Mcm10. Structure 16, 18921901.
Pape, T., Meka, H., Chen, S., Vicentini, G., van Heel, M., and Onesti, S. (2003).
Hexameric ring structure of the full-length archaeal MCM protein complex.
EMBO Rep. 4, 10791083.
Wessel, R., Schweizer, J., and Stahl, H. (1992). Simian virus 40 T-antigen DNA
helicase is a hexamer which forms a binary complex during bidirectional
unwinding from the viral origin of DNA replication. J. Virol. 66, 804815.
Randell, J.C., Fan, A., Chan, C., Francis, L.I., Heller, R.C., Galani, K., and Bell,
S.P. (2010). Mec1 Is One of Multiple Kinases that Prime the Mcm2-7 Helicase
for Phosphorylation by Cdc7. Mol. Cell 40, 353363.
Wohlschlegel, J.A., Dhar, S.K., Prokhorova, T.A., Dutta, A., and Walter, J.C.
(2002). Xenopus Mcm10 binds to origins of DNA replication after Mcm2-7
and stimulates origin binding of Cdc45. Mol. Cell 9, 233240.
Raschle, M., Knipscheer, P., Enoiu, M., Angelov, T., Sun, J., Griffith, J.D.,
Ellenberger, T.E., Scharer, O.D., and Walter, J.C. (2008). Mechanism of replication-coupled DNA interstrand crosslink repair. Cell 134, 969980.
Yardimci, H., Loveland, A.B., Habuchi, S., van Oijen, A.M., and Walter, J.C.
(2010). Uncoupling of sister replisomes during eukaryotic DNA replication.
Mol. Cell 40, 834840.
Wnt Regulates Spindle Asymmetry

to Generate Asymmetric
Nuclear b-Catenin in C. elegans
Kenji Sugioka,1,2 Kota Mizumoto,1,4 and Hitoshi Sawa1,2,3,*
1Laboratory
for Cell Fate Decision, RIKEN Center for Developmental Biology, Kobe 650-0047, Japan
of Biology, Graduate School of Science, Kobe University, Kobe 657-8501, Japan
3Multicellular Organization Laboratory, National Institute of Genetics, 1111 Yata, Mishima 411-8540, Japan
4Present address: Department of Biological Sciences and Neurosciences Program, Stanford University, Stanford, CA 94305, USA
*Correspondence: hisawa@lab.nig.ac.jp
DOI 10.1016/j.cell.2011.07.043
2Department
SUMMARY
Extrinsic signals received by a cell can induce remodeling of the cytoskeleton, but the downstream effects
of cytoskeletal changes on gene expression have not
been well studied. Here, we show that during telophase of an asymmetric division in C. elegans, extrinsic Wnt signaling modulates spindle structures
through APR-1/APC, which in turn promotes asymmetrical nuclear localization of WRM-1/b-catenin
and POP-1/TCF. APR-1 that localized asymmetrically
along the cortex established asymmetric distribution
of astral microtubules, with more microtubules found
on the anterior side. Perturbation of the Wnt signaling
pathway altered this microtubule asymmetry and led
to changes in nuclear WRM-1 asymmetry, gene expression, and cell-fate determination. Direct manipulation of spindle asymmetry by laser irradiation altered the asymmetric distribution of nuclear WRM-1.
Moreover, laser manipulation of the spindles rescued
defects in nuclear POP-1 asymmetry in wnt mutants.
Our results reveal a mechanism in which the nuclear
localization of proteins is regulated through the
modulation of microtubules.
INTRODUCTION
Extrinsic signals can regulate gene transcription in the nuclei as
well as the cytoskeletal network in the cytoplasm. Regulation of
the cytoskeleton is generally considered to be a process independent from gene transcription or a consequencerather
than causeof gene transcription. One example counter to
this prevailing notion is the ability of actin polymerization triggered by extrinsic signals to directly influence gene expression
and differentiation (Miralles et al., 2003). These effects are elicited by the MAL transcription factor, which binds to unpolymerized actin in the cytoplasm and is transported to the nucleus
upon actin polymerization where it regulates target genes (Miralles et al., 2003). The effects of regulating other major cytoskel942 Cell 146, 942954, September 16, 2011 2011 Elsevier Inc.
etal components, such as microtubules, on gene expression

have not been reported.
One example of an extrinsic signal that regulates both the cytoskeleton and gene transcription is the Wnt protein. This protein
plays important roles in many developmental processes, including self-renewing stem cell divisions (Clevers, 2006). The
canonical Wnt pathway regulates target gene expression in the
nucleus through the interaction of b-catenin and TCF. In this
pathway, b-catenin localization to the nucleus is crucial to activate transcription of the target genes (Cadigan and Peifer,
2009). However, it is not clear how b-catenin is directed to the
nucleus, as b-catenin does not have conventional nuclear import
or export signals (Eleftheriou et al., 2001; Fagotto et al., 1998;
Wiechens and Fagotto, 2001; Yokoya et al., 1999). One report
suggests that the APC tumor suppressor protein is involved in
the nuclear export of b-catenin, as well as its degradation (Brocardo and Henderson, 2008). Besides regulating transcription,
Wnt signals also stabilize microtubules through APC, possibly
through the APC binding to microtubule plus ends (MimoriKiyosue et al., 2000a; Zumbrunn et al., 2001). This function of
APC is required for cell migration and axon guidance (EtienneManneville and Hall, 2003; Purro et al., 2008), but its involvement
in b-catenin regulation has not been demonstrated.
In Caenorhabditis elegans, canonical Wnt pathway components, including WRM-1/b-catenin and APR-1/APC, control the
asymmetric divisions of a number of cells through the Wnt/
b-catenin asymmetry pathway (Mizumoto and Sawa, 2007b;
Phillips and Kimble, 2009). For example, the T cell in the larval
tail is polarized by LIN-44/Wnt expressed in the posterior side
of the cell (Goldstein et al., 2006; Herman et al., 1995). This polarization causes the asymmetric localization of Wnt signaling
components to the cell cortex; Dishevelled homologs (DSH-2
and MIG-5, two of three homologs in C. elegans) to the posterior
cell cortex, and WRM-1/b-catenin and APR-1/APC to the anterior cortex (Mizumoto and Sawa, 2007a). At telophase, in addition to localizing to the anterior cortex, WRM-1 also starts to
localize to the posterior nucleus (nuclear WRM-1 asymmetry)
(Takeshita and Sawa, 2005). Nuclear WRM-1 regulates the
nuclear export of POP-1/TCF, causing reciprocal nuclear asymmetry of WRM-1 and POP-1 (Lo et al., 2004). Similar asymmetric
localizations of DSH-2, WRM-1, and POP-1 have been observed
in the embryonic EMS cell, which also undergoes Wnt-dependent asymmetric division (Lin et al., 1995; Nakamura et al.,
2005; Walston et al., 2004). Experiments in which WRM-1 was
expressed uniformly on the T cell cortex showed that WRM-1
on the anterior cortex not only recruits APR-1 to the cortex, but
also inhibits WRM-1 itself from localizing to the anterior nucleus,
so that it accumulates at the posterior nucleus. In addition,
APR-1 is required for nuclear asymmetry and efficient nuclear
export of WRM-1 (Mizumoto and Sawa, 2007a). However, the
mechanism by which cortical WRM-1 and APR-1 regulate
nuclear WRM-1 asymmetry remains elusive.
Focusing on the EMS cell division, we found that microtubules
and kinesins are required for nuclear WRM-1 and POP-1 asymmetry. During this division, the number of astral microtubules
became higher in the anterior than posterior side. This spindle
asymmetry was disrupted in mutants of apr-1, as well as in those
of mom-2/wnt and wrm-1, which also showed defects in cortical
APR-1 localization. To determine the importance of spindle
asymmetry for WRM-1 asymmetry, we disrupted or enhanced
the spindle asymmetry by laser irradiation of the centrosomes,
and we found that the nuclear WRM-1 asymmetry was lost and
enhanced, respectively. Moreover, in wnt mutants, restoration
of the spindle asymmetry with laser treatment rescued the
nuclear POP-1 asymmetry. These results strongly suggest that
the nuclear WRM-1 and POP-1 asymmetries are regulated by
spindle asymmetry. Our study reveals a microtubule-mediated
link between extrinsic signals and gene expression.
RESULTS
Nuclear WRM-1 Asymmetry Requires Microtubules
In the early embryogenesis of C. elegans, asymmetric EMS cell
division is regulated by mom-2/wnt expressed in a P2 cell that
attaches to the EMS cell posterior cortex. After the division,
the anterior MS and posterior E daughters produce muscle
and gut, respectively. However, in mom-2 mutants, the posterior
E daughter has levels of nuclear WRM-1 and POP-1 similar to the
anterior MS daughter (Nakamura et al., 2005; Rocheleau et al.,
1997) and adopts an MS-like fate, resulting in the gutless phenotype (Rocheleau et al., 1997; Thorpe et al., 1997). To study the
role of microtubules in nuclear WRM-1 and POP-1 asymmetry,
we treated EMS cells with vinblastine or nocodazole at anaphase
to destabilize the microtubules without disrupting the cell division itself (see the Experimental Procedures; data not shown).
After division, we found that nuclear WRM-1 was symmetric
(Figures 1C and 1G). In addition, POP-1 signal in the anterior
nucleus was significantly decreased, resulting in more symmetric localization (Figures 1D and 1H). These results suggest
that microtubules are necessary to establish the nuclear WRM-1
and POP-1 asymmetries and are consistent with a previous
observation that a microtubule destabilizing drug inhibits gut
production (Goldstein, 1995a).
Astral Microtubule Numbers Become Asymmetric
during EMS Division
To determine how microtubules regulate the WRM-1 and POP-1
asymmetry, we examined whether the microtubule organization
is itself asymmetric. In embryos immunostained for endogenous
a-tubulin, we manually counted the number of astral microtubules emanating from each anterior or posterior centrosome
during EMS division. We found that the numbers of astral microtubules were asymmetric after metaphase and that the anterior
spindle pole had more astral microtubules than the posterior pole (Figure 2A and Figure S1 available online). This microtubule number asymmetry (hereafter called spindle asymmetry)
became most prominent at telophase (Figure 2A). Analyses of
the number of astral microtubules in living embryos expressing
GFP::b-tubulin confirmed the asymmetry to be strongest at telophase (Figure 2B; 0240 s). Consistent with this observation, the
amount of a-tubulin around each centrosome was asymmetric at
telophase, as measured by immunostaining and normalized to
the amount of g-tubulin (Figure S2B). To further confirm the
spindle asymmetry, we quantified GFP::b-tubulin intensities in
the entire anterior or posterior halves of the EMS cell and mathematically estimated the GFP signals corresponding to astral
microtubules. Astral microtubule intensity was defined as the
signals found between the strong signals of the centrosomal
regions and the weak background signals (estimated from signals in regions that do not contain microtubules). We found
that the numbers of pixels corresponding to microtubules are
larger in the anterior than posterior sides (Figure S2A). Based
on these results, we concluded that the astral microtubule
numbers are asymmetric during EMS division, most prominently
at telophase.
The Asymmetric Spindle Is Regulated by Wnt Signaling
Because the mitotic spindle in the EMS cell is oriented along
the anterior-posterior axis through the functions of MOM-2/
Wnt by metaphase (Goldstein, 1995b; Schlesinger et al., 1999),
we analyzed whether various Wnt signaling components are
involved in establishing the spindle asymmetry. In mom-2 or
mom-5/frizzled embryos, there appeared to be fewer microtubules in the anterior and more microtubules in the posterior
halves, albeit not significantly, than those in the wild-type.
Accordingly, the difference between the number of anterior
and posterior microtubules was significantly decreased in both
embryos (p value = 0.03 and 0.05 in mom-2 and mom-5, respectively), thereby disrupting the spindle asymmetry (Figures 2C and
2D and Figure S2C). In dsh-2(RNAi)mig-5(RNAi) [hereafter called
dsh(RNAi)], wrm-1(RNAi), and apr-1(RNAi) embryos, the anterior
microtubule numbers were significantly decreased compared to
those of control embryos, resulting in nearly symmetric microtubule numbers (Figure 2D and Figure S2C). Because apr-1 and
wrm-1 are not involved in spindle orientation (Schlesinger
et al., 1999), our results indicate that spindle asymmetry is regulated independently of spindle orientation by the Wnt/b-catenin
asymmetry pathway.
Spindle Asymmetry Is Likely to Be Regulated by Cortical
APR-1/APC
How does Wnt signaling regulate spindle asymmetry? APR-1/
APC may be a key regulator, since APC is known to regulate
microtubules (Munemitsu et al., 1994). In postembryonic seam
cells, Wnt signaling promotes the asymmetrical localization
of APR-1 (Mizumoto and Sawa, 2007a). Similarly, in the EMS
cell, we found that a GFP fusion protein of APR-1 localized
GFP::WRM-1/-catenin
GFP::POP-1/TCF
0.2% DMSO
MS
MS
E
100 nM vinblastine
MS
10 M ATA
10 M ATA
MS
H
200
(A.U.)
n.s.
n.s.
150
100
50
A P
DMSO
(n = 10)
A P
vinblastine
(n = 10)
A
P
ATA
(n = 10)
GFP::POP-1 in
intensity
GFP::WRM-1 intensity
**
G
(A.U.)
E
100 nM vinblastine
MS
MS
0.2% DMSO
2000
**
**
1500
1000
500
A
P
DMSO
(n = 10)
asymmetrically to the anterior cell cortex (APR-1 asymmetry)

(Figure 3A, arrowheads indicate the EMS cell cortex). We
analyzed the kinetics of GFP::APR-1 localization during division
by quantifying the signal intensities in the anterior and posterior
cell cortex (Figure S3A). Before the onset of mitosis, the APR-1
localization was not polarized. However, when the EMS cell
entered mitosis, APR-1 gradually became asymmetrical, with
the strongest asymmetry at telophase (Figure 3A and Figure S3A). Notably, the kinetics of APR-1 localization and of
spindle asymmetry were similar during division (compare Figure 2A and Figure S3A).
Next, we analyzed the APR-1 localization in Wnt-signaling
mutants or RNA interference (RNAi) embryos. In mom-2(or309)
or mom-5(ne12) embryos, although the signal intensities on the
anterior or posterior cortex were not significantly altered, the
differences in signal intensity were significantly smaller (Welchs
test; p < 0.01), resulting in symmetric APR-1 localization.
dsh(RNAi) caused symmetric localization as well as a significant
Figure 1. Microtubules Are Required for the

Nuclear WRM-1 and POP-1 Asymmetries
(AF) Drug effect on nuclear asymmetry. Embryos
were treated with 0.2% DMSO (control) (A and B),
100 nM vinblastine (microtubule disruptor) (C and
D), and 10 mM aurintricarboxylic acid (ATA; kinesin
inhibitor) (E and F) at EMS cell anaphase. Embryos
were imaged for GFP::WRM-1 (A, C, and E) or
GFP::POP-1 (B, D, and F) 5 min after furrowing
onset (1 min after completion of the division). The
scale bar represents 10 mm. White and whitedotted lines outline EMS daughters (MS and E) and
their nuclei, respectively. Anterior is to the left.
(G and H) Quantified anterior (gray bar, A) and
posterior (dotted bar, P) nuclear GFP::WRM-1 (G)
and GFP::POP-1 (H) fluorescence from images,
acquired as described in (A)(F). A.U., arbitrary
units. Error bars indicate 95% confidence intervals. Asterisk (*), double asterisk (**), and n.s. indicate the significance of differences between the
anterior and posterior sides (Welchs t test;
*p < 0.05; **p < 0.01; n.s., p > 0.05).
reduction in the anterior GFP::APR-1 levels (Figure 3B and Figure S3B). Although
wrm-1(RNAi) embryos still showed significant asymmetry in GFP::APR-1 levels,
the intensity of the anterior signal was
significantly reduced. As with the localization kinetics, the effect of these mutations
n.s.
*
or RNAi on GFP::APR-1 localization strongly correlated with their effects on spindle
asymmetry (compare Figure 2D and Figure 3B). Our results strongly suggest that
spindle asymmetry depends on the cortical GFP::APR-1 level.
A
P
A P
The number of microtubules depends
vinblastine
ATA
on
microtubule nucleation rate and
(n = 10)
(n = 8)
stability. To analyze which of the factors
is the cause for spindle asymmetry,
we used living embryos expressing
GFP::EBP-1. EBP-1 is a C. elegans homolog of the mammalian
EB1 protein, which binds to microtubule plus ends and was
used to analyze microtubule dynamics (Mimori-Kiyosue et al.,
2000b; Srayko et al., 2005). We measured the number and speed
of GFP::EBP-1 comets around the centrosomes for 5 s to
measure microtubule nucleation and polymerization rates,
respectively, and we found that these factors were neither significantly asymmetric nor affected by apr-1 knockdown (Figures
S3DS3F). Then, we examined the possibility that APR-1 on the
cell cortex affects the local microtubule stability similar to
mammalian APC. We measured the EBP-1 residency time on
the cortex and found that EBP-1 comets persisted for a longer
time on the anterior than on the posterior cortex (67% and
27% comets persisted longer than 5 s on the anterior and posterior cortex, respectively) (Figures 3C and 3D). In apr-1(RNAi),
EBP-1 comets on the anterior cortex rapidly detached themselves from the cortex (27% comets persisted longer than 5 s)
similar to the posterior cortex (Figures 3C and 3D). These results
E
B
80
**
n.s.
60
40
20
0
A P
prophase
(n = 4)
A P
metaphase
(n = 8)
telophase
20
A P
telophase
(n = 7)
Number of astral microtubules
15
Anterior
Posterior
10
5
(n = 10)
0
-160 -80 0 80 160 240

Time relative to furrowing onset (sec)
C
-tubulin
-tubulin
merge
WT
WT
WT
mom-2
mom-2
mom-2
D
80
**
n.s.
n.s.
n.s.
n.s.
n.s.
60
40
20
0
A P
WT
(n = 7)
A P
A P
A P
mom-2(or309) mom-5(ne12) dsh(RNAi)
(n = 6)
(n = 7)
(n = 5)
A P
A P
wrm-1(RNAi) apr-1(RNAi)
(n = 5)
(n = 11)
Figure 2. The Wnt Signaling Pathway Establishes Astral Microtubule Asymmetry

(A) Number of astral microtubules around the anterior (green; A) and posterior (blue; P) centrosome was counted in three different focal planes (the plane
containing the centrosome center, and those 1.5 mm from the center) in the EMS cell at each cell-cycle phase, after a-tubulin immunostaining. Microtubules
consisting of the central spindle were not counted.
(B) Number of astral microtubules around the anterior (green) and posterior (blue) centrosome was counted in a single focal plane containing the centrosome
center by the time-lapse recording of embryos expressing GFP::b-tubulin. Telophase is highlighted in gray.
(C) Immunostaining for endogenous a-tubulin (red) and g-tubulin (green) in wild-type (WT) and mom-2(or309) embryos at telophase. Blue is DAPI staining. White
dotted lines outline EMS cells. Arrows indicate centrosomes. The scale bar represents 10 mm.
(D) Number of astral microtubules in Wnt signaling mutants or RNAi embryos, counted as in (A). Error bars indicate 95% confidence intervals.
Asterisk (*), double asterisk (**), and n.s. indicate the significance of differences between the anterior and posterior sides (Welchs t test; *p < 0.05; **p < 0.01; n.s.,
p > 0.05). Red circles indicate significant reductions in number compared to wild-type (p < 0.05). See also Figures S1 and S2.
GFP::APR-1
cortical plane
mid-plane
interphase
GFP::EBP-1
control
anterior
posterior
cell cortex
cell cortex
0:00
0:00
apr-1(RNAi)
anterior
cell cortex
0:00
0:01
0:01
0:01
0:02
0:02
0:02
0:03
0:03
0:03
0:04
0:04
0:04
0:05
0:05
0:05
0:06
0:06
0:06
prophase
metaphase
anaphase
telophase
B (A.U.)
n.s.
n.s.
3500
3000
n.s.
2500
n.s.
2000
1500
1000
500
0
genotypes
transgene
A P A P A P A P A P A P
WT mom-2 mom-5 dsh wrm-1 WT
GFP::APR-1
none
(n = 5) (n = 5) (n = 5) (n = 5) (n = 5) (n = 3)
Frequency (%)
cortical signal intensity
4000
70
60
50
40
30
20
10
0
Microtubule plus-end residence

at the cell cortex
CtrlA
(n = 24)
CtrlP
(n = 22)
aprA
(n = 30)
aprP
(n = 30)
1-2
2-3
3-4
4-5
5>
GFP::EBP-1 residency time (sec)
Figure 3. Wnt-Dependent Asymmetric APR-1 Localization Is Required for Asymmetric Microtubule Stability on the Cell Cortex
(A) GFP::APR-1 localization in mid-planes (left) and cortical planes (right) of living embryos during EMS division at each cell-cycle phase. Arrowheads indicate the
boundary of EMS cell.
(B) Cortical GFP::APR-1 signal intensities in the anterior and posterior sides of the EMS cell with the indicated genotypes. Embryos without the GFP::APR-1
transgene were also quantified as background signals. Error bars indicate 95% confidence intervals. Asterisk (*), double asterisk (**), and n.s. indicate the
significance of differences in intensity between the anterior and posterior sides (Welchs t test; *p < 0.05; **p < 0.01; n.s., p > 0.05). Red circles indicate significant
reductions in intensity compared to wild-type (p < 0.05).
(C) Examples of time lapse images of cortical GFP::EBP-1 movement during telophase of EMS. The left, center, and right panels show the anterior cell cortex
(control), posterior cell cortex (control), and anterior cell cortex [apr-1(RNAi)], respectively. Arrows and arrowheads with distinct colors indicate individual EBP-1
comets and the positions where the comets disappeared, respectively.
suggest that APR-1 regulates local microtubule stability on the

anterior cell cortex. Consistent with this activity, 71% of the
GFP::APR-1 dots (n = 80) on the anterior cortex colocalized
with astral microtubules (n = 57), and 77% of the astral microtubules at the anterior cortex (n = 57) colocalized with APR-1 dots
(n = 44) (Figure S3C). These results suggest that anteriorly localized APR-1 binds and stabilizes plus ends of astral microtubules
at the anterior cortex, creating spindle asymmetry.
asymmetry, which is regulated by nuclear WRM-1, was also

significantly decreased by anterior OICD, although its enhancement by posterior OICD was not significant (Figures 5A and 5E).
Furthermore, when we created spindle asymmetry in mom2(or309) mutants by posterior OICD (Figure S4B), the nuclear
POP-1 asymmetry was restored (Figures 5B and 5F). These
results show that asymmetrically organized microtubules are
required to establish the nuclear WRM-1 asymmetry.
Correlation between Spindle Asymmetry

and Asymmetric Cell Fate
To reveal the importance of spindle asymmetry in cell fate, we
analyzed the relationship between defects in spindle asymmetry
and asymmetric division in mom-2(RNAi) embryos, which have
a weaker phenotype than mom-2(or309null). We first counted
the number of astral microtubules during EMS division at telophase using GFP::b-tubulin; then, (1) about 30 min later, we
analyzed expression levels of end-1::NLS::lacZ, which is normally expressed only in descendants of the posterior daughter
(E) of the EMS cell (Maduro et al., 2005) by immunostaining for
LacZ, or (2) about half a day later, we analyzed defects in asymmetric cell fate as evidenced by the absence of gut auto-fluorescence, which is normally produced by the E cell (Figure 4A). In
wild-type embryos in experiment 1, the ratio of the end-1 reporter expression, represented by log[(fluorescence level of
E daughters)/(that of MS daughters)] was higher than 0.38
(n = 16) (Figure 4B). mom-2(RNAi) embryos with ratios higher
or lower than 0.38 were grouped into END-1high and END1low classes, respectively (Figures 4B and 4C). We found that
END-1high embryos showed significant spindle asymmetry
(Figures 4C and 4D). In contrast, END-1low embryos showed
nearly symmetric spindle structures (Figures 4C and 4D). Consistently, in experiment 2, we found that embryos with and
without gut production (Gut+ and Gutless, respectively, in Figure 4D) showed asymmetric and nearly symmetric spindles,
respectively. These results suggest that spindle asymmetry
plays an important role in regulating asymmetric transcriptional
activities and asymmetric cell fates.
Kinesins Are Involved in Controlling WRM-1 Asymmetry

In mammalian cells, kinesin superfamily 3A (KIF3A)-KIF3B and
kinesin-associated protein KAP3 are involved in b-catenin targeting to the cell protrusion (Jimbo et al., 2002). To analyze the
involvement of kinesins, we first used aurintricarboxylic acid
(ATA). ATA is a kinesin inhibitor that was found by a structurebased computer screening (Hopkins et al., 2000). Although its
specificity in vivo has not been characterized precisely, ATA is
known to inhibit kinesin-dependent processes in neurons and
Xenopus (Batut et al., 2007; Lalli et al., 2003). We found that
treating embryos with ATA caused the symmetric localization
of nuclear WRM-1 and POP-1 (Figures 1E1H), suggesting that
kinesins are involved in WRM-1 asymmetry.
To identify the kinesin genes required for WRM-1 localization,
we individually inhibited 15 kinesin genes by mutations or RNAi,
and scored the Gutless and Psa (phasmid socket absent) phenotypes (Sawa et al., 2000) that probably result from defects in the
asymmetric divisions of EMS and postembryonic T cells, respectively (Table S1). We found that klp-18, bmk-1 and klp-15/klp-16
RNAi showed the Gutless phenotype (2.3%, 5%, and 4.4%,
respectively) and zen-4(or153ts) mutants showed the Psa
phenotype (47%; n > 50) (Table S1). Next, to clarify the involvement of these kinesins in asymmetric division, we observed the
localizations of WRM-1 and POP-1 in kinesin mutants or RNAitreated animals. In the postembryonic T cell at telophase, the
nuclear WRM-1 localization was symmetric in mutants of zen-4,
which encodes a homolog of mammalian kinesin-like protein
(MKLP) (50%; n = 22, Figure 6A). ZEN-4 is known to regulate cytokinesis with CYK-4/MgcRacGAP by forming a complex named
centralspindlin (Jantsch-Plunger et al., 2000; Mishima et al.,
2002). In fact, both zen-4 (22.7%; n = 22) and cyk-4 (42.1%;
n = 19) mutants showed cytokinesis defects in the T cell (Figure 6B). cyk-4 mutants, however, did not display defects in
WRM-1 nuclear localization (n = 11, Figure 6A), indicating that
the WRM-1 localization defect in zen-4 mutants was not an
indirect consequence of cytokinesis failure. Consistent with the
cytokinesis-independent function, ZEN-4 localized not only to
the mid-body, but also to the centrosomes and cell cortex during
T cell division (Figure 6C). These results suggest that ZEN-4
kinesin functions to regulate nuclear WRM-1 asymmetry.
In embryonic EMS division, although inhibition of klp-15/
klp-16, bmk-1, or zen-4 alone did not show defects in nuclear
POP-1 asymmetry (Figures 6D6F), bmk-1;zen-4 double inhibition showed a significant decrease of POP-1 in the anterior
nucleus. In bmk-1;zen-4 embryos but not in zen-4 embryos, we
Experimental Manipulation of Spindle Asymmetry

Affects Nuclear WRM-1 Asymmetry
To examine the importance of spindle asymmetry in WRM-1
nuclear localization, we manipulated the spindle asymmetry by
optically induced centrosome disruption (OICD) (Grill et al.,
2003), in which either the anterior or posterior centrosome
was irradiated with a laser in a dividing EMS cell at anaphase.
The treated embryos were immunostained for GFP::WRM-1,
GFP::POP-1 or endogenous POP-1 after the division. After anterior OICD, which disrupted spindle asymmetry (Figure S4B) but
not the APR-1 localization (Figure S4C), the WRM-1 localization
was symmetric (Figures 5A and 5D). In contrast, posterior OICD,
which enhanced spindle asymmetry (Figures S4A and S4B), also
enhanced the nuclear WRM-1 asymmetry (Figures 5A and 5D).
Consistent with the effects on the WRM-1 localization, POP-1
(D) Frequency of GFP::EBP-1 residency time on the cortex [CtrlA, control anterior; CtrlP, control posterior; aprA, apr-1(RNAi) anterior; and aprP, apr-1(RNAi)
posterior cortex].
Scale bars represent 5 mm. See also Figure S3.
Figure 4. Defects in Spindle Asymmetry Correlate

with Asymmetric Division in mom-2(RNAi)
Embryos
(A) Schematic representation of the experimental strategy.

In the wild-type, E cell descendants always express the
end-1 gene and differentiate into the gut. In mom-2/
wnt(RNAi) embryos, 29% of E cell descendants did not
produce a gut (Gutless phenotype) due to symmetric EMS
cell division. After imaging GFP::b-tubulin in control and
mom-2(RNAi) embryos, (1) expression of NLS::lacZ driven
by the end-1 promoter (end-1 reporter) were analyzed by
anti-LacZ immunostaining, or (2) the Gutless phenotype
was analyzed by observing gut autofluorescence.
(B) Ratios [log(E.x/Ms.x)] of expression levels of the end-1
reporter between EMS granddaughters (E.x versus MS.x)
in control (white circles) and mom-2(RNAi) embryos (red
circles). Based on the ratios in control embryos, mom2(RNAi) embryos were divided into END-1high (ratio >
0.38) and END-1low (ratio < 0.38) groups.
(C) Left: GFP::b-tubulin signals in END-1high and END1low embryos. Upper and lower panels for each END1high and END-1low are images of the anterior (A) and
posterior (P) poles, respectively, obtained from a single
embryo. Dotted white lines indicate EMS boundary.
Right: Expression of the end-1 reporter in END-1high (top)
and END-1low (bottom) embryos detected by anti-LacZ
staining (red). Blue is DAPI. Scale bars represent 10 mm.
(D) Numbers of astral microtubules from the anterior (A) or
posterior (P) poles in the EMS cell at telophase in the
control, END-1high and END-1low embryos in experiment
(1) or in the control and mom-2(RNAi) embryos with (Gut+)
or without (Gut) gut production in experiment (2). Error
bars indicate 95% confidence intervals. p values were
calculated by Welchs t test.
ZEN-4 regulate WRM-1/POP-1 through mediating effects of spindle asymmetry.

DISCUSSION
found that asymmetry of spindles was weaker than in control

embryos, suggesting that bmk-1 may also regulate spindle asymmetry. In contrast, inhibition of klp-18, which encodes a homolog
of Xenopus Xklp2, showed a significant decrease of POP-1 in the
anterior nucleus without affecting asymmetry of astral microtubules (Figures 6D, 6E, 6G, and 6H). Together, these results indicate that nuclear WRM-1/POP-1 asymmetry requires kinesins
(ZEN-4, BMK-1, and KLP-18), and that KLP-18 and possibly
Cortical APR-1/APC Localization

and Spindle Asymmetry
We have shown that APR-1 localizes to the
anterior cell cortex during EMS division, and
that Wnt is required for this localization. This is
consistent with the postembryonic T cell division, where APR-1 is recruited to the cortex by
cortical WRM-1, inhibiting the WRM-1/b-catenin nuclear localization (Mizumoto and Sawa,
2007a). The anterior cortical localization of
WRM-1 in the EMS cell is regulated by MOM-2/
Wnt signaling from the posterior P2 cell (Nakamura et al., 2005); therefore, mom-2 and mom-5/frizzled likely
regulate the APR-1 asymmetry through cortical WRM-1 asymmetry. We also showed that the APR-1 localization also requires
Dishevelled homologs DSH-2 and MIG-5. DSH-2 is known to
localize to the cell cortex, and to accumulate in the posterior
cortex of the EMS cell, at least before its division (Walston
et al., 2004). Although we did not show DSH-2s localization
during the EMS division and its localization is reported to be
GFP::WRM-1
control irradiation
anaphase
1 2
GFP::POP-1
control irradiation
1. Anterior irradiation
2. Control irradiation
3. Posterior irradiation
E (A.U.)
D (A.U.)
posterior irradiation
5000
GFP::WRM-1
4000
12000
3000
2000
1000
0
-1000
-2000
*
Ctrl A-ir P-ir
A-P difference in nuclear signals
anterior irradiation
P-A difference in nuclear signals
anterior irradiation
8000
6000
anti-POP-1
DAPI
4000
2000
0
-2000
(n = 7) (n = 8)(n = 11)
GFP::POP-1
10000
Ctrl A-ir P-ir
(n = 10)(n = 8) (n = 9)
mom-2(or309)
no irradiation
mom-2(or309)
A-P difference in nuclear POP-1 signals
Wild type
(A.U.)
6000
endogenous POP-1
5000
4000
3000
2000
1000
0
-1000
(n = 9)
Wild type
(n = 10)
(n = 5)
Posterior
No
irradiation irradiation
mom-2(or309)
Figure 5. Effect of Centrosomal Irradiation at Anaphase on the Nuclear Localization of WRM-1 and POP-1 at Telophase
(A and B) Wild-type or mom-2(or309) embryos irradiated at anaphase were immunostained for GFP::WRM-1 (A, left), GFP::POP-1 (A, right) or endogenous POP-1
(B) 5 min after furrowing onset. White lines and white dotted lines outline EMS daughter cells and their nuclei, respectively. Scale bars represent 5 mm.
(C) Schematic view of the EMS cell at anaphase with irradiation locations. Dark circles and white ellipses represent centrosomes and chromosomes, respectively.
(DF) Bar graphs represent the results of centrosomal irradiation using GFP::WRM-1 (D), GFP::POP-1 (E), and endogenous POP-1 staining (F) as shown in (A) and
(B). Differences in signal intensities of GFP::WRM-1 (D), GFP::POP-1 (E), or POP-1 (F) between the anterior and posterior nuclei are shown. Error bars indicate
95% confidence intervals. Asterisks indicate significant differences compared to the control irradiation (D and E) or mom-2(or309) embryos with no irradiation
(F) (Welchs t test; p < 0.05). A.U., arbitrary units.
See also Figure S4.
uniform after the late six-cell stage (Walston et al., 2004), DSH-2
might be enriched at the EMS cell posterior cortex during telophase, as seen in postembryonic seam cells (Mizumoto and
Sawa, 2007a). However, dsh(RNAi) experiments revealed that
its effects were observed more strongly in the anterior cortex
than in the posterior cortex. Therefore, DSH-2 appears to function at the anterior cortex to recruit APR-1 to the cortex and
this action may be inhibited in the posterior cortex, probably
by MOM-2. Another possibility is that DSH-2 has both positive
and negative roles in APR-1 localization. While DSH-2 functions
to localize APR-1 to the cortex, the DSH-2 that is recruited to the
posterior cortex by MOM-2 activity might inhibit APR-1 localizing

to that area.
We also found that APR-1 is required for asymmetry in microtubule numbers. APR-1 localizes strongly to the anterior cortex
and colocalizes with peripheral microtubules. Furthermore, the
microtubule plus ends reside for a longer period on the anterior
cell cortex than on the posterior cortex in an APR-1-dependent
manner. Because the dynamic instability of microtubules is regulated by microtubule growing and shrinking rates at their plus
ends (Mitchison and Kirschner, 1984), capturing microtubule
plus ends by APR-1 probably decreases the frequency of
E (A.U.)
Wildtype
Wildtype
cyk-4(ts)
cyk-4(ts)
zen-4(ts)
GFP::POP-1 intensity
400
WRM-1::GFP
22.5C
350
300
250
**
200
**
150
100
50
0
zen-4(ts)
control apr-1
AP
AP
kap-1 klp-18 bmk-1 klp-15/-16

A P
AP
A P
AP
(n = 10) (n = 10) (n = 10) (n = 6)
(A.U.)
250
GFP::POP-1 intensity
F
B
AJM-1::GFP
zen-4(ts)
cyk-4(ts)
ZEN-4::GFP
metaphase
telophase
(n = 9) (n = 13)
15C
25C
200
150
**
**
100
50
0
control zen-4 klp-18 bmk-1 klp-15/-16

AP
AP
A P
AP
AP
zen-4 background
(n = 7)
D
control
apr-1(RNAi)
kap-1(ok676)
zen-4(or153ts)
klp-18(RNAi)
zen-4;klp-18
GFP::POP-1
(n = 8)
(n = 8)
(n = 6) (n = 6)
klp-18(RNAi) , anterior
-tubulin/-tubulin/DAPI
klp-18(RNAi) , posterior
bmk-1(RNAi)
zen-4;bmk-1
H
80
70
60
50
40
30
20
10
0
22.5C
15C
25C
**
A P d
control
(n = 10)
A P d
klp-18
(n = 5)
A P d
A P d
zen-4
zen-4;bmk-1
(n = 7)
(n = 7)
Figure 6. Kinesins Are Required for Nuclear WRM-1 Asymmetry

(A) WRM-1::GFP localization in the postembryonic T cell in wild-type, zen-4(or153ts), and cyk-4(t1689ts) animals at telophase (left). Corresponding DIC images
with the T cell outlined in white, are also shown (right). All animals were grown at 15 C and shifted to 25 C at the three-fold stage, after the birth of the T cell.
(B) Merged DIC and fluorescent images of the T cell daughters in zen-4 and cyk-4 animals expressing AJM-1::GFP (apical membrane marker), showing cytokinesis failure (no visible cell-cell boundary between the sister cells).
shrinking, resulting in stabilization of microtubules and formation

of asymmetric spindles. In mammals, APC has been observed to
stabilize microtubules in migrating epithelial cells. During wound
healing, mammalian APC moves to the leading edges (Nathke
et al., 1996) and stabilizes microtubules, probably by binding
to their plus ends (Mimori-Kiyosue et al., 2000a; Zumbrunn
et al., 2001). This APC localization to the leading edges requires
Wnt5a (Schlessinger et al., 2007). Conversely, in growing axons,
Wnt3a inhibits APC localization to the tip of microtubules, and
the shape of the growth cone is remodeled (Purro et al., 2008).
In the C. elegans EMS cell, MOM-2 appears to recruit APR-1
to the anterior cortex, as seen in migrating epithelial cells, while
it inhibits APR-1 localization to the posterior cortex, as seen in
mammalian axons. Our results show that the regulation of microtubules by Wnts through APC plays important roles in animal
development.
Models for Establishing Nuclear WRM-1 Asymmetry

We have previously shown that the nuclear export rate of WRM1::GFP regulated by APR-1 is asymmetric at telophase, being
higher in the anterior than the posterior nucleus in a postembryonic cell (Mizumoto and Sawa, 2007a). Consistent with
this finding, when nuclear export is inhibited by imb-4/
exportin(RNAi), WRM-1 accumulates symmetrically in the E
and MS nuclei (Nakamura et al., 2005). Therefore, spindle asymmetry likely regulates the nuclear export of WRM-1. In mouse
embryonic fibroblasts and young adult colon cells, the inhibition
of kinesin superfamily proteins Kif3a or KHC, enhances nuclear
accumulation of b-catenin (Corbit et al., 2008; Cui et al., 2002).
Furthermore, in Madin-Darby canine kidney (MDCK) cells,
b-catenin and APC form a complex with KIF3A/B (kinesins)
through KAP3 (kinesin superfamily-associated protein 3). When
KAP3 is inhibited, b-catenin localization to the cell cortex is
disrupted, suggesting that b-catenin is transported toward the
cell cortex by kinesin (Jimbo et al., 2002). As KLP-18/kinesin is
required for asymmetric WRM-1 localization in C. elegans, a
similar kinesin-dependent transport system may be involved in
WRM-1 nuclear localization. In zebrafish and Xenopus embryos,
nuclear localization of the Smad2 protein is regulated by kinesindependent transport to the cell cortex, where Smad2 is phosphorylated by the activated TGF-b receptor (Batut et al., 2007);
this phosphorylation is required for its efficient nuclear localization. Applying this mechanism to WRM-1, WRM-1 transported
to the cell cortex by kinesins might also be modified to enhance
its nuclear export. Although LIT-1 that localizes to the anterior
cortex may be a candidate kinase to modify WRM-1 (Takeshita
and Sawa, 2005), it has been reported that LIT-1 is required for
the nuclear import rather than export of WRM-1 (Nakamura
et al., 2005). Another possibility, not exclusive to the above
Figure 7. A Model of Establishing Nuclear WRM-1 Asymmetry
B
C
WRM-1
KLP-18/kinesin
P
N
MT
[WRM-1N]
[WRM-1C]
[WRM-1P]
Kinesin
(A) Schematic illustration of the cellular compartments (N, nuclear region; P,

perinuclear region; C, cytoplasmic region) at telophase.
(B) A model for microtubule and kinesin regulation of WRM-1s nuclear
localization. The WRM-1 concentrations in each cellular compartment are
indicated by the sizes of the words: [WRM-1N], [WRM-1P], and [WRM-1C].
When kinesin transports WRM-1 toward the cell cortex, [WRM-1C] increases,
with a concomitant decrease in [WRM-1P], causing an equilibrium shift
between WRM-1 nuclear import and export, resulting in reduced [WRM-1N].
(C) Using the spindle asymmetry, kinesin transports WRM-1 more efficiently in
the anterior side, resulting in nuclear WRM-1 asymmetry. Red circles, blue
circles, and blue lines show WRM-1, kinesin, and microtubules (MT),
respectively. Black and yellow arrows indicate WRM-1 nuclear transport or
kinesin transport, respectively.
model, is that microtubule-dependent transport of WRM-1

toward the cell cortex reduces WRM-1 in the perinuclear region,
thereby promoting WRM-1 nuclear export due to a shift of equilibrium between nuclear import and export. The spindle asymmetry might cause this process to occur asymmetric manner,
resulting in an uneven distribution of nuclear WRM-1 (Figure 7).
Nuclear b-catenin asymmetry is also observed in two distantly
related bilaterian superphyla (Ecdysozoa, C. elegans; Lophotrochozoa, Platynereis dumerilii) (Schneider and Bowerman, 2007);
therefore, we expect that a similar mechanism for establishing
b-catenin asymmetry, which may have originated in an ancient
metazoan, may be conserved in a variety of organisms.
Microtubules Regulate Protein Nuclear Localization
Microtubules are required not only in the nuclear transport of
Smad2, as described above, but also in localizing the cancer
regulatory proteins p53, PTHrP, and Rb to the nucleus (Roth
et al., 2007). The tumor suppressor protein p53 forms a complex with polymerized microtubules and dynein. Introducing
(C) Localization of ZEN-4::GFP during T cell division. Arrowheads and arrows indicate T cell outlines and centrosomes, respectively.
(DF) GFP::POP-1 localization in kinesin gene mutants or RNAi embryos. Quantified fluorescence intensities were shown in (E) and (F) (A.U., arbitrary unit). zen4(or153ts) mutants and animals used in (F) were cultured at 15 C and shifted to 25 C just before EMS division. Other animals were cultured at 22.5 C. kap-1 is
a C. elegans homolog of mammalian KAP3.
(G) Immunostaining of a-tubulin (red) and g-tubulin (green) in klp-18(RNAi) embryos. Blue is DAPI.
(H) Number of astral microtubules in the anterior (A) or posterior (P) sides with their difference (d) in kinesin mutants or RNAi animals.
Error bars in (E), (F), and (H) indicate 95% confidence intervals. Asterisks in (E), (F), and (H) indicate the significant difference compared to control (Welchs t test;
p < 0.01). Lines and dotted lines in (A), (D), and (G) indicate cell boundaries and nuclei, respectively. Scale bars represent 10 mm. See also Table S1.
a microtubule-depolymerizing drug or an anti-dynein antibody

into the cell inhibits nuclear localization of p53, suggesting that
it is localized by dynein-dependent transport along microtubules
(Giannakakou et al., 2000). In all of the cases reported so far, the
microtubules themselves were not shown to be regulated, even
though they are required for regulating the nuclear localization of
proteins. Our results reveal a method of regulating protein
nuclear localization by controlling the microtubules. In the cells
of any organism, various signals often prompt the remodeling
of microtubule organizations through protein activity on the cell
cortex (Gundersen et al., 2004). Our results imply that such
microtubule reorganization may have a direct impact on gene
expression in the nucleus, as well as on cell polarity and shape.
containing less than 0.2% DMSO), were introduced into the space to displace
the egg salt buffer (Edgar, 1995). To introduce the drugs without affecting cell
division, we produced a hole on the eggshell by irradiating the trypan blue
particles on it specifically, when the dividing EMS cell was at anaphase. Permeabilization of the eggshell was confirmed by the slight swelling or shrinking
of the embryo.
Immunostaining
Embryos were freeze-cracked and fixed in 20 C methanol for a-tubulin,
g-tubulin, and GFP::APR-1 staining or, for GFP::WRM-1, GFP::POP-1 and
20 C acetone. For
LacZ staining, in
20 C methanol followed by
GFP::WRM-1 staining, after being fixed and washed, the embryos were incubated with Image-iT FX Signal Enhancer (Invitrogen) at room temperature for
30 min. The embryos were stained in the presence of 1% BSA (except for
GFP::WRM-1 staining) with the following primary antibodies: anti-a-tubulin
mouse monoclonal (clone DM1a; 1:1000; Sigma-Aldrich), anti-g-tubulin rabbit
polyclonal (LL-17; 1:1000; Sigma-Aldrich), anti-GFP rabbit polyclonal (1:1000;
Invitrogen), or anti-LacZ mouse monoclonal (Z378; 1:1000; Promega). Primary antibodies were detected by goat anti-mouse Rhodamine-X (1:1000;
Invitrogen), goat anti-rabbit Fluorescein (1:1000; Invitrogen), or goat antimouse Alexa 568 highly cross-adsorbed (1:1000; Invitrogen). For endogenous
POP-1 staining, embryos were fixed and immunostained as described previously (Sugioka and Sawa, 2010) with an anti-POP-1 antibody (Lin et al.,
1998).
C. elegans Culture and Strains

All strains used in this study were cultured by standard methods (Brenner,
1974). zen-4(or153ts) and cyk-4(t1689ts) mutants were grown at 15 C, and
four-cell and three-fold embryos for the embryonic and postembryonic experiment, respectively, were shifted to 25 C. The following integrated transgenic
lines were used: ruIs48 (Praitis et al., 2001) for GFP::b-tubulin; neIs2 (Nakamura
et al., 2005) for GFP::WRM-1; osIs5 (Mizumoto and Sawa, 2007a) for WRM1::GFP; wIs117 (Maduro et al., 2002) for GFP::POP-1; wIs27 (a gift from Joel
Rothman) for end-1 promoter driven LacZ::GFP; tjIs8 (Motegi et al., 2006) for
GFP::EBP-1; and kuIs46 (Chen and Han, 2001) for AJM-1::GFP. For generation
of the GFP::APR-1 construct, a genomic fragment of apr-1 containing the fulllength coding sequence was amplified by PCR and inserted into pID3.01
(DAgostino et al., 2006). The plasmid was integrated by microparticle
bombardment (Praitis et al., 2001) to create the GFP::APR-1 transgene osIs15.
RNAi
So that double-stranded RNAs (dsRNAs) could be produced, the following
complementary DNAs (cDNAs) were used as templates: yk40c12 (apr-1),
yk213d6 (wrm-1), yk714f1 (mom-2), yk55h11 (dsh-2), yk216a12 (mig-5),
yk1084e03 (pry-1), yk1134d03 (klp-6), yk735h1 (klp-19), yk495d7 (bmk-1),
yk616a7 (vab-8), yk1618h03 (klp-3), yk1466a11 (klp-15/klp-16), yk1547b05
(klp-18), and pMM414 (pop-1) (Maduro et al., 2002). Embryos were collected
2440 hr after the injection of dsRNAs or a transcription buffer as a control.
Microscopy and Analysis of Living Embryos
All embryos were dissected in an egg salt buffer from gravid hermaphrodites
(Edgar, 1995) and handled individually with mouth pipettes. For live imaging,
except for drug treatments, the embryos were mounted on 4% agar pads under
a coverslip and sealed with Vaseline. Embryos were observed at room temperature by a CSU10 spinning-disc confocal system (Yokogawa Electric) mounted
on an AxioPlan2 microscope (Carl Zeiss) with a Plan-Apochromat 1003 1.4 NA
oil immersion lens. The specimens were illuminated with a diode-pumped
solid-state 488 nm laser (HPU50100, 20 mW; Furukawa Electronic). Images
were acquired in 0.5 mm and 2 mm serial Z sections for fixed and live embryos,
respectively, with an Orca ER12-bit cooled CCD camera (Hamamatsu
Photonics), and the acquisition system was controlled by IP lab software
(2 3 2 binning; Scananalytics). Acquired images were processed with the IP
lab software and Adobe Photoshop (Adobe Systems). Microtubule numbers
were manually counted along three (center, 1.5 mm above and 1.5 mm below
the center of centrosome) and a single (center of centrosome) focal plane
selected from the serial images of fixed and live embryos, respectively.
Drug Treatments
Drugs were introduced into embryos as described previously (Hyman and
White, 1987; Lee and Goldstein, 2003). In brief, four-cell-stage embryos
coated with trypan blue were mounted in an egg salt buffer on 0.1% polyL-lysine coated coverslips, with Vaseline used as spacers. The drugs, (final
concentration, 100 nM vinblastine [Wako], 10 mg/ml Nocodazole [SigmaAldrich], and 10 mM aurintricarboxylic acid [Sigma-Aldrich] in an egg buffer
Optically Induced Centrosome Disruption

For centrosome irradiation, early-stage six-cell embryos were mounted on
0.1% poly-L-lysine-coated glass slides with Vaseline spacers, and irradiated
with several pulses of a 2 mW pulsed nitrogen laser (VSL-337; Laser Science)
as described (Labbe et al., 2004). Five minutes after the onset of furrowing in
the EMS cells, the slides were transferred into liquid nitrogen and immunostained as described below. Some embryos with abnormal nuclei, likely
caused by the misirradiation of chromosomes, were discarded.
Quantification of Fluorescence Intensities

The average fluorescence intensities of the GFP::WRM-1, GFP::POP-1, and
endogenous POP-1 signals in the nuclei were quantified by IP lab software
(Scanalytics), subtracting the background fluorescence, which was defined
as the average intensity in a cytoplasmic region of similar size. The values
were used to calculate the total nuclear fluorescence. The fluorescence intensities of cortical GFP::APR-1 were quantified using images of the mid-plane
(e.g., Figure 3, left panel), and the values were divided into the anterior and
posterior regions at the centers of the apical and basolateral cortex.
Supplemental Information includes four figures and one table and can be
ACKNOWLEDGMENTS
We thank Rueyling Lin for the anti-POP-1 antibody, Morris Maduro for
pMM414, Yuji Kohara for cDNAs, Geraldine Seydoux for the pID3.01B, Joel
Rothman, Francis J. McNally, Shohei Mitani, and the Caenorhabditis Genetics
Center (funded by the National Institutes of Health National Center for
Research Resources) for strains, Asako Sugimoto, Mika Toya, Bob Goldstein,
Kumiko Oishi, Yukinobu Arata, Yukimasa Shibata, and Shinji Ihara for technical
advice and helpful discussion, and Hazuki Hiraga for proofreading of our
manuscript. This work was supported by Grants-in-Aid for Scientific Research
from the Ministry of Education, Culture, Sports, Science, and Technology of
Japan to H.S. K.S. was supported by the Junior Research Associate Program,
RIKEN.
Received: October 18, 2010
REFERENCES
activating protein (GAP) required for central spindle formation and cytokinesis.
J. Cell Biol. 149, 13911404.
Batut, J., Howell, M., and Hill, C.S. (2007). Kinesin-mediated transport of
Smad2 is required for signaling in response to TGF-b ligands. Dev. Cell 12,
261274.
Jimbo, T., Kawasaki, Y., Koyama, R., Sato, R., Takada, S., Haraguchi, K., and
Akiyama, T. (2002). Identification of a link between the tumour suppressor APC
and the kinesin superfamily. Nat. Cell Biol. 4, 323327.
Brenner, S. (1974). The genetics of Caenorhabditis elegans. Genetics 77,

7194.
Brocardo, M., and Henderson, B.R. (2008). APC shuttling to the membrane,
nucleus and beyond. Trends Cell Biol. 18, 587596.
Cadigan, K.M., and Peifer, M. (2009). Wnt signaling from development to
disease: insights from model systems. Cold Spring Harb Perspect Biol 1,
a002881.
Chen, Z., and Han, M. (2001). Role of C. elegans lin-40 MTA in vulval fate specification and morphogenesis. Development 128, 49114921.
Clevers, H. (2006). Wnt/b-catenin signaling in development and disease. Cell
127, 469480.
Corbit, K.C., Shyer, A.E., Dowdle, W.E., Gaulden, J., Singla, V., Chen, M.H.,
Chuang, P.T., and Reiter, J.F. (2008). Kif3a constrains b-catenin-dependent
Wnt signalling through dual ciliary and non-ciliary mechanisms. Nat. Cell
Biol. 10, 7076.
Cui, H., Dong, M., Sadhu, D.N., and Rosenberg, D.W. (2002). Suppression of
kinesin expression disrupts adenomatous polyposis coli (APC) localization
and affects b-catenin turnover in young adult mouse colon (YAMC) epithelial
cells. Exp. Cell Res. 280, 1223.
DAgostino, I., Merritt, C., Chen, P.L., Seydoux, G., and Subramaniam, K.
(2006). Translational repression restricts expression of the C. elegans Nanos
homolog NOS-2 to the embryonic germline. Dev. Biol. 292, 244252.
Edgar, L.G. (1995). Blastomere culture and analysis. Methods Cell Biol. 48,
303321.
Eleftheriou, A., Yoshida, M., and Henderson, B.R. (2001). Nuclear export of
human b-catenin can occur independent of CRM1 and the adenomatous
polyposis coli tumor suppressor. J. Biol. Chem. 276, 2588325888.
Labbe, J.C., McCarthy, E.K., and Goldstein, B. (2004). The forces that position
a mitotic spindle asymmetrically are tethered until after the time of spindle
assembly. J. Cell Biol. 167, 245256.
Lalli, G., Gschmeissner, S., and Schiavo, G. (2003). Myosin Va and microtubule-based motors are required for fast axonal retrograde transport of tetanus
toxin in motor neurons. J. Cell Sci. 116, 46394650.
Lee, J.Y., and Goldstein, B. (2003). Mechanisms of cell positioning during
C. elegans gastrulation. Development 130, 307320.
Lin, R., Thompson, S., and Priess, J.R. (1995). pop-1 encodes an HMG box
protein required for the specification of a mesoderm precursor in early
C. elegans embryos. Cell 83, 599609.
Lin, R., Hill, R.J., and Priess, J.R. (1998). POP-1 and anterior-posterior fate
decisions in C. elegans embryos. Cell 92, 229239.
Lo, M.C., Gay, F., Odom, R., Shi, Y., and Lin, R. (2004). Phosphorylation by the
b-catenin/MAPK complex promotes 14-3-3-mediated nuclear export of TCF/
POP-1 in signal-responsive cells in C. elegans. Cell 117, 95106.
Maduro, M.F., Lin, R., and Rothman, J.H. (2002). Dynamics of a developmental
switch: recursive intracellular and intranuclear redistribution of Caenorhabditis
elegans POP-1 parallels Wnt-inhibited transcriptional repression. Dev. Biol.
248, 128142.
Maduro, M.F., Kasmir, J.J., Zhu, J., and Rothman, J.H. (2005). The Wnt
effector POP-1 and the PAL-1/Caudal homeoprotein collaborate with SKN-1
to activate C. elegans endoderm development. Dev. Biol. 285, 510523.
Mimori-Kiyosue, Y., Shiina, N., and Tsukita, S. (2000a). Adenomatous polyposis coli (APC) protein moves along microtubules and concentrates at their
growing ends in epithelial cells. J. Cell Biol. 148, 505518.
Etienne-Manneville, S., and Hall, A. (2003). Cdc42 regulates GSK-3b and

adenomatous polyposis coli to control cell polarity. Nature 421, 753756.
Mimori-Kiyosue, Y., Shiina, N., and Tsukita, S. (2000b). The dynamic behavior
of the APC-binding protein EB1 on the distal ends of microtubules. Curr. Biol.
10, 865868.
Fagotto, F., Gluck, U., and Gumbiner, B.M. (1998). Nuclear localization
signal-independent and importin/karyopherin-independent nuclear import of
b-catenin. Curr. Biol. 8, 181190.
Miralles, F., Posern, G., Zaromytidou, A.I., and Treisman, R. (2003). Actin
dynamics control SRF activity by regulation of its coactivator MAL. Cell 113,
329342.
Giannakakou, P., Sackett, D.L., Ward, Y., Webster, K.R., Blagosklonny, M.V.,
and Fojo, T. (2000). p53 is associated with cellular microtubules and is transported to the nucleus by dynein. Nat. Cell Biol. 2, 709717.
Mishima, M., Kaitna, S., and Glotzer, M. (2002). Central spindle assembly and
cytokinesis require a kinesin-like protein/RhoGAP complex with microtubule
bundling activity. Dev. Cell 2, 4154.
Goldstein, B. (1995a). An analysis of the response to gut induction in the

C. elegans embryo. Development 121, 12271236.
Mitchison, T., and Kirschner, M. (1984). Dynamic instability of microtubule

growth. Nature 312, 237242.
Goldstein, B. (1995b). Cell contacts orient some cell division axes in the
Caenorhabditis elegans embryo. J. Cell Biol. 129, 10711080.
Mizumoto, K., and Sawa, H. (2007a). Cortical b-catenin and APC regulate
asymmetric nuclear b-catenin localization during asymmetric cell division in
C. elegans. Dev. Cell 12, 287299.
Goldstein, B., Takeshita, H., Mizumoto, K., and Sawa, H. (2006). Wnt signals
can function as positional cues in establishing cell polarity. Dev. Cell 10,
391396.
Mizumoto, K., and Sawa, H. (2007b). Two betas or not two betas: regulation of
asymmetric division by b-catenin. Trends Cell Biol. 17, 465473.
Grill, S.W., Howard, J., Schaffer, E., Stelzer, E.H., and Hyman, A.A. (2003). The
distribution of active force generators controls mitotic spindle position.
Science 301, 518521.
Motegi, F., Velarde, N.V., Piano, F., and Sugimoto, A. (2006). Two phases of
astral microtubule activity during cytokinesis in C. elegans embryos. Dev.
Cell 10, 509520.
Gundersen, G.G., Gomes, E.R., and Wen, Y. (2004). Cortical control of microtubule stability and polarization. Curr. Opin. Cell Biol. 16, 106112.
Munemitsu, S., Souza, B., Muller, O., Albert, I., Rubinfeld, B., and Polakis, P.
(1994). The APC gene product associates with microtubules in vivo and
promotes their assembly in vitro. Cancer Res. 54, 36763681.
Herman, M.A., Vassilieva, L.L., Horvitz, H.R., Shaw, J.E., and Herman, R.K.
(1995). The C. elegans gene lin-44, which controls the polarity of certain asymmetric cell divisions, encodes a Wnt protein and acts cell nonautonomously.
Cell 83, 101110.
Hopkins, S.C., Vale, R.D., and Kuntz, I.D. (2000). Inhibitors of kinesin activity
from structure-based computer screening. Biochemistry 39, 28052814.
Nakamura, K., Kim, S., Ishidate, T., Bei, Y., Pang, K., Shirayama, M., Trzepacz,
C., Brownell, D.R., and Mello, C.C. (2005). Wnt signaling drives WRM-1/
b-catenin asymmetries in early C. elegans embryos. Genes Dev. 19, 1749
1754.
Hyman, A.A., and White, J.G. (1987). Determination of cell division axes in the
early embryogenesis of Caenorhabditis elegans. J. Cell Biol. 105, 21232135.
Nathke, I.S., Adams, C.L., Polakis, P., Sellin, J.H., and Nelson, W.J. (1996). The
adenomatous polyposis coli tumor suppressor protein localizes to plasma
membrane sites involved in active cell migration. J. Cell Biol. 134, 165179.
Jantsch-Plunger, V., Gonczy, P., Romano, A., Schnabel, H., Hamill, D., Schnabel, R., Hyman, A.A., and Glotzer, M. (2000). CYK-4: A Rho family gtpase
Phillips, B.T., and Kimble, J. (2009). A new look at TCF and b-catenin through
the lens of a divergent C. elegans Wnt pathway. Dev. Cell 17, 2734.
Praitis, V., Casey, E., Collar, D., and Austin, J. (2001). Creation of low-copy
integrated transgenic lines in Caenorhabditis elegans. Genetics 157, 1217
1226.
Srayko, M., Kaya, A., Stamford, J., and Hyman, A.A. (2005). Identification and
characterization of factors required for microtubule growth and nucleation in
the early C. elegans embryo. Dev. Cell 9, 223236.
Purro, S.A., Ciani, L., Hoyos-Flight, M., Stamatakou, E., Siomou, E., and Salinas, P.C. (2008). Wnt regulates axon behavior through changes in microtubule
growth directionality: a new role for adenomatous polyposis coli. J. Neurosci.
28, 86448654.
Sugioka, K., and Sawa, H. (2010). Regulation of asymmetric positioning of

nuclei by Wnt and Src signaling and its roles in POP-1/TCF nuclear asymmetry
in Caenorhabditis elegans. Genes Cells 15, 397407.
Rocheleau, C.E., Downs, W.D., Lin, R., Wittmann, C., Bei, Y., Cha, Y.H., Ali, M.,
Priess, J.R., and Mello, C.C. (1997). Wnt signaling and an APC-related gene
specify endoderm in early C. elegans embryos. Cell 90, 707716.
Takeshita, H., and Sawa, H. (2005). Asymmetric cortical and nuclear localizations of WRM-1/b-catenin during asymmetric cell division in C. elegans. Genes
Dev. 19, 17431748.
Roth, D.M., Moseley, G.W., Glover, D., Pouton, C.W., and Jans, D.A. (2007). A
microtubule-facilitated nuclear import pathway for cancer regulatory proteins.
Traffic 8, 673686.
Thorpe, C.J., Schlesinger, A., Carter, J.C., and Bowerman, B. (1997). Wnt
signaling polarizes an early C. elegans blastomere to distinguish endoderm
from mesoderm. Cell 90, 695705.
Sawa, H., Kouike, H., and Okano, H. (2000). Components of the SWI/SNF
complex are required for asymmetric cell division in C. elegans. Mol. Cell 6,
617624.
Schlesinger, A., Shelton, C.A., Maloof, J.N., Meneghini, M., and Bowerman, B.
(1999). Wnt pathway components orient a mitotic spindle in the early
Caenorhabditis elegans embryo without requiring gene transcription in the
responding cell. Genes Dev. 13, 20282038.
Schlessinger, K., McManus, E.J., and Hall, A. (2007). Cdc42 and noncanonical
Wnt signal transduction pathways cooperate to promote cell polarity. J. Cell
Biol. 178, 355361.
Schneider, S.Q., and Bowerman, B. (2007). b-Catenin asymmetries after all
animal/vegetal- oriented cell divisions in Platynereis dumerilii embryos
mediate binary cell-fate specification. Dev. Cell 13, 7386.
Walston, T., Tuskey, C., Edgar, L., Hawkins, N., Ellis, G., Bowerman, B., Wood,
W., and Hardin, J. (2004). Multiple Wnt signaling pathways converge to orient
the mitotic spindle in early C. elegans embryos. Dev. Cell 7, 831841.
Wiechens, N., and Fagotto, F. (2001). CRM1- and Ran-independent nuclear
export of b-catenin. Curr. Biol. 11, 1827.
Yokoya, F., Imamoto, N., Tachibana, T., and Yoneda, Y. (1999). b-catenin can
be transported into the nucleus in a Ran-unassisted manner. Mol. Biol. Cell 10,
11191131.
Zumbrunn, J., Kinoshita, K., Hyman, A.A., and Nathke, I.S. (2001). Binding of
the adenomatous polyposis coli protein to microtubules increases microtubule
stability and is regulated by GSK3 b phosphorylation. Curr. Biol. 11, 4449.
Regulation of the MEX-5 Gradient

by a Spatially Segregated
Kinase/Phosphatase Cycle
Erik E. Griffin,1 David J. Odde,2 and Geraldine Seydoux1,*
1Department of Molecular Biology and Genetics, Howard Hughes Medical Institute, Center for Cell Dynamics, Johns Hopkins School of
Medicine, 725 N. Wolfe Street, PCTB 706, Baltimore, MD 21205, USA
2Department of Biomedical Engineering, University of Minnesota, Minneapolis, MN 55455, USA
*Correspondence: gseydoux@jhmi.edu
DOI 10.1016/j.cell.2011.08.012
SUMMARY
Protein concentration gradients encode spatial information across cells and tissues and often depend
on spatially localized protein synthesis. Here, we report that a different mechanism underlies the MEX-5
gradient. MEX-5 is an RNA-binding protein that
becomes distributed in a cytoplasmic gradient
along the anterior-to-posterior axis of the one-cell
C. elegans embryo. We demonstrate that the MEX5 gradient is a direct consequence of an underlying
gradient in MEX-5 diffusivity. The MEX-5 diffusion
gradient arises when the PAR-1 kinase stimulates
the release of MEX-5 from slow-diffusive, RNA-containing complexes in the posterior cytoplasm. PAR1 directly phosphorylates MEX-5 and is antagonized
by the spatially uniform phosphatase PP2A. Mathematical modeling and in vivo observations demonstrate that spatially segregated phosphorylation
and dephosphorylation reactions are sufficient to
generate stable protein concentration gradients in
the cytoplasm. The principles demonstrated here
apply to any spatially segregated modification cycle
that affects protein diffusion and do not require
protein synthesis or degradation.
INTRODUCTION
Protein gradients are an efficient way to encode spatial information within cells and across tissues. The mechanisms that
generate and maintain protein gradients have been the subject
of extensive theoretical and experimental analyses (Wartlick
et al., 2009). Most studies have emphasized the role of a localized
protein source as the foundational asymmetry underlying
gradient formation. For example, in Drosophila embryos, the
Bicoid protein is synthesized at one end of the egg from a localized pool of bicoid mRNA. Diffusion away from the local source
and uniform protein degradation across the egg generate a
concentration gradient over the course of 2 hr (Ephrussi and
St Johnston, 2004; Little et al., 2011). Extracellular gradients

also depend on the localization of specialized cells that synthesize and secrete the signal (source) among cells that respond
to and internalize the signal (sink) (Wartlick et al., 2009).
A spatially segregated source/sink model can also account for
the formation of phosphorylation gradients or phosphogradients. Phosphogradients have been implicated in the spatial
organization of signal transduction pathways where phosphorylation modulates protein activity. Phosphogradients arise when
a diffusing substrate is acted upon by a kinase (source) and
phosphatase (sink) that are separated in space (Brown and Kholodenko, 1999). In phosphogradients, the ratio of unphosphorylated to phosphorylated substrate varies in space, but the overall
concentration of the substrate is uniform (Brown and Kholodenko, 1999; Coppey et al., 2008; Fuller et al., 2008; Kalab
et al., 2002; Maeder et al., 2007; Su et al., 1998). In 2008, Lipkow
and Odde predicted that, if phosphorylation changes the diffusivity of the substrate, spatially segregated kinase/phosphatase
cycles would also affect the overall distribution of the substrate
to generate a protein concentration gradient (Lipkow and
Odde, 2008). The spatial bias in the generation of the phosphorylated isoform generates a diffusion gradient that causes the
substrate to concentrate in regions of low diffusivity (Lipkow
and Odde, 2008). In the present study, we provide experimental
evidence in support of this model in C. elegans.
The C. elegans one-cell embryo (zygote) is a classic model for
the study of intracellular asymmetries (Goldstein and Macara,
2007; St Johnston and Ahringer, 2010). After fertilization, a group
of conserved polarity regulators, the PAR proteins, sort into
anterior (PAR-3/PAR-6/PKC-3) and posterior (PAR-2 and PAR1) domains in the actin-rich layer (or cortex) under the plasma
membrane (Kemphues, 2000). In response to PAR asymmetry
at the cortex, cell-fate determinants become asymmetrically
distributed in the cytoplasm. Among them is the RNA-binding
protein MEX-5, which redistributes in 10 min into an anteriorhigh/posterior-low gradient across the length of the 50 mm
zygote (Schubert et al., 2000; Tenlen et al., 2008). MEX-5, in
turn, partitions other factors such as PIE-1 to the posterior cytoplasm and PLK-1 to the anterior cytoplasm (Budirahardja and
Gonczy, 2008; Mello et al., 1996; Rivers et al., 2008; Schubert
et al., 2000). Consequently, during the first cell division, the
two daughter blastomeres inherit different determinants, which
help to specify their distinct fates (anterior/somatic and posterior/germline). Mutations in the PARs cause MEX-5 (and its
targets) to remain symmetrically distributed (Schubert et al.,
2000; Tenlen et al., 2008), but the mechanisms linking PAR
asymmetry to the MEX-5 gradient are not known.
Fluorescence recovery after photobleaching (FRAP) and
fluorescence correlation spectroscopy (FCS) experiments have
shown that, in polarized zygotes, GFP::MEX-5 diffuses faster
in the posterior cytoplasm, where MEX-5 protein concentration
is lowest (Daniels et al., 2010; Tenlen et al., 2008). Fast diffusion
requires par-1 activity and a C-terminal serine in MEX-5 (S458),
which is phosphorylated in a par-1- and par-4-dependent
manner in vivo (Tenlen et al., 2008). Phosphorylation of S458,
however, does not correlate with gradient formation or fast
diffusion, suggesting that other mechanisms regulate MEX-5
asymmetry (Tenlen et al., 2008). Two speculative models have
been proposed. The first model invokes dynamic binding of
MEX-5 to cytoskeletal elements asymmetrically distributed in
the cytoplasm (Tenlen et al., 2008). In this model, the PARs
localize MEX-5 indirectly by localizing factors, such as myosin,
that retard MEX-5 diffusion in the anterior cytoplasm (Tenlen
et al., 2008). A second model proposes that the PARs regulate
MEX-5 distribution by forming reactive surfaces in the anterior
and posterior cortices, which locally decrease and increase,
respectively, the rate of MEX-5 diffusion (Daniels et al., 2010).
How the PARs modify MEX-5 diffusion, and how differences
originated at the cortex are propagated through the cytoplasm,
however, is not known.
In this study, we present evidence that the MEX-5 gradient
arises as a direct consequence of a complementary PAR-1
kinase activity gradient in the cytoplasm. We demonstrate
that MEX-5 is a substrate of PAR-1 and identify PP2A as the
opposing phosphatase in the cytoplasm. Our findings reveal
an unexpected direct patterning role for PAR-1 in the cytoplasm
and provide experimental evidence for the theoretical model of
Lipkow and Odde (2008).
RESULTS
A MEX-5 Diffusion Gradient Underlies the MEX-5
Concentration Gradient
To examine MEX-5 dynamics in live zygotes, we generated
a Dendra::MEX-5 fusion. Dendra is a photoactivatable fluorescent protein that is photoconverted irreversibly from green to
red fluorescence by exposure to 405 nm light (Gurskaya et al.,
2006). Unlike FRAP, photoconversion is a positive marking technique that can be used to measure rates of protein degradation
and diffusion, without interference from new protein synthesis
(Lippincott-Schwartz and Patterson, 2008). We first photoconverted Dendra::MEX-5 throughout the zygote before polarization
(prior to appearance of the pronuclei). We found that photoconverted Dendra::MEX-5 (DendraR::MEX-5) formed an 3-fold
anterior-posterior gradient by nuclear envelope breakdown
(NEBD, first mitotic division), as is observed for endogenous
MEX-5 (Figures 1A and 1B). Total levels of DendraR::MEX-5 did
not change during gradient formation: levels increased in the
anterior half and decreased in the posterior half by 25% (Figure 1C). We conclude that formation of the MEX-5 gradient
involves redistribution of existing MEX-5 and does not require

MEX-5 synthesis or degradation.
Next, to compare mobility of MEX-5 between the anterior and
posterior, we photoconverted Dendra::MEX-5 in two stripes at
30% and 70% embryo length during polarization (Figure 1D).
DendraR::MEX-5 diffused symmetrically away from both stripes
with no directional bias (Figure 1D). The apparent diffusivity of
DendraR:MEX-5, however, appeared to differ between the two
stripes, with faster diffusion in the posterior stripe (Figure 1D).
These observations are consistent with earlier FRAP experiments, which showed that GFP::MEX-5 diffuses faster in the
posterior cytoplasm after polarization (Daniels et al., 2010; Tenlen et al., 2008).
To examine MEX-5 mobility systematically during polarization,
we measured the apparent diffusion coefficient (Dc) of DendraR::MEX-5 at 17 positions along the long (anterior-posterior)
axis and 3 positions along the short axis, before polarization
(before pronuclear formation), at the onset of polarization (pronuclear formation), and after polarization (NEBD). The apparent Dc
of DendraR::MEX-5 was uniformly slow before pronuclear formation (average Dc between 10% and 90% embryo length was
0.78 mm2/s) (Figure 1E). After pronuclear formation, the apparent
Dc of DendraR::MEX-5 increased to an average of 1.70 mm2/s.
This increase was observed throughout the central cytoplasm,
but not in the cytoplasm nearest the cortex (peripheral cytoplasm) where DendraR::MEX-5 diffusion remained slow (Figure 1E). By NEBD, the apparent Dc of DendraR::MEX-5 was
graded linearly throughout the cytoplasm, with the lowest value
at the anterior-most position and the highest value at the posterior-most position, mirroring the MEX-5 protein concentration
gradient (compare Figures 1B and 1E). We conclude that
redistribution of MEX-5 correlates temporally and spatially with
changes in MEX-5 diffusion.
par-1 Is Necessary and Sufficient to Increase MEX-5
Diffusion in Zygotes
To determine whether the anterior or posterior PARs regulate
MEX-5 dynamics, we monitored MEX-5 distribution and diffusion
at NEBD in zygotes defective for the anterior kinase aPKC/
PKC-3 or the posterior kinase PAR-1 (Figure 2A). pkc-3(RNAi)
embryos lack PKC-3 and have uniform PAR-1 (Figure 2A and
Figure S1A available online). The par-1 allele it51 inactivates
PAR-1 kinase activity but does not affect PAR-1 or PKC-3 localizations (Figures 2A and 2B; Figure S1B) (Cheeks et al., 2004;
Guo and Kemphues, 1995). Previous work has shown that in
par-1(RNAi) embryos, MEX-5 mobility does not increase in the
posterior cytoplasm and MEX-5 does not segregate (Tenlen
et al., 2008). We found that DendraR::MEX-5 remained symmetrically distributed in both pkc-3(RNAi) and par-1(it51) zygotes
(Figures S1A and S1B). Strikingly, the apparent Dc of DendraR::
MEX-5 was uniformly high in pkc-3(RNAi) zygotes and uniformly
low in par-1(it51) and par-1(it51);pkc-3(RNAi) zygotes (Figure 2C).
We conclude that PAR-1 functions downstream of PKC-3 and
is required to stimulate MEX-5 diffusion.
In polarized zygotes, PAR-1 kinase is present both in the cytoplasm and on the posterior cortex (Guo and Kemphues, 1995).
To examine PAR-1 dynamics during polarization, we imaged
zygotes expressing a full-length GFP::PAR-1 fusion. Before
5m
Endogenous MEX-5
DendraR::MEX-5
Anterior
Posterior
After PN
0%
100%
Embryo Length
Before PN
Formation
Relative Concentration
(AU)
NEBD
DendraR::MEX-5
PN
Meeting
NEBD
2 second
Relative Intensity
(AU)
Before PN
D
1 second
DendraR::MEX-5
Anterior
Total
Posterior
Time (minutes)
3 second
12
Anterior
Posterior
Diffusion Coefficient (m2/sec)
Before PN
10%
10%
50%
After PN
NEBD
90%
90%
Anterior
10%
Posterior
30%
50%
70%
90%
10% 50% 90%
Embryo Length
Embryo Length
Long Axis
Short Axis
Figure 1. A Gradient in Diffusivity Underlies the MEX-5 Gradient

(A) Time-lapse photomicrographs of a zygote expressing Dendra::MEX-5 photoconverted before pronuclear formation (before polarization). All embryos in this
and subsequent figures are oriented with anterior to the left and posterior to the right. PN stands for pronuclear formation, which marks the onset of polarity. NEBD
stands for nuclear envelope breakdown (mitosis) and occurs 10 min after pronuclear formation, by which time MEX-5 is maximally polarized.
(B) Graph plotting the relative signal intensity of DendraR::MEX-5 (red line; n = 7 embryos) and endogenous MEX-5 (blue line; n = 5 embryos) along the long axis of
the zygote after NEBD. Fluorescence intensity was averaged along a 20 pixel-wide box spanning the length of each zygote (0% anterior-most pole, 100%
posterior-most pole). Maximum values for each zygote were normalized to 1. Error bars represent the standard error of the mean (SEM).
(C) The relative concentration of DendraR::MEX-5 was measured in the anterior half, in the posterior half, and throughout the zygote (total) from before pronuclear
formation to 1 min following NEBD (just prior to cytokinesis). Mid-plane images were collected every 20 s. Embryos were normalized to each other by setting the
initial total value to 1 and averaged together (n = 5 embryos). Error bars represent SEM.
(D) Time-lapse photomicrographs of a zygote during polarity establishment (pronuclear migration) expressing Dendra::MEX-5 photoactivated in two stripes. Time
since photoactivation is indicated. Note that the signal from the posterior stripe diffuses more rapidly.
(E) Plot showing the apparent diffusion coefficient (Dc) of DendraR::MEX-5 at different positions along the long and short axes of the zygote and at different stages.
Embryo schematic shows the position of the photoconversion stripes along the long and short axes. Peripheral cytoplasm as mentioned in the text refers to
10% and 90% embryo length. Error bars represent SEM.
pronuclear formation, GFP::PAR-1 was uniformly distributed in

the cytoplasm and weakly on the cortex (data not shown). At
pronuclear formation, GFP::PAR-1 levels increased in the central
cytoplasm and decreased in the peripheral cytoplasm (Fig-
ure 2D). This relocalization coincided temporally and spatially

with an increase in DendraR::MEX-5 diffusion in the central cytoplasm (Figure 1E) and an increase in DendraR::MEX-5 levels
in the peripheral cytoplasm (Figure 2D). During pronuclear
Wild-type
PKC-3
MEX-5
Gradient:
pkc-3(RNAi)
par-1(it51)
par-1(b274)
par-2(RNAi), PN meeting
No
No
No
Yes
PAR-1
Yes
PAR-1
it51
R409K
b274
PKC-3
phosphorylation
site
Q819STOP
T983
1192
Kinase
Cortical localization
domain (aa965-1192)
Before PN Formation
KA1 domain
NEBD
Anterior
Posterior
WT
par-1(b274)
Wild-type
After PN Formation
WT
pkc-3(RNAi)
par-1(it51)
Wild-type
PN Meeting
par-1(it51)
pkc-3(RNAi)
par-1(b274)
par-1(b274)
par-1(RNAi)
par-1(b274)
pkc-3(RNAi)
par-2(RNAi)
par-2(RNAi)
PN Meeting
GFP::PAR-1
DendraR::MEX-5
Figure 2. PAR-1 Is Necessary and Sufficient to Increase MEX-5 Mobility

(A) Diagrams showing the distributions of PKC-3 (orange) and PAR-1 (purple on cortex and in cytoplasm) in zygotes of the indicated genotypes. MEX-5 localizes in
a gradient in wild-type and par-2(RNAi) embryos and remains symmetrically distributed in all other genotypes. par-1(it51) and par-1(b274) are mutations that,
respectively, inactivate PAR-1 kinase activity and truncate the PAR-1 protein (Figure S1E) (Guo and Kemphues, 1995). Also see Figures S1A, S1B, and S1E.
(B) PAR-1 schematic: T983 is a conserved aPKC/PKC-3 phosphorylation site. The C-terminal domain (9651192) contains the lipid-binding domain KA1
(Moravcevic et al., 2010) and localizes in a cytoplasmic gradient and to the posterior cortex (E.E.G., A.A. Cuenca, and G.S., unpublished data).
(C) Apparent Dc of DendraR::MEX-5 measured in the anterior (25% embryo length) and posterior (75% embryo length) cytoplasm in zygotes of the indicated
genotypes. Error bars represent SEM.
(D) Comparison of the distribution of GFP::PAR-1 and DendraR::MEX-5 in wild-type and par-2(RNAi) zygotes. Fluorescence intensity is represented by a rainbow
scale ranging from blue (low signal intensity) to red (high signal intensity). Arrows point to the subcortical region where GFP::PAR-1 is depleted and DendraR::MEX-5 accumulates after pronuclear formation. Note that in par-2(RNAi) zygotes, GFP::PAR-1 does not accumulate on the posterior cortex but still forms
a cytoplasmic gradient.
Also see Figures S1C, S1D, and S1F.
migration, GFP::PAR-1 levels remained low in the anteriorperipheral cytoplasm but increased in the posterior cytoplasm
and on the posterior cortex. By NEBD, GFP::PAR-1 was
enriched on the posterior cortex and formed a 3-fold anteriorlow/posterior-high gradient in the cytoplasm, paralleling the
gradient in MEX-5 diffusivity (Figure 2D and Figure S1C). Immunostaining of wild-type (WT) embryos with an anti-PAR-1 antibody confirmed the presence of a PAR-1 gradient in the cytoplasm of zygotes at NEBD (Figure S1D). We conclude that
PAR-1 dynamics in the cytoplasm correlate with MEX-5 diffusion
dynamics and that MEX-5 responds quickly to changes in PAR-1
distribution.
To explore whether cytoplasmic PAR-1 is sufficient to stimulate MEX-5 diffusion, we analyzed the par-1 allele b274.
par-1(b274) zygotes do not localize PAR-1 to the cortex and
do not segregate MEX-5 but are positive for pS458, suggesting
that this allele retains some par-1 kinase activity (Figure S1B)
(Guo and Kemphues, 1995; Tenlen et al., 2008). We sequenced
par-1(b274) and found a premature stop codon at residue
Q819 between the kinase domain and the domain that localizes
PAR-1 to the cortex (Figure 2B). Western blotting and immunofluorescence analyses confirmed the presence of a truncated
PAR-1 protein, expressed at 14% of wild-type levels and
uniformly cytoplasmic (Figures S1B and S1E) (Hurd and Kemphues, 2003). Before pronuclear formation, DendraR::MEX-5
mobility was uniformly low in par-1(b274) zygotes, as in wildtype and par-1(it51) zygotes. By NEBD, however, the apparent
Dc of DendraR::MEX-5 had increased throughout the cytoplasm
to a value intermediate between that of par-1(it51) and
pkc-3(RNAi) zygotes (Figure 2C). In par-1(b274) zygotes,
PKC-3 became enriched on the anterior cortex as in wild-type,
whereas DendraR::MEX-5 remained symmetrically distributed
(Figure S1B) (Tenlen et al., 2008). The intermediate DendraR::
MEX-5 diffusion rate in par-1(b274) zygotes was dependent on
PAR-1 but not on PKC-3 (Figure 2C). We conclude that PAR-1
kinase activity in the cytoplasm is sufficient to increase MEX-5
diffusivity after pronuclear formation.
We also examined the distribution of PAR-1 and MEX-5 in
par-2 zygotes, which localize anterior PARs to the anterior cortex
before, but not after, NEBD and which never enrich PAR-1 on the
posterior cortex (Boyd et al., 1996; Cuenca et al., 2003). We
found that GFP::PAR-1 still formed a cytoplasmic gradient by
pronuclear meeting in par-2(RNAi) zygotes (Figure 2D and Figure S1C). The GFP::PAR-1 gradient was transient and became
less pronounced following NEBD (Figure S1C). Remarkably,
DendraR::MEX-5 also formed a gradient by pronuclear meeting,
which weakened following NEBD (Figure 2D and Figure S1F).
The diffusivity of DendraR::MEX-5 was also asymmetric in par2(RNAi) zygotes (Figure 2C). We conclude that formation of
a cytoplasmic PAR-1 gradient is sufficient to change MEX-5
diffusion and drive the formation of a complementary MEX-5
gradient.
PAR-1 Phosphorylates MEX-5 on Two Residues: S458
and S404
Phosphorylation of S458 depends on par-1 activity in vivo,
raising the possibility that MEX-5 is a PAR-1 substrate (Tenlen
et al., 2008). To test this possibility directly, we expressed the
PAR-1 kinase domain (aa 1492) fused to maltose-binding

protein (MBP) in E. coli. We also included the activating mutation
T325E in the kinase activation loop (Lizcano et al., 2004).
MBP:PAR-1(aa 1492, T325E) phosphorylated MBP:MEX-5,
but not MBP or MBP:PIE-1 (Figure 3A). Replacement of S458
with alanine reduced, but did not abolish, phosphorylation of
MBP:MEX-5 (Figure 3A). Using a combination of deletion and
alanine mutagenesis, we identified S404 as a second PAR-1
phosphorylation site in MEX-5 (Figure 3A). MEX-5 mutated at
both S404 and S458 was no longer a substrate for MBP:PAR-1
(aa 1492, T325E) (Figure 3A). To determine whether S404 is
phosphorylated by PAR-1 in vivo, we generated an antibody
specific for pS404 (Figure S2A). Anti-pS404 immunoprecipitated
5% of total MEX-5 from extracts prepared from wild-type
hermaphrodites and only 1.7% from extracts prepared from
par-1(RNAi) hermaphrodites (Figure 3B). We conclude that
PAR-1 phosphorylates MEX-5 on S458 and S404 in vitro and
in vivo.
Reversible Phosphorylation of S404 Is Required
to Form the MEX-5 Gradient
To investigate the role of S404 and S458 phosphorylation in vivo,
we examined the distribution of MEX-5(S404A) and MEX5(S458A) fusions. As reported in Tenlen et al. (2008), the distribution of MEX-5(S458A) was variable from embryo to embryo,
with a minority of embryos forming a shallow MEX-5 gradient.
In contrast, DendraR::MEX-5(S404A) was symmetrically distributed in all embryos examined (Figure 3C). The double mutant
S404A/S458A behaved like S404A (data not shown). DendraR::
MEX-5(S404A) diffusion was slow, comparable to that of
wild-type DendraR::MEX-5 in par-1(it51) (Figure 3D and Figure 2C). DendraR::MEX-5(S404A) remained slow diffusing in
pkc-3(RNAi) and in par-1(b274) zygotes, indicating that this
fusion is no longer sensitive to changes in PAR-1 activity or localization (Figure 3D). We conclude that the MEX-5 protein and
diffusivity gradients depend primarily on phosphorylation of
S404 by PAR-1.
Immunofluorescence experiments using a phosphospecific
antibody have shown that S458 is phosphorylated during
oogenesis, and MEX-5 phosphorylated on S458 becomes enriched in the anterior in zygotes as does total MEX-5 (Tenlen
et al., 2008). These observations suggest that pS458 is relatively
stable and does not respond to changes in PAR-1 localization
during polarization. In contrast, we were not able to visualize
pS404 by immunofluorescence, even though our phosphospecific antibody could immunoprecipitate MEX-5 from extracts
(Figure 3B). We detected pS404 in extracts from fem-3(e2006)
females, which contain oocytes but no embryos, suggesting
that like S458, S404 is already phosphorylated during oogenesis
(data not shown). To examine phosphorylation and dephosphorylation dynamics at S458 and S404, we phosphorylated
MEX-5 in vitro using MBP::PAR-1(aa 1492; T325E) and incubated phosphorylated MEX-5 with embryo extract. Although
both sites were phosphorylated at similar rates in vitro, S404
was dephosphorylated significantly faster than S458 in embryo
extracts (Figure 3E). Dephosphorylation was inhibited by
200 nM okadaic acid, consistent with the presence of phosphatases in the extract (Figure 3E). We conclude that embryos
S458A
S404A
WT
MBP
S404A,
S458A
MBP:MEX-5
B
WT
MBP:PIE-1
par-1(RNAi)
IP: anti-pS404
Blot: anti-MEX-5
P-ATP
Input
32
Coomassie
MEX-5
PAR-1
D
Anterior
Posterior
Ratio (anterior/posterior)
WT
S458A
S404A
WT
let-92(RNAi)
WT
S458A
S404A
+ Embryonic Extract
Relative Immunoblot Signal
Relative Immunoblot Signal
+ MBP::PAR-1
S404A
S404A
WT
S404A
pkc-3(RNAi) par-1(b274) let-92(RNAi) let-92(RNAi)
pS458
pS404
pS458 + OA
pS458
pS404 + OA
pS404
01 5
20
60
120 min
01 5
20
60
120 min
Figure 3. PAR-1 Phosphorylates MEX-5 on S404 In Vitro and In Vivo

(A) SDS-PAGE gel of kinase reactions using MBP:PAR-1(aa 1492; T325E) and the indicated MBP substrates. Reactions were performed in the presence of
[32P]-ATP for 30 min. Top panel shows [32P] incorporation and bottom panel is Coomassie blue staining of the same gel.
(B) Western blot analysis of immunoprecipitates from whole worm extracts obtained with anti-MEX-5 pS404 antibodies. Five percent of extract used for
immunoprecipitation was loaded in input lanes. In the bottom panel, extract was probed with anti-PAR-1 antibodies to demonstrate depletion by par-1(RNAi).
(C) Ratio of mean anterior and posterior fluorescence intensities for embryos expressing the indicated DendraR::MEX-5 fusions at NEBD. Each dot represents an
individual embryo. Long bars represents the mean ratio and short bars represent the SEM.
(D) Apparent Dc of DendraR::MEX-5 mutants at NEBD. Dendra::MEX-5 was photoconverted along the anterior-posterior axis and apparent Dc was calculated at
25% (anterior) and 75% (posterior) embryo length. Error bars represent SEM. The results for wild-type embryos are also displayed in Figure 2C and Figure 4D.
Also see Figures S2B and S2C.
(E) Dynamics of pS404 and pS458 in vitro. Left panel: MBP:MEX-5 was incubated with MBP:PAR-1(aa 1492; T325E) and analyzed by western blot with
phosphospecific pS458 and pS404 antibodies at the indicated times. Right panel: Phosphorylated MBP::MEX-5 was incubated with embryonic extract in the
presence or absence of 200 nM okadaic acid (+ OA) and analyzed by western blot with pS404 and pS458 phosphospecific antibodies. Note the rapid
dephosphorylation at S404. Error bars represent SEM. Also see Figure S2A.
contain a phosphatase activity that efficiently reverses S404

phosphorylation.
The okadaic acid-sensitive phosphatase PP2A has been
implicated as a PAR-1 antagonist in Drosophila and C. elegans
(Kao et al., 2004; Krahn et al., 2009; Nam et al., 2007; Yoder
et al., 2004). PP2A is a heterotrimeric phosphatase consisting
of structural, catalytic, and regulatory subunits. In C. elegans,
the catalytic subunit, LET-92, is distributed throughout the

cytoplasm, on centrosomes, and on P granules (Schlaitz et al.,
2007). To test whether PP2A influences MEX-5 dynamics, we
analyzed let-92(RNAi) embryos. let-92(RNAi) increased the
mobility and decreased the asymmetry of wild-type DendraR::
MEX-5 (Figures 3C and 3D). Consistent with PP2A acting
primarily via S404, let-92(RNAi) only slightly increased the
PAR-1
Phosphorylation Sites
271
S458
S404
ZF2
K318
R274
MEX-5
M288
F294
ZF1
Y333
F339
342
468
Asymmetry
Diffusion
Anterior
Posterior
Full Length
Yes
1.03 (.11)
2.68 (.24)
aa1-355
No
1.02 (.07)
0.82 (.07)
aa1-245
No
13.3 (1.2)
12.3 (1.6)
aa245-468
Weak
3.22 (.30)
5.93 (.87)
aa345-468
No
> 10
> 10
C
Before PN Formation
NEBD
Ratio (anterior/posterior)
2.5
2.0
1.5
1.0
WT
WT
R274E,K318E R274E,K318E M288E,F294N, M288E,F294N,
mex-5/6(RNAi)
mex-5/6(RNAi) Y333E,F339N Y333E,F339N
mex-5/6(RNAi)
Anterior
Posterior
WT
R274E, K318E
WT
R274E, K318E R274E, K318E M288E,F294N,

Y343E,F339N
par-1(it51)
WT
Lat A
Figure 4. Regulation of MEX-5 Mobility by RNA Binding

(A) Schematic showing the MEX-5 truncations. Each construct was expressed as a Dendra fusion and its localization and apparent Dc (mm2/s) were determined
at NEBD (SEM in parentheses). The apparent Dc of DendraR::MEX-5(aa 345468) could not be determined accurately because of its rapid diffusion and relatively
low expression but exceeded 10 mm2/s.
(B) Ratio of anterior and posterior fluorescence intensities for embryos expressing the indicated Dendra::MEX-5 fusions. Each dot represents an individual
embryo. Long bars represent the mean ratio and short bars represent SEM.
(C) Apparent Dc of DendraR::MEX-5 mutants measured before pronuclear (PN) formation (before polarization) and at NEBD (after polarization). The results for wildtype embryos are also presented in Figure 2C and Figure 3D. Error bars represent SEM.
See also Figure S3.
mobility of DendraR::MEX-5 (S404A) (Figure 3D). let-92(RNAi)

did not affect the posterior localization of PAR-1 (Figures S2B
and S2C). We conclude that PP2A, and possibly other phosphatases, antagonizes PAR-1-dependent phosphorylation of MEX-5
to return MEX-5 to a slow-diffusing state.
RNA Binding Limits MEX-5 Diffusion
The apparent Dc of DendraR::MEX-5 before and after polarization
is 10- to 20-fold lower than that of DendraR alone (data not
shown). To determine which domains of MEX-5 retard its
mobility, we compared the localization and diffusion behavior
of a Dendra::MEX-5 deletion series (Figure 4A). A C-terminal
truncation lacking S404 and S458 (Dendra::MEX-5(aa 1355))
was symmetrically distributed and uniformly slow diffusing
even after polarization of the zygote (Figure 4A). An N-terminal
truncation (DendraR::MEX-5(aa 245468)) showed a moderate
increase in mobility in the anterior and posterior cytoplasm and
a shallower but still detectable gradient (Figure 4A). In contrast,
fusions lacking the CCCH fingers (DendraR::MEX-5(aa 1244)
and DendraR::MEX-5(aa 345468)) diffused >10 times faster
and lacked all asymmetry (Figure 4A). Consistent with these
findings, a GFP::MEX-5 fusion lacking only the CCCH fingers

was uniformly distributed and fast diffusing (Tenlen et al.,
2008). We conclude that MEX-5 localization and slow mobility
depend primarily on the CCCH fingers, with an additional contribution from the N-terminal domain.
The CCCH fingers of MEX-5 mediate RNA binding in vitro (Pagano et al., 2007). To test whether RNA binding retards MEX-5
mobility, we examined missense mutations in the CCCH fingers.
Studies on the TIS11 family of CCCH finger proteins identified
key amino acids that contact RNA, mutations in which disrupt
RNA binding (Hudson et al., 2004; Lai et al., 2002). The corresponding mutations in MEX-5 are M288E, M294N, Y333E,
F339N. In vitro, MEX-5 binds preferentially to poly-U tracks,
a sequence common in C. elegans 30 UTRs (Pagano et al.,
2007). R274E and K318E decrease MEX-5 affinity for poly-U
by 35-fold but only modestly reduce MEX-5 ability to bind to
a related sequence (UUAUUUAUU) (Pagano et al., 2007). We
found that both DendraR::MEX-5(M288E, M294N, Y333E,
F339N) and DendraR::MEX-5 (R274E, K318E) formed a shallower
gradient than wild-type and exhibited increased diffusivity
in both the anterior and posterior (Figures 4B and 4C). The
observations suggest that, in addition to RNA binding, interactions among MEX-5 and MEX-6 molecules can also contribute
to MEX-50 s diffusive behavior.
The actin cytoskeleton, which becomes enriched in the anterior cytoplasm during polarization, has been proposed as
another candidate for retarding MEX-5 mobility (Tenlen et al.,
2008). To test this possibility, we treated zygotes with Latrunculin
A, which depolymerizes F-actin and blocks polarization of the
PARs (Severson and Bowerman, 2003). Latrunculin A treatment
resulted in uniformly slow MEX-5 diffusion and blocked DendraR::MEX-5 gradient formation (Figure 4C and data not shown),
indicating that F-actin is not essential to retard MEX-5 mobility.
% of Total Dendra::MEX-5
40S
60S
80S
Fraction
WT
par-1(it51)
S404A
% Component
Component
Fast Slow
Fast Slow
Anterior Posterior
Fast Slow
Fast Slow
Anterior Posterior
Fast Slow
Fast Slow
Anterior Posterior
Figure 5. MEX-5 Associates with Multiple Complexes In Vivo

(A) Distribution of Dendra::MEX-5 fusions following sucrose gradient
fractionation and detection by anti-Dendra western blot. Light fractions are
on the left and heavy fractions are on the right. Approximate positions of the
40S, 60S, and 80S ribosomal subunits are indicated. Error bars represent
SEM. See Figure S4A for UV trace.
(B) Percentage of fast and slow MEX-5 complexes detected by FCS. Note that
fast and slow components were detected in all measurements. Error bars
represent SEM (wild-type, n = 24, par-1(it51), n = 5; S404A, n = 8 embryos). See
Figures S4B and S4C.
apparent Dc of DendraR::MEX-5 (R274E, K318E) was reduced

in par-1(RNAi) but remained higher than wild-type MEX-5, indicating that DendraR::MEX-5 (R274E, K318E) is still regulated
by PAR-1 but is intrinsically more mobile than wild-type DendraR::MEX-5 (Figure 4C). We conclude that RNA binding retards
MEX-5 mobility.
Tenlen et al. (2008) reported that cysteine-to-serine substitutions predicted to disrupt folding of the CCCH fingers
do not affect the MEX-5 gradient (Tenlen et al., 2008). We also
found that DendraR::MEX-5(C286S,C292S,C331S,C337S) forms
a gradient similar to wild-type DendraR::MEX-5. DendraR::MEX5(C286S,C292S,C331S,C337S), however, diffused faster than
DendraR::MEX-5 and was dependent on endogenous wild-type
MEX-5 and MEX-6 to form a gradient (Figures S3A and S3B)
(Tenlen et al., 2008). In contrast, the diffusive behaviors of DendraR::MEX-5(WT) and DendraR::MEX-5(R274E, K318E) were not
dependent on endogenous MEX-5 or MEX-6 (Figure 4B). These
MEX-5 Associates with Large Complexes

in an RNA-Dependent Manner
To directly investigate whether MEX-5 associates with RNA
in vivo, we examined the distribution of Dendra::MEX-5 in
worm extracts fractionated on a 10%45% sucrose gradient.
Dendra::MEX-5 was detected in both light and heavy fractions,
including fractions containing 80S ribosomes (fractions 8 and
9, Figure 5A and Figure S4A). In contrast, Dendra alone was
found primarily in the lightest fractions (Figure S4A). RNase treatment that eliminated the polysome RNA peaks, but not the 80S
peaks, caused the Dendra::MEX-5 to shift toward the lighter
fractions, indicating that the association of MEX-5 with large
complexes is RNA dependent (Figure 5A and Figure S4A).
To test whether MEX-50 s behavior on sucrose gradients
correlates with MEX-50 s diffusive behavior in vivo, we examined
Dendra::MEX-5(R274E, K318E) and Dendra::MEX-5(S404A). We
found that the profile of the fast-diffusing Dendra::MEX-5(R274E,
K318E) was shifted toward the lighter fraction, whereas the
profile of slow-diffusing MEX-5(S404A) was shifted toward the
heavy fractions (Figure 5A and Figure S4A). We conclude that
MEX-5 exists in both light and heavy complexes, and that association with the latter depends on RNA and correlates with
slower diffusion.
MEX-5 Exists in Multiple Diffusive Complexes In Vivo
The broad distribution of MEX-5 in sucrose gradients suggests
that MEX-5 exists in multiple complexes in vivo. To test this
hypothesis directly, we used fluorescence correlation spectroscopy (FCS) to measure the diffusive behavior of individual GFP::MEX-5 molecules in live zygotes. We monitored
GFP::MEX-5 at 30% and 70% embryo length in 24 zygotes at
NEBD. Autocorrelation curves were fit to three-dimensional
models containing one, two, or three diffusive components.
One-component models yielded Dc values that were significantly slower than observed with DendraR::MEX-5 in both
the anterior and posterior (anterior = 0.26 0.05 mm2/s;
posterior = 0.37 0.1 mm2/s) (Figure S4B). Two-component
models yielded fast and slow components with 100-fold
difference in Dc values, whose weighted averages (an estimate
of the population Dc) were in good agreement with those
observed with DendraR::MEX-5 (anterior = 1.40 0.29 mm2/s;
posterior = 3.13 0.37 mm2/s) (Figures S4B and S4C). The
concentration ratio of slow:fast components was significantly
higher in the anterior cytoplasm (66:34) compared to the posterior cytoplasm (50:50) (Figure 5B). We conclude that, as
suggested by the sucrose gradients, MEX-5 exists in multiple

complexes in the cytoplasm, with a bias toward slower
complexes in the anterior.
To examine the effect of phosphorylation by PAR-1, we
repeated the FCS measurements in par-1(it51) zygotes and in
wild-type zygotes expressing MEX-5(S404A). We obtained
similar FCS profiles for both genotypes. As described above
for wild-type GFP::MEX-5, one-component models yielded Dc
values that were significantly lower than those observed with
DendraR:MEX-5 in par-1(it51) embryos or DendraR::MEX5(S404A) in wild-type embryos (Figure S4B). Two-component
models, in contrast, yielded Dc values consistent with the DendraR values (Figure 5B and Figure S4B). The Dc and concentration ratios of fast and slow complexes were similar to those
observed in the anterior of wild-type embryos (Figure 5B and
Figure S4C). These results suggest that MEX-5 distributes
between slow- and fast-diffusing complexes even in the
absence of PAR-1, and that phosphorylation by PAR-1 shifts
the distribution of MEX-5 in favor of faster complexes.
Our FCS results indicate that MEX-5 fast and slow complexes
exhibit dramatically different rates of diffusion: 5.15 mm2/s
(1090th percentile range of 1.73 to 10.7 mm2/s) for the fast
class, and 0.086 mm2/s (1090th percentile range of 0.025 to
0.16 mm2/s) for the slow class. Daniels et al. (2010) reported
a similar range of mobilities for wild-type MEX-5 but did not
report the relative concentration of the slowest species and
only considered species within the fast range in their modeling
of the MEX-5 gradient (Daniels et al., 2010). Our analysis,
however, indicates that the slow species contributes significantly to the overall diffusive behavior of MEX-5 (70% of
MEX-5 complexes in the anterior). Omission of the slow-diffusing
species when calculating population Dc yields values that do not
match those observed experimentally using photoactivation
(this work) or FRAP (Daniels et al., 2010). We conclude that the
slow-diffusing species cannot be excluded from a description
of MEX-5s diffusive behavior.
Modeling of the MEX-5 Gradient
To determine whether our experimental results can be integrated into a self-consistent theoretical framework, we
developed a mathematical model for the reaction-diffusion
dynamics of MEX-5 (Figure 6A). The model is based on the
principle that steady-state protein gradients form if (1) different
phosphostates exhibit different diffusion coefficients and (2)
interconversion between phosphostates is mediated by
spatially segregated kinase and phosphatase reactions (Lipkow and Odde, 2008). We approximated MEX-5 diffusion
dynamics by allowing for a fast species (Dfast = 5 mm2/s) and
a slow species (Dslow = 0.07 mm2/s) whose interconversion is
regulated by a phosphorylation cycle mediated by PAR-1 and
PP2A (and possibly other phosphatases) (Figure 6A). Because
the relative activity of cortical and cytoplasmic PAR-1 are not
known, we independently considered how cytoplasmic and
cortical PAR-1 affect MEX-5 segregation. Phosphatase activity
was assumed to be uniform in the cytoplasm such that, in both
scenarios, the kinase and phosphatase activities are spatially
distinct from each other. Unsteady-state analysis and the
sensitivity of the cytoplasmic and cortical PAR-1 models to
changes in individual parameters are presented in Figure S5

and Figure S6 and described in Extended Experimental
Procedures.
We first considered a model in which PAR-1 activity exists in
a linear 5.5-fold gradient in the cytoplasm (low anterior, high
posterior). The PAR-1 and phosphatase rates were matched in
the posterior to yield slow:fast ratios of 1:1 in the posterior
and 2:1 in the anterior, as observed in our FCS measurements
(Figure 6B and Table 1). This model gives rise to a temporally
stable 2.9-fold MEX-5 concentration gradient as is observed
in vivo. Given a phosphatase rate of 0.1 s1 (within the range
reported in the literature of 0.1100 s1; Brown and Kholodenko,
1999), the timescale of gradient formation is 160 s (Figure S5B),
consistent with the kinetics observed in vivo (Figure 1). Coordinately changing the absolute kinase and phosphatase rates
over two orders of magnitude has little effect on the strength
of the gradient. For example, increasing or decreasing both
the kinase and phosphatase rates by a factor of 10 generates
3.0- and 2.8-fold MEX-5 gradients, respectively (Figures S5G
and S5H). If only the kinase or phosphatase rate is changed
(rather than changing them coordinately), the MEX-5 gradient
is lost (Figures S5CS5F). For example, reducing phosphatase
activity while maintaining PAR-1 activity increases the proportion
of fast-diffusing species and flattens the MEX-5 gradient (Figure 6C and Figure S5D), as observed in let-92(RNAi) embryos
(Figures 3C and 3D). Interestingly, the MEX-5 concentration
gradient is always weaker than the PAR-1 activity gradient (see
Discussion).
We next considered a model where PAR-1 is entirely cortical
and instantaneously phosphorylates MEX-5. In this corticalonly PAR-1 model, the extent of PAR-1s influence on MEX-5
is determined by the phosphatase rate. For example, at a kphos =
0.1 s1, the effect on MEX-5 drops off sharply within 10 mm
of the cortex (Figure 6D and Figure S6B). A phosphatase rate
of kphos = 0.01 s1 would generate an 3-fold MEX-5 gradient
(Figure 6E and Figure S6C). However, nearly all MEX-5 would
be in the slow-diffusing state, in contrast to our FCS observations. A gradient with the observed proportions of fast- and
slow-diffusing MEX-5 species is only obtained at a phosphatase
rate of kphos = 0.0001 s1. However, the approach to steady state
would be 17 min, far slower than what is observed in vivo (Figure 6K and Figure S6D). Thus, the cortical-only PAR-1 model is
not able to simultaneously explain the relative proportions of
fast and slow species while also maintaining a rapid approach
to steady state. Taken together, the modeling analyses support
a critical role for cytoplasmic PAR-1 and demonstrate that the
MEX-5 diffusion gradient is sufficient to account for the MEX-5
protein gradient.
DISCUSSION
In this study, we present evidence that the antagonistic activities of PAR-1 and PP2A regulate MEX-5 diffusion to establish
the MEX-5 protein gradient. We propose the following model.
MEX-5 is in dynamic, local equilibrium between different
diffusive RNA complexes in the cytoplasm. Phosphorylation
of S404 by PAR-1 biases MEX-5 toward faster-diffusing
complexes, and dephosphorylation by PP2A returns MEX-5
PAR-1
MEX-5 pS404
MEX-5
(5 m2/sec)
(0.07 m2/sec)
PP2A
Cytoplasmic PAR-1 +
kphos 0.1s-1
B
6
fast
slow
total
PAR-1
0.11s-1
PAR-1
0.02s-1
10
15
20
25
30
P O S IT IO N (m )
35
40
45
50
Cortical PAR-1 +
kphos 0.1s-1
PAR-1
0.02s-1
0
10
6
fast
slow
total
15
20
25
30
P O S IT IO N (m )
35
40
45
fast
slow
total
C O N C E N T R AT IO N (AU )
50
Cortical PAR-1 +
kphos 0.01s-1
PAR-1
0.11s-1
fast
slow
total
5
Cytoplasmic PAR-1 +
kphos 0.01s-1
10
15
20
25
30
35
40
45
50
P O S IT IO N (m )
10
15
20
25
30
P O S IT IO N (m )
35
40
45
50
Figure 6. Mathematical Modeling of the MEX-5 Gradient

(A) Model reactions. PAR-1 and PP2A are assumed to regulate interconversion between fast and slow MEX-5 species through a phosphorylation cycle. See Table
1 for assumptions used in the model.
(BE) Graphs showing the model-generated distribution of MEX-5 at steady state along the anterior-posterior axis (anterior end, 0 mm; posterior end, 50 mm).
See Extended Experimental Procedures and Figure S5 and Figure S6 for unsteady-state analysis.
(B) Cytoplasmic PAR-1 model. PAR-1 activity is assumed to be linearly distributed in the cytoplasm (low in anterior; high in posterior). This imposes an oppositely
oriented MEX-5 gradient with the fast and slow species approximately equal in concentration in the posterior and the slow species enriched in the anterior. The
total MEX-5 gradient primarily reflects the gradient in slow-diffusing MEX-5. The rapid diffusion of the fast-diffusing species effectively counteracts its asymmetric
formation in the posterior. See Figure S5B for unsteady-state analysis.
(C) PP2A depletion. Reducing the phosphatase rate by 10-fold weakens the MEX-5 gradient and increases the proportion of phosphorylated MEX-5. See
Figure S5D for unsteady-state analysis.
(D) Posterior cortical PAR-1 plus uniform phosphatase (kphos = 0.1 s1). See Figure S6B for unsteady-state analysis.
(E) Posterior cortical PAR-1 plus uniform phosphatase (kphos = 0.01 s1). See Figure S6C for unsteady-state analysis.
into slower-diffusing complexes. Before polarization, PP2A

activity dominates over PAR-1, pS404 levels are low, and the
majority of MEX-5 molecules are in slow-diffusing complexes.
At polarity onset, an unknown mechanism favors PAR-1 activity
over PP2A, causing pS404 levels to rise and MEX-5 to
enter faster complexes. During polarization, the PP2A/PAR-1
balance is changed along the anterior-posterior axis as PAR-1
becomes enriched in a posterior-to-anterior gradient in the

cytoplasm and on the posterior cortex, causing MEX-5 to
switch from phosphorylated (on average faster-diffusing) to
unphosphorylated (on average slower-diffusing), as it diffuses
down the PAR-1 gradient. As a result, MEX-5 redistributes in
a gradient opposite PAR-1. We consider each aspect of this
model in turn.
Table 1. Parameters and Variables for Cytoplasmic PAR-1 Model

Parameter/Variable
Symbol
Value
Units
Slow diffusion
coefficient
Dslow
0.07
mm2/s
Notes
Fast diffusion
coefficient
Dfast
mm2/s
Kinase (PAR-1)
rate constant
kkin(x)
0.020.11
s1
Linear rise
along A/P axis
Phosphatase rate
constant
kphos
0.1
s1
Uniform along
A/P axis
Embryo length
50
mm
Additional notes:
(1) Kinase and phosphatase rates can be varied coordinately over a range
of values (e.g., kkin(x) = 0.21.1 s1 and kphos = 1 s1 yields similar results).
Rate constants must be approximately equal in the posterior region to
obtain 1:1 slow:fast diffusing species, and kphos > kkin to obtain >1:1
slow:fast in the anterior region.
(2) Kinase rate constant gradient needs to be larger than the MEX-5
gradient. Here it is assumed that the PAR-1 activity gradient is 5.5-fold,
resulting in 2.9-fold MEX-5 concentration gradient.
(3) Posterior cortical-only PAR-1 case modeled with instantaneous
kinase reaction at the right boundary (i.e., x = L), kkin(x) = 0 s1 and kphos =
0.1 s1 (Figure 6D) or kphos=0.01s1 (Figure 6E).
MEX-5 Diffusion Is Retarded by Binding to RNA

throughout the Cytoplasm
Our FCS analysis indicates that MEX-5 distributes between
two classes of diffusive complexes: a fast class averaging
5.15 mm2/s and a slow class averaging 0.086 mm2/s. Both
classes are present throughout the cytoplasm, but the slow
class is distributed in an anterior-high to posterior-low gradient.
Because FCS analysis only constrains the minimum number of
diffusive species, we cannot distinguish whether MEX-5 participates in two or more complexes. The broad range of diffusion
coefficients for the fast and slow components and the broad
distribution of Dendra::MEX-5 in sucrose gradients suggest, in
fact, that MEX-5 may interact with a large range of complexes.
Several lines of evidence suggest that MEX-5s association
with slow-diffusive complexes depends on binding to RNA. First,
mutations in the CCCH fingers that reduce MEX-5 affinity for
RNA increase MEX-5 diffusion and reduce the steepness of the
MEX-5 gradient. Second, sucrose gradient fractionation demonstrates that MEX-5 associates with high-density complexes
(comparable to 80S ribosomes) in a manner dependent on
RNA and the MEX-5 RNA-binding domain. Third, the Dc for the
slow species is consistent with mRNP diffusion rates (0.01
0.09 mm2/s) in the cytoplasm of E. coli and in the nucleus of
mammalian cells (Golding and Cox, 2004; Shav-Tal et al.,
2004). Because mutations that block MEX-5 phosphorylation
(S404A) cause the slow MEX-5 species to be symmetrically
distributed even in wild-type zygotes, we do not favor a model
wherein MEX-5 diffusion is retarded by binding to a subclass
of asymmetrically localized mRNAs. Rather, we suggest that
MEX-5 interacts dynamically with many mRNAs throughout the
cytoplasm. Consistent with MEX-5 functioning as a broad-spectrum RNA-binding protein, MEX-5 binds to poly-U tracks, which
are common in C. elegans 30 untranslated regions (UTRs)
(Pagano et al., 2007), and activates maternal mRNA turnover in

somatic blastomeres after the two-cell stage (Gallo et al., 2008).
Our mutational analysis also indicates that the amino terminus
of MEX-5 contributes to, but is not sufficient for, slow MEX-5
diffusion. This region is rich in polyglutamine stretches, which
could mediate MEX-5 self-association. One possibility is that,
as proposed for Bruno in Drosophila, MEX-5 uses self-interactions and RNA binding to assemble into large ribonucleoprotein
particles with retarded diffusion (Chekulaeva et al., 2006).
Phosphorylation of S404 by PAR-1 Biases MEX-5 toward
Fast Complexes, and Dephosphorylation by PP2A
Returns MEX-5 into Slow Complexes
In the absence of PAR-1, fast and slow MEX-5 complexes are
distributed in a 30:70 constant ratio throughout the cytoplasm,
indicating that phosphorylation enhances, but is not essential
for, the formation of fast MEX-5 complexes. Because conditions
predicted to reduce (par-1(it51)) or increase (let-92(RNAi))
phosphorylation have opposite effects on MEX-5 diffusivity, we
suggest that phosphorylation promotes the shifting of MEX-5
from slow- to fast-diffusing complexes. Consistent with this
view, MEX-5(S404A) was enriched in the heavier sucrose
gradient fractions compared to wild-type MEX-5. In our simulation of the MEX-5 gradient, MEX-5 must switch multiple times
between phosphostates as it diffuses across the embryo (see
discussion in Extended Experimental Procedures). Consistent
with this possibility, we find that pS404 is highly labile in embryo
extracts. We suggest that the rapid turnover of pS404 renders
MEX-5 exquisitely sensitive to changes in PAR-1 distribution.
Cytoplasmic PAR-1 Is Essential for the Formation
of the MEX-5 Gradient
Our simulations also demonstrate the importance of cytoplasmic
PAR-1 in specifying the MEX-5 gradient. In the cortical-only
PAR-1 model, high kphos values generate MEX-5 gradients that
drop off sharply from the posterior cortex, whereas low kphos
values yield MEX-5 gradients that form too slowly. In contrast,
in the presence of a cytoplasmic PAR-1 activity gradient, a broad
range of kinase and phosphatase activities could generate the
MEX-5 gradient. Cytoplasmic PAR-1, therefore, eliminates the
trade-off between gradient scale and response time. Our in vivo
observations confirm that cytoplasmic PAR-1 is sufficient to
regulate MEX-5 distribution: most notably, MEX-5 forms a
gradient in par-2(RNAi) zygotes, which enrich PAR-1 in the
posterior cytoplasm but not on the cortex. That PAR-1 can
function off the cortex has also been suggested by Boyd et al.
(1996), who noted that par-2 mutant zygotes localize P granules,
a function requiring par-1 activity (Boyd et al., 1996; Cheeks
et al., 2004).
One striking aspect of our model is that the amplitude of the
MEX-5 gradient will always be smaller than the PAR-1 activity
gradient (an 2.9-fold MEX-5 gradient requires a 5.5-fold PAR-1
activity gradient). GFP::PAR-1 forms an 3- to 4-fold cytoplasmic concentration gradient, and regulation of PAR-1 kinase
activity along the anterior-posterior axis could also contribute to
an overall PAR-1 activity gradient. PAR-1 kinase activity has
been suggested to be regulated by several mechanisms (Marx
et al., 2010), including inhibition by aPKC (Hurov et al., 2004).
aPKC phosphorylates PAR-1 in vitro on a conserved threonine

(T983) required for PAR-1 asymmetry in vivo (E.E.G., Y. Hao,
and G.S., unpublished data). One possibility is that phosphorylation by anteriorly enriched PKC-3 regulates both PAR-1
activity and levels along the anterior-posterior axis. In polarized
T cells, PAR1b/EMK/MARK2 forms a cytoplasmic gradient
near the immunological synapse that depends on PKC phosphorylation sites (Lin et al., 2009). These observations raise the
possibility that the polarizing effects of the PAR network depend
on the formation of cytoplasmic PAR-1 gradients in several cell
types.
The reactive surface model of Daniels et al. (2010) proposes that MEX-5 diffusion is regulated at the cortex by interactions with both anterior and posterior PARs. Anterior PARs
convert phosphorylated MEX-5 into a slower-diffusive form
(0.41 mm2/s), which must be dephosphorylated before conversion back into a faster (15 mm2/s) form by the posterior PARs
(Daniels et al., 2010). The reactive surface model does not
consider the behavior of the slowest MEX-5 species, which in
our FCS analysis account for >50% of total MEX-5 (average
0.086 mm2/s, range 0.025 to 0.16 mm2/s). Furthermore, this
model predicts that loss of phosphatase activity should slow
MEX-5 diffusion, whereas we find that loss of the PP2A phosphatase increases MEX-5 diffusivity. This model also predicts
that, under conditions where MEX-5 is phosphorylated (PAR-1
active), loss of anterior PARs should increase MEX-5 diffusivity.
In contrast, we find that pkc-3(RNAi) has no effect on MEX-5
diffusivity in par-1(b274) zygotes (Figure 2C). Our genetic
analyses demonstrate that par-1 is fully epistatic to pkc-3 with
respect to MEX-5 diffusivity, making a direct contribution by
anterior PARs unlikely. Rather, our data indicate that anterior
PARs regulate MEX-5 diffusion indirectly, by controlling the
distribution (and possibly the activity) of PAR-1 along the
anterior-posterior axis.
Formation of Concentration Gradients by Spatially
Segregated Modification Enzymes
The model of Lipkow and Odde (2008) can be used to form
gradients at any cellular scale by varying diffusion and phosphatase rates. The MEX-5 gradient is established in an 50 mm
zygote, but the same principles could account for the apparent
CheY gradient that emerges in the cytoplasm of E. coli (5 mm)
upon uncoupling of the phosphatase/kinase pair CheZ/CheA
(Vaknin and Berg, 2004). Spatial segregation of kinase and phosphatase activities has been shown to lead to phosphogradients
in many cell types from bacteria to eukaryotic cells (Brown and
Kholodenko, 1999; Coppey et al., 2008; Fuller et al., 2008; Kalab
et al., 2002; Maeder et al., 2007; Su et al., 1998). Our modeling
analyses demonstrate that a spatially biased kinase and phosphatase cycle can give rise to protein concentration gradients
even under conditions where the phosphogradient is weak.
Despite higher PAR-1 activity in the posterior, phosphorylated
MEX-5 is predicted to distribute almost evenly across the zygote
due to its faster diffusion (Figure 6). In principle, any posttranslational modification cycle could generate a protein concentration
gradient, as long as the opposing enzymes are spatially segregated and the modification affects protein diffusion rates. We
suggest that the mechanism we uncover here for MEX-5 can
be applied broadly to understanding rapid changes in the distribution of cytoplasmic proteins in a variety of cell types.
Detailed experimental procedures are described in the Extended Experimental
Procedures.
C. elegans Strains
Transgenic worms used in this study are listed in Table S1.
Determination of DendraR::MEX-5 Diffusion Coefficients
Dendra::MEX-5 was photoconverted in a stripe with UV light and imaged on
a spinning disk confocal microscope. Intensity values were fit to Gaussian
distributions for each time point (GraphPad Prism), and the change in variance
over time was used to calculate Dc (Berg, 1993).
Recombinant Protein Purification, Kinase Assays,
and Dephosphorylation Assays
MBP:MEX-5 and MBP:PAR-1 (1492, T325E) were partially purified from
E. coli and incubated at 30 C in the presence of [32P]-ATP or cold ATP. For
nonisotopic phosphorylation and dephosphorylation assays, kinase reactions
were terminated with 20 nM staurosporine before embryonic extract was
added.
Immunoprecipitations
MEX-5 pS404 phosphospecific antibodies coupled to ProteinG dynabeads
were used to immunoprecipitate from whole worm extracts.
Sucrose Gradient Fractionation
Cycloheximide-treated whole worm extracts were fractionated over 10%
45% linear sucrose gradients at 39,000 rpm for 3 hr. Fractions were collected
after passing the gradient through a UV detector, and the distribution of
Dendra::MEX-5 was determined by western blot with anti-Dendra antibodies
(Axxora).
Fluorescence Correlation Spectroscopy
GFP::MEX-5 levels were reduced by partial GFP RNAi depletion prior to
imaging. Embryos were imaged on a Zeiss LSM 510 Confocal microscope
equipped with a Confocor 3 FCS. Autocorrelation curves were analyzed within
the Zeiss Confocor 3 software package.
Modeling of the MEX-5 Gradient
Parameters used in the models are listed in Table 1. A detailed description
of model and the contribution of individual parameters to the steady-state
and unsteady-state models are provided in the Extended Experimental
Procedures.
Supplemental Information includes Extended Experimental Procedures, six
figures, and one table and can be found with this article online at doi:10.
1016/j.cell.2011.08.012.
ACKNOWLEDGMENTS
We thank F. Motegi for providing the Dendra2 construct, S. He for help
with sucrose gradients, and N. Perkins and E. Pryce at the JHU Integrated
Imaging Center for assistance with FCS. We thank K. Kemphues for providing
strains and antibodies. We thank A. Cuenca for generation of the GFP::PAR-1
transgenic strain. We thank S. Kuo for helpful conversations and members of
the Seydoux lab for helpful comments on the manuscript. This work was
supported by the National Institutes of Health (Grants R01HD37047 [G.S.]
and R01 GM71522 [D.J.O.]) and the American Cancer Society (Grant PF-08158-01-DDC [E.E.G.]). G.S. is an Investigator of the Howard Hughes Medical
Institute.
Received: March 3, 2011

REFERENCES
Berg, H.C. (1993). Random Walks in Biology, Expanded Edition edn (Princeton,
New Jersey: Princeton University Press).
Boyd, L., Guo, S., Levitan, D., Stinchcomb, D.T., and Kemphues, K.J. (1996).
PAR-2 is asymmetrically distributed and promotes association of P granules
and PAR-1 with the cortex in C. elegans embryos. Development 122, 3075
3084.
Kao, G., Tuck, S., Baillie, D., and Sundaram, M.V. (2004). C. elegans SUR-6/
PR55 cooperates with LET-92/protein phosphatase 2A and promotes Raf
activity independently of inhibitory Akt phosphorylation sites. Development
131, 755765.
Kemphues, K. (2000). PARsing embryonic polarity. Cell 101, 345348.
Krahn, M.P., Egger-Adam, D., and Wodarz, A. (2009). PP2A antagonizes phosphorylation of Bazooka by PAR-1 to control apical-basal polarity in dividing
embryonic neuroblasts. Dev. Cell 16, 901908.
Lai, W.S., Kennington, E.A., and Blackshear, P.J. (2002). Interactions of CCCH
zinc finger proteins with mRNA: non-binding tristetraprolin mutants exert
an inhibitory effect on degradation of AU-rich element-containing mRNAs.
J. Biol. Chem. 277, 96069613.
Brown, G.C., and Kholodenko, B.N. (1999). Spatial gradients of cellular

phospho-proteins. FEBS Lett. 457, 452454.
Lin, J., Hou, K.K., Piwnica-Worms, H., and Shaw, A.S. (2009). The polarity
protein Par1b/EMK/MARK2 regulates T cell receptor-induced microtubuleorganizing center polarization. J. Immunol. 183, 12151221.
Budirahardja, Y., and Gonczy, P. (2008). PLK-1 asymmetry contributes to

asynchronous cell division of C. elegans embryos. Development 135, 1303
1313.
Lipkow, K., and Odde, D.J. (2008). Model for protein concentration gradients in
the cytoplasm. Cell. Mol. Bioeng. 1, 8492.
Cheeks, R.J., Canman, J.C., Gabriel, W.N., Meyer, N., Strome, S., and Goldstein, B. (2004). C. elegans PAR proteins function by mobilizing and stabilizing
asymmetrically localized protein complexes. Curr. Biol. 14, 851862.
Chekulaeva, M., Hentze, M.W., and Ephrussi, A. (2006). Bruno acts as a dual
repressor of oskar translation, promoting mRNA oligomerization and formation
of silencing particles. Cell 124, 521533.
Coppey, M., Boettiger, A.N., Berezhkovskii, A.M., and Shvartsman, S.Y.
(2008). Nuclear trapping shapes the terminal gradient in the Drosophila
embryo. Curr. Biol. 18, 915919.
Cuenca, A.A., Schetter, A., Aceto, D., Kemphues, K., and Seydoux, G. (2003).
Polarization of the C. elegans zygote proceeds via distinct establishment and
maintenance phases. Development 130, 12551265.
Daniels, B.R., Dobrowsky, T.M., Perkins, E.M., Sun, S.X., and Wirtz, D. (2010).
MEX-5 enrichment in the C. elegans early embryo mediated by differential
diffusion. Development 137, 25792585.
Ephrussi, A., and St Johnston, D. (2004). Seeing is believing: the bicoid
morphogen gradient matures. Cell 116, 143152.
Fuller, B.G., Lampson, M.A., Foley, E.A., Rosasco-Nitcher, S., Le, K.V., Tobelmann, P., Brautigan, D.L., Stukenberg, P.T., and Kapoor, T.M. (2008). Midzone
activation of aurora B in anaphase produces an intracellular phosphorylation
gradient. Nature 453, 11321136.
Gallo, C.M., Munro, E., Rasoloson, D., Merritt, C., and Seydoux, G. (2008).
Processing bodies and germ granules are distinct RNA granules that interact
in C. elegans embryos. Dev. Biol. 323, 7687.
Golding, I., and Cox, E.C. (2004). RNA dynamics in live Escherichia coli cells.
Proc. Natl. Acad. Sci. USA 101, 1131011315.
Goldstein, B., and Macara, I.G. (2007). The PAR proteins: fundamental players
in animal cell polarization. Dev. Cell 13, 609622.
Guo, S., and Kemphues, K.J. (1995). par-1, a gene required for establishing
polarity in C. elegans embryos, encodes a putative Ser/Thr kinase that is
asymmetrically distributed. Cell 81, 611620.
Gurskaya, N.G., Verkhusha, V.V., Shcheglov, A.S., Staroverov, D.B., Chepurnykh, T.V., Fradkov, A.F., Lukyanov, S., and Lukyanov, K.A. (2006).
Engineering of a monomeric green-to-red photoactivatable fluorescent protein
induced by blue light. Nat. Biotechnol. 24, 461465.
Hudson, B.P., Martinez-Yamout, M.A., Dyson, H.J., and Wright, P.E. (2004).
Recognition of the mRNA AU-rich element by the zinc finger domain of
TIS11d. Nat. Struct. Mol. Biol. 11, 257264.
Hurd, D.D., and Kemphues, K.J. (2003). PAR-1 is required for morphogenesis
of the Caenorhabditis elegans vulva. Dev. Biol. 253, 5465.
Hurov, J.B., Watkins, J.L., and Piwnica-Worms, H. (2004). Atypical PKC
phosphorylates PAR-1 kinases to regulate localization and activity. Curr.
Biol. 14, 736741.
Kalab, P., Weis, K., and Heald, R. (2002). Visualization of a Ran-GTP gradient
in interphase and mitotic Xenopus egg extracts. Science 295, 24522456.
Lippincott-Schwartz, J., and Patterson, G.H. (2008). Fluorescent proteins for

photoactivation experiments. Methods Cell Biol. 85, 4561.
ik, G., Kneeland, T.B., Wieschaus, E.F., and Gregor, T. (2011).
Little, S.C., Tkac
The formation of the Bicoid morphogen gradient requires protein movement
from anteriorly localized mRNA. PLoS Biol. 9, e1000596.
Lizcano, J.M., Goransson, O., Toth, R., Deak, M., Morrice, N.A., Boudeau, J.,
Hawley, S.A., Udd, L., Makela, T.P., Hardie, D.G., and Alessi, D.R. (2004).
LKB1 is a master kinase that activates 13 kinases of the AMPK subfamily,
including MARK/PAR-1. EMBO J. 23, 833843.
Maeder, C.I., Hink, M.A., Kinkhabwala, A., Mayr, R., Bastiaens, P.I., and Knop,
M. (2007). Spatial regulation of Fus3 MAP kinase activity through a reactiondiffusion mechanism in yeast pheromone signalling. Nat. Cell Biol. 9, 1319
1326.
Marx, A., Nugoor, C., Panneerselvam, S., and Mandelkow, E. (2010). Structure
and function of polarity-inducing kinase family MARK/Par-1 within the branch
of AMPK/Snf1-related kinases. FASEB J. 24, 16371648.
Mello, C.C., Schubert, C., Draper, B., Zhang, W., Lobel, R., and Priess, J.R.
(1996). The PIE-1 protein and germline specification in C. elegans embryos.
Nature 382, 710712.
Moravcevic, K., Mendrola, J.M., Schmitz, K.R., Wang, Y.H., Slochower, D.,
Janmey, P.A., and Lemmon, M.A. (2010). Kinase associated-1 domains drive
MARK/PAR1 kinases to membrane targets by binding acidic phospholipids.
Cell 143, 966977.
Nam, S.C., Mukhopadhyay, B., and Choi, K.W. (2007). Antagonistic functions
of Par-1 kinase and protein phosphatase 2A are required for localization of
Bazooka and photoreceptor morphogenesis in Drosophila. Dev. Biol. 306,
624635.
Pagano, J.M., Farley, B.M., McCoig, L.M., and Ryder, S.P. (2007). Molecular
basis of RNA recognition by the embryonic polarity determinant MEX-5. J.
Biol. Chem. 282, 88838894.
Rivers, D.M., Moreno, S., Abraham, M., and Ahringer, J. (2008). PAR proteins
direct asymmetry of the cell cycle regulators Polo-like kinase and Cdc25.
J. Cell Biol. 180, 877885.
Schlaitz, A.L., Srayko, M., Dammermann, A., Quintin, S., Wielsch, N.,
MacLeod, I., de Robillard, Q., Zinke, A., Yates, J.R., 3rd, Muller-Reichert, T.,
et al. (2007). The C. elegans RSA complex localizes protein phosphatase
2A to centrosomes and regulates mitotic spindle assembly. Cell 128, 115127.
Schubert, C.M., Lin, R., de Vries, C.J., Plasterk, R.H., and Priess, J.R. (2000).
MEX-5 and MEX-6 function to establish soma/germline asymmetry in early C.
elegans embryos. Mol. Cell 5, 671682.
Severson, A.F., and Bowerman, B. (2003). Myosin and the PAR proteins
polarize microfilament-dependent forces that shape and position mitotic
spindles in Caenorhabditis elegans. J. Cell Biol. 161, 2126.
Shav-Tal, Y., Darzacq, X., Shenoy, S.M., Fusco, D., Janicki, S.M., Spector,
D.L., and Singer, R.H. (2004). Dynamics of single mRNPs in nuclei of living
cells. Science 304, 17971800.
St Johnston, D., and Ahringer, J. (2010). Cell polarity in eggs and epithelia:
parallels and diversity. Cell 141, 757774.
Su, T.T., Sprenger, F., DiGregorio, P.J., Campbell, S.D., and OFarrell, P.H.
(1998). Exit from mitosis in Drosophila syncytial embryos requires proteolysis
and cyclin degradation, and is associated with localized dephosphorylation.
Genes Dev. 12, 14951503.
Tenlen, J.R., Molk, J.N., London, N., Page, B.D., and Priess, J.R. (2008).
MEX-5 asymmetry in one-cell C. elegans embryos requires PAR-4- and
PAR-1-dependent phosphorylation. Development 135, 36653675.
Vaknin, A., and Berg, H.C. (2004). Single-cell FRET imaging of phosphatase
activity in the Escherichia coli chemotaxis system. Proc. Natl. Acad. Sci.
USA 101, 1707217077.
Wartlick, O., Kicheva, A., and Gonzalez-Gaitan, M. (2009). Morphogen
gradient formation. Cold Spring Harb. Perspect. Biol. 1, a001255.
Yoder, J.H., Chong, H., Guan, K.L., and Han, M. (2004). Modulation of KSR
activity in Caenorhabditis elegans by Zn ions, PAR-1 kinase and PP2A phosphatase. EMBO J. 23, 111119.
Acetylation of Yeast AMPK

Controls Intrinsic Aging
Independently of Caloric Restriction
Jin-Ying Lu,1,2 Yu-Yi Lin,3 Jin-Chuan Sheu,4 June-Tai Wu,5 Fang-Jen Lee,5 Yue Chen,6 Min-I Lin,1 Fu-Tien Chiang,1,4
Tong-Yuan Tai,4 Shelley L. Berger,7 Yingming Zhao,6 Keh-Sung Tsai,1,2,4 Heng Zhu,8,10,* Lee-Ming Chuang,2,4,*
and Jef D. Boeke9,10,*
1Department
of Laboratory Medicine, National Taiwan University Hospital

Institute of Clinical Medicine, College of Medicine
3Institute of Biochemistry and Molecular Biology, College of Medicine
4Department of Internal Medicine, National Taiwan University Hospital
5Institute of Molecular Medicine, College of Medicine
National Taiwan University, Taipei 100, Taiwan
6Ben May Department for Cancer Research, University of Chicago, 929 East 57th Street, W421, Chicago, IL 60637, USA
7Gene Expression and Regulation Program, Wistar Institute, Philadelphia, PA 19104, USA
8Department of Pharmacology and Molecular Sciences
9Departments of Molecular Biology and Genetics
10The High Throughput Biology Center
Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA
*Correspondence: heng.zhu@jhmi.edu (H.Z.), leeming@ntu.edu.tw (L.-M.C.), jboeke@jhmi.edu (J.D.B.)
DOI 10.1016/j.cell.2011.07.044
2Graduate
SUMMARY
Acetylation of histone and nonhistone proteins is an

important posttranslational modification affecting
many cellular processes. Here, we report that NuA4
acetylation of Sip2, a regulatory b subunit of the
Snf1 complex (yeast AMP-activated protein kinase),
decreases as cells age. Sip2 acetylation, controlled
by antagonizing NuA4 acetyltransferase and Rpd3
deacetylase, enhances interaction with Snf1, the
catalytic subunit of Snf1 complex. Sip2-Snf1 interaction inhibits Snf1 activity, thus decreasing phosphorylation of a downstream target, Sch9 (homolog of
Akt/S6K), and ultimately leading to slower growth
but extended replicative life span. Sip2 acetylation
mimetics are more resistant to oxidative stress. We
further demonstrate that the anti-aging effect of
Sip2 acetylation is independent of extrinsic nutrient
availability and TORC1 activity. We propose a protein
acetylation-phosphorylation cascade that regulates
Sch9 activity, controls intrinsic aging, and extends
replicative life span in yeast.
INTRODUCTION
Reversible acetylation and deacetylation of histones are important in regulating chromatin structure and controlling transcription of genes that are crucial for maintenance of cell viability
(Lin et al., 2008). Among the histone deacetylases, Sir2 (yeast
homolog of mammalian SIRT1; Donmez and Guarente, 2010)
and Rpd3 (yeast homolog of mammalian HDAC1; Willis-Martinez

et al., 2010) are especially important in life span regulation in
yeast (Chang and Min, 2002; Jiang et al., 2002). Sir2 extends
replicative life span partially via deacetylating histone H4 lysine
16 that compromises transcriptional silencing (Dang et al.,
2009; Imai et al., 2000; Sinclair and Guarente, 1997). However,
the life span discrepancies between substrate histone mutants
and the acetyltransferase/deacetylase mutants (Dang et al.,
2009) suggest possible roles of nonhistone acetylation substrates in mediating life span regulation by these (de)acetylating
enzymes.
In a previous study, we identified many nonhistone substrates of
the NuA4 complex, of which the catalytic subunit, Esa1, is the only
essential histone acetyltransferase in yeast (Lin et al., 2009). We
show here that NuA4 catalytic mutants have replicative life span
defects caused by impaired acetylation of Sip2, a known replicative life span regulator (Ashrafi et al., 2000). Sip2 is one of three
b regulatory subunits of the Snf1 complex and the only b subunit
implicated in yeast replicative aging (Ashrafi et al., 2000; Lin
et al., 2003). The Snf1 complex contains (1) a catalytic a subunit,
Snf1, which is an AMP-activated serine/threonine protein kinase,
(2) a g subunit, Snf4, (3) and one of the three b regulatory subunitsSip1, Sip2, or Gal83each with a distinctive substrate
specificity (Schmidt and McCartney, 2000). In addition to being
required for transcription of glucose-repressed genes and utilization of carbon sources other than glucose (Amodeo et al., 2007),
Snf1 is also a key player in the response to cellular stress (Sanz,
2003). Importantly, Snf1 activity increases in aged cells even
when ambient glucose is abundant (Ashrafi et al., 2000; Hedbacker and Carlson, 2008). Null mutations in SIP2, the SNF1
repressor, decrease life span; these can be rescued by deletion
of SNF4, the SNF1 activator (Guarente and Kenyon, 2000).
Sip2
415
Ac
335
154
N-Myr
AcAcAc
247
C
HDAC
TSA
Rpd3 Hda1 -
Rpd3 Hda1
+
+
IP: -AcK
Probe: -GST
GBD
Input:
-GST-Sip2p
Snf1 interacon
D
Sip2p-TAP
WT
GST-Sip2p
sip2- sip2- sip23KR 4KR G2A
Sip2p-TAP
Sip2p-TAP
esa1
WT esa1 rpd3 rpd3
WT
rpd3
sip2- sip23KR 4KR
IP: -AcK
Probe: -TAP
Input:
-Sip2p-TAP
Figure 1. Sip2 Is Acetylated at Four Lysine Sites by Esa1, and Rpd3 Is the Counteracting Deacetylase
(A) Cartoon of Sip2 structural domains (adapted from Hedbacker and Carlson, 2008). Mass spectrometry identified four acetylated lysine residues of Sip2, K12,
K16, K17, and K256. Numbers indicate amino acid residues. (Vertical straight line) Myristoylation site; (downward arrows) acetylation sites; (right-left arrow)
regions mapped as sufficient for Snf1 interaction by deletion analysis (Amodeo et al., 2007). Ac, acetylation; N-Myr, N-myristoylation; GBD, glycogen-binding
domain.
(B) Chromosomally integrated sip2-3KR and sip2-4KR, but not sip2-G2A, mutants are hypoacetylated in vivo. The sip2-G2A mutant blocks myristoylation.
sip2-3KR, sip2-K12/16/17R; sip2-4KR, sip2-K12/16/17/256R.
(C) Rpd3, but not Hda1, removes Sip2 acetylation in vitro. Deacetylation reaction is inhibited by addition of trichostatin A (TSA). Asterisk indicates Hda1-TAP;
arrowhead indicates Rpd3-TAP.
(D) Sip2 is hypoacetylated in strains carrying the Ts allele of ESA1, esa1-531, but is hyperacetylated in rpd3D; deletion of RPD3 rescues the acetylation defect of
esa1-531.
(E) Increased Sip2 acetylation in rpd3D is blocked when lysines are mutated to arginines.
See also Figure S1 and Figure S2.
Here, we report that Sip2 acetylation is controlled by the NuA4

and its counteracting deacetylase, Rpd3 (Chang and Pillus,
2009). We examined replicative life span in various SIP2 acetylation mutants and found that Sip2 acetylation extends yeast life
span; besides, constitutive Sip2 acetylation mimetics nearly
totally rescue the life span-shortening phenotype of the NuA4
catalytic mutant, indicating the critical role of Sip2 acetylation
in life span modulation. We further investigated glucose limitation, peroxide sensitivity, and age-associated changes in Sip2
acetylation, established the role of Sip2 acetylation in controlling
Snf1 interactions and activities. Finally, we established that
Sch9, the yeast homolog of Akt and S6K (Madia et al., 2009),
was a common downstream target of two distinct replicative
life span regulating pathways: the intrinsic aging defense
pathway described here controlled by Snf1 kinase and the
extrinsic nutrient-sensing pathway regulated by TORC1.
RESULTS
Sip2 Acetylation Is Controlled by NuA4 and Rpd3
To investigate the function of Sip2 acetylation, we identified four
acetylated lysine residues (K12, K16, K17, and K256) (Figure 1A),
using tandem mass spectrometry (Figures S1AS1C available
online). We then created unacetylable lysine-to-arginine mutant
constructs at these four sites in various combinations by sitedirected mutagenesis and introduced these mutant SIP2
constructs into the endogenous chromosomal locus (Toulmay
and Schneiter, 2006). Using a previously described reverse IP
approach (Lin et al., 2009), we showed that Sip2 is hypoacetylated when the first three (K12, K16, and K17 [3KR]) or all
four (K12, K16, K17, and K256 [4KR]) lysines are mutated
(Sip2-3KR or -4KR; Figures S1D, Figure 1B). We also created
a chromosomally integrated SIP2 glycine 2-to-alanine mutant
(sip2-G2A) to mimic the short-lived, non-N-myristoylated
species of Sip2 (Ashrafi et al., 2000). We found that sip2-G2A
did not have acetylation defects (Figure 1B). An in vitro deacetylation reaction carried out with purified Rpd3-TAP and Hda1TAP (Figure S2A) revealed that Rpd3 treatment abolished the
acetylation signal of Sip2 (Figure 1C). We demonstrated the
activity of purified Hda1-TAP by showing that it deacetylated
lysine 14 of FLAG-Htz1 purified from hda1D strain (Figure S2B)
(Lin et al., 2008). As will be shown below, mutagenesis of the
Sip2 acetylation sites affects replicative aging. The findings are
consistent with a previous report that Rpd3, but not Hda1, is
involved in replicative life span regulation (Kim et al., 1999).
Previously we have shown that the acetylation signals of Sip2
are virtually completely dependent on NuA4/Esa1 both in vivo
and in vitro (Lin et al., 2009). Importantly, simultaneous deletion
of RPD3 reversed the hypoacetylation caused by NuA4 catalytic

subunit temperature-sensitive (Ts) allele esa1-531 when cells
were shifted to the nonpermissive temperature (Figure 1D).
Further, the increased acetylation in rpd3D was abolished by
sip2-3KR and -4KR mutations (Figure 1E). These results support
the simple hypothesis that K12, K16, K17, and K256 in Sip2 are
acetylated by Esa1 and are deacetylated by Rpd3.
Sip2 Acetylation Mimetics Increase Replicative Life
Span
We next examined replicative life span of the SIP2 acetylation
mutants (refer to Table S1 for detailed statistical results). We
found that the esa1-531 strain showed a shortened replicative
life span when grown at a semipermissive temperature, 30 C.
Importantly, in an esa1-531 strain, the lysine-to-glutamine
mutants of SIP2 (sip2-3KQ and sip2-4KQ), mimicking acetylated
Sip2, reversed the shortened life span of esa1-531 (Figure 2A).
Consistent with the above findings, deletion of RPD3 also
reversed the shortened life span of esa1-531 (Figure 2B). We
also found that sip2-4KQ significantly increased the life span,
whereas sip2-4KR and sip2D strains decreased life span to
a similar extent when compared to wild-type (WT) SIP2+ strains
(Figure 2C). Because the N-myristoylation on glycine 2 of Sip2 is
essential for long life span (Ashrafi et al., 2000; Lin et al., 2003),
the chromosomally integrated sip2-G2A mutant was used as
a positive control for shortened life span (Figure 2C). Finally,
we confirmed that simultaneous sip2-4KQ mutant did not further
extend the life span of the null mutant of RPD3 (Figure 2D). These
results provide strong evidence supporting Sip2 as the critical
downstream target of Esa1 and Rpd3 in regulating replicative
life span in yeast.
Sip2 Acetylation and Calorie Restriction
Because carbon substitutions greatly influence longevity in yeast
(Barker et al., 1999; Kaeberlein et al., 2005), we next asked
Figure 2. Sip2 Acetylation Increases the

Cellular Replicative Life Span
(A) Survival curves of the indicated strains with
each median life span in parentheses. The fractions of live cells are plotted as a function of age in
generations. Strains carrying esa1-531 exhibit life
span shortening that is rescued by SIP2 acetylation mimetics (sip2-3KQ and sip2-4KQ).
(B) Deletion of RPD3 reverses the short life span of
esa1-531.
(C) sip2-4KR mutants, as well as the sip2-G2A
mutants and deletion of SIP2, decrease cellular life
span, whereas sip2-4KQ mutants increase life span.
(D) Simultaneous sip2-4KQ mutants on the background of rpd3D fail to further increase life span of
rpd3D. Statistical significance was determined by
Mantel-Cox log-rank test and the details are presented in Table S1.
whether different carbon sources affect

Sip2 acetylation status. We monitored
Sip2 acetylation (Ac-Sip2) levels after
treating cells with 2% glucose, 2% galactose, 0.05% glucose, or 2% glycerol plus 3% ethanol. The only
significant difference in acetylation states was observed when
cells were treated with 0.05% glucose (low glucose, LG);
although Sip2 protein abundance was upregulated, the acetylation signal decreased significantly (Figures S3A and S3B). This
contradicts a simple model that calorie restriction extends
replicative life span via Sip2 acetylation. Indeed, calorie restriction was shown to increase replicative life span through Sir2
(Lin et al., 2002) and other mechanisms (Kaeberlein et al.,
2005). We therefore examined whether low glucose (LG) affected
life span in the SIP2 acetylation mutants and found that LG,
when compared to 2% glucose (normal glucose, NG), increased
life span by 50% in the sip2-4KR and sip2D strains and, to
somewhat a lesser extent, in WT (30%) (Figures 3A3C and
Table S1). However, LG failed to significantly increase the life
span of the sip2-4KQ mutant (Figure 3D and Table S1). Because
glucose limitation significantly improved the replicative life
span of sip2-4KR and sip2D, these findings imply that calorie
restriction extends life span through a mechanism largely independent of Sip2 acetylation. The fact that calorie restriction failed
to significantly extend the life span of sip2-4KQ, which was
similar to that reported in rpd3D (Jiang et al., 2002), suggests
the existence of a common effector(s) downstream of the Sip2
acetylation and calorie restriction modulating replicative life
span in yeast.
Aging Decreases Sip2 Acetylation and Interaction
with Snf1
To investigate the effect of replicative aging on Sip2 acetylation,
we examined Ac-Sip2 in young and old cells by sorting biotinlabeled mother cells (Smeal et al., 1996). The sorting efficiency
was confirmed by showing that old cells contained a mean of
seven bud scars, whereas young cells had none (Figure S3C).
We found that Ac-Sip2 levels decreased in old cells (Figure 4A).
Coimmunoprecipitation of Sip2-TAP and Snf1-HA demonstrated
Figure 3. Glucose Limitation and Replicative Life Span in SIP2 Acetylation Mutants
Survival curves of WT (A), sip2D (B), sip2-4KR (C),
and sip2-4KQ (D) grown in normal (2%, NG) versus
low glucose (0.05%, LG) media. The fractions of
live cells are plotted as a function of age in
generations. Median life span is shown in parentheses. Glucose limitation significantly increases
life span in WT, sip2D, and sip2-4KR, but not in
sip2-4KQ, as determined by Mantel-Cox log-rank
test. The statistical results were summarized in
Table S1. See also Figure S3.
that, although the interaction between Sip2 and Snf1 was readily
seen in young cells, it became undetectable in old cells
(Figure 4B).
To determine whether Sip2 acetylation dictates physical interaction between Sip2 and Snf1, we examined their association in
SIP2 acetylation mutants. Physical interaction between Sip2 and
Snf1 significantly decreased in esa1-531 but increased in rpd3D
(Figure 4C). More importantly, whereas the sip2-3KR and
sip2-4KR mutations almost abolished the interaction with Snf1,
the sip2-3KQ and sip2-4KQ mutations markedly enhanced it
(Figure 4D), supporting the idea Sip2 acetylation is required for
Snf1 interaction and defining the N terminus of Sip2 as an
enhancer of Snf1 interaction (Figure 1A). In contrast, the myristoylation null mutation sip2-G2A (Lin et al., 2003) did not affect
the physical interaction between Sip2 and Snf1 (Figure 4D), suggesting that myristoylation regulates Snf1 activity via a different
mechanism. Because Sip2 is a negative regulator of Snf1 (Ashrafi et al., 2000), these results at least partially explain why
Snf1 activity, which is stimulated under carbon stress, increases
with age and hence has detrimental effects on life span, even in
the presence of abundant ambient glucose (Lin et al., 2003).
To establish the connection between aging and Sip2 acetylation status in cells, we measured trehalose levels, a general
stress indicator in yeast, in young and old cells. This is based
on the previous findings that, during the quiescent phase, trehalose is stored in favor of glycogen presumably to fulfill its numerous stress-protectant functions (Shi et al., 2010). Indeed,
we found that young cells contained significantly less trehalose
than old cells (Figure 4E). Glucose limitation, a form of nutrient
stress, also significantly increased trehalose levels as reported
before (Figure S3D) (Pluskal et al., 2011). Measurement of trehalose levels revealed that esa1-531 contained elevated trehalose
levels that were partially reversed by concomitant RPD3 deletion
(Figure S3E). Sip2-4KR rpd3D and sip2D rpd3D increased treha972 Cell 146, 969979, September 16, 2011 2011 Elsevier Inc.
lose levels compared to rpd3D (Figure S3F). Also, sip2-4KR, sip2-G2A, and
sip2D mutants had increased trehalose
levels compared to WT, whereas sip24KQ had a lower level of trehalose (Figure 4F), supporting a role of Ac-Sip2 in
counteracting stress. We further showed
that sip2-4KR and sip2D mutants were
much more sensitive to hydrogen peroxide (H2O2), a form of oxidative stress
and a mediator of aging (Giorgio et al., 2007), when compared
to WT (Figure 4G); on the contrary, sip2-4KQ was resistant to
H2O2 (Figure 4G).
Consistent with previous findings that trehalose plays a key
role in fueling cell-cycle progression during growth and division
(Shi et al., 2010), we found that growth rates significantly
decreased in the rpd3D strain (Figure S4A) and the sip2-4KQ
strain on various genetic backgrounds (Figures S4B and S4C).
Conversely, sip2-4KR and sip2D mutants partially rescued the
rpd3D growth defect (Figure S4A). Cell volumes were similarly
affected by the mutations; all of the slower-growing variants
were smaller than normal cells (Figure S4D).
Sch9 Is the Downstream Target of Sip2-Snf1
Based on the above results and literature information, we
proposed that Snf1 might have a downstream target that,
when phosphorylated by Snf1, fuels rapid cell growth but results
in shortened life span. In a previous phosphorylation proteome
microarray study, 80 in vitro Snf1 substrates were identified (Ptacek et al., 2005). Among these, Sch9 (an Akt/S6K homolog)
kinase was a promising candidate because it had already been
shown to play an important role in yeast aging (Kaeberlein
et al., 2005). In addition to Snf1, Sch9 is the in vitro substrate
of two additional kinases (of 87 kinases tested): Tos3, the
upstream kinase of Snf1 (Kim et al., 2005) and Pho85 (Ptacek
et al., 2005). Sch9 is also a well-known in vivo kinase substrate
of target of rapamycin complex 1 (TORC1). Similar to TORC1,
it negatively regulates life span in response to nutrient availability
(Kaeberlein et al., 2005; Urban et al., 2007). A kinase itself, Sch9
phosphorylates a critically important ribosomal protein, Rps6
(Urban et al., 2007); rps6D mutants grow slowly and are small,
but they have increased replicative life span through general
inhibition of translation machinery (Chiocchetti et al., 2007) and
protein synthesis (Huber et al., 2009).
Sip2-TAP
Young Old
Young Old
IP
IgG HC
Input
WT
esa1 rpd3
IP: TAP
-HA
Input:
-Snf1-HA
IP: TAP
-HA
Input:
-Snf1-HA
Input:
-Sip2-TAP
Input:
-Sip2-TAP
IP: -Ac-K; input: -TAP

WT
Sip2-TAP
Snf1-HA
IP: TAP
-HA
+
+
sip2- sip2- sip2- sip2- sip23KR 4KR 3KQ 4KQ G2A
+
+
+
+
+
+
+
+
+
+
Trehalose
Input:
-Sip2-TAP
Input:
-Snf1-HA
12.5
10.0
7.5
5.0
2.5
0.0
-Tubulin
G0
3
2
1
Fracon survival
Trehalose
10
**
ns
G7
***
1
0.1
0.01
H2O2
0 min
60 min
0.001
Figure 4. Sip2 Acetylation and Physical Interaction between Sip2 and Snf1 Decrease as Cells Age
(A) Sip2 acetylation is significantly decreased in old cells.
(B and C) Physical interaction between Sip2 and Snf1 assessed by coimmunoprecipitation is decreased in old cells (B) and esa1-531 strain but increased in rpd3D
strain (C).
(D) Interaction between Sip2 and Snf1 is decreased in sip2D, and deacetylation mimetics (sip2-3KR and sip2-4KR) but is not changed in sip2-G2A mutants.
The interaction significantly increases in acetylation mimetics (sip2-3KQ and sip2-4KQ mutants).
(E) Cellular trehalose is significantly increased when cells age. Error bars show standard error of the mean. n = 3.
(F) Trehalose levels are significantly higher in sip2-4KR, sip2-G2A, and sip2D when compared to WT but are lower in sip2-4KQ. Error bars show standard error of
the mean. n = 3.
(G) Fractions of survival of WT, sip2-4KR, sip2-4KQ, and sip2D before and after 1 hr of treatment with 1.5 mM H2O2, are plotted on a log scale. Error bars
show standard error of the mean. n = 2. Statistical significance was assessed by two-way ANOVA with post-hoc test. ns, nonsignificant; *p < 0.05; **p < 0.01;
***p < 0.001.
See also Figure S3 and Figure S4.
We first showed that Sch9 is an in vitro and in vivo substrate of

Snf1. We purified GST-Sch9 and assessed the phosphorylation
signal by performing immunoblotting with antiphosphoserine
and antiphosphothreonine antibodies (Jablonowski et al.,
2004). As a proof of concept, the signal of phosphorylated
GST-Sch9 nearly totally disappeared after treatment with calf
intestine phosphatase, whereas sodium pervanadate (a phosphatase inhibitor) restored the phosphorylation signal (Figure S5A). We carried out an in vitro kinase assay with purified
Snf1-TAP (Figure S2A) from a sip2D strain and showed serine
and threonine phosphorylation of GST-Sch9 in vitro by Snf1
kinase (Figure 5A). We further confirmed that endogenous
Sch9-HA phosphorylation markedly decreased in snf1D especially after treatment with rapamycin to block TORC1 activity
(Figure 5B and whole gel images in Figure S5B). Moreover, a
mutation that abolishes the kinase activity of Snf1 (snf1-K84R)
fails to revert the decreased phosphorylation of endogenous
Sch9-HA in a snf1D strain (Figure S5C), which suggests that
Sch9 is a bona fide in vivo substrate of Snf1.
To detect Sch9 phosphorylation states in different SIP2
mutants, we purified GST-Sch9 proteins from WT, sip2-4KR,
sip2-4KQ, and sip2D strains and assessed their phosphorylation
signals. Both serine and threonine phosphorylation increased
slightly in sip2-4KR and sip2D compared to WT and sip2-4KQ
Snf1
ATP
GST-Sch9
+
+
-
Sch9-HA
+
+
Rapa
IP: HA
-P-Ser
IP: HA
-P-Thr
-P-Ser
-P-Thr
WT snf1 WT
+
-
In vitro
GST-Sch9 (+ Rapamycin)
snf1
+
WT sip2- sip2- sip2

4KR 4KQ
-P-Ser
-P-Thr
-HA
-GST
-GST
-Tubulin
In vivo
Sch9-HA
Young Old
IP: HA
-P-Ser
IP: HA
-P-Thr
-HA
YPD
YPD + Rapamycin
WT
sch9
sch9
sip2-4KR
sch9
sip2
WT (24)
sip2 sch9 (43.5)
1.0
0.8
sip2-4KR sch9 (43)
0.6
sch9 (51.5)
0.4
0.2
0.0
0
25
50
75
100
Figure 5. Acetylation of Sip2 Affects Snf1 Kinase Activity and Results in Hypophosphorylation of Sch9
(A) Snf1 phosphorylates GST-Sch9 in vitro at both serine and threonine sites. In vitro kinase assays were performed by incubating purified GST-Sch9 with or
without Snf1 or ATP as indicated and analyzed by phosphoserine antibody (a-P-Ser) and phosphothreonine antibody (a-P-Thr) to detect Sch9p phosphorylation.
(B) Endogenous Sch9-HA phosphorylation decreases in snf1D mutants after rapamycin (200 ng/ml) treatment to suppress the TOR pathway. Arrowheads
indicate the Sch9-HA band.
(C) GST-Sch9 phosphorylation increases in sip2-4KR and sip2D but decreases in sip2-4KQ, compared to WT after rapamycin (200 ng/ml) treatment.
(D) Endogenous Sch9-HA phosphorylation significantly increases in the old cells. Arrowheads indicate the Sch9-HA band.
(E) SCH9 is epistatic and thus downstream to SIP2 in regulating cellular growth. Ten-fold dilutions of the indicated strains were spotted and grown on YPD plates
without (2 days, 30 C) or with rapamycin (25 ng/ml, 4 days, 30 C).
(F) Deletion of SCH9 rescues the life span shortening of sip2D and sip2-4KR.
See also Figures S1 and Figure S5.
(Figure S5D); the differences were enhanced by inhibiting

TORC1 with rapamycin (Figure 5C), with the lowest level of
phosphorylation observed in the sip2-4KQ mutant. Finally, we
confirmed that endogenous Sch9-HA phosphorylation markedly
increased as cells aged (Figure 5D), which is consistent with
aging-associated increase in Snf1 kinase activity.
To demonstrate that Sch9 is the critical downstream mediator
of Sip2 in regulating growth and life span, we analyzed epistasis.
The growth defects of sch9D cells were not rescued by sip2-4KR
and sip2D (Figure S5E) as expected for a downstream effector.
We showed that sch9D was hypersensitive to rapamycin as
reported (Urban et al., 2007), which remained unaltered in the
concomitant SIP2 mutants (Figure 5E). We also observed that
the phenotype of shortened life span in the sip2D and sip2-4KR
mutant cells was nearly completely reversed by deleting SCH9
(sip2D sch9D and sip2-4KR sch9D in Figure 5F). Consistent
with these observations, sch9D cells also showed comparable
life span extension to both double-mutant strains (Figure 5F).
These data support Sch9 as the downstream target of Sip2-Snf1.
Sip2-Snf1 and TORC1 Function in Parallel Upstream

of Sch9
By comparing growth patterns of Sip2 acetylation mutants in
YPD and rapamycin, we further examined whether Sip2-Snf1
and TORC1 function in parallel upstream of Sch9 in regulating
growth. Both rpd3D and sip2-4KQ showed growth defects in
YPD, consistent with reduced Sch9 activity in these mutants.
The growth differences among the mutants and WT were dampened by rapamycin treatment (Figures 6A6C) through suppressing TORC1, an essential activator of Sch9; rapamycin did
not further inhibit growth of the already Sch9-suppressed
rpd3D and sip2-4KQ. Both sip2-4KR and sip2D partially rescued
the growth defects of rpd3D, possibly through Snf1 activation of
the downstream Sch9. Again, the growth differences disappeared when treated by rapamycin. In contrast, the growth
defects of rpd3D were not rescued by sip2-G2A, another
short-lived mutant (Figure S6A).
Sip2-Snf1 and TORC1 might similarly function in parallel
upstream of Sch9 to regulate replicative life span. We first
Figure 6. Sch9 Is a Parallel Downstream Target of TORC1 and Sip2-Snf1

(AC) Comparison of growth in YPD with and without rapamycin to reveal genetic interactions between RPD3 and SIP2. Ten-fold dilutions of the indicated strains
were spotted and grown on YPD plates without (2 days, 30 C) or with rapamycin (25 ng/ml, 4 days, 30 C).
(D) The life span of sip2-4KR and sip2D in rapamycin is shorter than that of WT in rapamycin (25 ng/ml).
See also Figure S6.
confirmed that rapamycin increased life span of WT (Figure S6B)

(Kaeberlein et al., 2005). We further showed that life span of
sip2-4KR and sip2D, although mildly increased after rapamycin
treatment (Figures S6C and S6D), remained significantly shorter
than WT in rapamycin, suggesting the importance of Sip2 acetylation in controlling cellular life span independent of TORC1
activity (Figure 6D). Life span of sip2-4KQ and rpd3D were not
extended by rapamycin treatment (Figure S6E), consistent with
the fact that Sch9 serves as a common downstream effector.
These results support a protein acetylation/phosphorylation
signaling cascade distinct from the one mediated by TORC1
regulating cellular growth and life span through Sch9 as the
common kinase effector.
Finally, we showed that SCH9 mutants of seven known PDK
(phosphoinositide-dependent protein kinase)-dependent and
TORC1-dependent phosphorylation sites (Urban et al., 2007) still
showed increased phosphorylation signal upon treatment with
Snf1 kinase (Figure S6F), supporting the hypothesis that Snf1
catalyzes Sch9 phosphorylation at sites distinct from those
used by PDK and TORC1. These results also corroborate our
previous findings (Figure 3F) that glucose limitation and Sip2
acetylation seem to operate on the same effector protein in
affecting replicative life span.
DISCUSSION
Yeast longevity studies have provided instructive models for
understanding basic cellular processes in human aging (Bitterman et al., 2003). Among the conserved longevity pathways,
acetyltransferase and deacetylases are well established as links

between cell metabolism, growth, and aging processes (Chang
and Min, 2002; Nakamura et al., 2010). Similar to their opposite
effects on chromatin silencing (Rundlett et al., 1996; Zhou
et al., 2009), mutations in deacetylases RPD3 and SIR2 showed
disparities in extending replicative life span previously thought to
be related to distinct histone tail targets (Chang and Min, 2002).
However, although histone acetylation changes do occur in
aging yeast cells, specific histone acetylation site mutants,
such as those of H4K16 and H3K56, do not phenocopy the
effects of the corresponding acetyltransferases and deacetylases on life span regulation (Dang et al., 2009; Ehrentraut
et al., 2010). This suggests a role for critical nonhistone acetylation substrates in life span regulation (Close et al., 2010).
We report here an in-depth replicative life span analysis of a
nonhistone substrate, Sip2, a b-regulatory subunit of yeast
AMPK (Snf1 complex) that was previously identified in a largescale proteomic screen of NuA4 acetyltransferase (Lin et al.,
2009). Four functionally important acetylation sites were identified; one falls within the C-terminal half within a region of Sip2
that was previously cocrystallized in a heterotrimer with Snf1
and its g subunit (Amodeo et al., 2007). The remaining three sites
cluster in the N terminus in a region absent from the structure
(and not phylogenetically conserved). Because mutation of these
three lysines is sufficient to confer most of the interaction, life
span, and other phenotypes observed in vivo, we suggest that
Sip2 N-terminal acetylation might change Sip2 conformation,
allowing binding to Snf1 and/or the g subunit to stabilize the
Sip2-Snf1 complex.
Figure 7. Model for the Effects of Sip2 Acetylation on Life Span

Regulation
Red lines indicate the intrinsic aging defense pathway identified in this study.
Blue lines indicate the extrinsic nutrient-sensing pathway regulating Sch9
activity. Not shown here for clarity is evidence of a connection between the two
pathways in that we also observed a modest decrease in Sip2 acetylation in
glucose limitation. Ac, lysine acetylation; DeAc, lysine deacetylation; Ph,
phosphorylation of serine and threonine.
NuA4 acetyltransferase, along with the corresponding Sip2

deacetylase Rpd3, are directly involved in replicative life span
regulation. Not only did Sip2 acetylation-mimic mutants live
significantly longer, they nearly totally rescued the severely
shortened life span of NuA4 catalytic mutants. This life span
extension validates their importance as key substrates regulating aging. Furthermore, unacetylable Sip2 mutants, like a
complete deletion, led to a shortened life span compared to
WT, further supporting the critical role of Sip2 acetylation in
Sip2 function. The fact that Sip2 (de)acetylation mimetics phenocopy the corresponding catalytic mutants of NuA4 and Rpd3
strongly supports the importance of Sip2 as the critical downstream nonhistone substrate of these enzymes in controlling
replicative life span.
Sip2 is a regulatory subunit of the Snf1 complex that is homologous to the AMPK complex in higher organisms and plays a
critical role in metabolism and other cellular processes in
response to energy supply (Scott et al., 2009). As a key regulator
of energy homeostasis controlled by AMP/ATP ratios and
various upstream kinases (Lee et al., 2007; Sanz, 2008), we
show here that Snf1 kinase activity can be independently
controlled by acetylation of Sip2. These findings help interpret
the fact that Snf1, required for transcription of glucose-repressed genes (Hedbacker and Carlson, 2008) and a key player
in response to cellular stress (Sanz, 2003), exhibits increased
activity with aging and detrimental effects on life span (Lin
et al., 2003). Our results suggest that Sip2 acetylation enhances
physical interaction with Snf1 and thereby antagonizes catalytic
activity. An increase in the disaccharide trehalose is observed as
a common response to glucose limitation and replicative aging,
representing two distinct stressors in yeast (Shi et al., 2010;
Wang et al., 2010). Decreased Sip2 acetylation observed in
these distinct stress conditions helps explain increased Snf1
activity. The antagonistic effect of Sip2 acetylation on Snf1

activity explains how Sip2 suppresses the detrimental effects
of Snf1 on replicative life span extension (Lin et al., 2003).
Hydrogen peroxide (H2O2), yet another form of stress and aging
mediator, accumulates as aging occurs (Giorgio et al., 2005;
Migliaccio et al., 1999). We showed here that Sip2 acetylation
was critical for H2O2 resistance, supporting the hypothesis that
Sip2 acetylation protects yeast cells from oxidative stress.
Through in vitro and in vivo kinase assays and epistasis
analyses, we confirmed that Sch9, the yeast homolog of
mammalian Akt/S6K (Powers, 2007), is a critical downstream
effector of both TORC1 (Kaeberlein et al., 2005) and the Snf1
complex. In addition to being boosted by sensing nutrient accessibility via TORC1, we observed that Sch9 can be independently
phosphorylated and activated by the Snf1 kinase. This agedependent regulation of Snf1 kinase activity is governed by
progressive deacetylation of Sip2, which releases Snf1 from
Sip2 inhibition in old cells. However, it is possible that this progression is only one dimension of this regulatory pathway and
that there might be dynamic changes in Sip2 acetylation state
in response to certain stresses in younger cells. The Snf1-mediated phosphorylation and activation of Sch9 associated with
replicative aging processes might ultimately lead to dysregulated protein homeostasis, representing a type of intrinsic
aging. Intrinsic aging might represent exhaustion of a metabolite
or regulatory factor or the unavoidable accumulation of damage
(proteotoxicity, oxidative, metabolic, DNA damage, etc.) during
aging (Cohen and Dillin, 2008; Haigis and Yankner, 2010; Silva
and Conboy, 2008; Vijg and Campisi, 2008). This intrinsic aging
is counteracted by a series of posttranslational modifications
on several critical enzymes that works in parallel to an extrinsic
nutrient-sensing pathway through common downstream
machinery to coordinate vegetative growth and replicative life
span in the yeast cells (Figure 7).
AMPK activation promotes longevity under calorie restriction
in metazoans (Evans et al., 2011; Mair et al., 2011; Williams
et al., 2009). The finding that AMPK in yeast seemingly plays an
opposite role to that observed previously in metazoans is
intriguing. However, AMPK activity has also been found to
increase during senescence in various cellular models, including
fibroblasts (Wang et al., 2003), skeletal muscle cells (Thomson
et al., 2009), and aortic endothelial cells (Zu et al., 2010). Although
transient activation of AMPK protects cells against various
internal and external stresses (Narbonne and Roy, 2009), prolonged activation of AMPK is paradoxically correlated with
irreversible senescence and has detrimental effects on normal
physiological functions in mammalian cells (Jones et al., 2005;
Marino et al., 2008; Wang et al., 2011). Also, in a normal mouse
brain, AMPK activity increased with age (Liu et al., 2011). These
findings on senescent mammalian cells may be related to the
end-stage phenotype observed in old yeast mother cells.
The acetylation-phosphorylation signaling cascade is largely
independent of calorie restriction but could potentially integrate
information of ambient nutrient availability and intracellular metabolic status through Sch9, a major downstream substrate of
TORC1 kinase. Whether calorie restriction is the only means of
life span extension remains a debated issue (Colman and Anderson, 2011). Previous reports suggest that calorie restriction
might lead to unwelcome health concerns, especially in elderly

and nonobese subjects (Dirks and Leeuwenburgh, 2006). Thus
modulators of the intrinsic aging might represent more attractive
pharmacologic candidates than calorie restriction mimetics
for life span extension. Although details might differ between
yeast and metazoans, we note that a similar signaling cascade
potentially exists in higher organisms because all of the major
components are evolutionarily conserved.
Strains and Plasmids
Yeast strains used in this study are derived from BY4741 (MATa his3D1 leu2D0
met15D0 ura3D0) and listed in Table S2. Integrated site-directed mutagenesis
was performed as previously described (Toulmay and Schneiter, 2006) and
verified by sequencing. The primers used for site-directed mutagenesis and
sequencing are listed in Table S3. Refer to the Extended Experimental Procedures for mass spectrometer, in vivo acetylation, enzyme purification, and
in vitro deacetylation.
Replicative Life Span Determination and Statistical Analysis
Replicative life span was carried out as described (Kaeberlein et al., 1999). In
brief, cells were freshly streaked to YPD (1% yeast extract, 2% peptone, 2%
dextrose, and 2% agar) from glycerol stock and recovered at room temperature for 2 days and then lightly patched on a new YPD plate and grown overnight. An appropriate number of individual cells was randomly picked under
a microscope and aligned in isolated areas with a micromanipulator. After
about 2 hr of incubation at 30 C, virgin cells (newly budded cells) were separated and left at the original site, and the mother cells were removed to the
grave yard. The life span of these cells was determined by counting the total
number of daughter cells produced. The old mother cells were defined dead
when the budding process stopped and the refractility was completely lost.
For life span analyses in low glucose (LG), cells were grown on YEP plate
(1% yeast extract, 2% peptone, 2% agar) containing 0.05% dextrose.
For life span analyses in rapamycin, cells were grown on YPD containing
25 ng/ml rapamycin. The nonparametric Mantel-Cox log-rank test was performed using GraphPad Prism (version 5.00 for Windows, GraphPad Software,
San Diego California USA) to assess statistical significance of life span differences, as listed in Table S1.
H2O2 Stress Tolerance Test

The test was performed as previously described (Jamieson, 1992) with slight
modification. Exponential phase cultures of WT, sip2-4KR, sip2-4KQ, and
sip2D strains growing at 30 C in YPD (A600 0.25, equals to 5.4 3 106 per ml)
were divided into two aliquots (duplicated) and treated for 1 hr at 30 C with
1.5 mM H2O2. About 200 cells were placed on YPD plates before (0 min)
and after (60 min) H2O2 treatment. Number of colonies formed was counted
after overnight culture at 30 C. Fraction of survival was calculated by dividing
the number of colonies formed after H2O2 treatment by the number before
treatment, with the fraction of survival of each strain before treatment being
set as 1. Statistical significance was assessed by two-way ANOVA plus
post-hoc tests. Refer to Extended Experimental Procedures for cell size
determination.
In Vitro Kinase Assay
Snf1-TAP in a sip2D strain was purified as described (Puig et al., 2001). Kinase
reactions were done in kinase buffer (25 mM HEPES [pH 7.5], 10 mM MgCl2,
2 mM DTT, 0.1 mM EDTA, 0.1 mM EGTA, 50 mM KCl, with or without 1 mM
ATP) for 30 min at 30 C using 1 mg substrate and 100 ng Snf1-TAP.
Reactions were stopped by boiling in 23 SDS sample buffer and then separated by SDS-PAGE. Phosphorylation signals were assessed by immunoblot
blotting using phosphoserine antibody (QIAGEN, 37430) and phosphothreonine antibody (QIAGEN, 37420) following essentially the manufacturers
manual as described (Jablonowski et al., 2004). To verify that the antibodies
are phosphorylation specific, GST-Sch9 purified from WT was treated with
calf intestinal alkaline phosphatase (New England Biolabs, M0290S) following
the manufacturers protocol, with or without adding Na3VO4 (Sigma-Aldrich,
S6508), a phosphatase inhibitor. Reactions were separated by SDS-PAGE
and probed with phosphoserine and phosphothreonine antibodies as
described above. In vitro kinase assay using GST-Sch9 mutated at four serine
or three threonine phosphorylation sites previously identified as PDK and
TORC1 targets (Urban et al., 2007) was performed to test whether these sites
were Snf1 targets. Refer to Extended Experimental Procedures for in vivo Sch9
phosphorylation.
Supplemental information includes Extended Experimental Procedures,
six figures, and three tables and can be found with this article online at
doi:10.1016/j.cell.2011.07.044.
ACKNOWLEDGMENTS
Sorting Old Cells

Yeast old mother cells (generation 7, G7) were isolated from young cells
(generation 0, G0) as described (Smeal et al., 1996). In brief, fresh cells were
grown at 24 C in YPD (1% yeast extract, 2% peptone, and 2% dextrose) to
an A600 of 0.6, and then 10 A600 units of the cells was labeled with 7 mg freshly
prepared biotin (EZ-link Sulfo-NHS-LC-LC; Pierce), cultured to 1 L YPD from
A600 0.01 to A600 0.8 at 24 C in YPD, followed by affinity purification using
Dynabeads MyOne and DynaMag-50 (Invitrogen) according to the manufacturers suggestions. An aliquot of sorted cells was stained with 0.1 mg/ml
Calcofluor White (Sigma-Aldrich, F3543) and observed using a Zeiss Axioskop
fluorescent microscope equipped with a Cool Snap FX camera for bud scar
counting.
Trehalose Assay
A trehalose assay was performed as previously described (Shi et al., 2010). In
brief, 10 A600 units of cells grown in synthetic dropout media to exactly A600 1.2
were pelleted and quickly washed with 1.2 ml ice-cold H2O and then resuspended in 0.3 ml 0.25 M Na2CO3. The samples were boiled for 4 hr, and
then 0.18 ml of 1 M acetic acid and 0.72 ml of 0.2 M sodium acetate was
added to each sample. For controls, half (0.6 ml) of each sample was transferred to a microfuge tube, and the remaining half (0.6 ml) of the sample was
incubated overnight with 0.025 U/ml trehalase (Sigma-Aldrich, T8778) at
37 C. Samples were then centrifuged at top speed for 3 min and assayed
for glucose using a Glucose Assay Kit (Sigma-Aldrich, GAGO20).
We thank Sheng-Ce Tao and Chien-Sheng Chen for their early contribution
to this work. We are grateful to Dr. Brian K. Kennedy for providing sch9D
strain, Dr. Robbie Loewith for providing SCH9 phosphorylation mutants, and
Dr. Marian B. Carlson for providing snf1-K84R construct. We thank
Dr. Fang-Jen Lee for kindly sharing the laboratory space and reagents. We
thank Department of Medical Science, National Taiwan University for technical
help with sequencing. We thank Dr. Yi-Juang Chern for critical suggestions on
this work. Work was supported by National Science Council (NSC 98-2314-B002-031-MY3 to J.-Y.L.), National Taiwan University Hospital (099-001376,
to J.-Y.L.), National Taiwan University (99C101-603 to J.-Y.L., Y.-Y.L., and
L.-M.C.), Liver Disease Prevention & Treatment Research Foundation (Taiwan)
(to J.-Y.L. and Y.-Y.L.), and NIH Common Fund grant (USA) (U54-RR020839
to H.Z. and J.D.B.).
Received: January 18, 2011
Published online: September 8, 2011
REFERENCES
Amodeo, G.A., Rudolph, M.J., and Tong, L. (2007). Crystal structure of the
heterotrimer core of Saccharomyces cerevisiae AMPK homologue SNF1.
Nature 449, 492495.
Ashrafi, K., Lin, S.S., Manchester, J.K., and Gordon, J.I. (2000). Sip2p and its
partner snf1p kinase affect aging in S. cerevisiae. Genes Dev. 14, 18721885.
Barker, M.G., Brimage, L.J., and Smart, K.A. (1999). Effect of Cu,Zn superoxide dismutase disruption mutation on replicative senescence in Saccharomyces cerevisiae. FEMS Microbiol. Lett. 177, 199204.
Bitterman, K.J., Medvedik, O., and Sinclair, D.A. (2003). Longevity regulation in
Saccharomyces cerevisiae: linking metabolism, genome stability, and heterochromatin. Microbiol. Mol. Biol. Rev. 67, 376399.
Chang, C.S., and Pillus, L. (2009). Collaboration between the essential Esa1
acetyltransferase and the Rpd3 deacetylase is mediated by H4K12 histone
acetylation in Saccharomyces cerevisiae. Genetics 183, 149160.
Chang, K.T., and Min, K.T. (2002). Regulation of lifespan by histone deacetylase. Ageing Res. Rev. 1, 313326.
Chiocchetti, A., Zhou, J., Zhu, H., Karl, T., Haubenreisser, O., Rinnerthaler, M.,
Heeren, G., Oender, K., Bauer, J., Hintner, H., et al. (2007). Ribosomal proteins
Rpl10 and Rps6 are potent regulators of yeast replicative life span. Exp.
Gerontol. 42, 275286.
Close, P., Creppe, C., Gillard, M., Ladang, A., Chapelle, J.P., Nguyen, L., and
Chariot, A. (2010). The emerging role of lysine acetylation of non-nuclear
proteins. Cell. Mol. Life Sci. 67, 12551264.
Cohen, E., and Dillin, A. (2008). The insulin paradox: aging, proteotoxicity and
neurodegeneration. Nat. Rev. Neurosci. 9, 759767.
Colman, R.J., and Anderson, R.M. (2011). Nonhuman primate calorie restriction. Antioxid. Redox Signal. 14, 229239.
Dang, W., Steffen, K.K., Perry, R., Dorsey, J.A., Johnson, F.B., Shilatifard, A.,
Kaeberlein, M., Kennedy, B.K., and Berger, S.L. (2009). Histone H4 lysine 16
acetylation regulates cellular lifespan. Nature 459, 802807.
Dirks, A.J., and Leeuwenburgh, C. (2006). Caloric restriction in humans:
potential pitfalls and health concerns. Mech. Ageing Dev. 127, 17.
Jamieson, D.J. (1992). Saccharomyces cerevisiae has distinct adaptive

responses to both hydrogen peroxide and menadione. J. Bacteriol. 174,
66786681.
Jiang, J.C., Wawryn, J., Shantha Kumara, H.M., and Jazwinski, S.M. (2002).
Distinct roles of processes modulated by histone deacetylases Rpd3p,
Hda1p, and Sir2p in life extension by caloric restriction in yeast. Exp. Gerontol.
37, 10231030.
Jones, R.G., Plas, D.R., Kubek, S., Buzzai, M., Mu, J., Xu, Y., Birnbaum, M.J.,
and Thompson, C.B. (2005). AMP-activated protein kinase induces a
p53-dependent metabolic checkpoint. Mol. Cell 18, 283293.
Kaeberlein, M., McVey, M., and Guarente, L. (1999). The SIR2/3/4 complex
and SIR2 alone promote longevity in Saccharomyces cerevisiae by two
different mechanisms. Genes Dev. 13, 25702580.
Kaeberlein, M., Powers, R.W., III, Steffen, K.K., Westman, E.A., Hu, D., Dang,
N., Kerr, E.O., Kirkland, K.T., Fields, S., and Kennedy, B.K. (2005). Regulation
of yeast replicative life span by TOR and Sch9 in response to nutrients.
Science 310, 11931196.
Kim, S., Benguria, A., Lai, C.Y., and Jazwinski, S.M. (1999). Modulation of
life-span by histone deacetylase genes in Saccharomyces cerevisiae. Mol.
Biol. Cell 10, 31253136.
Kim, M.D., Hong, S.P., and Carlson, M. (2005). Role of Tos3, a Snf1 protein
kinase kinase, during growth of Saccharomyces cerevisiae on nonfermentable
carbon sources. Eukaryot. Cell 4, 861866.
Lee, J.H., Koh, H., Kim, M., Kim, Y., Lee, S.Y., Karess, R.E., Lee, S.H., Shong,
M., Kim, J.M., Kim, J., and Chung, J. (2007). Energy-dependent regulation of
cell structure by AMP-activated protein kinase. Nature 447, 10171020.
Lin, S.J., Kaeberlein, M., Andalis, A.A., Sturtz, L.A., Defossez, P.A., Culotta,
V.C., Fink, G.R., and Guarente, L. (2002). Calorie restriction extends Saccharomyces cerevisiae lifespan by increasing respiration. Nature 418, 344348.
Donmez, G., and Guarente, L. (2010). Aging and disease: connections to

sirtuins. Aging Cell 9, 285290.
Lin, S.S., Manchester, J.K., and Gordon, J.I. (2003). Sip2, an N-myristoylated
beta subunit of Snf1 kinase, regulates aging in Saccharomyces cerevisiae by
affecting cellular histone kinase activity, recombination at rDNA loci, and
silencing. J. Biol. Chem. 278, 1339013397.
Ehrentraut, S., Weber, J.M., Dybowski, J.N., Hoffmann, D., and EhrenhoferMurray, A.E. (2010). Rpd3-dependent boundary formation at telomeres by
removal of Sir2 substrate. Proc. Natl. Acad. Sci. USA 107, 55225527.
Lin, Y.Y., Qi, Y., Lu, J.Y., Pan, X., Yuan, D.S., Zhao, Y., Bader, J.S., and Boeke,
J.D. (2008). A comprehensive synthetic genetic interaction network governing
yeast histone acetylation and deacetylation. Genes Dev. 22, 20622074.
Evans, D.S., Kapahi, P., Hsueh, W.C., and Kockel, L. (2011). TOR signaling
never gets old: aging, longevity and TORC1 activity. Ageing Res. Rev. 10,
225237.
Lin, Y.Y., Lu, J.Y., Zhang, J., Walter, W., Dang, W., Wan, J., Tao, S.C., Qian, J.,
Zhao, Y., Boeke, J.D., et al. (2009). Protein acetylation microarray reveals that
NuA4 controls key metabolic target regulating gluconeogenesis. Cell 136,
10731084.
Giorgio, M., Migliaccio, E., Orsini, F., Paolucci, D., Moroni, M., Contursi, C.,
Pelliccia, G., Luzi, L., Minucci, S., Marcaccio, M., et al. (2005). Electron transfer
between cytochrome c and p66Shc generates reactive oxygen species that
trigger mitochondrial apoptosis. Cell 122, 221233.
Giorgio, M., Trinei, M., Migliaccio, E., and Pelicci, P.G. (2007). Hydrogen
peroxide: a metabolic by-product or a common mediator of ageing signals?
Nat. Rev. Mol. Cell Biol. 8, 722728.
Guarente, L., and Kenyon, C. (2000). Genetic pathways that regulate ageing in
model organisms. Nature 408, 255262.
Haigis, M.C., and Yankner, B.A. (2010). The aging stress response. Mol. Cell
40, 333344.
Hedbacker, K., and Carlson, M. (2008). SNF1/AMPK pathways in yeast. Front.
Biosci. 13, 24082420.
Huber, A., Bodenmiller, B., Uotila, A., Stahl, M., Wanka, S., Gerrits, B.,
Aebersold, R., and Loewith, R. (2009). Characterization of the rapamycinsensitive phosphoproteome reveals that Sch9 is a central coordinator of
protein synthesis. Genes Dev. 23, 19291943.
Liu, F., Benashski, S.E., Persky, R., Xu, Y., Li, J., and McCullough, L.D. (2011).
Age-related changes in AMP-activated protein kinase after stroke. Age
(Dordr). Published online March 1, 2011. 10.1007/s11357-011-9214-8.
Madia, F., Wei, M., Yuan, V., Hu, J., Gattazzo, C., Pham, P., Goodman, M.F.,
and Longo, V.D. (2009). Oncogene homologue Sch9 promotes age-dependent
mutations by a superoxide and Rev1/Polzeta-dependent mechanism. J. Cell
Biol. 186, 509523.
Mair, W., Morantte, I., Rodrigues, A.P., Manning, G., Montminy, M., Shaw,
R.J., and Dillin, A. (2011). Lifespan extension induced by AMPK and calcineurin
is mediated by CRTC-1 and CREB. Nature 470, 404408.
Marino, G., Ugalde, A.P., Salvador-Montoliu, N., Varela, I., Quiros, P.M.,
Cadinanos, J., van der Pluijm, I., Freije, J.M., and Lopez-Otn, C. (2008).
Premature aging in mice activates a systemic metabolic response involving
autophagy induction. Hum. Mol. Genet. 17, 21962211.
Migliaccio, E., Giorgio, M., Mele, S., Pelicci, G., Reboldi, P., Pandolfi, P.P.,
Lanfrancone, L., and Pelicci, P.G. (1999). The p66shc adaptor protein controls
oxidative stress response and life span in mammals. Nature 402, 309313.
Imai, S., Armstrong, C.M., Kaeberlein, M., and Guarente, L. (2000). Transcriptional silencing and longevity protein Sir2 is an NAD-dependent histone
deacetylase. Nature 403, 795800.
Nakamura, A., Kawakami, K., Kametani, F., Nakamoto, H., and Goto, S. (2010).
Biological significance of protein modifications in aging and calorie restriction.
Ann. N Y Acad. Sci. 1197, 3339.
Jablonowski, D., Fichtner, L., Stark, M.J., and Schaffrath, R. (2004). The yeast
elongator histone acetylase requires Sit4-dependent dephosphorylation for
toxin-target capacity. Mol. Biol. Cell 15, 14591469.
Narbonne, P., and Roy, R. (2009). Caenorhabditis elegans dauers need

LKB1/AMPK to ration lipid reserves and ensure long-term survival. Nature
457, 210214.
Pluskal, T., Hayashi, T., Saitoh, S., Fujisawa, A., and Yanagida, M. (2011).
Specific biomarkers for stochastic division patterns and starvation-induced
quiescence under limited glucose levels in fission yeast. FEBS J. 278, 1299
1315.
Thomson, D.M., Brown, J.D., Fillmore, N., Ellsworth, S.K., Jacobs, D.L.,
Winder, W.W., Fick, C.A., and Gordon, S.E. (2009). AMP-activated protein
kinase response to contractions and treatment with the AMPK activator AICAR
in young adult and old skeletal muscle. J. Physiol. 587, 20772086.
Powers, T. (2007). TOR signaling and S6 kinase 1: Yeast catches up. Cell
Metab. 6, 12.
Toulmay, A., and Schneiter, R. (2006). A two-step method for the introduction
of single or multiple defined point mutations into the genome of Saccharomyces cerevisiae. Yeast 23, 825831.
Ptacek, J., Devgan, G., Michaud, G., Zhu, H., Zhu, X., Fasolo, J., Guo, H., Jona,
G., Breitkreutz, A., Sopko, R., et al. (2005). Global analysis of protein phosphorylation in yeast. Nature 438, 679684.
Puig, O., Caspary, F., Rigaut, G., Rutz, B., Bouveret, E., Bragado-Nilsson, E.,
Wilm, M., and Seraphin, B. (2001). The tandem affinity purification (TAP)
method: a general procedure of protein complex purification. Methods 24,
218229.
Rundlett, S.E., Carmen, A.A., Kobayashi, R., Bavykin, S., Turner, B.M., and
Grunstein, M. (1996). HDA1 and RPD3 are members of distinct yeast histone
deacetylase complexes that regulate silencing and transcription. Proc. Natl.
Acad. Sci. USA 93, 1450314508.
Sanz, P. (2003). Snf1 protein kinase: a key player in the response to cellular
stress in yeast. Biochem. Soc. Trans. 31, 178181.
Sanz, P. (2008). AMP-activated protein kinase: structure and regulation. Curr.
Protein Pept. Sci. 9, 478492.
Schmidt, M.C., and McCartney, R.R. (2000). beta-subunits of Snf1 kinase are
required for kinase function and substrate definition. EMBO J. 19, 49364943.
Urban, J., Soulard, A., Huber, A., Lippman, S., Mukhopadhyay, D., Deloche,
O., Wanke, V., Anrather, D., Ammerer, G., Riezman, H., et al. (2007). Sch9 is
a major target of TORC1 in Saccharomyces cerevisiae. Mol. Cell 26, 663674.
Vijg, J., and Campisi, J. (2008). Puzzles, promises and a cure for ageing.
Nature 454, 10651071.
Wang, J., Jiang, J.C., and Jazwinski, S.M. (2010). Gene regulatory changes in
yeast during life extension by nutrient limitation. Exp. Gerontol. 45, 621631.
Wang, W., Yang, X., Lopez de Silanes, I., Carling, D., and Gorospe, M. (2003).
Increased AMP:ATP ratio and AMP-activated protein kinase activity during
cellular senescence linked to reduced HuR function. J. Biol. Chem. 278,
2701627023.
Wang, Y., Liang, Y., and Vanhoutte, P.M. (2011). SIRT1 and AMPK in regulating
mammalian senescence: a critical review and a working model. FEBS Lett.
585, 986994.
Scott, J.W., Oakhill, J.S., and van Denderen, B.J. (2009). AMPK/SNF1 structure: a menage a trois of energy-sensing. Front. Biosci. 14, 596610.
Williams, D.S., Cash, A., Hamadani, L., and Diemer, T. (2009). Oxaloacetate
supplementation increases lifespan in Caenorhabditis elegans through an
AMPK/FOXO-dependent pathway. Aging Cell 8, 765768.
Shi, L., Sutter, B.M., Ye, X., and Tu, B.P. (2010). Trehalose is a key determinant
of the quiescent metabolic state that fuels cell cycle progression upon return to
growth. Mol. Biol. Cell 21, 19821990.
Willis-Martinez, D., Richards, H.W., Timchenko, N.A., and Medrano, E.E.

(2010). Role of HDAC1 in senescence, aging, and cancer. Exp. Gerontol. 45,
279285.
Silva, H., and Conboy, I.M. (2008). Aging and stem cell renewal (Cambridge,
MA: In StemBook).
Zhou, J., Zhou, B.O., Lenzmeier, B.A., and Zhou, J.Q. (2009). Histone
deacetylase Rpd3 antagonizes Sir2-dependent silent chromatin propagation.
Nucleic Acids Res. 37, 36993713.
Sinclair, D.A., and Guarente, L. (1997). Extrachromosomal rDNA circles

a cause of aging in yeast. Cell 91, 10331042.
Smeal, T., Claus, J., Kennedy, B., Cole, F., and Guarente, L. (1996). Loss of
transcriptional silencing causes sterility in old mother cells of S. cerevisiae.
Cell 84, 633642.
Zu, Y., Liu, L., Lee, M.Y., Xu, C., Liang, Y., Man, R.Y., Vanhoutte, P.M., and
Wang, Y. (2010). SIRT1 promotes proliferation and prevents senescence
through targeting LKB1 in primary porcine aortic endothelial cells. Circ. Res.
106, 13841393.
Endothelial Cells Are Central

Orchestrators of Cytokine Amplification
during Influenza Virus Infection
John R. Teijaro,1,5 Kevin B. Walsh,1,5 Stuart Cahalan,2 Daniel M. Fremgen,1 Edward Roberts,3 Fiona Scott,4
Esther Martinborough,4 Robert Peach,4 Michael B.A. Oldstone,1,* and Hugh Rosen2,*
1Department
of Immunology and Microbial Science

of Chemical Physiology
3Department of Chemistry
The Scripps Research Institute, La Jolla, CA 92037, USA
4Receptos, Inc., La Jolla, CA 92037, USA
*Correspondence: mbaobo@scripps.edu (M.B.A.O.), hrosen@scripps.edu (H.R.)
DOI 10.1016/j.cell.2011.08.015
2Department
SUMMARY
Cytokine storm during viral infection is a prospective

predictor of morbidity and mortality, yet the cellular
sources remain undefined. Here, using genetic and
chemical tools to probe functions of the S1P1 receptor, we elucidate cellular and signaling mechanisms that are important in initiating cytokine storm.
Whereas S1P1 receptor is expressed on endothelial
cells and lymphocytes within lung tissue, S1P1 agonism suppresses cytokines and innate immune cell
recruitment in wild-type and lymphocyte-deficient
mice, identifying endothelial cells as central regulators of cytokine storm. Furthermore, our data reveal
immune cell infiltration and cytokine production as
distinct events that are both orchestrated by endothelial cells. Moreover, we demonstrate that suppression of early innate immune responses through
S1P1 signaling results in reduced mortality during
infection with a human pathogenic strain of influenza
virus. Modulation of endothelium with a specific
agonist suggests that diseases in which amplification of cytokine storm is a significant pathological
component could be chemically tractable.
INTRODUCTION
Morbidity and mortality caused by severe influenza infections
reflect properties that are intrinsic to the virus strain, including
replication potential, receptor usage, and cytopathic effects on
pulmonary epithelial cells (Garca-Sastre, 2010; Tscherne and
Garca-Sastre, 2011). In addition to viral-intrinsic factors, hostspecific traits such as divergent susceptibilities to infection as
well as differences in host immune responses may ameliorate
or exacerbate both infection and clinical outcome. An overly
aggressive innate response, with early recruitment of inflamma980 Cell 146, 980991, September 16, 2011 2011 Elsevier Inc.
tory leukocytes to the lung, was a key contributor to the

morbidity of the 1918 influenza infection (Kobasa et al., 2007).
More recent clinical literature on avian H5N1 infection documented a significant association between excessive early cytokine responses, immune cell recruitment, and poor outcome
(de Jong et al., 2006). Public health approaches to influenza
pandemics have relied primarily on preventative vaccine strategies and supportive measures, including the use of antiviral therapies. Nevertheless, the speed at which the 2009 H1N1 influenza
virus swine pandemic spread during the lag of vaccine availability highlighted the need to identify additional mechanisms
for amelioration of influenza virus infection (Openshaw and
Dunning, 2010). Antiviral drugs inhibiting virus replication may
select for mutational escape, rendering the therapy ineffective.
Modulation of the host immune response has the potential
advantage of exerting less-selective pressure on viral populations. Though the prospect of blunting cytokine storm is enticing,
a major limitation to treating diseases in which cytokine storm
contributes to pathogenesis is the limited understanding of the
cellular triggers of this process. Thus, we sought to define
cellular signaling pathways that are chemically tractable to test
the hypothesis that modulating cytokine storm would provide
insight into influenza pathogenesis, with potential therapeutic
implications.
Chemical modulators of the sphingosine-1-phosphate (S1P)
signaling system have provided insight into immune cell trafficking and immune responses (Rivera et al., 2008; Rosen
et al., 2009). Due to the immunomodulatory properties of S1P
receptor signaling, chemical agonists have been used successfully for the treatment of relapsing and remitting multiple sclerosis (Brinkmann et al., 2010). Though the systems biology of
the S1P receptors is complex, with five receptor subtypes, differential expression, coupling, attenuation, and catabolism (Rosen
and Liao, 2003), the availability of selective chemical probes
together with mouse genetic models allows detailed insights
into immunopathogenesis. We previously showed that a nonselective S1P receptor agonist inhibited cytokine storm,
dendritic cell migration, and subsequent antigen-specific T cell
pg/mL BALF of C57BL/6J Mice
1200
VEH
AAL-R
CYM-5442
***
900
600
***
300
40
*
***
*
***
20
0
***
**
**
IFN-
CCL2
IL-6
TNF-
IFN-
Cytokines/chemokines
3
***
VEH
AAL-R
CYM-5442
2
1
**
0.8
***
0.6
**
0.4
***
0.2
0
*
Macrophages/
monocytes
Neutrophils
NK cells
Cell Populations
CD69 MFI on
macrophages/monocytes
C
% of Max
4000
***
**
3000
2000
1000
D
CD69 MFI on NK cells
Administration of an S1P1 Agonist Blunts Cytokine

Storm
Previously, we reported that treatment of influenza virus-infected
mice with AAL-R, a promiscuous S1P receptor agonist for S1P1
and S1P35 receptors, inhibits early proinflammatory cytokine
expression and innate immune cell accumulation within the
lung (Marsolais et al., 2008; Marsolais et al., 2009). To assess
the contribution of S1P1 receptor signaling in inhibition of influenza virus-induced early inflammation, infected mice were
treated with the S1P1 receptor-specific agonist CYM-5442 (Gonzalez-Cabrera et al., 2008). Treatment with 2 mg/kg of CYM5442 twice daily significantly inhibited secretion of cytokines
and chemokines associated with influenza virus-induced pathology, including IFN-a, CCL2, IL-6, TNF-a, and IFN-g (Figure 1A)
in addition to CCL3, CCL5, CXCL2, and IL-1a (Figure S1A),
compared to vehicle-treated mice 48 hr postinfection. CYM5442 reduction of IFN-a, CCL2, CCL3, CCL5, IL-1a, and IL-6
expression was as complete as treatment with the promiscuous
agonist; however, it was not as effective as AAL-R in suppressing
CXCL2, TNF-a, and IFN-g (Figure 1A and Figure S1A), suggesting a role for other S1P receptors in modulating those cytokines. In addition to inhibiting cytokine/chemokine production,
AAL-R and CYM-5442 administration to influenza virus-infected
mice blunted the accumulation of innate inflammatory infiltrate
characterized as macrophages/monocytes (CD11b+, F480+,
Ly6G ), neutrophils (CD11b+, LyG6+, F480 ), and NK cells
(NK1.1+, CD3 ), although AAL-R was more effective at inhibiting
macrophage/monocyte and NK cell accumulation in the lung
(Figure 1B). We also observed significantly reduced CD69
expression on macrophage/monocytes and NK cells following
treatment with either CYM-5442 or AAL-R 48 hr postinfection,
demonstrating diminished cell activation (Figures 1C and 1D).
Despite significant blunting of innate immune cell recruitment
and cytokine/chemokine responses, we observed no differences
in viral titers following AAL-R or CYM-5442 treatment compared
1500
% of Max
RESULTS
Cell Numbers in the Lung

of C57BL/6J mice x (10 6)
proliferation, protecting the lung tissue from host-mediated

injury. Alleviation of immunopathology by the nonselective S1P
receptor agonist did not affect virus clearance nor the production
of neutralizing antibodies, demonstrating that, although cytokine
storm was diminished, retention of a sufficient cellular and
humoral immune response, as well as long-term protective
immunity, still occurs (Marsolais et al., 2008; Marsolais et al.,
2009).
Here, through the use of an S1P1 receptor subtype-selective
agonist as well as genetic and biochemical tools, we define
a crucial endothelial signaling loop that is important for the initiation of cytokine storm. We reveal that cytokine secretion and
immune cell infiltration are separable events that are both regulated by the pulmonary endothelium. Further, we demonstrate
that suppression of early innate immune responses through
S1P1 signaling results in reduced mortality during human pathogenic influenza virus challenge. Thus, S1P1 receptor signaling
in endothelium provides a mechanism for attenuation of influenza virus-induced morbidity and reveals an unexpected role
for endothelial cells as regulators of cytokine storm.
4000
**
3000
2000
1000
0
VEH
AAL-R CYM-5442
CD69
Figure 1. S1P1 Receptor Agonism Suppresses Early Proinflammatory Cytokine and Chemokine Production and Innate
Immune Cell Recruitment during Influenza Virus Infection
Mice were infected with 1 3 104 PFU WSN influenza virus, and vehicle (water),
AAL-R (0.2 mg/kg) (1 hr postinfection), or CYM-5442 (2 mg/kg) (1,13, 25, and
37 hr postinfection) were administered i.t. to mice.
(A) Proinflammatory cytokines and chemokines were measured 48 hr postinfection in BALF by ELISA.
(B) Total numbers of innate immune cells were quantified from collagenasedigested lungs by flow cytometry at 48 hr postinfluenza virus infection.
(C) Histograms (left) and mean fluorescent intensity (MFI) (right) of CD69
expression on macrophages was quantified on vehicle, AAL-R-, or CYM-5442treated mice 48 hr postinfluenza virus infection by flow cytometry staining.
(D) Histograms (left) and MFI (right) of NK cell CD69 expression quantified as
in (C).
Data represent average SEM from four to mice per group. *p < 0.05;
**p < 0.005; ***p < 0.0005. Results are representative of greater than six
independent experiments. See also Figure S1 and Figure S2.
pg/ml in the BALF

of C57BL/6J mice
2000
Vehicle
CYM-5442
1500
1000
500
400
300
200
100
0
**
**
**
**
***
IFN- CCL2 CCL5 CXCL10 IL-6
Total # of cells in the lung

of C57BL/6J mice (x 105)
1.0
**
0
CD69+ Macrophages
& monocytes
Cytokines/Chemokines
CD69+ NK cells
Vehicle
RP-002
1500
1000
500
400
300
200
100
0
***
**
*
**
**
**
IFN- CCL2 CCL5 CXCL10 IL-6
IL-1

of C57BL/6J mice (x105)
D
2000
pg/ml in the BALF

of C57BL/6J mice
0.5
IL-1
Vehicle
CYM-5442
1.5
2.0
Vehicle
RP-002
1.5
1.0
**
0.5
0
CD69+ Macrophages/
monocytes
CD69+ NK cells
Figure 2. S1P1 Receptor Agonism Suppresses Early Proinflammatory Cytokine and Chemokine Production and Recruitment of Activated
Innate Immune Cells during Human Pathogenic H1N1:2009 Swine Influenza Virus Infection
(AD) Mice were infected with 1 3 105 PFU A/Wisconsin/WSLH34939/09 influenza virus, and either vehicle (water), CYM-5442 (2 mg/kg) (1, 13, 25, and 37 hr
postinfection), or RP-002 (2 mg/kg on 1 and 25 hr postinfection) were administered i.t. to mice. Proinflammatory cytokines and chemokines were measured 48 hr
postinfection in BALF by ELISA in either CYM-5442- (A) or RP-002-treated mice (C). Total numbers of innate immune cells were quantified from collagenasedigested lungs by flow cytometry at 48 hr postinfluenza virus infection in mice treated with either CYM-5442 (B) or RP-002 (D). Data represent average SEM from
four to five mice per group. *p < 0.05; **p < 0.005; ***p < 0.0005. Results are representative of two independent experiments. See also Figure S3.
to vehicle-treated mice (Figure S2), demonstrating that S1P1

agonism neither inhibits nor enhances viral replication. Collectively, these results reveal that activation of a single sphingosine-1-phosphate receptor, S1P1, is sufficient to blunt global
innate inflammation following mouse-adapted influenza virus
infection in mice.
We next asked whether S1P1 receptor agonism could suppress early innate cytokine and chemokine responses following
infection with a human pathogenic isolate of influenza virus. To
test this, we infected mice with a virulent isolate of the 2009
H1N1 pandemic influenza virus that was isolated from a hospitalized patient and has never been passaged through mice
(A/Wisconsin/WSLH34939/09) (Itoh et al., 2009). Infection of
C57BL/6J mice with this strain causes severe disease, resulting
in rapid mortality beginning between days 4 and 6 postinfection
(data not shown). Similar to infection with mouse-adapted WSN
influenza virus, treatment with CYM-5442 following infection with
H1N1:2009 significantly suppressed cytokine and chemokine
responses and the accumulation of activated innate immune
cells 48 hr postinfection (Figures 2A and 2B). We also tested
an additional S1P1 receptor-selective agonist (RP-002) to provide support for S1P1 receptor specificity and to show directly
that S1P1-selective agonists share this modulation of innate
immunopathology. We infected mice with H1N1:2009 pandemic
influenza virus as above and treated mice with RP-002 at 1
and 25 hr postinfection. Similar to what we observed with
CYM-5442, RP-002 treatment significantly inhibited the production of multiple proinflammatory cytokine and chemokines
(Figure 2C) and suppressed the accumulation of activated
(CD69+) macrophages/monocytes and NK cells in the infected
lung 48 hr postinfection (Figure 2D). Moreover, RP-002-mediated suppression of innate immune cell recruitment and cytokine/chemokine production occurred without altering lung viral
titers (Figure S3), demonstrating that S1P1 agonist-mediated
suppression of cytokines and chemokines is not due to direct
effects on influenza virus replication.
S1P1 Agonist-Mediated Suppression of Early Innate
Immune Responses Results in Protection to Human
Pathogenic Influenza Virus Challenge in Mice
Early dysregulated innate immune responses in the lung have
been associated with morbidity and mortality during infection
with highly pathogenic strains of influenza virus (Cilloniz et al.,
2009; Kobasa et al., 2007). Therefore, we asked whether blunting
early innate cytokine/chemokine responses using an S1P1
agonist could protect mice from lethal infection with a virulent
human isolate of pandemic 2009 H1N1 that had not been
passaged in mice (A/Wisconsin/WSLH/34939/09). In order not
to impose additional stress on the infected lung and because
oral delivery is a popular route for drug administration, we
orally administered the S1P1 agonist, RP-002, at 6 mg/kg
by gavage to C57BL/6J mice infected with 1 3 105 PFU of
S1P1 Is Expressed on Pulmonary Endothelium

and Lymphocytes
To identify cells in the lung that express the S1P1 receptor, we
utilized an eGFP-S1P1 receptor knockin mouse in which the
native S1P1 receptor was homologously replaced with a functional enhanced green fluorescent protein (eGFP)-tagged S1P1
receptor (S1P1-eGFP) (Cahalan et al., 2011). This mouse model
allowed direct detection of eGFP-S1P1 receptor protein expression by flow cytometry, with additional biochemical confirmation
using highly specific antibodies for GFP followed by western
blotting that distinguishes the fused S1P1-eGFP receptor by
specific molecular weight. High levels of S1P1-eGFP were detected on lung lymphatic (CD45 , CD31+, gp38+) and vascular
(CD45 , CD31+, gp38 ) endothelium, whereas pulmonary epithelium (CD45 , CD31 , EpCAM+) was negative for S1P1-eGFP
(Figure 4A). As reported previously, CD4+ T cells (CD4+, CD3+),
CD8+ T cells (CD4+, CD3+), and B cells (B220+, CD19+) expressed S1P1-eGFP (Figure 4B), whereas leukocytes, including
macrophages/monocytes, a subset of alveolar macrophages
(F480+, CD11c+, CD11b ), dendritic cells (CD11c+, I/A-I/E+,
CD205+, F480 ), neutrophils, NK cells (NK1.1+, CD3-), and innate
lymphoid cells (Lin Sca1+), expressed negligible levels of
S1P1-eGFP (Figure 4C and data not shown). Cells from uninfected mice expressed similar levels of S1P1-eGFP as infected
mice, demonstrating that S1P1 receptor expression is not altered
following influenza virus infection (Figure 4D). To confirm the
S1P1-eGFP expression pattern, a western blot for S1P1-eGFP
was performed on cells FACS sorted from the lungs of infected
800
Vehicle
RP-002
pg/ml in the BALF

of C57Bl/6J mice
600
400
200
150
100
**
50
**
IFN-
CCL5
IL-1
IL-6
TNF-
B
of C57Bl/6J mice (x105)
10
Vehicle
RP-002
8
6
**
4
2
***
0
CD69+ Macrophages/
monocytes
C
Percent Survival
A/Wisconsin/WSLH/34939/09 at 1 and 25 hr postinfection.

RP-002 significantly suppressed cytokine amplification (Figure 3A) as well as recruitment of myeloid cells to the infected
lung, as measured by flow cytometry (Figure 3B).
Early administration of RP-002 resulted in enhanced survival
time after lethal challenge with A/Wisconsin/WSLH/34939/09,
with death initiating in vehicle-treated mice on day 6 postinfection compared to day 11 postinfection for RP-002-treated
mice (Figure 3C, p < 0.005). Moreover, RP-002 treatment resulted in significant improvement in overall survival compared
to vehicle-treated mice (20% mortality in RP-002 versus
80% mortality for vehicle, p < 0.005; Figure 3C). These findings
demonstrate a significant biological phenotype resulting from
early S1P1 receptor agonist treatment following pathogenic
influenza virus infection. By showing directly that aborting an
early step in cytokine amplification significantly protects against
an ordinarily lethal human pathogenic H1N1:2009 infection,
the findings expand previous observations of an association
between humans and animal models in which dysregulation of
innate immune responses contributes to morbidity and mortality.
Importantly, treatment with RP-002 ameliorated morbidity and
mortality associated with human pathogenic H1N1:2009 severe
influenza infection. Effects upon survival are in keeping with our
previously published data (Marsolais et al., 2009) that this mechanism does not alter virus-specific neutralizing antibody response, affinity maturation, and class switching. This documents
that the cytokine storm plays a direct and cardinal role in influenza-mediated lung disease. We then sought to delineate the
mechanism of S1P1 suppression of cytokine amplification.
CD69+ NK cells
100
75
50
Vehicle
RP-002
6 mg/kg p.o.
25
0
0
10
12
14
16
Days post-infection
Figure 3. Oral Administration of an S1P1 Agonist Suppresses Early
Innate Cytokine and Chemokine Production and Significantly
Improves Survival to Lethal Infection with H1N1 2009 Swine Influenza Virus
Mice were infected with 2 3 105 PFU A/Wisconsin/WSLH34939/09 and treated
by gavage with either vehicle (water) or RP-002 (6 mg/kg on 1 and 25 hr
postinfection for cytokine and cell recruitment assays at 48 hr, with a third dose
administered at 49 hr for the 16 day survival experiment).
(A) Levels of proinflammatory cytokines and chemokines were analyzed in the
BALF 48 hr postinfection by ELISA.
(B) Total numbers of activated (CD69+) macrophages/monocytes and NK cells
were detected in collagenase-digested lungs 48 hr postinfection by flow
cytometry.
(C) Mice were monitored daily for survival for 16 days postinfection.
For (A) and (B), data represent the average SEM of four to five mice per group,
whereas (C) had ten mice per group. *p < 0.05; **p < 0.005; ***p < 0.0005.
Data are representative of three independent experiments. See also Figure S3.
Lymphatic Endothelium
Vascular Endothelium
Epithelium
4.59
0.95
S1P1-eGFP
expression
% of Max
5.47
S1P1-eGFP
>
B
CD4 T Cells
CD8 T Cells
3.00
B Cells
1.77
S1P1-eGFP
expression
% of Max
3.98
C57BL/6J
S1P1 - eGFP
S1P1-eGFP
>
C
Macrophage/Monocytes
Alveolar Macrophages
0.96
1.20
NK Cells
1.29
1.12
S1P1-eGFP
expression
% of Max
1.44
Neutrophils
Dendritic Cells
S1P1-eGFP
>
Relative S1P1-eGFP
expression
8
Prior to Infection
48-Hours post-infection
6
4
2
lls
ce
ils
N
ph
ro
eu
t
ph
ag
e
ce
lls
M
ac
ro
8
D
C
ce
ce
l
ls
lls
m
el
iu
ith
Ep
ph
m
Ly
Va
sc
ul
ar
at
ic
Cell Types
Figure 4. S1P1 Receptor Is Expressed on Pulmonary Endothelium and Lymphocytes, but Not on Pulmonary Epithelial Cells
(AC) Flow cytometry histograms showing eGFP fluorescence from heterozygous S1P1-eGFP on lung endothelial and epithelial cells. (A) CD4 and CD8 T cells and
B cells. (B) Macrophages, neutrophils, dendritic cells, and NK cells (C) prior to influenza virus infection.
(D) Relative expression of S1P1-eGFP on lung cell populations before or 48 hr after influenza virus infection. The numbers in the upper-right corner of each plot
represent the relative eGFP expression as calculated by dividing the MFI of eGFP on S1P1-eGFP transgenic mice over eGFP MFI of C57BL/6J littermate controls.
Data represent average SD from two to five mice per group. Results are representative of three independent experiments. See also Figure S4.
Inhibition of Cytokine Storm Is Not Due to Lymphocyte

S1P1 Receptor Activation
Lymphocytes and pulmonary endothelium are the only cells
within the lung that express measurable amounts of S1P1-eGFP
as detected by flow cytometry and western blot. It is plausible
that lymphocytes within the lung provide a bystander effect
during the innate immune response to influenza virus infection. To rule out a role for lymphocytes during S1P1 receptormediated inhibition of inflammation following influenza virus infection, lymphocyte-deficient Rag2 / mice were infected and
treated with vehicle or CYM-5442. Administration of CYM5442 to influenza virus-infected Rag2 / mice resulted in significant inhibition of expression of IFN-a, CCL2, IL-6, TNF-a, and
IFN-g (Figure 5A), in addition to CCL5, CXCL2, and CXCL10
(Figure S1B). Atypical lymphocyte populations that could,
in principal, contribute to suppression of cytokine production
did not express S1P1-eGFP and thus also do not contribute to this mechanism. Inflammatory cell infiltrates, including
macrophages/monocytes and NK cells, were significantly
reduced in CYM-5442-treated Rag2 / mice (Figure 5B).
Reduced numbers of neutrophils were also observed within
the infected lung, but the difference was not statistically significant (Figure 5B). Moreover, lung-infiltrating macrophages and
NK cells in Rag2 / mice were less activated, as measured by
CD69 surface expression (Figure S5). CYM-5442 treatment of
influenza virus-infected lymphocyte-deficient mice inhibits the
innate inflammatory response, excluding a role for lymphocytes
as key regulators of influenza virus-induced cytokine storm.
S1P1 Receptor Agonism Suppresses Immune Cell
Recruitment through Downregulation of Chemokine
Production by Lung Endothelial Cells
The strong expression of S1P1-eGFP receptor on pulmonary
endothelium, coupled with the inhibition of cytokine storm in
RAG2 / mice, suggests that S1P1 receptor signaling in lung
endothelial cells suppresses cytokine storm. Endothelial cells
have been shown to produce various cytokines and chemokines
during inflammatory processes and may be a source of cytokines and chemokines in the lung during influenza virus infection.
To evaluate the effects of CYM-5442 treatment on pulmonary
A
pg/mL in the BALF of
RAG2-/- mice
800
Vehicle
CYM-5442
600
400
*
200
***
***
**
0
IFN-
CCL2
IL-6
TNF-
IFN-
Chemokine/Cytokine
B
Number of cells in
6
RAG2 -/- mice (x10 )
and uninfected mice. Endothelial cells, T cells, and B cells were

positive for S1P1-eGFP in uninfected mice, confirming our flow
cytometry results (Figure S4A). However, though S1P1-eGFP
was detected by western blot in B cells from uninfected mice,
we did not detect S1P1-eGFP expression by western Blot in
influenza virus-infected lung-resident B cells (Figure S4A). We
did not detect S1P1-eGFP expression by western blot in epithelial cells, macrophages/monocytes, and alveolar macrophages
either before or after influenza virus infection (Figure S4A). Moreover, CYM-5442 treatment of infected S1P1-eGFP knockin mice
did not diminish expression of S1P1-eGFP on endothelial cells,
T cells, or B cells, demonstrating that administration of this
S1P1-specific agonist does not induce degradation of the S1P1
receptor (Figure S4B). Therefore, CYM-5442 does not exert an
antagonistic effect due to receptor degradation, demonstrating
functional agonism of S1P1 as the mechanism by which CYM5442 suppresses cytokine storm.
Vehicle
CYM-5442
**
*
0
Macrophages/
monocytes
Neutrophils
NK cells
Cell Types
Figure 5. S1P1 Receptor Agonism Suppresses Early Proinflammatory Immune Responses Independently of Lymphocytes
Rag2 / mice were infected and treated with either vehicle or CYM-5442, as in
Figure 1.
(A) Proinflammatory cytokines and chemokines were measured in BALF by
ELISA 48 hr postinfluenza infection.
(B) Total numbers of innate immune cell recruitment were quantified from
collagenase-digested lungs by flow cytometry at 48 hr postinfluenza virus
infection.
Data represent average SEM from five mice per group. *p < 0.05; **p < 0.005;
***p < 0.0005. Results are representative of two independent experiments. See
also Figure S1 and Figure S5.
endothelial cell cytokine and chemokine production, we purified

lung endothelial cell populations and performed mRNA and
protein analyses. Endothelial cells expressed elevated levels of
the chemokines CCL2, CCL5, and CXCL10 at the mRNA level
(Figure 6A). Importantly, treatment with CYM-5442 significantly
reduced mRNA expression of chemokines in lung endothelial
cells compared to vehicle-treated mice early after influenza virus
infection (Figure 6A). Examination of chemokine production in
purified lung vascular and lymphatic endothelial cells revealed
a significant reduction in chemokine production at the protein
level after CYM-5442 treatment (Figure 6B). Analysis of pulmonary endothelial cell integrin expression did not show appreciable differences in mice infected with influenza virus treated
with vehicle compared to CYM-5442 (Figure S6).
Because chemokine presentation on endothelial cells is
crucial for leukocyte activation and extravasation into infected
tissue, we asked whether reduction of chemokine expression
following CYM-5442 treatment resulted in reduced entry of inflammatory cells into the infected lung. For these experiments,
CXCL10 (pg/2.5x105 cells) CCL2 (pg/2.5x105 cells)
50
40
30
20
10
Vehicle
CYM-5442
6
5
4
3
2
1
CCL2
CCL5
Lymphatic
Endothelium
Vascular
Endothelium
Ct on lung endothelial cells
Relative mRNA Expression
CXCL10
25
25
20
20
**
15
15
10
10
800
1200
600
800
400
***
400
200
0
**
VEH
CYM-5442
VEH
CYM-5442
Treatment
Treatment
Total # of cells in the lung (x10 6)
E
C
Treat with CYM-5442

or vehicle
2 hours postinfection
Harvest lung
tissue 2 days
post-infection
and run FACS
48 hours post
infection
LysM-GFP
C57Bl/6
2.5
***
2.0
1.5
Vehicle
CYM -5442
Uninfected Vehicle
Unifected CYM -5442
1.0
***
0.5
0
Granulocytes
Macrophages/
Monocytes
**
VEH
CYM-5442
CYM-5442 + CCL2
***
**
**
Macrophages/
monocytes
***
NK cells
Neutrophils
Cell Types
3000
pg/ml in the BALF
GFP+ Cells/Lung (x104)
VEH
CYM-5442
CYM-5442 + CCL2
2000
1000
400
300
***
**
*
***
*
200
100
0
Figure 6. S1P1 Receptor Agonism Actively Suppresses Recruitment of Innate Immune Cells through Downregulation of Chemokine
Expression on Lung Endothelial Cells
(A) Relative mRNA expression of chemokines from purified lung endothelial cell populations at 36 hr postinfluenza virus infection.
(B) Protein expression of CCL2 and CXCL10 measured by ELISA on FACS-purified vascular and lymphatic lung endothelial cells 48 hr postinfluenza virus
infection.
(C) LysM-GFP mice were infected with 1 3 104 PFU of influenza virus. At 2 days later, bone marrow cells were harvested from infected LysM-GFP mice and
transferred into C57BL/6J mice that were either left uninfected or infected with 1 3 104 PFU of influenza virus 2 hr prior. Mice receiving bone marrow cells from LysMGFP-positive mice were treated with either vehicle or CYM-5442 (2 mg/kg 1, 13, 25, and 37 hr postinfection), and at 48 hr postinfection, total lung cells were harvested
from lung homogenates and the total numbers of GFP-expressing granulocytes (CD11b+, Ly6G+) and macrophages/monocytes (CD11b+F480+) were quantified.
(D) Total number of LysM-GFP-positive granulocytes and macrophages/monocytes per lung of uninfected or infected mice treated with either vehicle or
CYM-5442.
(E) Exogenous intratracheal administration of chemokine restores recruitment of innate immune cells. However, it does not resurrect cytokine/chemokine
production after CYM-5442 treatment. Mice were infected with 1 3 104 PFU WSN influenza virus, and either vehicle (water) or CYM-5442 (2 mg/kg 1, 13, 25, and
37 hr postinfection) were administered to mice i.t. Following administration of CYM-5442, recombinant mouse CCL2 was administered (50 mg) i.t. directly into the
lungs. The bar graph shows the total numbers of macrophages/monocytes, neutrophils, and NK cells in the lung 48 hr postinfluenza virus infection.
(F) Proinflammatory cytokines and chemokines were measured in BALF 48 hr postinfection by ELISA in mice treated with vehicle, CYM-5442, or CYM-5442 +
rmCCL2, as in (E).
Data represent average SEM from five mice/group. *p < 0.05; **p < 0.005; ***p < 0.0005. Results are representative of two independent experiments. See also
Figure S1 and Figure S6.
we used LysM-GFP mice in which macrophages, monocytes,

and neutrophils express green fluorescent protein, allowing us
to track these cell populations after adoptive transfer into congenic C57BL/6J mice. We infected LysM-GFP mice with influenza virus and 48 hr later purified bone marrow cells for adoptive
transfer. 5 3 106 LysM-GFP bone marrow cells were transferred
into influenza virus-infected C57BL/6J mice treated with either
vehicle or CYM-5442 (Figure 6C). The lungs of infected or
uninfected C57BL/6J mice receiving LysM-GFP cells were
then harvested 48 hr postinfection, and the numbers of GFPpositive cells in the lung were quantified. Treatment with CYM5442 resulted in significantly reduced recruitment of both GFPpositive neutrophils (CD11b+ F480 Ly6G+) and macrophages/
monocytes (CD11b+ Ly6G F480+) into the lung (Figure 6D).
Therefore, CYM-5442 treatment inhibits the infiltration of circulating inflammatory cells into the lung.
S1P1 receptor signaling is associated with enhanced capillary
integrity (Rosen et al., 2007), whereby S1P1 receptor antagonists
induce vascular leakage (Rosen et al., 2008; Sanna et al., 2006),
and S1PR-selective agonists protect from vascular leakage
induced by exogenous administration of VEGF (Sanchez et al.,
2003). Though myelomonocytic recruitment is not modulated
by endothelial permeability changes, we formally assessed
whether suppression of inflammatory cell recruitment to the
lung reflected a passive inhibition of leukocyte recruitment or
whether suppression of chemokines on lung endothelial cells
limits innate immune cell recruitment. Rescue experiments
were performed in which mice were infected with influenza in
the presence or absence of intratracheal administration of recombinant murine CCL2 (rmCCL2), with or without CYM-5442
treatment. Administration of rmCCL2 to infected mice treated
with CYM-5442 restored macrophage/monocyte and NK cell
numbers in the lung to levels equivalent to infected mice treated
with vehicle (Figure 6E) 48 hr postinfection. These data demonstrate that CYM-5442 inhibition of cellular infiltration into the
lung is not due to the enhancement of endothelial cell barrier
function but due to suppression of chemokine production.
Therefore, S1P1 receptor agonism suppresses endothelial cell
chemokine expression, resulting in diminished cell infiltration.
Despite the rescue of CCL2 levels in the BALF and the restoration of macrophage/monocyte recruitment into the lung of
CYM-5442-treated mice, global cytokine and chemokine production was not restored (Figure 6F and Figure S1C). Influenza
virus infection of CCR2-deficient mice results in substantial
reductions of infiltrating macrophages/monocytes (Dawson
et al., 2000). To rule out infiltrating macrophages, we infected
CCR2-deficient mice with WSN influenza virus as done above
and treated with either vehicle or CYM-5442. Infection of
CCR2-deficient mice resulted in significantly reduced numbers
of macrophages/monocytes without altering neutrophil or NK
cell numbers in the lung 48 hr postinfection (Figure S7A). Despite
reduced levels of macrophages/monocytes in the lung, we still
detected significant levels of IFN-a, CCL2, CCL3, CCL5,
CXCL2, CXCL10, IL-1a, IL-6, and IFN-g (Figure S7B). More
importantly, treatment of CCR2-deficient mice with CYM-5442
still significantly reduced all cytokines and chemokines tested
at 48 hr postinfection except for CCL5 (Figure S7B). Failure to
inhibit cytokine and chemokine production in CCR2-deficient
mice or to re-establish cytokine and chemokine production by

the induction of macrophage/monocyte cell infiltration through
rmCCL2 administration following CYM-5442 treatment suggests
that monocytes and macrophages are not major sources of early
cytokine and chemokine production. Furthermore, these data
indicate that cell recruitment and cytokine responses may be
uncoupled events.
Proinflammatory Cytokine Responses Are Independent
of Innate Immune Cell Recruitment and Dependent on
Type I Interferon Signaling
Infiltration of macrophages and NK cells alone does not appear
to be associated with cytokine storm. Therefore, we evaluated
the contribution of total inflammatory cell infiltration to cytokine
production during influenza virus infection. Mice were treated
with an anti-CD11b antibody (M7/80) that has been shown to
inhibit the recruitment of CD11b-expressing cells into inflamed
tissues (Rosen and Gordon, 1987). Though treatment with antiCD11b significantly inhibited the recruitment of macrophages/
monocytes, neutrophils, and NK cells into the lung (Figure 7A),
we observed no reduction in levels of proinflammatory cytokines/chemokines, other than IFN-g (Figure 7B), which is probably a direct result of diminished infiltration of NK cells, the
primary source of IFN-g at this time point. Moreover, treatment
with CYM-5442 significantly inhibited the production of IFN-a,
CCL2, IL-6, TNF-a, and IFN-g (Figure 7B), in addition to CCL3,
CCL5, CXCL2, CXCL10, and IL-1a (Figure S1D) in anti-CD11btreated mice. These results further demonstrate that innate cell
recruitment and cytokine production are independent events
early after influenza virus infection, both which are inhibited by
S1P1 receptor agonism of pulmonary endothelial cells. Importantly, our data demonstrate that inflammatory cell infiltration
into the lung is not required for cytokine and chemokine production. An additional line of evidence excluding hematopoietic cells
from S1P1 agonist-mediated reduction of early innate cytokine
and chemokine responses following influenza virus infection
was provided by irradiation resistance of S1P1 agonist suppression. C57BL/6J mice were irradiated with 500 rads and 24 hr later
infected with WSN influenza virus and treated with either vehicle
or CYM-5442. Irradiation reduced CD45+ hematopoietic cells in
lung by 90% (vehicle: 2 3 107 cell/lung versus 500 rads: 2 3 106
cells/lung), and importantly, treatment of irradiated mice with
CYM-5542 significantly reduced the levels of IFN-a (irradiated
vehicle: 86 pg/ml versus irradiated CYM-5442: 17.9 pg/ml,
p = 0.0006), CCL2 (irradiated vehicle: 944 pg/ml versus irradiated
CYM-5442: 452 pg/ml, p = 0.00002), CXCL10 (irradiated vehicle:
204 pg/ml versus irradiated CYM-5442: 83 pg/ml, p = 0.00003),
and IL-6 (irradiated vehicle: 170 pg/ml versus irradiated CYM5442: 70 pg/ml, p = 0.0001), demonstrating that a radiationresistant cell population mediated S1P1 receptor-mediated suppression of the inflammatory response. These results, coupled
with the restricted S1P1 receptor expression pattern, support
an important role for endothelial cells in regulating pulmonary
innate immune responses to influenza virus.
Type I interferons, predominantly IFN-a species, are elevated
early after respiratory virus infection and are thought to be crucial
for the production of proinflammatory cytokines and chemokines. Our results thus far have demonstrated that CYM-5442
Figure 7. Proinflammatory Cytokine Responses Are Independent of Innate Immune Cell Recruitment and Dependent on Type I Interferon
Signaling
(A) Total numbers of innate immune cells were quantified from lung digests by flow cytometry at 48 hr postinfluenza virus infection in mice treated 1 hr postinfection with vehicle or CYM-5442 in the presence of either anti-CD11b or isotype control antibody (M7/80) (0.5 mg/mouse) 0 and 24 hr postinfection.
(B) Proinflammatory cytokines and chemokines were measured 48 hr postinfection by ELISA in BALF in mice treated as in (A).
(C) C57BL/6J or IFN ab receptor-deficient mice were infected with 1 3 104 PFU of influenza virus and treated with either vehicle or CYM-5442 (2 mg/kg 1, 13, 25,
and 36 hr postinfection. Total numbers of innate immune cells were quantified from lung digests by flow cytometry at 48 hr postinfluenza virus infection, and
(D) proinflammatory cytokines and chemokines were measured 48 hr postinfection by ELISA in BALF fluid.
Data represent average SEM from five mice per group. *p < 0.05; **p < 0.005; ***p < 0.0005. Results are representative of two independent experiments. See
also Figure S1 and Figure S7.
treatment consistently inhibits the production of IFN-a in the lung

early after influenza virus infection (Figure 1, Figure 2, Figure 5,
and Figure 6). Therefore, we postulated that blunting IFN-a
production may be a mechanism by which CYM-5442 inhibits
cytokine storm early after influenza virus infection. To address
this, we infected IFNa/b receptor knockout mice with influenza
virus, treated these mice with either vehicle or CYM-5442, and
measured innate cell recruitment and cytokine/chemokine
production 48 hr postinfection. Significant differences were not
observed in inflammatory cell recruitment of macrophages/
monocytes, neutrophils, and NK cells into the lung of IFNa/b
receptor-deficient mice when compared to C57Bl/6J mice (Figure 7C). As seen with infected C57BL/6J mice, CYM-5442
treatment inhibited innate immune cell recruitment in IFNa/b
receptor-deficient mice 48 hr postinfection (Figure 7C). Despite
recruitment of innate immune cells into the lung of IFNa/b
receptor-deficient mice 48 hr postinfection, these mice exhibited
significantly reduced levels of IFN-a, CCL2, IL-6, and IFN-g (Figure 7D) in addition to CCL5 and CXCL10 in the BALF fluid (Figure S1E). Taken together, our data demonstrate that regulation
of cellular recruitment into the pulmonary tissue is mediated by

endothelial cells and is independent of type I interferon signaling.
Moreover, S1P1 receptor agonism of endothelial cells inhibits
IFN-a production and results in the dampening of global proinflammatory cytokine responses.
DISCUSSION
The recruitment of innate immune cells into the lungs combined
with excessive proinflammatory cytokine and chemokine production are hallmarks of influenza virus infection (La Gruta
et al., 2007). Contrary to current dogma, which places lung
epithelium and inflammatory infiltrate at the center of influenza
virus-induced cytokine storm (La Gruta et al., 2007), we demonstrate here a central role for pulmonary endothelium in regulating
this process. Furthermore, we determined that both innate
immune cell recruitment and early innate cytokine and chemokine production are uncoupled events, with endothelial cells
at the center of both processes. The chemical and genetic
approaches shown here have a potentially broad impact on
understanding disease pathogenesis. Early dysregulation of

innate cellular and cytokine responses predict disease severity
and death during highly pathogenic influenza virus infection
(Bermejo-Martin et al., 2009; Cilloniz et al., 2009; de Jong
et al., 2006; Kobasa et al., 2004, 2007). The early induction of
the cytokines IFN-a, TNF-a, IL-1a, and IL-6 and the chemokines
CCL2, CCL3, CXCL2 (IL-8), and CXCL10 are associated with
symptom formation in humans (Hayden et al., 1998; Kaiser
et al., 2001). TNF-a, IL-1, and IL-6 possess multifunctional activities and are associated with morbidity during influenza virus
infection. Chemokines such as CCL2, CCL3, CXCL2, and
CXCL10 induce the recruitment of innate immune cells into the
lung, which can release more cytokines exacerbating cytokine
storm and further damage the lung. We demonstrate that
suppression of early innate immune responses through S1P1
signaling on endothelial cells results in significantly reduced
mortality during infection of mice with a human pathogenic strain
of influenza virus (Figure 3). Importantly, our results identify
a pulmonary cell type, endothelial cells, as potential targets for
suppressing excessive innate inflammatory responses.
The ability of CYM-5442 to diminish the production of IFN-a
early following influenza virus infection is broadly relevant. The
requirement of type I interferon signaling for the early production
of multiple cytokines and chemokines after influenza virus infection is striking. Moreover, production of proinflammatory cytokines requires type I interferon signaling, but the type I interferon
pathway is dispensable for cell recruitment. Though type I interferon signaling is well known to inhibit viral replication (GarcaSastre and Biron, 2006), evidence also points to pathogenic
roles for IFN-a during viral infection. Several proinflammatory
cytokines and chemokines are downstream of type I interferon
receptor signaling. Moreover, disease onset correlates directly
with local respiratory production of IFN-a in humans (Hayden
et al., 1998). Thus, type I interferon signaling may play dual roles
in viral pathogenesis and viral clearance. S1P1 receptor signaling
on pulmonary endothelial cells in vivo blunts but does not abolish
IFN-a production and may explain why pathology is lessened
without compromising the hosts ability to clear the virus.
The ability of S1P1 receptor activation to suppress cytokine
and chemokine production in the lung in irradiated mice, coupled
to the pattern of S1P1 receptor expression in the lung, strongly
indicates a central endothelial component to early innate immune responses. An interesting question is whether lung lymphatic, vascular, or both endothelial cell populations are directly
responsible for the observed immune modulation. Moreover, of
the large receptor reserve in blood, endothelium ensures that
S1P1 receptor expression is maintained on both basolateral
and luminal surfaces of the targeted endothelial cell populations
(Cahalan et al., 2011) and retains a responsiveness to agonist.
Additional approaches to separate blood and lymphatic contributions are still needed. Though we know that endothelial cells
produce chemokines, endothelial-mediated regulation of cytokine production in the lung still needs to be determined. It is
possible that endothelial cells may regulate cytokine production
in the lung through a complex crosstalk mechanism with lung
epithelium or resident hematopoietic cells. Nevertheless, the
identification of endothelial cells as central orchestrators of early
innate immune-mediated inflammation is of fundamental interest
and has broad implications for treating multiple diseases. The

contribution of aberrant proinflammatory cytokine and chemokine production to pathogenesis has been reported for viral
and bacterial diseases, including HIV (Stacey et al., 2009), Hantavirus (Borges et al., 2006), severe acute respiratory syndrome
(SARS) (Thiel and Weber, 2008), and pneumococcal bacterial
pneumonia (Bergeron et al., 1998). Further, the etiology of several autoimmune conditions has been directly associated with
excessive innate immune responses (Kawane et al., 2010; Link,
1998). Thus, understanding the cellular pathways that regulate
cytokine storm and developing appropriate chemical signaling
tools to identify pathophysiological points of control not only
provide insight into microbial-host interactions, but also may ultimately reveal additional approaches to achieve effective immunotherapy in multiple diseases. In addition, our results present a
nonlymphopenic mechanism by which sphingosine analog treatment suppresses pathogenic immune responses. Lastly, these
data suggest that endogenous S1P acting on endothelial S1P1
receptor could be a negative regulator of cytokine amplification
and raise the possibility that heterogeneities in S1P metabolism
between individuals could contribute to the advantages or
disadvantages conferred by genetic individuality to host survival
in a number of diseases.
Mice, Virus, Compounds, and Reagents
Six- to eight-week-old C57Bl/6 male mice and S1P1-eGFP mice described
elsewhere (Cahalan et al., 2011) were bred and maintained in a closed
breeding facility at The Scripps Research Institute. Influenza A/WSN/33
(WSN; H1N1) and the human H1N1 2009 isolate A/Wisconsin/WSLH34939/
09 (a kind gift from Yoshihiro Kawaoka, University of Madison, WI) were amplified and plaqued on Madin-Darby Canine Kidney (MDCK) cells. Mice were infected intratracheally (i.t.) with 1 3 104 PFU of influenza A/WSN/33 virus or
intranasally with 1 3 105 PFU of A/Wisconsin/WSLH34939/09 under isoflurane
anesthesia. At 1 hr postinfection, mice were anesthetized by isoflurane inhalation for i.t. delivery of vehicle (100 ul of water), AAL-R (0.2 mg/kg dissolved in
water), CYM-5442 (2 mg/kg dissolved in water), or RP-002 (3 mg/kg i.t. or
6 mg/kg orally dissolved in water). Multiple doses of compound were administered at the specific times listed in the figurte legends. (AAL-R and CYM5442 were synthesized according to published methods (Jo et al., 2005).
RP-002, (R)-2-(4-(5-(3-cyano-4-isopropoxyphenyl)-l,2,4-oxadiazol-3-yl)-2,3dihydro-lH-inden-l-ylamino)-N,N-dimethylacetamide hydrochloride, was synthesized according to the published method (Martinborough et al., 2011).
The compound had an EC50 for S1P1 of 0.13 nM, was > 100-fold selective
versus S1P5, and was 10,000-fold selective versus S1P2, 3, and 4 respectively, when assayed as described (Cahalan et al., 2011). Recombinant murine
CCL2 (rmCCL2) was purchased from Shenandoah Biotechnology Inc.
(Warwick, PA).
Cytokine and Chemokine Analysis
The trachea of euthanized mice was exposed, transected, and incubated with
a blunt 18 gauge needle. One milliliter of phosphate-buffered saline supplemented with Complete Mini, EDTA-free Protease Inhibitor Cocktail (Roche)
was infused and recovered four times. The recovered bronchoalveolar lavage
fluid was spun at 3000 3 g for 3 min at 4 C and stored at 80 C until use. Multiplex ELISA was performed on supernatant by Quansys Biosciences (Logan,
UT) to detect IL-1a, IL-1b, IL-2, IL-3, IL-4, IL-5, IL-6, IL-9, IL-10, IL-12p70,
TNF-a, MIP-1a, MCP-1, GM-CSF, IFN-g, and RANTES. ELISAs were also performed using CCL2 (MCP-1), CCL5 (RANTES), CXCL10 (IP-10), IL-1a, IL-6,
TNF-a, and IFN-g Duoset kits (R&D systems) as well as the VeriKine Mouse
Interferon-Alpha and Interferon-Beta ELISA Kits (Pestka Biomedical Laboratories, Inc). For the quantification of mRNA chemokine expression, lung
endothelial cells were FACS purified to > 90%95% purity, and mRNA was
purified using the RNeasy mini kit from QIAGEN. Prior to real-time PCR,
genomic DNA was digested and cDNA was made using the RT2 First Strand
Kit (C-03) according to manufacturers instructions (SA Biosciences, Frederick, MD). cDNA (200 ng) was added to individual wells and quantified using
the RT2 profiler PCR Array according to the manufacturers instructions (SA
Biosciences, Frederick, MD).
Cellular Analysis by Flow Cytometry
Lungs were harvested from PBS-perfused mice and mechanically diced into
small tissue pieces using surgical scissors. Diced lungs were suspended in
4 ml of CDTI buffer (0.5 mg/ml collagenase from Clostridium histolyticum
type IV [Sigma], 0.1 mg/ml Dnase I from bovine pancreas grade II [Roche],
1 mg/ml trypsin inhibitor type Ii-s [Sigma] in DMEM) for 1 hr at 37 C. Lung
was then disrupted mechanically through a 100 mm filter, and red blood cells
were lysed using red blood cell lysis buffer (0.02 Tris-HCL and 0.14 NH4Cl).
Inflammatory cells were purified by centrifugation in 35% PBS-buffered Percoll
(GE Healthcare Life Sciences) at 1,500 rpm for 15 min. Cell pellets were resuspended in staining buffer, and Fc receptors were blocked using 25 mg/ml antimouse CD16/32 (BD Biosciences). Cells were stained with the following
anti-mouse antibodies: AlexaFluor 488-conjugated gp38 (eBioscience; clone
eBio8.1.1), PE-conjugated (BioLegend, Inc.; clone ME13.3) and APCconjugated (eBioscience; clone 390) CD31, PE-Cy7-conjugated EpCAM
(BioLegend, Inc.; clone 68.8), Pacific blue-conjugated CD45.2 (BioLegend,
Inc.; clone 104), PerCP-Cy5.5-conjugated NK1.1 (BD Biosciences; clone
PK136), PE-Cy7-conjugated CD3e (eBioscience; clone 145-2C11), e450conjugated CD4 (eBioscience; clone L3T4), PE-conjugated CD8a (BD Biosciences; clone 53-6.1), Pacific blue-conjugated B220 (BD Biosciences; clone
RA3-6B2), PE-conjugated CD19 (BD Biosciences, clone 1D3), PE-Cy7conjugated CD11b (eBiosciences; clone M1/70), PerCP-Cy5.5-conjugated
CD11c (eBiosciences; clone N418), APC-conjugated Gr-1 (BD Biosciences;
clone RB6-8C5), Pacific blue and PE-conjugated Ly6G (BD Biosciences; clone
IA8), APC-conjugated F480 (eBioscience; clone BM8), PE-conjugated 7/4
(AbD Serotec; clone 7/4), PE-conjugated I/A-I/E (BD Biosciences; clone M5/
114.15.2), PE-Cy7 conjugated CD205 (eBiosciences; clone 205yekta), Fitcconjugated CD69 (BD Biosciences; clone H1.2F3), and APC-conjugated
CD25 (eBiosciences; clone PC61.5). Flow cytometry acquisition was performed with BD FACSDiva-driven BD LSR II flow cytometer (Becton, Dickinson
and Company). Data were then analyzed with FlowJo software (Treestar Inc.).
Western Blot
FACS-purified lung cell populations (1 3 105 cells) using the antibodies
described above were homogenized in RIPA buffer supplemented with
protease inhibitors (Pierce). Lysates were centrifuged at 50,000 3 g for
30 min, and the protein concentration in the supernatant was determined by
BCA assay (Pierce). Equal amounts of protein from cell lystates were loaded
in nondenaturing conditions and separated by SDS-PAGE in 4%12%
NuPAGE (Novex) Bis-Tris gels. Gels were transferred to PVDF membranes
followed by probing for GFP using an anti-GFP antibody (Abcam) and an
anti-rabbit Ig light-chain HRP secondary (ELC Biosciences) using chemiluminescence autoradiography.
Supplemental Information includes seven figures and can be found with this
article online at doi:10.1016/j.cell.2011.08.015.
ACKNOWLEDGMENTS
This is Publication Number 21112 from the Department of Immunology and
Microbial Science and the Department of Chemical Physiology and The
Scripps Research Institute Molecular Screening Center, The Scripps Research
Institute (TSRI). This work was supported, in part, by USPHS grants AI074564
(M.B.A.O., H.R., K.B.W., and J.R.T.), AI009484 (M.B.A.O.), AI05509 (H.R.),
MH084512 (H.R.), and NIH training grants NS041219 (K.B.W.), AI007244
(K.B.W.), and AI007364 (J.R.T). We thank Marcus Boehm, Li-ming Huang,
and Bryan Clemons (Receptos, Inc.) for helping provide RP-002 as a chemical
tool. Hugh Rosen is a founder of Receptos. Edward Roberts is a consultant

to Receptos. Fiona Scott, Esther Martinborough, and Robert Peach are
employees of Receptos.
Received: February 23, 2011
Revised: March 27, 2011
REFERENCES
Bergeron, Y., Ouellet, N., Deslauriers, A.M., Simard, M., Olivier, M., and
Bergeron, M.G. (1998). Cytokine kinetics and other host factors in response
to pneumococcal pulmonary infection in mice. Infect. Immun. 66, 912922.
Bermejo-Martin, J.F., Ortiz de Lejarazu, R., Pumarola, T., Rello, J., Almansa,
R., Ramrez, P., Martin-Loeches, I., Varillas, D., Gallegos, M.C., Seron, C.,
et al. (2009). Th1 and Th17 hypercytokinemia as early host response signature
in severe pandemic influenza. Crit. Care 13, R201.
Borges, A.A., Campos, G.M., Moreli, M.L., Souza, R.L., Aquino, V.H.,
Saggioro, F.P., and Figueiredo, L.T. (2006). Hantavirus cardiopulmonary
syndrome: immune response and pathogenesis. Microbes Infect. 8, 2324
2330.
Brinkmann, V., Billich, A., Baumruker, T., Heining, P., Schmouder, R., Francis,
G., Aradhye, S., and Burtin, P. (2010). Fingolimod (FTY720): discovery and
development of an oral drug to treat multiple sclerosis. Nat. Rev. Drug Discov.
9, 883897.
Cahalan, S.M., Gonzalez-Cabrera, P.J., Sarkisyan, G., Nguyen, N., Schaeffer,
M.T., Huang, L., Yeager, A., Clemons, B., Scott, F., and Rosen, H. (2011).
Actions of a picomolar short-acting S1P1 agonist in S1P1-eGFP knock-in
mice. Nat. Chem. Biol. 7, 254256.
Cilloniz, C., Shinya, K., Peng, X., Korth, M.J., Proll, S.C., Aicher, L.D., Carter,
V.S., Chang, J.H., Kobasa, D., Feldmann, F., et al. (2009). Lethal influenza virus
infection in macaques is associated with early dysregulation of inflammatory
related genes. PLoS Pathog. 5, e1000604.
Dawson, T.C., Beck, M.A., Kuziel, W.A., Henderson, F., and Maeda, N. (2000).
Contrasting effects of CCR5 and CCR2 deficiency in the pulmonary inflammatory response to influenza A virus. Am. J. Pathol. 156, 19511959.
de Jong, M.D., Simmons, C.P., Thanh, T.T., Hien, V.M., Smith, G.J., Chau,
T.N., Hoang, D.M., Chau, N.V., Khanh, T.H., Dong, V.C., et al. (2006). Fatal
outcome of human influenza A (H5N1) is associated with high viral load and
hypercytokinemia. Nat. Med. 12, 12031207.
Garca-Sastre, A. (2010). Influenza virus receptor specificity: disease and
transmission. Am. J. Pathol. 176, 15841585.
Garca-Sastre, A., and Biron, C.A. (2006). Type 1 interferons and the virus-host
relationship: a lesson in detente. Science 312, 879882.
Gonzalez-Cabrera, P.J., Jo, E., Sanna, M.G., Brown, S., Leaf, N., Marsolais,
D., Schaeffer, M.T., Chapman, J., Cameron, M., Guerrero, M., et al. (2008).
Full pharmacological efficacy of a novel S1P1 agonist that does not require
S1P-like headgroup interactions. Mol. Pharmacol. 74, 13081318.
Hayden, F.G., Fritz, R., Lobo, M.C., Alvord, W., Strober, W., and Straus, S.E.
(1998). Local and systemic cytokine responses during experimental human
influenza A virus infection. Relation to symptom formation and host defense.
J. Clin. Invest. 101, 643649.
Itoh, Y., Shinya, K., Kiso, M., Watanabe, T., Sakoda, Y., Hatta, M., Muramoto,
Y., Tamura, D., Sakai-Tagawa, Y., Noda, T., et al. (2009). In vitro and in vivo
characterization of new swine-origin H1N1 influenza viruses. Nature 460,
10211025.
Jo, E., Sanna, M.G., Gonzalez-Cabrera, P.J., Thangada, S., Tigyi, G., Osborne,
D.A., Hla, T., Parrill, A.L., and Rosen, H. (2005). S1P1-selective in vivo-active
agonists from high-throughput screening: off-the-shelf chemical probes of
receptor interactions, signaling, and fate. Chem. Biol. 12, 703715.
Kaiser, L., Fritz, R.S., Straus, S.E., Gubareva, L., and Hayden, F.G. (2001).
Symptom pathogenesis during acute influenza: interleukin-6 and other cytokine responses. J. Med. Virol. 64, 262268.
Kawane, K., Tanaka, H., Kitahara, Y., Shimaoka, S., and Nagata, S. (2010).
Cytokine-dependent but acquired immunity-independent arthritis caused by
DNA escaped from degradation. Proc. Natl. Acad. Sci. USA 107, 19432
19437.
Kobasa, D., Jones, S.M., Shinya, K., Kash, J.C., Copps, J., Ebihara, H., Hatta,
Y., Kim, J.H., Halfmann, P., Hatta, M., et al. (2007). Aberrant innate immune
response in lethal infection of macaques with the 1918 influenza virus. Nature
445, 319323.
Kobasa, D., Takada, A., Shinya, K., Hatta, M., Halfmann, P., Theriault, S.,
Suzuki, H., Nishimura, H., Mitamura, K., Sugaya, N., et al. (2004). Enhanced
virulence of influenza A viruses with the haemagglutinin of the 1918 pandemic
virus. Nature 431, 703707.
La Gruta, N.L., Kedzierska, K., Stambas, J., and Doherty, P.C. (2007).
A question of self-preservation: immunopathology in influenza virus infection.
Immunol. Cell Biol. 85, 8592.
Link, H. (1998). The cytokine storm in multiple sclerosis. Mult. Scler. 4, 1215.
Marsolais, D., Hahm, B., Edelmann, K.H., Walsh, K.B., Guerrero, M., Hatta, Y.,
Kawaoka, Y., Roberts, E., Oldstone, M.B., and Rosen, H. (2008). Local not
systemic modulation of dendritic cell S1P receptors in lung blunts virusspecific immune responses to influenza. Mol. Pharmacol. 74, 896903.
Marsolais, D., Hahm, B., Walsh, K.B., Edelmann, K.H., McGavern, D., Hatta,
Y., Kawaoka, Y., Rosen, H., and Oldstone, M.B. (2009). A critical role for the
sphingosine analog AAL-R in dampening the cytokine response during
influenza virus infection. Proc. Natl. Acad. Sci. USA 106, 15601565.
Rosen, H., and Gordon, S. (1987). Monoclonal antibody to the murine type 3
complement receptor inhibits adhesion of myelomonocytic cells in vitro and
inflammatory cell recruitment in vivo. J. Exp. Med. 166, 16851701.
Rosen, H., and Liao, J. (2003). Sphingosine 1-phosphate pathway therapeutics: a lipid ligand-receptor paradigm. Curr. Opin. Chem. Biol. 7, 461468.
Rosen, H., Sanna, M.G., Cahalan, S.M., and Gonzalez-Cabrera, P.J. (2007).
Tipping the gatekeeper: S1P regulation of endothelial barrier function. Trends
Immunol. 28, 102107.
Rosen, H., Gonzalez-Cabrera, P., Marsolais, D., Cahalan, S., Don, A.S., and
Sanna, M.G. (2008). Modulating tone: the overture of S1P receptor immunotherapeutics. Immunol. Rev. 223, 221235.
Rosen, H., Gonzalez-Cabrera, P.J., Sanna, M.G., and Brown, S. (2009). Sphingosine 1-phosphate receptor signaling. Annu. Rev. Biochem. 78, 743768.
Sanchez, T., Estrada-Hernandez, T., Paik, J.H., Wu, M.T., Venkataraman, K.,
Brinkmann, V., Claffey, K., and Hla, T. (2003). Phosphorylation and action of
the immunomodulator FTY720 inhibits vascular endothelial cell growth
factor-induced vascular permeability. J. Biol. Chem. 278, 4728147290.
Sanna, M.G., Wang, S.K., Gonzalez-Cabrera, P.J., Don, A., Marsolais, D.,
Matheu, M.P., Wei, S.H., Parker, I., Jo, E., Cheng, W.C., et al. (2006). Enhancement of capillary leakage and restoration of lymphocyte egress by a chiral
S1P1 antagonist in vivo. Nat. Chem. Biol. 2, 434441.
Martinborough, E., Boehm, M., Yeager, A., Yamiyo, J., Huang, L., Brahmachary, E., Moorjani, M., Timony, G., Brooks, J., Peach, R., et al. (2011). Selective
sphingosine 1-phosphate receptor modulators and methods of chiral
synthesis, World International Patent Organization: WO/2011/060392 p. 92.
Stacey, A.R., Norris, P.J., Qin, L., Haygreen, E.A., Taylor, E., Heitman, J.,
Lebedeva, M., DeCamp, A., Li, D., Grove, D., et al. (2009). Induction of a striking
systemic cytokine cascade prior to peak viremia in acute human immunodeficiency virus type 1 infection, in contrast to more modest and delayed
responses in acute hepatitis B and C virus infections. J. Virol. 83, 37193733.
Openshaw, P.J., and Dunning, J. (2010). Influenza vaccination: lessons

learned from the pandemic (H1N1) 2009 influenza outbreak. Mucosal Immunol. 3, 422424.
Thiel, V., and Weber, F. (2008). Interferon and cytokine responses to SARScoronavirus infection. Cytokine Growth Factor Rev. 19, 121132.
Rivera, J., Proia, R.L., and Olivera, A. (2008). The alliance of sphingosine1-phosphate and its receptors in immunity. Nat. Rev. Immunol. 8, 753763.
Tscherne, D.M., and Garca-Sastre, A. (2011). Virulence determinants of

pandemic influenza viruses. J. Clin. Invest. 121, 613.
Hunger States Switch a Flip-Flop

Memory Circuit via a Synaptic
AMPK-Dependent Positive Feedback Loop
Yunlei Yang,1 Deniz Atasoy,1 Helen H. Su,1 and Scott M. Sternson1,*
1Janelia Farm Research Campus, Howard Hughes Medical Institute, 19700 Helix Drive, Ashburn, VA 20147, USA
*Correspondence: sternsons@janelia.hhmi.org
DOI 10.1016/j.cell.2011.07.039
SUMMARY
Synaptic plasticity in response to changes in physiologic state is coordinated by hormonal signals across
multiple neuronal cell types. Here, we combine celltype-specific electrophysiological, pharmacological,
and optogenetic techniques to dissect neural circuits
and molecular pathways controlling synaptic plasticity onto AGRP neurons, a population that regulates
feeding. We find that food deprivation elevates
excitatory synaptic input, which is mediated by a
presynaptic positive feedback loop involving AMPactivated protein kinase. Potentiation of glutamate
release was triggered by the orexigenic hormone
ghrelin and exhibited hysteresis, persisting for hours
after ghrelin removal. Persistent activity was reversed by the anorexigenic hormone leptin, and optogenetic photostimulation demonstrated involvement of opioid release from POMC neurons. Based
on these experiments, we propose a memory storage
device for physiological state constructed from
bistable synapses that are flipped between two
sustained activity states by transient exposure to
hormones signaling energy levels.
INTRODUCTION
Neurons that express Agouti-related protein (Agrp) are a molecularly defined population localized in the hypothalamic arcuate
nucleus, and increasing their electrical activity is sufficient to
rapidly induce voracious feeding behavior (Aponte et al.,
2011). They are intermingled with a separate and functionally
opposed population that is delineated by expression of Proopiomelanocortin (Pomc), which inhibits feeding (Aponte et al.,
2011). These neurons form a core circuit regulating food intake
and energy expenditure in which AGRP neurons inhibit POMC
neurons (Cowley et al., 2001). Both populations behave as interoceptive sensory neurons, modulating their electrical activity
in response to hormonal signals of metabolic state. In addition,
AGRP and POMC neurons receive synaptic inputs that are also
sensitive to the adipocyte-derived anorexigenic hormone leptin
(Pinto et al., 2004). Thus, there is a close interaction between

synaptic and hormonal regulation of these neurons.
Nevertheless, the influence of synaptic input on the activity of
AGRP neurons is not known. In light of the capacity of AGRP
neurons to stimulate feeding (Aponte et al., 2011), the relationship between excitatory synaptic input and neuronal activity is
expected to be behaviorally important. Synaptic regulation of
neuron activity is also of general interest in physiological circuits
because synapses have a potential memory capacity through
synaptic plasticity (Gordon and Bains, 2006). Here, we investigate these issues by first mapping a signaling pathway for
synaptic plasticity at AGRP neurons and then selectively manipulating this pathway to test its influence on AGRP neuron electrical activity. The properties of this pathway endow this circuit
with a memory mechanism.
We find that food deprivation results in persistent upregulation
of excitatory synaptic input to AGRP neurons, which is also
sufficient to increase their firing rate. We further examined this
persistent synaptic activity and identified presynaptic signaling
pathways and neural circuits that are responsible for its induction, maintenance, and offset. Plasticity was induced by the
gut-derived orexigenic hormone ghrelin, which triggered a positive feedback loop dependent on AMP-activated kinase (AMPK)
that maintained elevated synaptic activity even after ghrelin
was removed. Conversely, persistent activity was switched
off by leptin, acting through a previously undescribed opioid
receptor-dependent pathway, which optogenetic experiments
showed could be mediated by opioid release from POMC
neurons. Based on these experiments, we propose a memory
device for physiological state constructed from bistable synapses that are flipped between sustained high- and low-activity
states by transient exposure to hormones associated with
energy deficit and surfeit, respectively.
RESULTS
Deprivation-Induced Synaptic Plasticity
To investigate a potential role for synaptic plasticity in hunger,
synaptic inputs onto AGRP neurons were compared between
food-deprived and ad libitum fed mice during the early light
period, when mice consume little. AGRP neurons were identified for whole-cell patch clamp recordings in brain slices as
the subpopulation of neuropeptide Y (NPY)-expressing neurons
Figure 1. Deprivation-Induced Synaptic Plasticity in AGRP and POMC Neurons

(A) mEPSCs from fed and food-deprived (dep) mice.
(B) fmEPSC in AGRP neurons from deprived mice in the light period or from fed mice at the transition to the dark period (DP) were both significantly increased over
fmEPSC from fed mice in the light period.
(C) Fluorescence micrograph of the arcuate nucleus from a Pomc-topazFP, Npy-sapphireFP double-transgenic mouse. POMC neurons (green) and AGRP
neurons (blue) are intermingled.
(D) In POMC neurons, fmEPSC is decreased by food deprivation.
(E) Elevated spontaneous firing rate in AGRP neurons from deprived mice (n = 11) is reduced to the level of fed mice (n = 12) by CNQX.
***p < 0.001. Data are represented as mean SEM. See also Figure S1.
(Hahn et al., 1998) located in the arcuate nucleus of NpysapphireFP transgenic mice (Pinto et al., 2004). Because two
measures of postsynaptic plasticity, AMPA-R/NMDA-R synaptic current ratio (Figure S1A available online) and the rectification of glutamatergic synaptic currents (Figure S1B), did not
show significant differences between fed and food-deprived
mice, we focused on presynaptic plasticity.
We examined presynaptic function by measuring the frequency of miniature excitatory postsynaptic currents (fmEPSC).
mEPSCs reflect spontaneous neurotransmitter release from
the presynaptic terminal (Fatt and Katz, 1952) under pharmacologic conditions that block action potentials and, by extension,
network activity (tetrodotoxin, TTX, 1 mM). For AGRP neurons,
we found that fmEPSC was approximately doubled after deprivation or when measured at the beginning of the dark period (DP)
in fed mice (fed: 1.4 0.1 s 1, n = 62; deprived: 3.0 0.2 s 1,
n = 62; fed(DP): 3.1 0.3 s 1, n = 26; F2,147 = 30.5, p < 0.001;
Figures 1A and 1B), whereas mEPSC amplitudes were not significantly different (F2,145 = 2.2, p = 0.11; Figure S1C). Thus, in
AGRP neurons, fmEPSC increased in association with the animals
tendency to consume food, either after food deprivation or at the
start of the dark period. Furthermore, deprivation-induced
synaptic upregulation was cell type selective as POMC neurons,
which are intermingled with AGRP neurons (Figure 1C) but
are functionally opposed, experienced marked reduction in
fmEPSC with deprivation (fed: 3.3 0.3 s 1, n = 30; deprived:
1.8 0.2 s 1, n = 22; unpaired t test, p < 0.001; Figure 1D),
consistent with previous experiments measuring synaptic inputs
to these neurons (Sternson et al., 2005).
Activation of AGRP neurons increases feeding behavior
(Aponte et al., 2011), so we considered the impact of elevated
spontaneous synaptic input on AGRP neuron firing rate.
Because neuron firing is measured in the absence of TTX, we
first recorded the frequency of spontaneous excitatory synaptic
currents under this condition, which showed a significant deprivation-induced increase (fed: 2.4 0.3 s 1, n = 9; deprived: 5.4
0.7 s 1, n = 10; unpaired t test, p = 0.009). In line with elevated
excitatory synaptic input, AGRP neurons from deprived mice
also have a higher spontaneous firing rate (fAP) than neurons
from fed mice. This difference was eliminated by blocking excitatory synaptic activity with the glutamate receptor antagonist
CNQX (2 mM) (CNQX: F1,21 = 15.8, p < 0.001; fed/dep: F1,21 =
3.2, p = 0.09; interaction: F1,21 = 6.2, p = 0.02; Figure 1E). A
previous report showed that AGRP neuron firing rate was
elevated after food deprivation in the absence of all synaptic
input (Takahashi and Cone, 2005). For pharmacological conditions similar to those used in this prior study, we find comparable fAP in slices from deprived mice (fAP, CNQX/picrotoxin:
1.2 0.4 s 1, n = 10); however, our results indicate that in the
presence of synaptic inhibition, a more physiologically relevant
condition, excitatory synaptic upregulation is necessary for
increased electrical activity. Together with AGRP neuron photostimulation-evoked feeding (Aponte et al., 2011), these experiments suggest a relationship between synaptic activity, AGRP
neuron firing, and feeding behavior. Because of the potential
behavioral importance of this synaptic control point, we further
investigated synaptic regulation of these neurons.
Synaptic Upregulation Requires Ryanodine-Sensitive
Calcium Stores
To probe the mechanism of synaptic upregulation at AGRP
neurons, we first manipulated intracellular Ca2+ buffering in brain
slices with the membrane-permeable Ca2+ buffer, BAPTA-AM
(25 mM). This treatment reduced fmEPSC in slices from deprived
mice to the level observed in BAPTA-AM-treated slices from
fed mice (fed: 1.1 0.1 s 1, n = 24; deprived: 1.4 0.1 s 1,
n = 27; Figure 2A, left). These data implicate Ca2+-dependent
processes in synaptic plasticity observed after food deprivation.
One potential pathway for Ca2+ is through voltage-gated Ca2+
channels (VGCCs). Blockade of VGCCs with CdCl2 (200 mM)
reduced fmEPSC in both fed and deprived mice (fed: 0.7
0.1 s 1, n = 26; deprived: 1.8 0.3 s 1, n = 21; Figure 2A, right);
however, fmEPSC in AGRP neurons from deprived mice was still
significantly greater than that observed in fed mice. Thus,
VGCCs contribute to fmEPSC in AGRP neurons, but they are not
required for deprivation-induced synaptic upregulation. Moreover, in the presence of CdCl2, fmEPSC in AGRP neurons from
deprived mice was still reduced by BAPTA-AM to the level
Figure 2. Role of Calcium and Ghrelin in Synaptic

Plasticity at AGRP Neurons
(A) Comparison of fmEPSC in AGRP neurons from fed and
food-deprived (dep) mice with no treatment (nt, data from
Figure 1B) or treated with BAPTA-AM, ryanodine, or
ghrelin (left panel). The right panel shows treatments in the
presence of CdCl2, which blocks VGCCs. All pairwise
interactions were tested and p values were corrected with
Holms method. Significant differences are denoted by
any interaction across the red dashed line; interactions on
the same side of the line are not significant (p > 0.05). Left
and right panels were analyzed separately.
(B) Caffeine increases fmEPSC (n = 8), which is blocked by
ryanodine pretreatment (red, n = 3).
(C) Time course of ghrelin-mediated fmEPSC increase in an
AGRP neuron.
(D) Ghrelin increases fmEPSC in AGRP neurons (n = 6),
which is blocked by D-Lys3-GHRP6 (blue, n = 5) or ryanodine (red, n = 7).
(E) For fed mice, ghrelin injection increased fmEPSC relative
to saline treatment.
(F) i.c.v. D-Lys3-GHRP6 (blue) during deprivation blocked
the fmEPSC increase observed with i.c.v. saline (black).
***p < 0.001. Data are represented as mean SEM.
observed in slices from fed mice (0.6 0.2 s 1, n = 11; Figure 2A,
right), which indicated that a different source of Ca2+ was
responsible for elevated fmEPSC in deprived mice.
We next perturbed Ca2+ release from internal stores by blocking ryanodine receptors (RyR) with ryanodine (10 mM). Under
these conditions, fmEPSC in AGRP neurons from deprived mice
was reduced to the level observed for fed mice (fed: 1.5
0.1 s 1, n = 26; deprived: 1.6 0.1 s 1, n = 21; Figure 2A, left),
and this was also the case in the presence of CdCl2 (deprived:
0.8 0.1 s 1, n = 10; Figure 2A, right). This indicates that internal
stores serve as a source of Ca2+ required for elevated fmEPSC in
deprived mice. In addition, caffeine (10 mM), which activates
RyR-mediated Ca2+ release, increased fmEPSC in AGRP neurons
from fed mice to the level observed after deprivation, and this
was blocked by ryanodine pretreatment (Figure 2B). It has
been shown previously that increased spontaneous neurotransmitter release is associated with Ca2+ release from internal
stores in the presynaptic terminal (Emptage et al., 2001). Our
experiments are consistent with a role for Ca2+ release from
internal stores in deprivation-induced plasticity.
we tested the role of the gut-derived orexigenic

hormone ghrelin, which is elevated in circulation
after food deprivation (Tschop et al., 2000).
We found that ghrelin treatment (30 nM) of brain
slices from fed mice (n = 23) increased fmEPSC
to levels similar to those observed in fooddeprived mice (n = 17, Figure 2A). This synaptic
upregulation was rapid (Figures 2C and 2D)
and was blocked by pretreatment with a ghrelin
receptor (Ghsr1) antagonist, D-Lys3-GHRP6 (100 mM) (Howard
et al., 1996) (Figure 2D). Moreover, pretreatment with ryanodine
also blocked the ghrelin-mediated increase in fmEPSC (Figure 2D),
consistent with our proposal that RyR-mediated Ca2+ release
is needed for this pathway.
Next, we tested the necessity and sufficiency of ghrelin for
deprivation-induced synaptic plasticity. We found that the
upregulation of fmEPSC could also be induced by intraperitoneal
(i.p.) administration of ghrelin (1 mg/g) to ad libitum fed mice in
the early light period, only 30 min before preparing brain slices
(fed/saline: 1.4 0.1 s 1, n = 20; fed/ghrelin: 2.8 0.3 s 1,
n = 24; unpaired t test, p < 0.001; Figure 2E). In addition, the
deprivation-induced increase of fmEPSC onto AGRP neurons
was blocked by intracerebroventricular (i.c.v.) administration of
D-Lys3-GHRP6 (0.35 nmol, 3 times during 24 hr deprivation
period; saline: 2.5 0.5 s 1, n = 8; D-Lys3-GHRP6: 1.2
0.1 s 1, n = 14; unpaired t test, p = 0.003; Figure 2F). Together,
these results are consistent with a key role for ghrelin and
Ghsr1 signaling for upregulation of synaptic activity in AGRP
neurons after food deprivation.
Ghrelin Mediates Deprivation-Induced Synaptic

Plasticity
Synaptic plasticity was associated with conditions that increase
food intake, such as deprivation or dark-period onset. Therefore,
Presynaptic AMPK Signaling Mediates Synaptic

Upregulation and AGRP Neuron Activation
Ghrelin-evoked feeding requires both AMPK signaling (Andrews
et al., 2008; Lopez et al., 2008) and AGRP neurons (Luquet et al.,
2007). Furthermore, we showed that, in brain slices, excitatory

synaptic input is necessary for elevated firing in AGRP neurons
after food deprivation. Thus, the influence of ghrelin on presynaptic activity suggested the possibilities that (1) AMPK regulates
presynaptic activity and (2) presynaptic upregulation of glutamate release activates AGRP neuron firing. To test this, we
used pharmacological activation and inhibition of AMPK
signaling to examine the role of AMPK for ghrelin- and deprivation-induced synaptic activation and its impact on electrical
activity in AGRP neurons.
A small-molecule activator of AMPK, AICAR, is transported
into cells and phosphorylated to generate ZMP (Corton et al.,
1995), a mimetic of adenosine monophosphate, which allosterically activates AMPK. AICAR (500 mM) increased fmEPSC in AGRP
neurons in brain slices from fed but not food-deprived mice
(AICAR: F1,14 = 8.5, p = 0.011; fed/dep: F1,14 = 1.7, p = 0.21;
interaction: F1,14 = 14.8, p = 0.002; Figure 3A, dose response
in Figure S2A). Conversely, the AMPK antagonist, Compound
C (Cpd C, 150 mM), reduced fmEPSC in brain slices from deprived
but not fed mice and also blocked subsequent ghrelin-mediated
activation of fmEPSC (treatment: F2,12 = 5.4, p = 0.012; fed/dep:
F1,12 = 3.5, p = 0.09; interaction: F2,12 = 5.2, p = 0.013; Figure 3B,
dose response in Figure S2B). Pharmacological control experiments testing alternative modes of action for AICAR and Cpd
C on fmEPSC did not reveal off-target effects (Figures S2C
S2H). These results show that AMPK is a necessary and sufficient signaling component mediating synaptic upregulation after
food deprivation or application of ghrelin in brain slices.
Modulation of fmEPSC indicated that AMPK may act presynaptically. Because presynaptic AMPK signaling has not been reported previously, we manipulated AMPK in the postsynaptic
patch-clamped AGRP neuron in additional experiments to rule
out a postsynaptic mechanism. Dialysis of AGRP neurons with
the membrane-impermeant AMPK activator ZMP (3 mM) (Corton
et al., 1995) in the patch pipette did not increase fmEPSC in brain
slices from fed mice (1.2 0.1 s 1, n = 11; Figure 3C and
Figure S2I). In brain slices from deprived mice, dialysis with
a high concentration of Cpd C (500 mM) did not significantly
decrease fmEPSC (break-in: 3.1 0.4 s 1, n = 7; 15 min cell dialysis: 3.0 0.4 s 1, n = 7; paired t test, p = 0.25; Figure 3D). These
results indicate that AMPK signaling in the postsynaptic AGRP
neuron does not influence fmEPSC, which is consistent with
a presynaptic AMPK signaling pathway.
Because deprivation-induced upregulation of excitatory synaptic input increases the action potential firing rate (fAP) of
AGRP neurons (Figure 1E), we next determined the contribution
of AMPK-mediated presynaptic activation to electrical activity
in AGRP neurons. For this, we blocked AMPK signaling in the
postsynaptic AGRP neuron with dialysis of Cpd C from the patch
pipette while perturbing the presynaptic pathway with subsequent bath application of AMPK activators and inhibitors. In
brain slices from fed mice, AICAR increased fAP in AGRP neurons
under conditions of postsynaptic AMPK inhibition (control:
0.11 0.05 s 1; AICAR: 0.9 0.3 s 1, n = 5; paired t test, p =
0.033; Figure 3E). Conversely, AMPK blockade by extracellular
but not intracellular Cpd C reduced fAP in slices from deprived
mice (control: 1.3 0.4 s 1; Cpd C: 0.2 0.1 s 1, n = 5; paired
t test, p = 0.037; Figure 3F). Together, these experiments
show that AMPK signaling is sufficient to increase AGRP neuron

firing rate to levels observed in deprived mice, and that this is
not due to AMPK in the postsynaptic AGRP neuron.
Ghrelin activation of synaptic activity, which is sensitive to Cpd
C-mediated AMPK inhibition (Figure 3B), was also unaffected by
Cpd C targeted to the postsynaptic AGRP neuron (control: 1.4
0.1 s 1, ghrelin: 2.5 0.2 s 1, n = 6; paired t test, p < 0.001; Figure 3G), consistent with a presynaptic effect. Further supporting
this, mEPSC amplitude was not significantly changed by ghrelin
treatment (control: 17.8 0.9 pA; ghrelin: 19.2 1.6 pA, n = 6;
paired t test, p = 0.34). In addition, ghrelin elevated fmEPSC
in AGRP neurons dialyzed with high BAPTA (10 mM, control:
1.7 0.3 s 1, ghrelin: 2.9 0.1 s 1, n = 4; paired t test, p =
0.007, Figure S2J). These experiments indicate that neither postsynaptic AMPK nor postsynaptic Ca2+ signaling was necessary
for upregulation of fmEPSC by ghrelin, which is in agreement
with a presynaptic mechanism.
Consistent with elevated excitatory synaptic input frequency,
AGRP neuron firing was increased by ghrelin in slices from fed
but not food-deprived mice with AMPK signaling blocked in
the postsynaptic AGRP neuron (ghrelin: F1,13 = 5.1, p = 0.041;
fed/dep: F1,13 = 8.1, p = 0.014; interaction: F1,13 = 6.9, p =
0.014; Figure 3H). Also, ghrelin-mediated activation of fAP in
AGRP neurons was prevented by excitatory synaptic blockade
with CNQX (control/CNQX: 0.1 0.03 s 1; ghrelin/CNQX: 0.1
0.05 s 1, n = 9; paired t test, p = 0.57; Figure 3I). Both of these
results indicate that ghrelin acts directly on presynaptic terminals, which is consistent with a report showing ghrelin-binding
sites associated with axonal boutons in the arcuate nucleus
(Cowley et al., 2003). Our data also indicate a presynaptic site
of action for ghrelin activation of AGRP neuron firing rate and
not a direct effect on AGRP neurons as has been previously
suggested (Andrews et al., 2008; Cowley et al., 2003).
To further examine the presynaptic mechanism described
here, we also considered the influence of this signaling pathway
on activity-evoked synaptic release. The ratio of synaptic responses (EPSC2/EPSC1, paired-pulse ratio, PPR) from a pair of
electrical stimuli (50 ms interval) to afferent axons is a property
associated with presynaptic function, where lower PPR values
(synaptic depression) are indicative of higher synaptic release
probability (Dobrunz and Stevens, 1997). A positive relationship
might be expected between release probability and elevated
fmEPSC. However, despite divergent fmEPSC levels, we found
that PPR in AGRP neurons was not significantly different in brain
slices from fed and food-deprived mice (fed: 0.67 0.09, n = 15;
dep: 0.59 0.08, n = 12; unpaired t test, p = 0.53; Figure 3J). In
both conditions, low PPR was not due to glutamate receptor
desensitization, as similar results were obtained under partial
receptor block by g-D-glutamylglycine (2 mM), a competitive
glutamate receptor antagonist (Figure S2K). Next, we considered the possibility that because synapses onto AGRP neurons
from fed mice show high release probability, they may have
limited capacity for further reduction of PPR after food deprivation or treatment with ghrelin. To test this, we shifted synaptic
responses from depressing to facilitating by reducing external
[Ca2+] from our standard conditions (2 mM) to 0.5 mM, which
resulted in PPR > 1 in AGRP neurons from fed mice (Figure 3K).
We found that treatment of brain slices with ghrelin under these
Figure 3. AMPK Signaling Mediates Deprivation- and Ghrelin-Induced Synaptic Activity

(A) AICAR increases fmEPSC in AGRP neurons from fed (n = 8) but not food-deprived (n = 8) mice.
(B) Cpd C reduces fmEPSC in AGRP neurons from deprived (n = 8) but not fed (n = 6) mice and blocks the ghrelin-mediated increase in fmEPSC.
(C) AGRP neuron dialysis with the AMPK activator ZMP (3 mM) in the patch pipette internal solution (int.) does not significantly increase fmEPSC relative to AGRP
neurons recorded with standard internal solution (nt, data from Figure 1B).
(D) Targeted AMPK inhibition with Cpd C (int.) in AGRP neurons does not significantly (p > 0.05) change fmEPSC after neuron dialysis (15 min).
(E) AGRP neuron firing rate is increased by bath-applied AICAR after intracellular blockade of AMPK with Cpd C (int.).
(F) AGRP neuron firing is decreased by bath-applied Cpd C after dialysis with Cpd C (int.).
(G) Targeted AMPK inhibition with Cpd C (int.) does not block ghrelin-mediated fmEPSC increase.
(H) Ghrelin activates firing in AGRP neurons from fed (n = 8) but not deprived (n = 7) mice with postsynaptic AMPK blockade by Cpd C (int.).
(I) Glutamate receptor blockade prevents ghrelin activation of AGRP neuron firing.
(J) Paired-pulse ratio (PPR) in AGRP neurons from fed and food-deprived mice in 2 mM Ca2+.
(K) PPR in AGRP neurons from fed mice (0.5 mM external Ca2+) is reduced by ghrelin, and this is reversed by Cpd C. Average EPSC responses from two cells
are also shown (inset).
(L and M) PPR in AGRP neurons from deprived mice (0.5 mM external Ca2+) is unaffected by ghrelin and increased by Cpd C. (M, inset) Average EPSC response
from one cell is shown.
(N) Inhibition of CAMKK with STO-609 in AGRP neurons from fed mice blocks ghrelin-mediated but not AICAR-mediated increase of fmEPSC.
(O) AICAR does not significantly increase fmEPSC in the presence of ryanodine.
(P) 8-Br-cADP ribose blocks the ghrelin-mediated increase of fmEPSC.
(Q) Diagram of the signaling pathway supported by these experiments. Pointed and T arrows represent activation and inhibition, respectively.
n.s., p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001. Data are represented as mean SEM. See also Figure S2.
conditions reduced PPR and switched synaptic responses from

facilitating to depressing, and that this was reversed by treatment with Cpd C (control: 1.3 0.05; ghrelin: 0.88 0.03; ghrelin+Cpd C: 1.4 0.07, n = 7; Figure 3K). Conversely, even in
0.5 mM Ca2+, synaptic depression was observed in AGRP
neurons from deprived mice and was not significantly altered
by ghrelin (control: 0.80 0.09; ghrelin: 0.80 0.06, n = 3; paired
t test, p = 0.98; Figure 3L). Notably, AMPK inhibition with Cpd C
significantly increased PPR, which produced facilitating synaptic
responses in AGRP neurons from food-deprived mice (control:
0.85 0.05; Cpd C: 1.3 0.1, n = 4; paired t test, p = 0.007; Figure 3M). These experiments indicate that a ghrelin- and AMPKsensitive signaling pathway in presynaptic terminals influences
both spontaneous release and evoked release; however, the
observable effect on evoked release might be limited at physiologic [Ca2+] due to high baseline release probability.
A Signaling Pathway for AMPK-Mediated
Synaptic Activity
Elevated AMPK activity is typically associated with energy deficit
and a high cellular AMP/ATP ratio. In our experiments, brain
slices were maintained in an energy-rich, high-glucose solution
for several hours. Thus, it is unlikely that cellular depletion of
ATP after food deprivation is the mechanism for AMPK activation, and we considered other pathways. Ca2+/calmodulindependent protein kinase kinase (CAMKK) is a direct activator
of AMPK (Hawley et al., 2005; Hurley et al., 2005; Woods et al.,
2005) and has been shown to be important for ghrelin-mediated
feeding (Anderson et al., 2008). Inhibition of CAMKK with
the inhibitor STO-609 (315 mM, dose response in Figure S2L)
blocked the ghrelin-mediated increase in fmEPSC onto AGRP
neurons from fed mice; however, AICAR was still capable of
increasing fmEPSC after treatment with STO-609, confirming
that CAMKK signaling is upstream of AMPK (control: 1.7
0.2 s 1; STO-609: 1.8 0.2 s 1; ghrelin: 1.8 0.1 s 1; AICAR:
2.6 0.2 s 1, n = 10; Figure 3N). We note that higher concentrations of STO-609 blocked AICAR-mediated synaptic upregulation (Figure S2L), in line with known off-target AMPK inhibition
by high STO-609 doses (Hawley et al., 2005). However, doseresponse curves for fmEPSC and western blots of brain slice
lysates show a dosage window in brain slice preparations,
consistent with STO-609 inhibition of CAMKK without AMPK
blockade (Figures S2LS2N).
We also investigated the downstream signaling pathway by
which AMPK increases fmEPSC. Ryanodine blockade of AICAR
upregulation of fmEPSC showed that RyRs are downstream of
AMPK (Figure 3O). Furthermore, AMPK has been shown previously to elicit Ca2+ release in smooth muscle tissue via the
endogenous RyR potentiator, cADP ribose (Evans et al., 2005).
The cADP ribose antagonist, 8-Br-cADP ribose (100 mM),
blocked the ghrelin-mediated increase in fmEPSC (Figure 3P),
indicating that cADP ribose production is likely the link from
AMPK to RyR activation.
Collectively, these experiments are consistent with a deprivation-induced, ghrelin-sensitive pathway wherein signaling
through Ghsr1, a Gaq/11-coupled pathway that mobilizes Ca2+
from IP3-sensitive internal stores (Smith, 2005), leads to activation of CAMKK, which signals through AMPK to release Ca2+
from ryanodine-sensitive internal stores, thereby increasing

spontaneous excitatory synaptic activity and release probability
(Figure 3Q). Elevated synaptic glutamate release results in
increased AGRP neuron firing, a response that has been shown
to elicit elevated food intake in mice (Aponte et al., 2011).
Persistently Elevated Synaptic Activity Results from
an AMPK-Dependent Positive Feedback Loop
In light of this ghrelin-dependent pathway, it was puzzling that
elevated synaptic activity in brain slices from food-deprived
mice persisted for hours after preparation and incubation of
brain slices in ghrelin-free artificial cerebrospinal fluid (aCSF).
We found that persistent activity could be sustained even after
Ghsr1 blockade by D-Lys3-GHRP6 or by the Ghsr1 inverse
agonist [D-Arg1,D-Phe5,D-Trp7,9,Leu11]-Substance P (SP*,
100 mM) (Holst et al., 2003) (Figure S3), indicating that ghrelinmediated or ghrelin-independent constitutive Ghsr1 signaling
were not responsible for the sustained activity.
To explore this further, we used transient exposure to ghrelin
(5 min) followed by washout and treatment with SP* to block
subsequent Ghsr1 signaling in order to examine the potential
for ghrelin to elicit persistent activity in brain slices from fed
mice. This sequence led to elevation of fmEPSC and demonstrated
marked hysteresis, remaining high after ghrelin washout and
treatment with SP* (n = 6; Figure 4A).
Hysteresis, a memory capacity observed in diverse biological
processes, frequently involves positive feedback (Ferrell, 2002).
We considered the possibility that, in this system, hysteresis resulted from positive feedback in the AMPK-dependent signaling
pathway (Figure 3Q). This ghrelin-responsive pathway results in
Ca2+ release from ryanodine-sensitive internal stores but also
involves the upstream Ca2+-sensitive kinase CAMKK, raising
the possibility for positive feedback to sustain the pathway in
the absence of ghrelin. Consistent with the high cooperativity
associated with positive feedback (Ferrell, 2002), dose-response
curves for AICAR, Cpd C, and STO-609 showed switch-like ultrasensitivity (Koshland et al., 1982) (Figures S2A, S2B, and S2L).
To further explore the possibility of positive feedback, we
reasoned that persistent synaptic activity should also be sensitive to activation and blockade at any node in the loop. To test
this, we released Ca2+ from ryanodine-sensitive internal stores
using a brief exposure (5 min) to caffeine, which we have shown
elevates fmEPSC (Figure 2B). Under these conditions, synaptic
activity also exhibited hysteresis; after washout of caffeine,
fmEPSC was still elevated (Figure 4B). Subsequent blockade of
the upstream AMPK node of the pathway with Cpd C reversed
fmEPSC to precaffeine levels (Figure 4B; compare with Figure S2F), as expected from a positive feedback loop. Toggling
from the high- to the low-activity state in the absence of ghrelin
shows that these synapses are bistable.
Duration of the Persistently Activated State
To measure the duration of the persistently activated state, brain
slices from fed mice were treated briefly with ghrelin (30 nM,
5 min), washed by transfer to a ghrelin-free bathing solution,
and then transferred again for incubation with the Ghsr1 antagonist, D-Lys3-GHRP6 (100 mM), to ensure the absence of ongoing
ghrelin signaling for 35 hr, after which they were used for
different from those observed in ghrelin treatment experiments

that involved continuous exposure to the hormone (Figure 2D
and Figure 3G). Furthermore, fmEPSC was still rapidly reversed
by AMPK blockade with Cpd C (ghrelin/wash/D-Lys3-GHRP6:
2.7 0.2 s 1; +Cpd C: 1.1 0.1 s 1, n = 3; paired t test, p =
0.023; Figure 4D) or CAMKK inhibition by STO-609 (3 or
15 mM; Figure 4E). As a control, we incubated slices with
D-Lys3-GHRP6 without pre-exposure to ghrelin; these showed
low fmEPSC and no effect of Cpd C (D-Lys3-GHRP6, 35 hr:
1.3 0.02 s 1; +Cpd C: 1.2 0.1 s 1, n = 3; paired t test,
p = 0.51; Figure 4D).
In addition, we tested brief caffeine exposure (10 mM, 5 min) in
brain slices from fed mice followed by incubation in caffeine-free
aCSF for 35 hr (Figure 4C). This treatment also elevated fmEPSC,
which was rapidly reversed by Cpd C (caffeine/wash: 2.9
0.3 s 1; +Cpd C: 1.9 0.1 s 1, n = 6; paired t test, p = 0.026; Figure 4F). Thus, synaptic activation persists for at least several
hours after induction, is AMPK sensitive throughout this timeframe, and can be induced upstream and downstream of
AMPK signaling.
Figure 4. Synaptic Hysteresis Resulting from an AMPK-Dependent

Positive Feedback Loop
(A) Ghrelin upregulation of fmEPSC shows hysteresis; fmEPSC is sustained after
ghrelin washout and Ghsr1 blockade with SP*. Duration of each transition is in
parentheses (minutes).
(B) Caffeine increases fmEPSC, which remains elevated after washout. Synaptic
activity returned to baseline after treatment with Cpd C, consistent with a
positive feedback loop. Significant differences are denoted by any interaction
across the red dashed line.
(CF) Procedure for testing duration of persistent activity by transient exposure
of brain slices to ghrelin or caffeine (5 min), transfer to a wash solution (10 min),
and transfer again to a solution containing either aCSF alone or with D-Lys3GHRP6 (35 hr). Slices were subsequently transferred to a recording chamber
for electrophysiology.
(D and E) After transient ghrelin exposure and prolonged D-Lys3-GHRP6
incubation, fmEPSC remained elevated but was rapidly (10 min) reduced by (D)
Cpd C or (E) STO-609. Control brain slices that were not treated (nt) with
ghrelin but otherwise incubated as above (D) had low fmEPSC.
(F) After transient exposure to caffeine and prolonged incubation in caffeinefree aCSF (35 hr), fmEPSC was still elevated but was rapidly (10 min) reduced
by Cpd C, consistent with the operation of a positive feedback loop.
n.s., p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001. Data are represented as
mean SEM. See also Figure S3.
electrophysiology in D-Lys3-GHRP6-containing solution (Figure 4C). Remarkably, AGRP neurons from these slices still
showed elevated fmEPSC levels that were not significantly
Counter-regulation of Persistent Synaptic Activity

Next, we investigated the pathway by which persistent synaptic
upregulation is switched off. We observed that fmEPSC from
mice that were food deprived and then refed was still elevated
24 hr after refeeding but was reduced to baseline by 48 hr
(Figure 5A); notably, mEPSC amplitude did not change significantly throughout this time course (data not shown, F4,187 =
1.22, p = 0.31). Consistent with the elevated excitatory synaptic
input, the action potential firing rate of AGRP neurons was also
elevated 24 hr after refeeding, which required glutamatergic
synaptic input (24 hr refed: 1.5 0.2 s 1; 24 hr refed/CNQX:
0.2 0.1 s 1, n = 6; paired t test, p = 0.002; Figure 5B).
This timescale led us to examine the role of the hormone leptin,
which is associated with long-term regulation of energy homeostasis. Leptin reduces ghrelin-induced feeding (Nakazato et al.,
2001) and deprivation-induced AMPK phosphorylation in the
hypothalamus (Minokoshi et al., 2004). Leptin administration
(1 mg/g) to deprived mice 3 hr before brain slice preparation
decreased fmEPSC to levels found in fed mice (i.p. saline: 2.8
0.4 s 1, n = 9; i.p. leptin: 1.7 0.2 s 1, n = 10; unpaired t test,
p = 0.024; Figure 5C). Leptin (100 nM) treatment of brain slices
from deprived mice did not affect fmEPSC onto AGRP neurons
(2.7 0.3 s 1, n = 11), so we considered an indirect interaction.
One possibility is that leptin activation of POMC neurons in
the arcuate nucleus (Cowley et al., 2001) could release a neuropeptide modulator of AMPK activity. Because a-MSH is a POMC
neuron-derived neuropeptide that reduces feeding through
melanocortin receptor signaling, we tested the influence of
MTII (500 nM), a synthetic agonist of melanocortin receptors,
on fmEPSC. However, this treatment did not decrease fmEPSC
onto AGRP neurons in brain slices from deprived mice (control:
3.0 0.7 s 1; MTII: 3.1 0.6 s 1, n = 6; paired t test, p = 0.93).
We next considered the possibility that an opioid such as
b-endorphin could reverse persistent synaptic upregulation in
AGRP neurons. b-endorphin is also produced in POMC neurons
and is a m-opioid receptor (MOR) agonist (Silva et al., 2001). Brain
slices from food-deprived mice were treated with a synthetic
Figure 5. Persistent Synaptic Upregulation Is

Reversed by Leptin-Mediated Opioid Release
(A) Synaptic activity before, during, and after food deprivation and refeeding (fed and deprived data from Figure 1C, 1 hr: n = 21, 24 hr: n = 27, 48 hr: n = 23). p values for
multiple pairwise comparisons were adjusted with Holms
correction. The significant differences are denoted by any
interaction across the red dashed line.
(B) AGRP neuron firing rate 24 hr after refeeding was still
elevated, which was dependent on glutamatergic synaptic
input.
(C) Injection of leptin in deprived mice reduced fmEPSC in
AGRP neurons relative to saline injection.
(D) DAMGO reduced fmEPSC in AGRP neurons under
VGCC block (CdCl2). The fmEPSC remained at this level
during a 15 min wash but was increased by treatment with
AICAR.
(E) fmEPSC in AGRP neurons from deprived mice treated
with DAMGO is insensitive to Cpd C.
(F) NTX pretreatment of deprived mice blocks leptinmediated reduction in fmEPSC observed with saline
pretreatment.
(G) Coinjection of ghrelin and NTX, but not ghrelin alone,
leads to elevated fmEPSC in brain slices prepared after 3 hr.
*p < 0.05, **p < 0.01, ***p < 0.001. Data are represented as
mean SEM.
selective MOR agonist, DAMGO (1 mM). This was done in the

presence of CdCl2 because MORs can also influence synaptic
release by inhibition of VGCCs (Endo and Yawo, 2000), which
we have shown to be unrelated to deprivation-induced plasticity
(Figure 2A). Under these conditions, we found that, after treatment and washout of DAMGO (see Experimental Procedures),
fmEPSC in AGRP neurons was similar to levels from fed mice.
Furthermore, these synapses were now sensitive to AICAR,
which increased fmEPSC (control: 2.2 0.3 s 1; DAMGO: 0.9
0.2 s 1; wash: 1.0 0.2 s 1; AICAR: 1.8 0.3 s 1, n = 9; Figure 5D). Conversely, treatment with Cpd C after washout of
DAMGO did not change fmEPSC, indicating that AMPK had
been fully inactivated by MOR signaling (control: 2.3 0.2 s 1;
DAMGO: 1.1 0.2 s 1; wash: 1.1 0.3 s 1; Cpd C: 1.1
0.3 s 1, n = 5; Figure 5E).
The negative regulation of AMPK by opioid receptor signaling
raised the question of whether tonic opioid activity was required
to prevent ghrelin-independent activation of the AMPK-mediated positive feedback loop. Incubation of brain slices from ad
libitum fed mice with the opioid receptor antagonist naltrexone
(NTX, 500 nM) for 35 hr did not increase fmEPSC or induce
sensitivity to Cpd C (NTX: 1.3 0.2 s 1; NTX/Cpd C: 1.4
0.1 s 1, n = 3; paired t test, p = 0.84), indicating that AMPK is
not disinhibited by NTX blockade of opioid

receptors. This suggests that additional control
pathways restrain this presynaptic positive
feedback loop in the absence of ghrelin.
We also tested the leptin-opioid relationship
in vivo by treating food-deprived mice with
NTX (10 mg/g) followed by leptin. This combination blocked leptin reduction of fmEPSC in brain
slices from deprived mice (vehicle/leptin: 1.6
0.1 s 1, n = 26; NTX/leptin: 3.0 0.2 s 1, n = 34; unpaired
t test, p < 0.001; Figure 5F), which confirms that leptin downregulates AMPK-mediated synaptic activity through an opioid
receptor-dependent mechanism.
Next, we examined the persistence of the ghrelin-induced upregulation of fmEPSC after ghrelin injection to ad libitum fed mice.
Because injected ghrelin levels are reported to decline to baseline over 12 hr (Tschop et al., 2000), we looked at the effect of
ghrelin treatment on fmEPSC 3 hr post-injection, during which
time mice were not exposed to food. Based on our analysis of
persistent activity at these synapses, we expected fmEPSC to
be elevated as it is 30 min after ghrelin administration (Figure 2E).
Instead, we found that fmEPSC was similar to that in uninjected fed
mice (ghrelin, 3 hr: 1.5 0.2 s 1, n = 22). We tested the possibility
that the opioid tone in fed mice was sufficient to switch off
elevated synaptic activity once injected ghrelin was cleared.
Consistent with this, cotreatment of ad libitum fed mice with
ghrelin and NTX led to fmEPSC that was still elevated 3 hr after
ghrelin treatment (ghrelin/NTX 3 hr: 2.7 0.4 s 1, n = 10;
unpaired t test, p = 0.009; Figure 5G). Fed mice treated solely
with NTX did not increase fmEPSC after 3 hr (1.5 0.3 s 1,
n = 11). These experiments are consistent with the persistence
of ghrelin-induced synaptic upregulation and indicate that the
Figure 6. POMC Neurons Release an Opioid that

Resets Persistent Synaptic Activity
(A) Epifluorescence micrograph of brain slice with POMC
neurons expressing ChR2-tdtomato. Blue circles: photostimulation sites; ARC: arcuate nucleus; 3V: third ventricle;
D: dorsal; V: ventral.
(B) ChR2-mediated photostimulation of POMC neurons
in brain slices from deprived mice in the presence or
absence of NTX. Photostimulation of POMC neurons
reduces fmEPSC unless performed in the presence of NTX.
(C) Subset of neurons in (B) subjected to AICAR
after photostimulation. POMC neuron photostimulation
in the absence of NTX (n = 5) rendered fmEPSC in AGRP neurons sensitive to AICAR, whereas those photostimulated in the presence of NTX (n = 5)
were insensitive to AICAR, indicating that a POMC neuron-derived opioid inactivates AMPK.
**p < 0.01. Data are represented as mean SEM.
baseline opioid tone in fed mice can reverse this after ghrelin
levels fall.
POMC Neurons Release an Opioid that Resets
Persistent Synaptic Activity
Although opioid receptors are involved in the leptin-mediated
inactivation of presynaptic AMPK, it was unclear whether
POMC neurons could serve as the source of these opioids. We
tested this using an optogenetic approach. We expressed the
light-activated cation channel channelrhodopsin-2 (ChR2) (Boyden et al., 2005) in POMC neurons, rendering them selectively
photoexcitable. POMC neurons were targeted using the Cre recombinase (Cre)-dependent viral vector rAAV-FLEX-rev-ChR2tdtomato, which we have described previously for POMC neuron
photostimulation (Aponte et al., 2011; Atasoy et al., 2008).
In brain slices from food-deprived Pomc-Cre;Npy-sapphireFP
double-transgenic mice expressing ChR2, we photostimulated
the area containing the arcuate nucleus with focused laser light
in a 4 3 10 spatial pattern of stimulation sites (Figure 6A). Our
expectation was that a neuromodulator would be released that
could potentially downregulate AMPK signaling. Next, the slice
was treated with TTX to record mEPSCs and CdCl2 in order to
eliminate any contribution of VGCCs. POMC neuron photostimulation led to a significant reduction in fmEPSC in AGRP neurons,
and this effect was blocked if photostimulation was performed
in the presence of NTX (500 nM) (control: 1.3 0.2 s 1, n = 15;
NTX: 2.2 0.2 s 1, n = 14; unpaired t test, p = 0.003; Figure 6B).
When NTX was present during photostimulation, AGRP neurons
from deprived mice maintained insensitivity to AICAR (Figure 6C).
However, after photostimulation in the absence of NTX, fmEPSC
could be increased in response to subsequent AICAR administration (AICAR: F1,8 = 43.3, p < 0.001; NTX: F1,8 = 3.6, p =
0.096; interaction: F1,8 = 36.8, p < 0.001; Figure 6C). These
results indicate that a POMC neuron-derived opioid, likely
b-endorphin, reverses AMPK-mediated persistent synaptic
upregulation.
DISCUSSION
Homeostasis requires increased food intake as energy stores
are depleted. In this study, we provide evidence for a dynamic
neural circuit (Figure 7A) with a reversible memory storage
capacity that regulates feeding behavior in response to energy
deficit. Previous work showed a core circuit in which AGRP

neurons inhibit POMC neurons and electrical activity is regulated
by hormones that signal physiological state (Cowley et al., 2001,
2003). We report here that a key control point for this circuit is
excitatory synapses onto AGRP neurons. These synapses
endow this circuit with unexpected memory properties based
on a presynaptic AMPK-dependent positive feedback loop and
regulation by a previously undescribed POMC neuron-derived
opioid pathway.
The investigation of synaptic regulation of AGRP neurons
provided insight into the mechanism of ghrelin-evoked feeding.
AGRP neurons are required for ghrelins orexigenic properties
(Luquet et al., 2007). Previous reports concluded that ghrelin
activates AGRP neurons directly (Andrews et al., 2008; Cowley
et al., 2003). A model by Andrews et al. (2008) of ghrelin-induced
feeding suggests that Ghsr1 signaling in AGRP neurons elicits
firing by AMPK-mediated alteration of fatty-acid oxidation and
regulation of mitochondrial uncoupling proteins. However, it is
unclear how these metabolic pathways influence electrical
activity, except perhaps by increasing available intracellular
energy stores.
Our findings support an alternative model in which ghrelin acts
at presynaptic receptors to increase glutamate release and
activate AGRP neurons through ionotropic glutamate receptors.
This is unlikely to be related to increased excitatory synapse
number based on previous ultrastructural quantification following ghrelin treatment (Pinto et al., 2004). We suggest that
metabolic pathways for ghrelin action in AGRP neurons (Andrews et al., 2008) are likely responsible for altering neuropeptide gene expression or addressing metabolic requirements to
maintain AGRP neuron firing but are not directly involved in the
increase of neuron firing following ghrelin administration.
We also find that AMPK participates in a positive feedback
loop, resulting in synaptic hysteresis. Hysteresis for AMPK
activation was investigated previously in Xenopus oocytes,
but, although nonreversing activation was observed with transient antimycin treatment, this was attributed to irreversible
metabolic block, not bistability, and other activators of AMPK
did not show hysteresis (Martianez et al., 2009). Thus, hysteresis
for AMPK activity is not elicited in all cells.
Hysteresis is a hallmark of bistability (Bhalla and Iyengar,
1999), which confers memory capacity in biological systems,
often in conjunction with positive feedback loops (Ferrell,
Figure 7. SR Flip-Flop Model of a Neural Circuit with Synaptic

Memory of Physiological State
(A) A core circuit in which AGRP neurons synaptically inhibit POMC neurons
and are regulated by circulating hormones is controlled by ghrelin-responsive
excitatory synapses. These synapses give this circuit a memory property
based on an AMPK-dependent positive feedback loop (inset), which can be
reversed by POMC neuron output, likely b-endorphin.
(B) A heuristic for the logic of this circuit is the SR flip-flop memory storage
circuit. In the analogy with the neural circuit reported here, the set signal is
ghrelin, which activates the green NOR gate, representing the conglomeration
of AGRP neurons and their ghrelin-sensitive excitatory presynaptic terminals.
The reset signal is leptin, which interacts with POMC neurons represented as
the blue NOR gate. Notably, when R and S are both high, the circuit does not
support memory, and this condition is consistent with the case of ghrelin
treatment of fed mice where opioid signaling is sufficiently high to prevent
persistent synaptic upregulation (Figure 5G).
2002). Other kinase-dependent positive feedback loops have

been considered previously for long-term information storage
in the central nervous system (Bhalla and Iyengar, 1999; Tanaka
and Augustine, 2008). The AMPK-mediated feedback described
here persistently upregulates activity in presynaptic terminals for
at least 5 hr, which could serve as a memory of the hormone
ghrelin in the form of sustained synaptic activity onto AGRP
neurons.
Synapses with positive feedback also require a separate
inhibitory signal to reverse synaptic upregulation, for example
once energy balance is restored. Leptin is well suited for this
role as it signals energy reserves by falling rapidly during food
deprivation and rises with refeeding (Boelen et al., 2006). We
find that leptin is sufficient to counter-regulate elevated presynaptic activity onto AGRP neurons through an opioid
receptor-dependent pathway. Leptin activates POMC neurons
(Cowley et al., 2001), and it is likely that b-endorphin is the
opioid released from POMC neurons to downregulate elevated synaptic activity. Interestingly, rapid synaptic plasticity at
AGRP neurons was previously described in leptin-deficient

ob/ob mice after leptin administration (Pinto et al., 2004); future
work could investigate whether this involves the opioid-dependent mechanism described here.
Previous studies showed that selective ablation of only the
b-endorphin-encoding portion of the Pomc gene yields mice
that are hyperphagic and overweight (Appleyard et al., 2003).
This was unexpected in the context of b-endorphin intracranial
administration, which activates multiple opioid pathways and
leads to overeating (Grandison and Guidotti, 1977), but it is
consistent with the anorexigenic function of POMC neurons. In
light of the experiments reported here, b-endorphin ablation
appears to disable a key regulatory system that may include
ghrelin-mediated synaptic activity onto AGRP neurons, consequently potentiating ghrelins feeding stimulatory function.
The circuit described here could be further elaborated by
considering the origin of the ghrelin-sensitive glutamatergic
inputs. Currently, these neurons are not known, but they should
originate from Ghsr1-expressing populations, which are clustered in several brain regions (Zigman et al., 2006). Retrograde
tracing studies (Li et al., 1999; DeFalco et al., 2001) indicate
that a subset of these brain regions project to the arcuate
nucleus and to AGRP neurons. Future work could examine
the identity of these presynaptic neurons, which would provide
an entry point for genetically encoded tools to perturb presynaptic pathways with kinase mutants or optogenetic tools. In
addition, upstream populations projecting to AGRP neurons
might also be expected to influence feeding behavior, possibly
extending control of AGRP neuron electrical activity beyond the
contexts of hormonal and homeostatic regulation discussed in
this study.
Why Might Neural Circuits Responding to Hormonal
Cues Require a Memory Capacity?
These experiments provide evidence of a circuit that has the
capacity to store a memory of specific hormonal states (Figure 7A). Heuristically, its properties are analogous to the set/
reset (SR) flip-flop memory storage circuit from digital electronics, which consists of two interconnected NOR logic gates
(Horowitz and Hill, 1989) (Figure 7B). Such a memory mechanism
is consistent with the logic for hormonal control of synaptic upregulation and counter-regulation described in this study.
SR flip-flop memory has useful features for physiologic regulation. In this model, two hormones (set: ghrelin and reset: leptin) separately signal deficit or surfeit. The memory properties of
the SR flip-flop act as an until loop, maintaining activity in
response to S (even after S is turned off) until R is true. By using
two discrete signals that reflect different physiological states, the
circuit detects a state change. This model does not have a set
point, but instead, it has a set range that is defined by the respective thresholds for activation of S and R. The operation of this
circuit differs from homeostatic set point control theory in which
a physiological parameter is continuously compared to a reference value and results in commensurate adjustment of physiological or behavioral output. This two-component circuit model
is, however, evocative of Walter Cannons original formulation
of the homeostasis concept, which referred to the balance of
opposing factors (Cannon, 1929). In this circuit, though, we
add the provision that these signals operate as discrete Boolean

logical operations (see Figure 7B) due to an AMPK-dependent
positive feedback loop. Importantly, this circuit displays memory
properties only under sufficient deprivation to significantly
reduce R (leptin levels or opioid tone), which is what we observe
experimentally.
A potential advantage of a physiological set range for energy
balance is that it avoids constant readjustment of energy stores
as required for defending a set point, which may facilitate pursuit
of other behaviors when energy repletion is not critical. Moreover, modulation of the response threshold for ghrelin, leptin,
or opioids could shift this range and, potentially, food intake
and body weight. Another implication of this state-sensing property is that it could, in principle, simplify requirements for the
regulatory control of ghrelin and leptin release from peripheral
endocrine cells such that they need not reflect the integrated
energy balance of the organism but instead simply report a state
of deficit and surfeit, respectively. For example, high ghrelin
levels resulting from food deprivation have been shown to fall
within 12 hr of refeeding, due to negative regulation by nutrients
such as carbohydrates (Tschop et al., 2000), even though energy
homeostasis may not be achieved in this short time. However,
the ghrelin-initiated synaptic upregulation and the corresponding firing rate elevation in AGRP neurons were maintained after
24 hr of refeeding (Figures 5A and 5B), during which energy
stores are restored. The operation of this memory system
extends the responsiveness of the circuit beyond the lifetime
of the triggering hormone until a signal of energy surfeit resets
the circuit, in this case by opioid receptor activation. Such
a mechanism may also be important in neural circuits governing
behavioral responses to other physiological conditions. Furthermore, the relative ubiquity of these signaling pathways may
extend the findings of positive feedback and set/reset bistability
to other neuromodulators in the brain.
Experimental protocols were conducted according to U.S. National Institutes
of Health guidelines for animal research and were approved by the Institutional
Animal Care and Use Committee at Janelia Farm Research Campus. Food
deprivation was for 24 hr.
For several experiments, the experimenter was blinded to the identity of the
pharmacological reagent or to the deprivation state of the animal (Figures 1B,
1D, and 1E; Figure S1; Figure 2A [except ghrelin treatment]; Figures 3A and 3B;
Figure 5F; Figure 6).
Electrophysiology and photostimulation methods are provided in Extended
Experimental Procedures.
In figures, bars represent separate groups of cells, and symbols connected
by lines represent manipulations applied across a group of cells. Data are represented as mean standard error of the mean (SEM). p values for pairwise
comparisons were calculated by two-tailed Students t test. p values for
comparisons across more than two groups were adjusted with Holms correction (Holm, 1979). Tests involving one-way ANOVA or two-way ANOVA with
one factor repeated-measures were calculated with SigmaPlot (Systat). Not
significant (n.s.): p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001.
Supplemental Information includes Extended Experimental Procedures and
three figures and can be found with this article online at doi:10.1016/j.cell.
2011.07.039.
ACKNOWLEDGMENTS
This research was funded by the Howard Hughes Medical Institute. We thank
L. Scheffer for discussion of digital electronic circuits; J. Cox for mouse
husbandry; Y. Aponte for assistance with animal studies; Y. Li and Z. Zhang
for assistance with biochemistry; and K. Moses, S. Eddy, K. Svoboda,
G. Murphy, and C. Zuker for comments on the manuscript. Leptin was
provided by Amylin Pharmaceuticals. Y.Y. performed the experiments. H.S.
performed western blots. D.A. made the initial finding of deprivation-induced
plasticity. Y.Y. and S.M.S. designed the study, analyzed the data, and wrote
the paper.
Received: November 10, 2010
REFERENCES
Anderson, K.A., Ribar, T.J., Lin, F., Noeldner, P.K., Green, M.F., Muehlbauer,
M.J., Witters, L.A., Kemp, B.E., and Means, A.R. (2008). Hypothalamic
CaMKK2 contributes to the regulation of energy balance. Cell Metab. 7,
377388.
Andrews, Z.B., Liu, Z.W., Walllingford, N., Erion, D.M., Borok, E., Friedman,
J.M., Tschop, M.H., Shanabrough, M., Cline, G., Shulman, G.I., et al. (2008).
UCP2 mediates ghrelins action on NPY/AgRP neurons by lowering free
radicals. Nature 454, 846851.
Aponte, Y., Atasoy, D., and Sternson, S.M. (2011). AGRP neurons are sufficient
to orchestrate feeding behavior rapidly and without training. Nat. Neurosci. 14,
351355.
Appleyard, S.M., Hayward, M., Young, J.I., Butler, A.A., Cone, R.D.,
Rubinstein, M., and Low, M.J. (2003). A role for the endogenous opioid
beta-endorphin in energy homeostasis. Endocrinology 144, 17531760.
Atasoy, D., Aponte, Y., Su, H.H., and Sternson, S.M. (2008). A FLEX switch
targets Channelrhodopsin-2 to multiple cell types for imaging and long-range
circuit mapping. J. Neurosci. 28, 70257030.
Bhalla, U.S., and Iyengar, R. (1999). Emergent properties of networks of biological signaling pathways. Science 283, 381387.
Boelen, A., Kwakkel, J., Vos, X.G., Wiersinga, W.M., and Fliers, E. (2006).
Differential effects of leptin and refeeding on the fasting-induced decrease
of pituitary type 2 deiodinase and thyroid hormone receptor beta2 mRNA
expression in mice. J. Endocrinol. 190, 537544.
Boyden, E.S., Zhang, F., Bamberg, E., Nagel, G., and Deisseroth, K. (2005).
Millisecond-timescale, genetically targeted optical control of neural activity.
Nat. Neurosci. 8, 12631268.
Cannon, W.B. (1929). Organization for physiological homeostasis. Physiol.
Rev. 9, 399431.
Corton, J.M., Gillespie, J.G., Hawley, S.A., and Hardie, D.G. (1995). 5-aminoimidazole-4-carboxamide ribonucleoside. A specific method for activating
AMP-activated protein kinase in intact cells? Eur. J. Biochem. 229, 558565.
Cowley, M.A., Smart, J.L., Rubinstein, M., Cerdan, M.G., Diano, S., Horvath,
T.L., Cone, R.D., and Low, M.J. (2001). Leptin activates anorexigenic POMC
neurons through a neural network in the arcuate nucleus. Nature 411, 480484.
Cowley, M.A., Smith, R.G., Diano, S., Tschop, M., Pronchuk, N., Grove, K.L.,
Strasburger, C.J., Bidlingmaier, M., Esterman, M., Heiman, M.L., et al.
(2003). The distribution and mechanism of action of ghrelin in the CNS
demonstrates a novel hypothalamic circuit regulating energy homeostasis.
Neuron 37, 649661.
DeFalco, J., Tomishima, M., Liu, H., Zhao, C., Cai, X., Marth, J.D., Enquist, L.,
and Friedman, J.M. (2001). Virus-assisted mapping of neural inputs to
a feeding center in the hypothalamus. Science 291, 26082613.
Dobrunz, L.E., and Stevens, C.F. (1997). Heterogeneity of release probability,
facilitation, and depletion at central synapses. Neuron 18, 9951008.
Emptage, N.J., Reid, C.A., and Fine, A. (2001). Calcium stores in hippocampal
synaptic boutons mediate short-term plasticity, store-operated Ca2+ entry,
and spontaneous transmitter release. Neuron 29, 197208.
Endo, K., and Yawo, H. (2000). mu-Opioid receptor inhibits N-type Ca2+
channels in the calyx presynaptic terminal of the embryonic chick ciliary
ganglion. J. Physiol. 524, 769781.
Evans, A.M., Mustard, K.J., Wyatt, C.N., Peers, C., Dipp, M., Kumar, P.,
Kinnear, N.P., and Hardie, D.G. (2005). Does AMP-activated protein kinase
couple inhibition of mitochondrial oxidative phosphorylation by hypoxia to
calcium signaling in O2-sensing cells? J. Biol. Chem. 280, 4150441511.
Fatt, P., and Katz, B. (1952). Spontaneous subthreshold activity at motor nerve
endings. J. Physiol. 117, 109128.
Ferrell, J.E., Jr. (2002). Self-perpetuating states in signal transduction: positive
feedback, double-negative feedback and bistability. Curr. Opin. Cell Biol. 14,
140148.
Gordon, G.R., and Bains, J.S. (2006). Can homeostatic circuits learn and
remember? J. Physiol. 576, 341347.
Grandison, L., and Guidotti, A. (1977). Stimulation of food intake by muscimol
and beta endorphin. Neuropharmacology 16, 533536.
Hahn, T.M., Breininger, J.F., Baskin, D.G., and Schwartz, M.W. (1998).
Coexpression of Agrp and NPY in fasting-activated hypothalamic neurons.
Hawley, S.A., Pan, D.A., Mustard, K.J., Ross, L., Bain, J., Edelman, A.M.,
Frenguelli, B.G., and Hardie, D.G. (2005). Calmodulin-dependent protein
kinase kinase-beta is an alternative upstream kinase for AMP-activated
protein kinase. Cell Metab. 2, 919.
Holm, S.A. (1979). A simple sequentially rejective multiple test procedure.
Scand. J. Stat. 6, 6570.
Holst, B., Cygankiewicz, A., Jensen, T.H., Ankersen, M., and Schwartz, T.W.
(2003). High constitutive signaling of the ghrelin receptoridentification of
a potent inverse agonist. Mol. Endocrinol. 17, 22012210.
Horowitz, P., and Hill, W. (1989). The Art of Electronics, Second Edition
(Cambridge, UK: Cambridge University Press).
Howard, A.D., Feighner, S.D., Cully, D.F., Arena, J.P., Liberator, P.A.,
Rosenblum, C.I., Hamelin, M., Hreniuk, D.L., Palyha, O.C., Anderson, J.,
et al. (1996). A receptor in pituitary and hypothalamus that functions in growth
hormone release. Science 273, 974977.
Hurley, R.L., Anderson, K.A., Franzone, J.M., Kemp, B.E., Means, A.R., and
Witters, L.A. (2005). The Ca2+/calmodulin-dependent protein kinase kinases
are AMP-activated protein kinase kinases. J. Biol. Chem. 280, 2906029066.
Lopez, M., Lage, R., Saha, A.K., Perez-Tilve, D., Vazquez, M.J., Varela, L.,
Sangiao-Alvarellos, S., Tovar, S., Raghay, K., Rodrguez-Cuenca, S., et al.
(2008). Hypothalamic fatty acid metabolism mediates the orexigenic action
of ghrelin. Cell Metab. 7, 389399.
Luquet, S., Phillips, C.T., and Palmiter, R.D. (2007). NPY/AgRP neurons are not
essential for feeding responses to glucoprivation. Peptides 28, 214225.
Martianez, T., France`s, S., and Lopez, J.M. (2009). Generation of digital
responses in stress sensors. J. Biol. Chem. 284, 2390223911.
Minokoshi, Y., Alquier, T., Furukawa, N., Kim, Y.B., Lee, A., Xue, B., Mu, J.,
Foufelle, F., Ferre, P., Birnbaum, M.J., et al. (2004). AMP-kinase regulates
food intake by responding to hormonal and nutrient signals in the hypothalamus. Nature 428, 569574.
Nakazato, M., Murakami, N., Date, Y., Kojima, M., Matsuo, H., Kangawa, K.,
and Matsukura, S. (2001). A role for ghrelin in the central regulation of feeding.
Nature 409, 194198.
Pinto, S., Roseberry, A.G., Liu, H., Diano, S., Shanabrough, M., Cai, X.,
Friedman, J.M., and Horvath, T.L. (2004). Rapid rewiring of arcuate nucleus
feeding circuits by leptin. Science 304, 110115.
Silva, R.M., Hadjimarkou, M.M., Rossi, G.C., Pasternak, G.W., and Bodnar,
R.J. (2001). Beta-endorphin-induced feeding: pharmacological characterization using selective opioid antagonists and antisense probes in rats.
J. Pharmacol. Exp. Ther. 297, 590596.
Smith, R.G. (2005). Development of growth hormone secretagogues. Endocr.
Rev. 26, 346360.
Sternson, S.M., Shepherd, G.M., and Friedman, J.M. (2005). Topographic
mapping of VMH > arcuate nucleus microcircuits and their reorganization
by fasting. Nat. Neurosci. 8, 13561363.
Takahashi, K.A., and Cone, R.D. (2005). Fasting induces a large, leptin-dependent increase in the intrinsic action potential frequency of orexigenic arcuate
nucleus neuropeptide Y/Agouti-related protein neurons. Endocrinology 146,
10431047.
Tanaka, K., and Augustine, G.J. (2008). A positive feedback signal transduction loop determines timing of cerebellar long-term depression. Neuron 59,
608620.
Tschop, M., Smiley, D.L., and Heiman, M.L. (2000). Ghrelin induces adiposity
in rodents. Nature 407, 908913.
Koshland, D.E., Jr., Goldbeter, A., and Stock, J.B. (1982). Amplification and
adaptation in regulatory and sensory systems. Science 217, 220225.
Woods, A., Dickerson, K., Heath, R., Hong, S.P., Momcilovic, M., Johnstone,
S.R., Carlson, M., and Carling, D. (2005). Ca2+/calmodulin-dependent protein
kinase kinase-beta acts upstream of AMP-activated protein kinase in mammalian cells. Cell Metab. 2, 2133.
Li, C., Chen, P., and Smith, M.S. (1999). Identification of neuronal input to the
arcuate nucleus (ARH) activated during lactation: implications in the activation
of neuropeptide Y neurons. Brain Res. 824, 267276.
Zigman, J.M., Jones, J.E., Lee, C.E., Saper, C.B., and Elmquist, J.K. (2006).
Expression of ghrelin receptor mRNA in the rat and the mouse brain.
J. Comp. Neurol. 494, 528548.
Driving Opposing Behaviors with Ensembles

of Piriform Neurons
Gloria B. Choi,1 Dan D. Stettler,1 Benjamin R. Kallman,1 Shakthi T. Bhaskar,1 Alexander Fleischmann,1,2
and Richard Axel1,*
1Department of Neuroscience and the Howard Hughes Medical Institute, College of Physicians and Surgeons, Columbia University,
New York, NY 10032, USA
2Present address: Center for Interdisciplinary Research in Biology (CIRB) Colle
` ge de France 11, place Marcelin Berthelot,
75231 Paris Cedex 05, France
*Correspondence: ra27@columbia.edu
DOI 10.1016/j.cell.2011.07.041
SUMMARY
Anatomic and physiologic studies have suggested

a model in which neurons of the piriform cortex receive convergent input from random collections of
glomeruli. In this model, odor representations can
only be afforded behavioral significance upon experience. We have devised an experimental strategy
that permits us to ask whether the activation of an
arbitrarily chosen subpopulation of neurons in piriform cortex can elicit different behavioral responses
dependent upon learning. Activation of a small subpopulation of piriform neurons expressing channelrhodopsin at multiple loci in the piriform cortex, when
paired with reward or shock, elicits either appetitive or aversive behavior. Moreover, we demonstrate
that different subpopulations of piriform neurons expressing ChR2 can be discriminated and independently entrained to elicit distinct behaviors. These
observations demonstrate that the piriform cortex
is sufficient to elicit learned behavioral outputs in
the absence of sensory input. These data imply that
the piriform does not use spatial order to map odorant identity or behavioral output.
INTRODUCTION
Olfactory sensory systems transmit information to the brain
where it is processed to create an internal representation of
odors in the external world. This internal representation must
then be translated into appropriate behavioral output. Sensory
systems do not passively represent the external world. Rather,
they actively interpret features of the world that are combined
in higher cortical centers to construct meaningful sensory representations. In vision, touch, and hearing, features central to
perception are topographically ordered in the sense organ.
These features, which include spatial location or sound frequency, are continuously variable in at least one dimension in
the external world. The persistence of topographic order from
the periphery to primary sensory cortices (Marshall et al., 1941;
Talbot and Marshall, 1941; Woolsey and Walzl, 1942) has led
to the view that early sensory processing is mediated by developmentally programmed neural circuits. In contrast, olfactory
features cannot be meaningfully represented along continuous
dimensions in the physical world and are not topographically
organized in the olfactory sensory epithelium (Ressler et al.,
1993; Vassar et al., 1993). Olfactory information from the nose
is transmitted to the olfactory bulb and then to higher centers.
The vast majority of odors drive behavior only after learning,
but the brain regions responsible for these learned behaviors
remain elusive. The piriform cortex receives extensive input
from the bulb (Price and Powell, 1970) and projects to areas
implicated in behavioral output (Schwabe et al., 2004), posing
the question as to whether the piriform is a substrate for olfactory
learning. We have developed an experimental strategy that
permits us to ask whether exogenous activation of the same subpopulation of piriform neurons can be entrained to elicit distinct
behavioral responses depending on the learning paradigm.
Olfactory perception is initiated by the recognition of odorants
by a large repertoire of receptors in the sensory epithelium. Individual sensory neurons in mice express only one of 1500 different receptor genes (Buck and Axel, 1991; Godfrey et al.,
2004; Zhang and Firestein, 2002). An odorant can interact with
multiple distinct receptors resulting in the activation of an ensemble of sensory neurons (Araneda et al., 2004; Malnic et al.,
1999; Oka et al., 2006). Discrimination among odorants then
requires that the brain determine which of the sensory neurons
have been activated by a given odorant. Neurons expressing
a given receptor are distributed within zones of the epithelium
but project with precision to two spatially invariant glomeruli in
the olfactory bulb (Mombaerts et al., 1996; Ressler et al., 1993,
1994; Vassar et al., 1994, 1993). Thus a transformation in the
representation of olfactory information is apparent in the bulb
where the dispersed population of active neurons in the sense
organ is consolidated into a discrete spatial map of glomerular
activity.
If an odorant activates a unique ensemble of glomeruli, then
the recognition of an odor requires integration of information
from multiple glomeruli by higher olfactory centers. The projection neurons of the olfactory bulb, mitral and tufted cells, extend
an apical dendrite into a single glomerulus and send axons to
several telencephalic areas, including a significant input to the
piriform cortex (Price and Powell, 1970). Anatomic tracing

reveals that axonal projections from individual glomeruli diffusely
innervate the piriform and do not exhibit the segregated pattern
of the bulb (Ghosh et al., 2011; Miyamichi et al., 2011; Sosulski
et al., 2011). In accord with this anatomy, electrophysiological
and optical imaging studies demonstrate that individual odorants activate subpopulations of neurons distributed across the
piriform without apparent spatial preference (Illig and Haberly,
2003; Poo and Isaacson, 2009; Rennaker et al., 2007; Stettler
and Axel, 2009; Sugai et al., 2005; Zhan and Luo, 2011). Moreover, individual piriform neurons respond to multiple structurally
dissimilar odorants, and no similarity in response properties
is observed among neighboring neurons (Stettler and Axel,
2009). Therefore, piriform representations differ from those of
other neocortical sensory areas where cells are tuned for stimulus features and show macroscopic spatial patterning (Hubel
and Wiesel, 1959; Mountcastle et al., 1957).
These observations are consistent with a model in which individual piriform cells receive convergent input from random
collections of glomeruli (Stettler and Axel, 2009). In this model,
odor representations in piriform can only be afforded behavioral
significance upon learning. Furthermore, the piriform cortex is
anatomically poised to mediate learned olfactory behaviors. Piriform output to the amygdala, basal ganglia, and hippocampus
(Schwabe et al., 2004) may link sensory representations to
behavioral output. In this study, we demonstrate that exogenous
activation of an arbitrarily chosen ensemble of piriform neurons
can elicit multiple behaviors of contrasting valence dependent
on learning. Thus the piriform cortex can mediate learned behaviors in the absence of sensory input. Ensembles composed of
less than 500 piriform neurons can be entrained to elicit different
behaviors. Moreover, aversive behavior can be elicited by these
ensembles independent of their position across the piriform, indicating that the piriform does not use spatial order to map either
olfactory input (Ghosh et al., 2011; Miyamichi et al., 2011; Sosulski et al., 2011; Stettler and Axel, 2009) or behavioral output.
RESULTS
results in the activation of ChR2-expressing neurons with millisecond precision in awake, behaving animals (Aravanis et al.,
2007; Boyden et al., 2005). In initial experiments, channelrhodopsin expression was driven by the human Synapsin1 promoter
(Kugler et al., 2003) in both excitatory and inhibitory cells in the
piriform cortex (Figure 1A). In a second set of experiments,
a ChR2 gene in the reverse orientation was flanked by pairs of
loxP sites, such that expression of ChR2 was dependent upon
Cre recombinase (Atasoy et al., 2008). Injection of this virus
into the piriform cortex of Emx1-IRES-Cre mice in which Cre recombinase is restricted to excitatory neurons (Gorski et al.,
2002), resulted in ChR2 expression in pyramidal neurons but
not inhibitory interneurons (Figure 1B). These two strategies for
ChR2 delivery resulted in abundant expression of ChR2 in
over 50% of the neurons in an injection site that ranged from
500 mm-1000 mm in diameter (Figures 1A and 1B).
We have achieved sparser distribution of active neurons
by coinjection of high-titer Cre-dependent ChR2 virus with
a range of dilutions of lentivirus encoding Cre. Under these
conditions (dual virus strategy), only a small subset of the cells
were coinfected with both viruses and high-level ChR2 expression was therefore sparse, with about 10% (8.25 0.33%,
n = 3) of neurons expressing ChR2 at the injection site (Figure 1C). Analysis of c-Fos expression, a marker of neuronal activation (Morgan and Curran, 1991), revealed that all three genetic
approaches to effect channelrhodopsin expression resulted in
robust neural activation upon exposure to light (Figures 1A, 1B,
and 1D). We observed a threefold (3.06 0.46, n = 4) increase
in the number of c-Fos+ cells at injection sites using the first
two expression strategies that produced dense populations of
ChR2-expressing neurons. In the third expression strategy that
generated sparse populations of ChR2-expressing neurons,
the coexpression of nuclear Cherry (Figure 1C) allowed us to
identify the incidence of c-Fos expression among ChR2+ cells.
We observed that about 40% of cells expressing ChR2 also expressed c-Fos (37.76 11.7%, n = 6), whereas 5% of cells
were positive for c-Fos expression in uninjected control hemispheres (6.09 0.36%, n = 3).
Expression of Channelrhodopsin in an Ensemble

of Piriform Neurons
We introduced channelrhodopsin (ChR2), a light-activated
cation channel (Aravanis et al., 2007; Boyden et al., 2005), into
a small subpopulation of neurons in the piriform cortex of the
mouse. In these mice, light should activate the ensemble of
ChR2-expressing neurons, independent of mitral cell input. Activation of the ensemble of piriform neurons with light served as
a conditioned stimulus (CS) that was paired with either an aversive or appetitive unconditioned stimulus (US). We then asked
whether subsequent exposure to the CS alone (light) would elicit
a behavioral response consistent with the conditioning paradigm. This experimental strategy permitted us to ask whether
the same random ensemble of neurons was capable of eliciting
different behaviors depending on the nature of the unconditioned stimulus.
Channelrhodopsin was expressed in layer 2 and 3 piriform
neurons by infection with three different genetic variants of lentivirus (Dittgen et al., 2004). Photostimulation via an optical fiber
An Ensemble of Neurons Trained to Elicit Aversive

Behavior
In initial experiments, we asked whether photostimulation of
ChR2-expressing neurons in the pirform cortex could serve as
a conditioned stimulus, eliciting avoidance behavior in an aversive
conditioning paradigm. We adopted a conditioning paradigm
(Yan et al., 2008) to discern whether photostimulation of a subpopulation of piriform neurons could recapitulate the ability of
odor to elicit aversive behavior. Training was carried out in a
custom-designed rectangular arena in which the animal was
allowed to move freely. Foot shock was applied only to the side
where the animal was located at the time of CS presentation,
allowing the animal to flee from the aversive stimulus by running
toward the opposite side of the arena (Figure 2A). The CS-US
presentation was randomly applied to either side. After two training sessions (one session = 10 pairings of the CS with US), the
animals were returned to the same arena to determine whether
the CS alone, photostimulation of ChR2-expressing neurons,
was sufficient to elicit flight behavior. In mice expressing ChR2,
Figure 1. Expression of ChR2 in Layer 2 and 3 of Piriform Cortex after

Injection with Different Variants of Lentivirus Encoding ChR2
(A) Lentivirus carrying ChR2 fused to a fluorescent reporter (XFP = Cherry or
EYFP) under control of the hSynapsin1 promoter was stereotactically injected
into the piriform. The hSynapsin1 promoter drives ChR2:XFP expression in
both excitatory and inhibitory neurons. Coronal sections through the injection
site reveal expression of ChR2:XFP (red) in dense populations of layer 2 and 3
piriform neurons. The labeled cells are shown at higher magnification on the
right. c-Fos expression after in vivo photostimulation is shown in green. NT,
Neurotrace (blue). Scale bar on left represents 50 mm, and on right represents
100 mm.
(B) Lentivirus carrying ChR2:XFP flanked by loxP sites (two different variants
indicated by colored triangles) and under control of the EF1 alpha promoter
was injected into the piriform of Emx1-IRES-Cre mice. ChR2:XFP (red) expression is restricted to dense populations of excitatory neurons in these mice.
(C) Lentivirus carrying ChR2:EYFP-IRES-nCherry (nuclear Cherry) flanked by
loxP sites and under control of the EF1alpha promoter was coinjected into
piriform with a second lentivirus carrying the hSynapsin1 promoter driving
we observed that photostimulation of piriform cortex served as an

effective CS, resulting in robust flight behavior (Figure 2B). In at
least 95% of the trials, mice exhibited flight behavior in response
to photostimulation after two training sessions (% flight behavior:
hSynapsin1 = 96.92 10.8%, n = 7; Emx1 = 100 0%, n = 4; dual
virus strategy with > 300 ChR2+ cells = 95.71 9.64%, n = 10).
This conditioned response was retained 5 days after training
(% flight behavior: 83.33 33.33%, n = 4, data not shown).
Photostimulation of ChR2-expresing piriform neurons in
CNG2A knock-out animals, which lack odor-evoked activity in
the main olfactory epithelium (Brunet et al., 1996), could be entrained to elicit conditioned flight behaviors (100 0%, n = 2,
data not shown). These experiments demonstrate that an ensemble of piriform neurons can elicit conditioned aversive
behavior in the absence of olfactory sensory input.
Animals that were not infected with virus (0 0%, n = 3) or
animals infected with lentivirus encoding EGFP (0 0%, n = 6)
in the absence of ChR2 failed to exhibit an aversive behavioral
response upon photostimulation (Figure 2B). In addition, mice expressing ChR2 in piriform neurons but not subjected to photostimulation during training (0 0%, n = 3), mice exposed to unpaired
CS and US presentation (0 0%, n = 4) and mice exposed to CS
and US in reverse order (0 0%, n = 3) did not exhibit flight
behavior (Figure 2B). We have also demonstrated that light
activation of ChR2-expressing ensembles did not elicit a behavioral preference without conditioning. Animals expressing ChR2
were placed in the central compartment of a three-chambered
arena and were photostimulated when the animal entered one
side chamber. Animals spent the same amount of time in each
chamber (Figure 2C), demonstrating that light activation of
ChR2-expressing neurons did not elicit approach or avoidance
behaviors ([+]Photostimulation = 36.3 10.46%, Middle =
29.04 7.77%, [ ]Photostimulation = 34.19 12.14%, n = 10).
We next asked whether photostimulation of ChR2+ ensembles
exhibited the properties of an odorant component in a mix.
Animals expressing ChR2 in piriform were trained using the aversive conditioning paradigm with a CS consisting of a mix of two
odorants (ethyl acetate and citronellol) plus photostimulation.
The simultaneous delivery of these three stimuli was paired
with a foot shock and animals were subsequently tested with
the complete CS or its components. Animals exhibited flight
behavior in response to the mixture of two odors, either odor
alone, as well as to light (Figure 2D) (Odorant mix + photostimulation = 100 0%, Odorant mix = 78.33 20.21%, Odorant
component = 79.63 22.43, Photostimulation = 80.57
17.33%, Air = 0 0%, n = 3). These results indicate that light activation of an ensemble of ChR2-expressing neurons exhibit properties similar to a component of odorant mix. The inclusion of
odorant during training does not interfere with an animals ability
to generalize a conditioned response to the photostimulation
component, nor does the inclusion of photostimulation interfere
with generalized responses to odorant components.
Cre:EGFP. This dual virus strategy was used to generate sparse labeling of
piriform neurons. nCherry (red) labels the cell bodies whereas EYFP (green)
labels both cell bodies and processes.
(D) c-Fos expression (blue) after in vivo photostimulation for the same animal
shown in (C).
Figure 2. Ensembles of ChR2-Expressing

Piriform Neurons Entrained to Elicit Aversive Behavior
(A) Schematic of the apparatus used for the
aversive conditioning paradigm. During training,
photostimulation of ChR2-expressing neurons
in piriform, the conditioned stimulus (CS), was
paired with foot shock, the unconditioned stimulus (US), applied only on the side of the arena
where the animal was located at the time of photostimulation. The animals escaped foot shock by
running to the opposite side.
(B) The percentage of trials in which animals exhibited flight behavior in response to the CS alone
during the testing phase. hSynapsin1 = ChR2
expression driven from the human Synapsin1
promoter (n = 7); Emx1 = ChR2 expression driven
by the Emx1 promoter (n = 4); dual virus = ChR2
expression generated by coinfection of Cre and
Cre-dependent ChR2 viruses (animals with > 300
ChR2+ neurons, n = 10); ( ) virus = no viral injection
(n = 3); EGFP = virus encoding EGFP but not ChR2
was injected into piriform (n = 6); ( ) photostimulation = ChR2 expression driven by the
hSynapsin1 promoter without photostimulation
during training (n = 3); Unpaired CS/US = ChR2
expression was driven from the human Synapsin1
promoter but foot shock application was not
contingent upon photostimulation (i.e., equal
numbers of CSs and USs were presented in random order with delays always exceeding 1 min,
n = 4); Reversed CS/US = ChR2 expression was
generated by the dual virus strategy but foot shock
application preceded photostimulation (n = 3).
(C) The percentage of time naive ChR2-expressing
animals spent in each chamber during a 5 min (n = 2) or 10 min (n = 8) testing period. One of the side chambers in a three-chambered arena was chosen as the (+)
photostimulated compartment. Photostimulation was applied only when the animals entered the (+) photostimulated chamber. Training was not involved. ChR2
was densely expressed using the dual virus strategy.
(D) The percentage of trials in which ChR2-expressing animals (n = 3) exhibited flight behavior in response to a complete multi-component CS or its components
after training in the aversive conditioning paradigm. The complete CS was an odorant mix (ethyl acetate + citronellol) codelivered with photostimulation. Odorant
component was either ethyl acetate or citronellol. ChR2 was densely expressed using the dual virus strategy.
We have determined the number of ChR2-expressing cells

required to elicit conditioned aversive behavior. It was possible
to titrate the frequency of ChR2-expressing cells in mice in which
the ChR2 expression was dependent upon dual infection with
virus encoding Cre-dependent ChR2 and a second Creexpressing virus. Animals with fewer than 200 piriform neurons
expressing ChR2 failed to exhibit a behavioral response to the
photostimulation after training (4.71 7.39%, n = 6) (Figure 3A).
Mice with about 300 ChR2-expressing cells exhibited flight
behavior in 30% of the trials (28.56 20.19%, n = 4). Mice with
more than 500 ChR2-expressing cells exhibited this behavior
in 95% of the trials (93.1 13.79%, n = 4). This behavioral scaling
with cell number was also observed when we scored the
distance run in response to photostimulation (Figure 3B).
Each odor activates about 100,000 neurons distributed across
the piriform cortex without spatial preference (Stettler and Axel,
2009). ChR2-expressing neurons, however, localize to restricted
domains that occupy less than 10% of the piriform (0.35 to
1.0 mm in diameter). We have demonstrated that each of 18
independent ensembles at different locations across about
30% of piriform is capable of eliciting aversive behavior (Fig-
ure 3C and Table S1 available online). The observation that aversive behavior can be entrained at multiple loci distributed across
the piriform indicates that valence of behavioral output is not
spatially segregated in the piriform cortex.
We have compared the efficiency with which odor and photostimulation served as a conditioned stimulus to elicit flight
behavior. Pairing of odor exposure with foot shock resulted in
a consistent aversive response to odor alone after 10 CS-US
pairings (10.25 0.96 pairings, n = 4). Approximately twice as
many pairings were required to elicit flight behavior in mice expressing ChR2 in a subpopulation of piriform neurons (18
4.02 pairings, n = 11) (Figure 3D). Thus, the activation of an
ensemble of 500 piriform neurons approaches the efficacy of
odor activation of 100,000 neurons (Stettler and Axel, 2009), in
eliciting conditioned aversion.
An Ensemble of Neurons Trained to Elicit Appetitive
Behavior
We next asked whether the photostimulation of an ensemble of
piriform neurons expressing ChR2 could elicit appetitive behavioral responses if paired with a rewarding US. We modified
Figure 3. Efficiency of ChR2-Expressing Piriform Neurons in Eliciting Conditioned Behavior

(A) Relationship between the number of ChR2-expressing neurons and incidence of flight behavior in the aversive conditioning paradigm. The number of ChR2+
neurons was determined and correlated with the percentage of trials in which animals exhibited flight behavior for mice expressing ChR2 using the dual virus
strategy.
(B) Relationship between the number of ChR2-expressing neurons and distance traveled in the aversive conditioning paradigm. Same animals as in (A).
(C) Spatial distribution of conditioned ensembles. The centers of ChR2-expressing ensembles are mapped on a schematic showing the borders of the piriform
cortex for mice expressing ChR2 using the dual virus strategy and trained in the aversive conditioning behavioral paradigm. The borders of the piriform were
drawn by referring to the Paxinos atlas. Only animals with > 300 ChR2+ neurons are included. Percent flight behavior for each injection site is documented in
Table S1.
(D) Comparison of the number of CS-US pairings required for the onset of flight behavior in response to the CS alone in the aversive conditioning paradigm when
the CS was either an odorant (n = 4) or photostimulation of ChR2+ neurons (hSynapsin1: n = 2, Emx1: n = 2, dual virus with > 300 ChR2+ neurons: n = 7).
(E) Comparison of the number of blocks of trials required to reach a fraction of correct licks (# of licks following CS+ / total # of licks) exceeding 0.7 for two
consecutive blocks in the appetitive go/no go discrimination assay when the CS was either an odorant (n = 6) or photostimulation of ChR2+ neurons (dual virus
with > 300 ChR2+ neurons: n = 6).
a go/no-go odor discrimination assay (Bodyak and Slotnick,

1999) in which water-restricted mice were exposed to two odors,
one of which was followed by a water reward (Figure 4A, top).
During this pre-training period, mice learned to sample the
odor stimuli and lick only in response to the rewarded odor
(data not shown). This behavioral paradigm was then modified
such that water-restricted mice received a water reward only
after photostimulation of ChR2-expressing piriform neurons
(CS+) but not in the absence of photostimulation (CS-) (Figure 4A,
bottom). Both the CS+ and CS- were accompanied by a pulse of
air to mimic the pre-training condition. After switching from odor
to photostimulation, mice initially licked in anticipation of water
reward upon presentation of either the CS+ or CS- (Figures 4C
and 4D). As the trials progressed, animals expressing ChR2
reliably learned to lick after photostimulation and suppressed
licking in its absence (fraction correct licks (number of licks
following CS+ / total number of licks) = 0.83 0.07, n = 7) (Figures
4B4D). Control mice that did not express ChR2 continued to
perform at chance levels (fraction correct licks = 0.51 0.04,
n = 7) (Figures 4B and 4C). When photostimulation served as
the CS+, approximately twice as many blocks of trials were
required to elicit robust conditioning compared with odor

(odorant = 14.67 9.58 blocks, n = 6; photostimulation =
32.67 16.85 blocks, n = 6) (Figure 3E). Thus the activation of
an arbitrarily chosen ensemble of piriform neurons could be
entrained to drive appetitive as well as aversive behavioral
responses.
We also demonstrated that an ensemble of ChR2-expressing
neurons in piriform could be entrained to a socially rewarding
unconditioned stimulus. We designed a behavioral paradigm
in which males were exposed to odor in the presence or
absence of a female (Figure 5A, left). Training was performed in
a three-chambered arena housing a female and odor (CS+) on
one side and odor alone (CS-) on the other side. A male was
introduced to the middle chamber and allowed to freely explore
the arena. As previously described, the male spent most of
the training time exploring the female (Nadler et al., 2004) (data
not shown). After training, the male was returned to the same
arena that now contained only the CS+ and CS- odors. We
observed that males spent two to threefold more time in the
chamber with the CS+ odor than in the other chambers (CS+
chamber = 50.76 5.31%; Middle chamber = 22.87 4.84%;
Figure 4. Ensembles of ChR2-Expressing Piriform Neurons Entrained to Elicit Appetitive Behavior

(A) Mice expressing ChR2 using the dual virus strategy were trained in an appetitive behavioral conditioning paradigm. Mice pre-trained to sample and lick only in
response to a rewarded odor (CS+) subsequently received a water reward after photostimulation of ChR2-expressing neurons (CS+) but not in the absence of
photostimulation (CS-). The CS+ and CS- were accompanied by a pulse of air to cue discrimination. Each training block consisted of 10 CS+ and 10 CS- trials.
(B) The average fraction of correct licks over the last three training blocks for ChR2-expressing animals (Emx1: n = 1, dual virus with > 300 ChR2+ neurons: n = 6)
and control animals (in which EGFP but not ChR2 was expressed or a Cre-dependent ChR2 virus was injected without a second Cre-expressing virus) trained
using the same paradigm that included photostimulation (n = 7). Fraction correct licks = number of licks following CS+ / total number of licks.
(C) Performance plotted as the fraction correct licks per block number for a ChR2-expressing mouse using the dual virus strategy and a control mouse in which
EGFP but not ChR2 was expressed. Start of the training session on each day in (C) and (D) is marked with an arrow.
(D) Same data for the ChR2-expressing mouse shown in (C) plotted as the number of licks following the CS+ and CS-. The decrease in licks at the end of the first
training session is typical and is likely due to the animal reaching satiety.
CS- chamber = 26.37 3.17%, n = 6) (Figure 5A, right). These data

demonstrate that odor can be associated with a socially
rewarding US to elicit conditioned approach behavior.
We then asked whether an ensemble of ChR2-expressing piriform neurons could also be entrained to a socially rewarding US
by replacing the CS+ odor with photostimulation during training
(Figure 5B, left). In this paradigm, photostimulation was applied
when males investigated the female. Testing was performed in
the absence of a female and photostimulation was delivered in
only one of the chambers (CS+ chamber). The percentage of
time trained males spent in the CS+ chamber was threefold
higher than in the other chambers (Figure 5B, right) (CS+
chamber = 57.6 0.87%; Middle chamber = 17.05 7.07%;
CS- chamber = 25.34 6.41%, n = 3), demonstrating that activa-
tion of ChR2+ neurons can be associated with a socially

rewarding US to elicit a learned approach behavior. This behavioral paradigm extends the repertoire of appetitive behaviors that
can be elicited by an arbitrarily chosen ensemble of neurons in
the piriform.
The Same Ensemble of Neurons Can Elicit Aversive
and Appetitive Behavior
If the representation of odor in the piriform results from the
random convergence of glomerular inputs, then its valence cannot be developmentally programmed and is likely to be imposed
by experience. Consistent with this reasoning, the same odor
can elicit appetitive or aversive responses dependent upon
learning (Abraham et al., 2004; Yan et al., 2008). We therefore
elicited freezing, another form of aversive behavior, but failed

to produce flight behavior (data not shown). These experiments
demonstrate that a given ensemble of piriform neurons can be
sequentially entrained to elicit appetitive and aversive behavioral
responses. After the sequential training, the aversive behavior
appears to dominate the response to photostimulation.
Figure 5. Ensembles of ChR2-Expressing

Entrained to a Socially Rewarding Stimulus
Piriform
Neurons
(A) Entrainment of an odorant with a social reward. A male was trained in

a three-chamber arena, in which the CS+ odor was paired with a female in
a randomly selected side chamber (left). The other side chamber contained the
CS odor without a female. For testing, the animal was returned to the same
arena with side chambers containing only the CS+ and CS odors. On right, the
percentage of time animals spent in each chamber during a 5 min testing
period is plotted for when CS+ and CS were odors (n = 6). CS+, chamber with
CS+ odor; middle, middle chamber; CS , chamber with CS odor.
(B) Entrainment of ChR2-expressing ensembles with a social reward. During
training, photostimulation was applied when the males actively investigated
the female in a randomly selected side chamber (left). Upon testing in the
absence of a female, photostimulation was delivered in one of the side
chambers (CS+ chamber) but not the other (CS chamber). On right, the
percentage of time animals spent in each chamber during a 5 min testing
period when the CS+ was photostimulation is plotted for animals expressing
ChR2 (ChR2, n = 3) and for control animals without ChR2 (n = 3). ChR2 was
densely expressed using the dual virus strategy.
asked whether the same population of ChR2-expressing neurons in piriform could be entrained to sequentially elicit behaviors
of different valence (Figure 6A). Mice expressing ChR2 in piriform
were first trained in the appetitive paradigm and consistently exhibited conditioned licking responses upon photostimulation
(Figure 4B). The animals were then conditioned in the aversive
foot shock paradigm and now displayed robust flight behavior
(Figure 6B). Sequentially trained animals acquired the aversive
behavior as quickly as naive animals (data not shown) and exhibited flight behavior in 90% of the trials (ChR2+ animals =
88.57 25.55%, n = 5; control animals = 3.33 8.16%, n = 6).
We next asked whether the response to photostimulation in
these sequentially trained animals was context-dependent.
Sequentially trained animals were returned to the appetitive
conditioning context. Upon photostimulation, they no longer exhibited appetitive responses in anticipation of water reward.
Rather, the level of licking to the CS+ approached that of the
CS- (Figure 6C). Moreover, photostimulation occasionally
Distinct Ensembles of Active Neurons

Can Be Discriminated
Distinct odors activate unique ensembles of neurons in the piriform cortex (Stettler and Axel, 2009) and can be linked by conditioning to different behavioral responses (Abraham et al., 2004;
Yan et al., 2008). We have therefore asked whether different
subpopulations of piriform neurons expressing ChR2 could be
discriminated and independently entrained to elicit distinct
behaviors. Lentivirus encoding ChR2 was injected into the piriform cortex of each hemisphere to generate two anatomically
distinct populations of neurons that could be independently
photostimulated by two optical fibers. Stimulation of one ensemble (CS1) was paired with shock on the side of the arena
where the animal received the photostimulation. As described
earlier, this resulted in robust flight to the opposite side of the
training arena that was free of shock. Stimulation of the second
ensemble (CS2) was paired with shock to the opposite side of
the arena and the mice remained within the vicinity of the site
of CS presentation (Figure 7A, left). Thus, two different ensembles of ChR2-expressing neurons could be behaviorally discriminated: CS1 elicited reliable flight whereas CS2 resulted in
a stationary behavioral response (% flight behavior: CS1 =
100 0% and CS2 = 2.86 6.39%, n = 5) (Figure 7B).
We then asked whether each ensemble retained the potential
to elicit both behavioral responses. We trained the mice in a
reversal-learning paradigm in which the shock contingency
was switched between the two ensembles (Figure 7A, right:
Reversal Learning). The CS2, which initially elicited stationary
behavior, resulted in flight after reversal learning indicating that
both ChR2-expressing ensembles were capable of eliciting flight
behavior after appropriate training (CS2 = 85.71 0%, n = 3) (Figure 7C). Immediately upon reversal, the CS1 that initially elicited
flight continued to produce aversive behavior. However, upon
further training, animals learned that flight resulted in shock and
CS1 ultimately elicited stationary behavior (% flight behavior:
CS1 = 14.2 14.29%, n = 3) (Figure 7C). These observations
demonstrate that different ensembles of ChR2-expressing
neurons in the piriform can be discriminated and can be entrained
to elicit distinct behavioral outputs.
DISCUSSION
We have devised an experimental strategy that permits us to ask
whether the activation of an arbitrarily chosen subpopulation of
neurons in piriform cortex can elicit different behavioral responses dependent upon experience. Activation of a small subpopulation of as few as 300 neurons at multiple loci in the piriform
cortex, when paired with different unconditioned stimuli elicits
either appetitive or aversive behavior. Moreover, we demonstrate that different subpopulations of piriform neurons expressing ChR2 can be discriminated and independently entrained to
Figure 6. The Same Ensemble of ChR2-Expressing Neurons Can Be Entrained to Elicit Appetitive and Aversive Behaviors
(A) A schematic of the sequential training of ChR2-expressing animals to produce appetitive and aversive behaviors.
(B) A subset of mice shown in Figure 4B, which were trained in an appetitive water reward behavior, was subsequently trained in the aversive foot shock paradigm.
The percentage of trials in which animals exhibited flight behavior in response to photostimulation alone during the testing phase is plotted for ChR2-expressing
(Emx1: n = 1, dual virus with > 300 ChR2+ neurons: n = 4) and control animals (n = 6).
(C) The average lick number over the last three training blocks of sequentially trained animals (from Figure 4B and Figure 6B) before and after aversive conditioning
with the same ensemble. ChR2 before aversive: number of licks following CS+ and CS- for ChR2-expressing mice during initial appetitive conditioning (CS+ =
58.68 9.76 licks and CS- = 14.43 9.99 licks, n = 5). ChR2 After Aversive: number of licks following CS+ and CS- for these mice after sequential appetitiveaversive conditioning (CS+ = 12.81 15.96 licks and CS- = 11.39 8.21 licks, n = 5). Control After Aversive: number of licks following CS+ and CS- for control
animals after sequential appetitive-aversive conditioning (CS+ = 35.58 13.05 licks and CS- = 34.29 4.94 licks, n = 5).
elicit distinct behaviors. Thus, an experimentally generated

network comprised of a small ensemble of piriform neurons, activated in the absence of sensory input, is sufficient to elicit one of
multiple, learned behavioral outputs.
Each odorant activates 3%15% of the neurons distributed
across the piriform cortex without spatial preference (Stettler
and Axel, 2009). Photostimulation of as few as 300 spatially
localized neurons in piriform, about 0.5% the number of neurons activated by odor, is capable of eliciting both appetitive
and aversive behaviors. Moreover, we demonstrate that photostimulation of neuronal ensembles at several positions across
the piriform is capable of eliciting aversive behavior, indicating
that valence is not spatially segregated in the piriform. Taken
together, these observations imply that the piriform cortex
does not use spatial order to map olfactory input (Ghosh et al.,
2011; Miyamichi et al., 2011; Sosulski et al., 2011; Stettler and
Axel, 2009) or behavioral output.
How does the same ensemble of piriform neurons elicit behavioral outputs of different valence? In one model, each neuron
within an ensemble connects with multiple, different behavioral
outputs and learning reinforces only one of these outputs to
assure an appropriate response. Alternatively, different subsets
of neurons within the ensemble may be connected with distinct

behavioral outputs and learning will enhance the output of only
one subset of neurons. If the piriform is comprised of subsets
of neurons dedicated to distinct behavioral outputs, our data
indicate that these neurons do not reside within gross, spatially
defined domains. We have not determined the brain regions
responsible for these behavioral outputs. Piriform projects to
multiple downstream areas, including amygdala, tubercle and
prefrontal cortex that have been implicated in motivated
behavior. Piriform also sends recurrent projections to the olfactory bulb that arborize in the granule cell layer (Shepherd,
2004). It remains possible that this feedback also participates
in eliciting the observed behaviors.
This experimental scenario may provide insight into the neural
processing that transforms olfactory sensory input into behavioral output. In the piriform, projections from individual glomeruli
are distributed throughout the cortex without apparent topographic order (Ghosh et al., 2011; Miyamichi et al., 2011; Sosulski et al., 2011). Individual odors activate a distributed subpopulation of neurons across the cortex without spatial preference
(Illig and Haberly, 2003; Poo and Isaacson, 2009; Rennaker
et al., 2007; Stettler and Axel, 2009; Sugai et al., 2005; Zhan
Figure 7. Distinct Ensembles of ChR2-Expressing

Piriform Neurons Can Be Entrained to Elicit
Different Behaviors
(A) A schematic of the apparatus used for the conditioning
paradigm. Initially, stimulation of one ensemble (CS1) was
paired with shock on the side of the arena where the
animal received the photostimulation, and stimulation of
the second ensemble (CS2) was paired with shock to the
opposite side of the arena. For a reversal-learning paradigm, the shock contingency was switched between the
two ensembles (reversal learning).
(B) The percentage of trials in which animals exhibited
flight behavior in response to CS1 and CS2 after training
with the CS-shock contingencies described in (A), left
(hSynapsin1: n = 1, dual virus: n = 4).
(C) The percentage of trials in which flight behavior was
elicited by the CS1 and CS2 for a subset of animals shown
in (B) after they were subsequently trained with reversed
CS-shock contingencies described in (A), right (reversal
learning, hSynapsin1: n = 1, dual virus: n = 2).
and Luo, 2011). One model of piriform processing consistent

with these anatomic and physiologic observations invokes the
random convergence of a combination of excitatory inputs
from multiple mitral cells onto piriform neurons (Stettler and
Axel, 2009). In this model, a given odor will activate a different
ensemble of piriform neurons in different individuals. However,
in an individual, a given odor will consistently activate the
same ensemble and this representation will acquire coherence
to dictate a specific behavioral output. This model is supported
by recent experiments demonstrating that a cell in piriform can
be activated by stimulating a random combination of glomeruli
(Davison and Ehlers, 2011). Despite this evidence, it remains
possible that piriform odorant representations will reveal an
undiscovered order.
If the connections from bulb to cortex are indeed random, then
the quality of an odorant or its valence in the piriform must be
imposed by experience. This experience-dependent relation
between representation and valence is observed upon light activation of an arbitrarily chosen ensemble of piriform neurons: a
given ensemble of neurons can elicit different behaviors dependent upon the conditioning paradigm. However, we cannot conclude that the ChR2 network we have generated recapitulates
the neural processing of the circuit elicited by an odorant. An
odorant representation comprises far more neurons in piriform
than does the ChR2 network we have generated. Moreover,
odorants activate the bulb that in turn projects to several other
brain areas in addition to piriform.
Earlier models have been elaborated in which features of the
external world are transmitted from the sense organ to the cortex
via ordered, genetically determined pathways (Changeux et al.,
1973; Edelman, 1987). Topologic order is then reorganized
at higher centers, creating a degenerate network with variability
in cortical connections among individuals. This degenerate
network is then subject to selection by experience over the life of an organism. Selection reinforces connections from neurons that represent
sensory objects of behavioral significance. The
ability to entrain populations of piriform neurons
to elicit specific behavioral responses is in accord with these
models.
A similar conceptual organization in which the random convergence of entorhinal inputs creates a distributed ensemble may
also be operative for hippocampal place cells. Place cells in
the hippocampus, a three-layered cortical structure like piriform,
exhibit no apparent relationship between their positions in brain
space and their firing field in the external world (OKeefe et al.,
1998; Redish et al., 2001). As a consequence, the spatial map
in the hippocampus is likely to differ in different individuals
residing in the same environment. Moreover, place cells remap:
they alter their firing properties in response to changes in spatial
environment (Colgin et al., 2008). This suggests models in which
inputs to individual place cells are randomly chosen during
development such that a given location is represented by a
distributed ensemble of active neurons.
The exogenous activation of neurons in other sensory cortices, either by microstimulation (Doty, 1969; Murphey and
Maunsell, 2007; Yang et al., 2008) or photostimulation (Huber
et al., 2008), has been shown to elicit behaviors following
training. Microstimulation of loci within visual, auditory, and
somatosensory cortex suggests that neuronal activation at
many neocortical levels of sensory processing is capable of influencing perceptual tasks. In more recent experiments, photostimulation of sparse ensembles of ChR2-expressing neurons in
somatosensory cortex was conditioned to drive appetitive
behavior (Huber et al., 2008). These sensory neocortices maintain topographic order that represents stimulus features in at
least one dimension across the cortex. As a consequence, microstimulation or photostimulation of a locus in these cortices
results in the activation of a topographically constrained subpopulation of neurons that is likely to encode specific features
of a sensory stimulus. In these experiments, focal activation
may exploit the underlying topographic organization of sensory

neocortices to elicit specific behavioral output. In the piriform,
features of an olfactory stimulus are not topographically organized (Ghosh et al., 2011; Miyamichi et al., 2011; Sosulski
et al., 2011; Stettler and Axel, 2009) and the ability of an arbitrarily
chosen ensemble of ChR2-expressing neurons to elicit specific
outputs cannot exploit an underlying spatial order. Rather, the
behavioral entrainment of an ensemble of ChR2+ neurons may
reflect the ability of piriform to elicit odor-evoked behavior by
activating a distributed ensemble of neurons without regard to
spatial order.
It may be argued that it is possible to elicit specific behaviors
by the activation of any collection of neurons in cortex independent of the underlying neural organization. The postulated ability
of any arbitrarily chosen ensemble to elicit learned behaviors
would reflect a striking, inherent property of neural populations
in the brain. Our data suggest that piriform cortex may exploit
this property to translate olfactory input into learned behavioral
output.
In vision, touch, and sound, features central to perception are
topographically ordered in the sense organ and this representation is maintained in the primary sensory neocortices (Marshall
et al., 1941; Talbot and Marshall, 1941; Woolsey and Walzl,
1942). Moreover, these features, such as spatial location and
sound frequency, are continuously variable in at least one dimension in the external world. A meaningful representation of a
sensory object, however, may require that these features are
combined at higher cortical centers. It is difficult to conceive of
a developmentally programmed strategy to encode the inordinately large number of complex stimuli that can be discriminated. A strategy involving selection from a random combination
of features could, however, accommodate the complex problem
of sensory discrimination. However, experience-dependent selection from randomness would be most apparent, not in primary
sensory cortex but at higher processing centers. In contrast,
olfactory features cannot be meaningfully represented along
continuous dimensions in the physical world and are not topographically organized in the piriform cortex (Ghosh et al., 2011;
Miyamichi et al., 2011; Sosulski et al., 2011; Stettler and Axel,
2009). In the olfactory system, features of an odorant may be
encoded by the receptors themselves such that the random
combination of these features is already apparent in primary
olfactory cortex.
If the connections from bulb to cortex are indeed random, then
the representation of an odorant or its valence in the piriform
must be imposed by experience. A small subset of odorants,
however, elicit stereotyped behaviors that are likely to be mediated by genetically-determined projections from the olfactory
bulb to other olfactory centers (Kobayakawa et al., 2007).
Spatially invariant projections from the olfactory bulb to cortical
amygdala implicate this structure in the generation of innate
behaviors (Miyamichi et al., 2011; Sosulski et al., 2011). This
bifurcation in the olfactory circuit in the mouse is analogous to
the architecture of the olfactory system in Drosophila despite
the six hundred million years of evolution that separates the
two organisms. In Drosophila, information from the antennal
lobe (olfactory bulb equivalent) bifurcates with one branch exhibiting spatially invariant projections to the lateral horn, a brain
region mediating innate olfactory behavior. A second branch

projects to the mushroom body, a structure that may receive
random convergent input and is required for learned olfactory
responses (Marin et al., 2002; Murthy et al., 2008; Wong et al.,
2002). Thus, innate, olfactory-driven behaviors are likely to
derive from determined neural circuits that result from Darwinian
selection over evolutionary time whereas learned behaviors may
be mediated by the selection and reinforcement of random
ensembles over the life of an organism.
Histochemistry
Immunofluorescence was performed on coronal sections of brain perfused
with 4% paraformaldehyde, following standard protocols. The prepared slices
were labeled with the following antibodies: chicken anti-GFP (Abcam,
ab5450), goat anti-c-Fos (Santa Cruz, sc-52-G), rabbit anti-c-Fos (Santa
Cruz, sc-7270), or rabbit anti-DsRed (Clontech, 632496).
Stereotaxic Injections
All procedures were carried out according to the approved protocols at
Columbia University. Wild-type C57 BL6/J, heterozygous Emx1-IRES-Cre or
homozygous CNG2A mice were injected with lentivirus carrying ChR2, Credependent ChR2, or Cre-dependent ChR2 mixed with lentivirus carrying Cre.
Aversive Behaviors
The conditioning paradigm consisted of 3-4 s of laser activation (photostimulation = 20 Hz with 25 ms pulses) followed immediately by a 0.7 mA foot-shock.
Photostimulation/shock pairings were spaced 3-4 min apart. Each of the two
training sessions consisted of 10 photosimulation/shock pairings, for a total of
20 pairings. The testing session was identical in set-up to the training sessions,
but only photostimulation was applied when the animal was located in either
end of the apparatus.
Appetitive Go/No-Go Discrimination Assay
Training and testing were performed using the Slotnick operant conditioning
paradigm and a liquid-dilution, eight-channel olfactometer (Knosys, Lutz,
FL). The animals were trained to discriminate between the photostimulation
of ChR2+ piriform neurons as CS+ (photostimulation = 20 or 30 Hz with
25 ms pulses, for 3 s) and absence of photostimulation as CS-. Both the CS+
and CS- were accompanied by a pulse of air to mimic the pre-training condition. The fraction correct licks were calculated as number of licks following the
CS+ / total number of licks.
Social Approach Behavioral Paradigm
Behavioral training and testing were carried out in a custom-built three-chambered arena. During training, a wire cage containing a female was placed in
one side chamber while an empty wire cage was placed in the opposite side
chamber. Placement of the female-containing cage was randomly selected
for each trial. Photostimulation was applied when the male actively investigated the female and lasted for total of 30 s per trial. A minimum of 10 trials
was completed, with an inter-trial interval of 3 min. During testing in the
absence of a female, photostimulation was delivered when the male was in
one of the arbitrarily chosen side chambers (CS+ chamber).
Supplemental Information includes Extended Experimental Procedures and
one table and can be found with this article online at doi:10.1016/j.cell.2011.
07.041.
ACKNOWLEDGMENTS
We thank Karl Deisseroth for ChR2 reagents; Nadia Propp for setting up lentiviral production method; Meredith Glinka, Justin Schwarz, and Dan Feng Mei
for help with behavioral assays; Monica Mendelsohn, Jennifer Kirkland, and
Nataliya Zabello for help with the mice; Anmo Kim for help with laser and olfactometer machines; Phyllis Kisloff for assistance in preparation of the manuscript; David J. Anderson, Larry Abbott, Thomas Jessell, Eric R. Kandel, and
members of the Axel lab for critical reading of the manuscript and discussions;
and Miriam Gutierrez for general laboratory support. This work was supported
by the Howard Hughes Medical Institute and the Mathers Foundation. G.B.C.
was supported by the Damon Runyon Cancer Research Foundation Postdoctoral Fellowship. A.F. was supported by long-term postdoctoral fellowships
from EMBO and the Human Frontiers Science Program.
Received: March 4, 2011
REFERENCES
Abraham, N.M., Spors, H., Carleton, A., Margrie, T.W., Kuner, T., and
Schaefer, A.T. (2004). Maintaining accuracy at the expense of speed: stimulus
similarity defines odor discrimination time in mice. Neuron 44, 865876.
Araneda, R.C., Peterlin, Z., Zhang, X., Chesler, A., and Firestein, S. (2004). A
pharmacological profile of the aldehyde receptor repertoire in rat olfactory
epithelium. J. Physiol. 555, 743756.
Aravanis, A.M., Wang, L.P., Zhang, F., Meltzer, L.A., Mogri, M.Z., Schneider,
M.B., and Deisseroth, K. (2007). An optical neural interface: in vivo control of
rodent motor cortex with integrated fiberoptic and optogenetic technology.
J. Neural Eng. 4, S143S156.
Atasoy, D., Aponte, Y., Su, H.H., and Sternson, S.M. (2008). A FLEX switch
targets Channelrhodopsin-2 to multiple cell types for imaging and long-range
circuit mapping. J. Neurosci. 28, 70257030.
Bodyak, N., and Slotnick, B. (1999). Performance of mice in an automated
olfactometer: odor detection, discrimination and odor memory. Chem. Senses
24, 637645.
Boyden, E.S., Zhang, F., Bamberg, E., Nagel, G., and Deisseroth, K. (2005).
Millisecond-timescale, genetically targeted optical control of neural activity.
Brunet, L.J., Gold, G.H., and Ngai, J. (1996). General anosmia caused by
a targeted disruption of the mouse olfactory cyclic nucleotide-gated cation
channel. Neuron 17, 681693.
Buck, L., and Axel, R. (1991). A novel multigene family may encode odorant
receptors: a molecular basis for odor recognition. Cell 65, 175187.
Changeux, J.P., Courrege, P., and Danchin, A. (1973). A theory of the epigenesis of neuronal networks by selective stabilization of synapses. Proc. Natl.
Acad. Sci. USA 70, 29742978.
Colgin, L.L., Moser, E.I., and Moser, M.B. (2008). Understanding memory
through hippocampal remapping. Trends Neurosci. 31, 469477.
Davison, I.G., and Ehlers, M.D. (2011). Neural circuit mechanisms for pattern
detection and feature combination in olfactory cortex. Neuron 70, 8294.
Dittgen, T., Nimmerjahn, A., Komai, S., Licznerski, P., Waters, J., Margrie,
T.W., Helmchen, F., Denk, W., Brecht, M., and Osten, P. (2004). Lentivirusbased genetic manipulations of cortical neurons and their optical and electrophysiological monitoring in vivo. Proc. Natl. Acad. Sci. USA 101, 1820618211.
Gorski, J.A., Talley, T., Qiu, M., Puelles, L., Rubenstein, J.L., and Jones, K.R.
(2002). Cortical excitatory neurons and glia, but not GABAergic neurons, are
produced in the Emx1-expressing lineage. J. Neurosci. 22, 63096314.
Hubel, D.H., and Wiesel, T.N. (1959). Receptive fields of single neurones in the
cats striate cortex. J. Physiol. 148, 574591.
Huber, D., Petreanu, L., Ghitani, N., Ranade, S., Hromadka, T., Mainen, Z., and
Svoboda, K. (2008). Sparse optical microstimulation in barrel cortex drives
learned behaviour in freely moving mice. Nature 451, 6164.
Illig, K.R., and Haberly, L.B. (2003). Odor-evoked activity is spatially distributed
in piriform cortex. J. Comp. Neurol. 457, 361373.
Kobayakawa, K., Kobayakawa, R., Matsumoto, H., Oka, Y., Imai, T., Ikawa, M.,
Okabe, M., Ikeda, T., Itohara, S., Kikusui, T., et al. (2007). Innate versus learned
odour processing in the mouse olfactory bulb. Nature 450, 503508.
Kugler, S., Kilic, E., and Bahr, M. (2003). Human synapsin 1 gene promoter
confers highly neuron-specific long-term transgene expression from an
adenoviral vector in the adult rat brain depending on the transduced area.
Gene Ther. 10, 337347.
Malnic, B., Hirono, J., Sato, T., and Buck, L.B. (1999). Combinatorial receptor
codes for odors. Cell 96, 713723.
Marin, E.C., Jefferis, G.S., Komiyama, T., Zhu, H., and Luo, L. (2002). Representation of the glomerular olfactory map in the Drosophila brain. Cell 109,
243255.
Marshall, W.H., Woolsey, C.N., and Bard, P. (1941). Observations on cortical
somatic sensory mechanisms of cat and monkey. J. Neurophysiol. 4, 124.
Miyamichi, K., Amat, F., Moussavi, F., Wang, C., Wickersham, I., Wall, N.R.,
Taniguchi, H., Tasic, B., Huang, Z.J., He, Z., et al. (2011). Cortical representations of olfactory input by trans-synaptic tracing. Nature 472, 191196.
Mombaerts, P., Wang, F., Dulac, C., Chao, S.K., Nemes, A., Mendelsohn, M.,
Edmondson, J., and Axel, R. (1996). Visualizing an olfactory sensory map. Cell
87, 675686.
Morgan, J.I., and Curran, T. (1991). Stimulus-transcription coupling in the
nervous system: involvement of the inducible proto-oncogenes fos and jun.
Annu. Rev. Neurosci. 14, 421451.
Mountcastle, V.B., Davies, P.W., and Berman, A.L. (1957). Response properties of neurons of cats somatic sensory cortex to peripheral stimuli.
J. Neurophysiol. 20, 374407.
Murphey, D.K., and Maunsell, J.H. (2007). Behavioral detection of electrical
microstimulation in different cortical visual areas. Curr. Biol. 17, 862867.
Murthy, M., Fiete, I., and Laurent, G. (2008). Testing odor response stereotypy
in the Drosophila mushroom body. Neuron 59, 10091023.
Nadler, J.J., Moy, S.S., Dold, G., Trang, D., Simmons, N., Perez, A., Young,
N.B., Barbaro, R.P., Piven, J., Magnuson, T.R., et al. (2004). Automated apparatus for quantitation of social approach behaviors in mice. Genes Brain
Behav. 3, 303314.
OKeefe, J., Burgess, N., Donnett, J.G., Jeffery, K.J., and Maguire, E.A. (1998).
Place cells, navigational accuracy, and the human hippocampus. Philos.
Trans. R. Soc. Lond. B Biol. Sci. 353, 13331340.
Oka, Y., Katada, S., Omura, M., Suwa, M., Yoshihara, Y., and Touhara, K.
(2006). Odorant receptor map in the mouse olfactory bulb: in vivo sensitivity
and specificity of receptor-defined glomeruli. Neuron 52, 857869.
Poo, C., and Isaacson, J.S. (2009). Odor representations in olfactory cortex:
sparse coding, global inhibition, and oscillations. Neuron 62, 850861.
Doty, R.W. (1969). Electrical stimulation of the brain in behavioral context.

Annu. Rev. Psychol. 20, 289320.
Price, J.L., and Powell, T.P. (1970). The mitral and short axon cells of the olfactory bulb. J. Cell Sci. 7, 631651.
Edelman, G. (1987). Neural Darwinism. The Theory of Neuronal Group

Selection (New York: Basic Books).
Redish, A.D., Battaglia, F.P., Chawla, M.K., Ekstrom, A.D., Gerrard, J.L., Lipa,
P., Rosenzweig, E.S., Worley, P.F., Guzowski, J.F., McNaughton, B.L., et al.
(2001). Independence of firing correlates of anatomically proximate hippocampal pyramidal cells. J. Neurosci. 21, RC134.
Ghosh, S., Larson, S.D., Hefzi, H., Marnoy, Z., Cutforth, T., Dokka, K., and
Baldwin, K.K. (2011). Sensory maps in the olfactory cortex defined by longrange viral tracing of single neurons. Nature 472, 217220.
Godfrey, P.A., Malnic, B., and Buck, L.B. (2004). The mouse olfactory receptor
gene family. Proc. Natl. Acad. Sci. USA 101, 21562161.
Rennaker, R.L., Chen, C.F., Ruyle, A.M., Sloan, A.M., and Wilson, D.A. (2007).
Spatial and temporal distribution of odorant-evoked activity in the piriform
cortex. J. Neurosci. 27, 15341542.
Ressler, K.J., Sullivan, S.L., and Buck, L.B. (1993). A zonal organization of
odorant receptor gene expression in the olfactory epithelium. Cell 73,
597609.
Vassar, R., Chao, S.K., Sitcheran, R., Nunez, J.M., Vosshall, L.B., and Axel, R.
(1994). Topographic organization of sensory projections to the olfactory bulb.
Cell 79, 981991.
Ressler, K.J., Sullivan, S.L., and Buck, L.B. (1994). Information coding in the
olfactory system: evidence for a stereotyped and highly organized epitope
map in the olfactory bulb. Cell 79, 12451255.
Vassar, R., Ngai, J., and Axel, R. (1993). Spatial segregation of odorant
receptor expression in the mammalian olfactory epithelium. Cell 74, 309318.
Schwabe, K., Ebert, U., and Loscher, W. (2004). The central piriform cortex:
anatomical connections and anticonvulsant effect of GABA elevation in the
kindling model. Neuroscience 126, 727741.
Shepherd, G.M. (2004). The Synaptic Organization of the Brain (New York:
Oxford University Press).
Sosulski, D.L., Lissitsyna Bloom, M., Cutforth, T., Axel, R., and Datta, S.R.
(2011). Distinct representations of olfactory information in different cortical
centres. Nature 472, 213216.
Stettler, D.D., and Axel, R. (2009). Representations of odor in the piriform
cortex. Neuron 63, 854864.
Sugai, T., Miyazawa, T., Fukuda, M., Yoshimura, H., and Onoda, N. (2005).
Odor-concentration coding in the guinea-pig piriform cortex. Neuroscience
130, 769781.
Talbot, S.A., and Marshall, W.H. (1941). Physiological studies on neural mechanisms of visual localization and discrimination. Am. J. Ophthalmol. 24, 1255
1263.
Wong, A.M., Wang, J.W., and Axel, R. (2002). Spatial representation of the
glomerular map in the Drosophila protocerebrum. Cell 109, 229241.
Woolsey, C.N., and Walzl, E.M. (1942). Topical projection of nerve fibers from
local regions of the cochlea to the cerebral cortex of the cat. Bull. Johns
Hopkins Hosp. 71, 315344.
Yan, Z., Tan, J., Qin, C., Lu, Y., Ding, C., and Luo, M. (2008). Precise circuitry
links bilaterally symmetric olfactory maps. Neuron 58, 613624.
Yang, Y., DeWeese, M.R., Otazu, G.H., and Zador, A.M. (2008). Millisecondscale differences in neural activity in auditory cortex can drive decisions.
Nat. Neurosci. 11, 12621263.
Zhan, C., and Luo, M. (2011). Diverse patterns of odor representation by
neurons in the anterior piriform cortex of awake mice. J. Neurosci. 30,
1666216672.
Zhang, X., and Firestein, S. (2002). The olfactory receptor gene superfamily of
the mouse. Nat. Neurosci. 5, 124133.
Resource
Identification of 67 Histone Marks

and Histone Lysine Crotonylation
as a New Type of Histone Modification
Minjia Tan,1,6 Hao Luo,1,6 Sangkyu Lee,1,6 Fulai Jin,2 Jeong Soo Yang,1 Emilie Montellier,3 Thierry Buchou,3
Zhongyi Cheng,1 Sophie Rousseaux,3 Nisha Rajagopal,2 Zhike Lu,1 Zhen Ye,2 Qin Zhu,4 Joanna Wysocka,5 Yang Ye,4
Saadi Khochbin,3 Bing Ren,2 and Yingming Zhao1,*
1Ben
May Department of Cancer Research, The University of Chicago, Chicago, IL 60637, USA
Institute for Cancer Research and Department of Cellular and Molecular Medicine, University of California San Diego School of
Medicine, 9500 Gilman Drive, La Jolla, CA 92093, USA
3INSERM, U823; Universite
Joseph Fourier - Grenoble 1; Institut Albert Bonniot, Faculte de Medecine, Domaine de la Merci,
38706 La Tronche Cedex, France
4Shanghai Institute of Materia Medica, Chinese Academy of Sciences, 555 Zu Chong Zhi Road, Shanghai 201203, P.R. China
5Department of Chemical and Systems Biology, Stanford University School of Medicine, Stanford, CA 94305, USA
*Correspondence: yingming.zhao@uchicago.edu
DOI 10.1016/j.cell.2011.08.008
2Ludwig
SUMMARY
We report the identification of 67 previously undescribed histone modifications, increasing the current
number of known histone marks by about 70%.
We further investigated one of the marks, lysine crotonylation (Kcr), confirming that it represents an
evolutionarily-conserved histone posttranslational
modification. The unique structure and genomic
localization of histone Kcr suggest that it is mechanistically and functionally different from histone
lysine acetylation (Kac). Specifically, in both human
somatic and mouse male germ cell genomes, histone
Kcr marks either active promoters or potential enhancers. In male germinal cells immediately following meiosis, Kcr is enriched on sex chromosomes
and specifically marks testis-specific genes, including a significant proportion of X-linked genes
that escape sex chromosome inactivation in haploid
cells. These results therefore dramatically extend
the repertoire of histone PTM sites and designate
Kcr as a specific mark of active sex chromosomelinked genes in postmeiotic male germ cells.
INTRODUCTION
Mounting evidence suggests that histone PTMs play a crucial
role in diverse biological processes, such as cell differentiation
and organismal development, and that aberrant modification of
histones contributes to diseases such as cancer (Berdasco
and Esteller, 2010; Fullgrabe et al., 2011). At least eleven types
of PTMs have been reported at over 60 different amino acid residues on histones, including histone methylation, acetylation,
propionylation, butyrylation, formylation, phosphorylation, ubiquitylation, sumoylation, citrullination, proline isomerization, and
ADP ribosylation (Martin and Zhang, 2007; Ruthenburg et al.,
2007).
Histone PTMs are thought to contribute to the regulation of
chromatin-templated processes via two major mechanisms
(Kouzarides, 2007; Ruthenburg et al., 2007). First, histone
PTMs can directly modulate the packaging of chromatin either
by altering the net charge of histone molecules or by altering
inter-nucleosomal interactions, thereby regulating chromatin
structure and the access of DNA-binding proteins such as transcription factors. Second, histone PTMs regulate chromatin
structure and function by recruiting PTM-specific binding proteins, which recognize modified histones via specialized structural folds such as bromo-, chromo- and PHD domains (Wysocka
et al., 2005, 2006; Zeng and Zhou, 2002). Alternatively, histone
PTMs can also function by inhibiting the interaction of specific
binders with chromatin. PTM-induced changes in protein interactions between chromatin and its binding partners are in turn
translated into biological outcomes (Margueron et al., 2005).
While the majority of known histone PTMs are located within
the N-terminal tail domain of core histones, PTMs of crucial
importance for histone-DNA and histone-histone interactions
have also been found in the globular domain of core histones
(Cosgrove et al., 2004; Garcia et al., 2007c; Mersfelder and Parthun, 2006). Novel PTM sites occurring outside of the N-terminal
tails continue to be discovered, generally with the aid of sequence
and modification-specific antibodies or by unbiased mass spectrometry (MS) methods (Chu et al., 2006; Garcia et al., 2007b;
Johnson et al., 2004; Wisniewski et al., 2007). The recent
discovery of O-GlcNAc modification (Sakabe et al., 2010)
suggests that additional histone PTMs may yet be discovered.
Here, we used an integrated, mass spectrometry-based
proteomics approach, which takes advantage of in vitro propionylation, efficient peptide separation using isoelectric focusing
(OFFGEL), and the high sensitivity of the LTQ Orbitrap Velos

mass spectrometer to carry out a comprehensive analysis of
histone PTMs. With this approach, we achieved high sequence
coverage of peptide mapping in core and linker histones,
ranging from 87%100%, which in turn resulted in the identification of 67 new PTM sites. These histone marks expand
the total number of known histone PTMs by about 70%. Interestingly, our results show that histones are intensely modified
at various residues not only in the N-terminal tail, but also
within globular domains. Among the modifications, we identified tyrosine hydroxylation (Yoh) and lysine crotonylation (Kcr)
as two novel histone mark types. Finally, we demonstrated
that histone Kcr is a robust indicator of active promoters and
could be an important signal in the control of male germ cell
differentiation.
RESULTS
Experimental Design
Histone proteins are characterized by a high ratio of both lysine
and arginine residues (Garcia et al., 2007a; Zee et al., 2010). As
a result, tryptic digestions of histones tend to yield peptides
that are relatively small and hydrophilic, which are difficult for
subsequent detection by MS due to poor retention by the C18
RP-HPLC column. This problem can be addressed by chemical
derivatization (e.g., lysine propionylation) of amine groups in the
protein (N-terminal amines, and free and monomethylated lysine
-amino groups) before or after tryptic digestion (Garcia et al.,
2007a). Similarly, lysine propionylation of core histones, before
or after tryptic digestion, can generate complementary peptide
sequences that boost the sequence coverage of peptide
mapping by MS. Additionally, IEF separation of the tryptic digest
into 12 fractions will further reduce peptide complexity and
improve dynamic range.
Using this rationale, we designed an integrated approach
for the systematic analysis of histone PTMs which maximizes
both sequence coverage and sensitivity, leading to the identification of many novel PTM sites. In this method, MS analysis
was carried out in histone proteolytic peptides that were generated by four parallel methods (Figure 1A): (1) Tryptic digestion
of core histones without an in vitro chemical derivatization reaction; (2) tryptic peptides that were in vitro propionylated after
tryptic digestion; (3) tryptic peptides that were generated by
tryptic digestion of in vitro propionylated histone proteins; and
(4) tryptic peptides that were generated by tryptic in-gel digestion of the individual histone proteins.
We used PTMap, an algorithm capable of identifying all
possible PTMs of a protein (Chen et al., 2009), to analyze all
acquired MS/MS data and identify histone peptides with or
without a PTM. As anticipated, sequence coverage by MS
mapping was significantly improved after in vitro propionylation,
either before or after tryptic digestion (Figure 1B). Among the four
methods, Method III (in vitro propionylation before tryptic digestion of histones) achieved the highest sequence coverage of
histones H1.2 (100%), H2A (90.7%), and H2B (94.4%). Method
IV gave the best coverage for histones H3 (87.3%) and H4
(82.3%). In aggregate, we achieved sequence coverage of
100% of H1.2, 90.7% of H2A, 100% of H2B, 91% of H3, and
87.3% of H4. To our knowledge, this represents the highest

reported sequence coverage for peptide mapping in histones.
Using this approach, we identified 130 unique PTM sites,
which not only confirmed 63 previously known histone PTMs,
but also revealed 67 novel ones, including 28 Kcr sites, 18 lysine
monomethylation (Kme) sites, 1 lysine dimethylation (Kme2) site,
4 lysine formylation (Kfo) sites, 2 lysine acetylation (Kac) sites, 8
arginine monomethylation (Rme) sites, and 6 tyrosine hydroxylation (Yoh) sites (Figure 1C).
A summary of the non-Kcr modification sites and Kcr sites
identified in this study is shown in Figures 1D and 1E, respectively. All the MS/MS spectra for the identified histone PTM
peptides were carefully verified as previously reported (Chen
et al., 2005). We confirmed the identification of 10 novel nonKcr PTM sites by high-resolution MS/MS (Figure S1 available
online). Identification and validation of these non-Kcr PTMs are
included in Extended Experimental Procedures.
Characterization of the Novel Histone PTM sites
A core histone protein typically consists of an unstructured
N-terminus, a globular core including a central histone-fold
domain, and a conformationally mobile C-terminal tail (Garcia
et al., 2007c; Mersfelder and Parthun, 2006). The central
histone-fold domain consists of three a helices and two loops
that are known to be involved in histone pair-pair and histoneDNA interaction sites (McGhee and Felsenfeld, 1980). The
majority of known histone PTM sites were previously identified
in the N-terminal regions of histones.
In this study, 39 novel non-Kcr PTM sites were identified. Interestingly, among these sites, only four sites were mapped to the
N-terminal domains, while 25 were mapped to central histonefold domains, and another ten were mapped to the C-terminal
domains (Figure 1D). Six PTM sites (including three monomethylated residues at H2BR79, H3R63, and H4K77, 1 acetylated
residue at H3K122, and 2 hydroxylated sites at H2BY83 and
H4Y88) are located at the histone-fold domains (Figure S1M).
H3R63 and H4K77 participate in DNA interactions (Arents and
Moudrianakis, 1993; Luger et al., 1997; Mersfelder and Parthun,
2006), while H2BR79, H2BY83, and H4Y88 participate in the
H2B-H4 interaction. While the amino acid residues located on
the outer nucleosome surface do not contact DNA, they are
known to regulate chromatin structure (Mersfelder and Parthun,
2006). Five PTM sites were mapped at the outer nucleosome
surface, among which we found 3 novel PTM sites, including
monomethyllysine and formyllysine at H2BK116, and dimethyllysine at H4K59 (Figure S1M). Given the important roles performed
by these residues in nucleosome structure and DNA binding, it is
highly likely that these PTM sites will have significant impacts on
transcriptional and epigenetic regulation.
Identification and Validation of Kcr Residues in Histones
A PTM will induce a structural change in the substrate residue
and therefore a change of its molecular weight. Interestingly,
on 28 lysine residues of core histone peptides, our analysis
identified a mass shift of + 68 Da that does not match the shift
associated with any known PTM (Figure 1E). This result suggested the possible presence of a previously unreported histone
mark.
Histones
PTMs
Linker
histone
Trypsin
Propionylation
Trypsin
Trypsin
yp
Propionylation
Method I
In-gel digestion
In-sol digestion
Method II
Method IV
Method III
IEF fractionation
12 fractions
Validation
# of
identified
sites
# of novel
sites
SDS-PAGE
Sequence coverage (%)
In-sol digestion
Kme
Core histones
Total
H1.2
H2A
H2B
H3
H4
31
12
31
33
23
17
10
19
10
11
67
18
130
Kme2
Kfo
4
2
Protein sequence
alignment
Kac
Rme
Yoh
Kcr
28
HPLC-MS/MS
me
fo
Nac
H1.2
fo
me
fo me
ac fo
me
oh
fo
me
fo
ac
fo
me
me
me
me
me
me
me
me
SKASGPKKKKYKKKKLNKKKKKKKKK
1
36
54
48
65 66
me
ac
73
oh
87
83
92
99
131
150
me
fo
me
me
108
170
201
186
226
me
SGRGKQGGKAR GNYSER HLQLAIR KKTESHHK

H2A SGRGKQGGKAR...GNYSERHLQLAIRKKTESHHK
1
me
ac ac
fo
ac
H2B
H3
39
118 119
125
me
me
ac
me
ac ac
88
42
oh
fo
me
fo
me me
oh
me
me
fo
fo
fo
...KKKGSKKAVTKAQKK..Y..K..K..RYNKR..KKAVTK
11 12
1516
20
ac
me
me2
me
me
ac
34
23
me
ac
me
me2
me3
ac
me
ac
46
37
57
79
83
me
fo
me2
me ac
me fo
me2 ac
85
99 108
me
fo
ac
116
120
: Known sites
X : Novel sites
me
ARTKQTARKKKKAARKK..K..R..K...KDIQLR
4
ac
14
me
ac
fo
ac
ac
23
18
27
me
me
fo
36
56
122
63 79
oh
fo
me
me2 me
me
128
me
ox
fo
fo
ac
H4 SGRGKK..K...KRHRKK..RYEETRK..RKRKYALK
5
12
16
20
Kcr: lysine
y
crotonylation
y
Mouse:
Human:
31
35
51
55
59
67
77
79
88
91
Kac: lysine
y
acetylation
y
Kcr
Kcr
Kcr
Kcr
K cr
Kcr
Kcr
Kcr
K cr
Kcr
Kcr
K cr
H1.2 Human: NH2-SETPRKASGPALKKALGLKSLVSKGTLVQTKGKAKKATVTKKVAKK-COOH

84
63
33
Mouse:
Human:
89
96
Kcr
Kcr
158
167
Kcr
KcrKcr
Kcr
118 119
125
H2A Human: NH2-SGRGKQGGKARAKAKTRLLRKGNYAVLLPKKTESHHKAKGK-COOH

5
Kac
Mouse:
Human:
13
Kac
Kcr
36
Kac Kac
KcrKcr
KcrKcr
Kcr
15
Kac
Kcr Kcr
Kcr Kcr
Kcr
Kcr
Kcr
Kcr
Kcr
K cr
H2B Human: NH2-PEPAKSAPAPKKGSKKAVTKAQKKRSRKESYSITKYTSSK-COOH

11 12
Kac
Mouse:
Human:
Kcr
Kcr
Kac
15 16
20
Kac
23 24
Kac
Kcr
Kcr
Kcr
Kcr
34
Kac
Kcr
Kcr
Kcr
Kcr
Kcr
Kcr
H3 Human: NH2-ARTKQTARKSTGGKAPRKQLATKAARKSRYQKSTRIRGERA-COOH
4
Mouse:
Human:
14
Kac
Kac
Kcr
Kcr
Kcr
Kcr
18
Kac
23
Kac
Kac
27
Kac
56
Kac
K cr
Kcr
H4 Human: NH2-SGRGKGGKGLGKGGAKRHRKVLRDNIQGTLYGFGG-COOH
5
Kac
Kac
12
Kac
16
Kac
20
Kac
Figure 1. Experimental Strategy and Results for Identified Histone PTM Sites
(A) Schematic diagram of the experimental design for comprehensive mapping of PTM sites in linker and core histones from HeLa cells. Histone extracts were
in-solution trypticly digested without chemical propionylation (Method I), chemically propionylated after in-solution tryptic digestion (Method II), chemically
propionylated before in-solution tryptic digestion (Method III), and in-gel digested after SDS-PAGE gel separation. Samples from Methods I and II were further
subjected to IEF fractionation to generate 12 fractions.
(B) Peptide sequence coverage of linker and core histones in each of the four methods is shown.
(C) A table summarizing all the PTM sites identified by this study. Abbreviations: me, monomethylation; me2, dimethylation; me3, trimethylation; fo, formylation;
ac, acetylation; oh, hydroxylation; and cr, crotonylation.
(D) A diagram showing sites of histone PTMs other than Kcr identified in this study. Amino acid residue number is indicated below its sequence. Gray and blank
boxes indicate N-terminal and globular core domains, respectively.
+ X-CoA
Crotonyllysine
X = acetyl, crotonyl
Lysine
Acetyllysine
Crotonyllysine
Acetyllysine
C
Multiple steps
Figure 2. Short-Chain Lysine Acylations

(A) An illustration of the enzymatic reactions for lysine acetylation by lysine acetyltransferases (KATs) using acetyl-CoA as a cofactor, and a hypothesized
mechanism for Kcr using crotonyl-CoA as a cofactor.
(B) Ball-and-stick models of a crotonyl group and an acetyl group. The three-dimensional arrangement of four carbons and one oxygen of the crotonyl group are
rigid and located in the same plane (left). The two olefinic carbons of the crotonyl group are shown in yellow. In contrast, the tetrahedral CH3 in the acetyl group
(right) can be rotated such that it is structurally very different from the crotonyl group.
(C) Crotonyl-CoA metabolism pathways. Crotonyl-CoA was generated from butyryl-CoA or glutaryl-CoA, and oxidized to acetyl-CoA through multiple steps.
To determine the nature of this modification, we selected one

of these peptides, PEPAK+68SAPAPK (modified at H2BK5),
for further analysis. After manual inspection of the high-resolution MS data (precursor ion mass at m/z 580.8181), we determined the accurate mass shift of this modification was +
68.0230 Da. By setting the mass tolerance to 0.01 Da
(9 ppm, which is within the mass accuracy of our mass spectrometer), and specifying a maximum of 2 nitrogen atoms, we
were able to deduce the possible element compositions of the
modification group as either C4H4O or H6NO3. The former,
C4H5O (mass shift plus one proton), is the only reasonable
molecular formula of this modification. There were 4 possible

structures consistent with the element composition: Kcr (Figures
2A and 2B), vinylacetyllysine (3-butenoyllysine), methacryllysine,
and cyclopropanecarboxyllysine (Figure S2A). As crotonyl-CoA
is an important and abundant intermediate (Figure 2C) in metabolic pathways of butyryl-CoA and acetyl-CoA, we focused on
Kcr as the PTM candidate most likely to cause the mass shift.
To test if the identified mass shift of + 68.0230 Da was caused
by Kcr, we synthesized the Kcr peptide, PEPAKcrSAPAPK, and
compared its MS/MS spectrum with that of the in vivo-derived
peptide. The in vivo modified peptide bearing a lysine residue
(E) Illustrations of histone Kcr sites in human HeLa cells and mouse MEF cells. All Kcr sites are shown in red and underlined. Previously reported Kac
sites are shown in blue. See also Figure S1 and Supplemental Information 1, Supplemental Information 2, Supplemental Information 3, and Supplemental
Information 4.
y 10 y 9 y 8
y 9++
580.8181
z=2
100
y7
y6 y5
y4 y3
y2
P E P A Kcr S A P A P K
467.7691
b2
b4
b5
b6 b7
b9
In Vivo
y6
580 582
934.5306
570.3210
b2
y2
227.1013
244.1640
b4
y4
395.1901
y5
412.2520 483.2897
y3
100
y7
766.4409
500
y8
b9
917.4662
837.4792
678.3420
200
100
b7
b 6 749.3759
b5
591.3103
315.2015
y9
y 10
1063.5884
800
1100
467.7692
580.8174
z=2
Synthetic
0
580 582
227.1013
244.1644
934.5308
570.3215
412.2527
395.1904
483.2896
315.2018
0
200
500
1063.5768
800
1100
467.7691
580.8176
z=2
100
766.4418
591.3106
678.3412 749.3792
917.4660
837.4786
Mixture
934.5307
580 582
570.3215
412.2524
591.3104
483.2893
395.1922
227.1014
244.1642
766.4417
749.3783
917.4668
837.4816
678.3448
0
200
500
1063.5651
800
1100
m/z
D
61.82
61.12
60.08
In Vivo
40
80
Synthetic
40
80
Mixture
40
80
Retention Time (min)
Figure 3. Identification and Verification of a Kcr Peptide, PEPAKcrSAPAPK

Kcr indicates a crotonyllysine residue.
(AC) High-resolution MS/MS spectrum of a tryptic peptide, PEPAKSAPAPK, with a mass of +68.0230 Da at its Lys5 residue identified from in vivo histone H2B
(A), its synthetic Kcr counterpart (B), and a peptide mixture of the in vivo-derived tryptic peptide and its synthetic counterpart (C), each showing the same MS/MS
fragmentation patterns and the same precursor ion mass. Inset shows their precursor ion masses.
(D) Extracted ion chromatograms (XICs) of the in vivo-derived PEPAK+68.0230SAPAPK peptide, the synthetic Kcr counterpart, and their mixture by nano-HPLC/
MS/MS analysis using a reversed-phase HPLC column, showing the coelution of the two peptides.
with a mass shift of + 68.0230 Da, the synthetic Kcr peptide with
the same peptide sequence (PEPAKcrSAPAPK), and the mixture
of the two peptides exhibited almost identical parent masses
and high-resolution MS/MS spectra (Figures 3A3C). In addition,

the mixture of the in vivo and synthetic peptides coeluted in
HPLC/MS analysis (Figure 3D). These results indicated that
C
Competition
Kcr
H1
Peptide
Kac
Kpr
Kbu
WB: anti-Kcr
Kcr
Core Histones
1 ng
Crotonate (mM)
5 ng
50
100
WB: anti-Kcr
H1
25 ng
Blue staining
Blue staining
Core Histones
D
y 10 y 9 y 8
y 9++
100
b2
581
b2
y4
412.2546
y5
570.3226
y2
b9
y8
b7
y9
938.5569
b5
y 7 841.4987 b 9
753.4058
595.3361 b
921.4927
770.4672
6
682.3664
483.2908
0
200
b5 b6 b7
y6
227.1019
y2
y6 y5 y4
D 4- crotonate labeling
585
244.1656
y7
P E P A KD4-cr S A P A P K
469.7821
582.8303
z=2
500
800
y10
1067.5991
1100
m/z
Competition
Kcr
WB: anti-Kcr
Blue staining
Figure 4. Detection of Kcr in Histones by Western Blotting.

(A) Specificity of pan anti-Kcr antibody demonstrated by dot-spot assay using five peptide libraries with indicated amount (ng). Each peptide library contains 13
residues CXXXXXKXXXXXX, where X is a mixture of 19 amino acids (excluding cysteine), C is cysteine, and the 7th residue is a fixed lysine residue: unmodified
lysine (K), Kac, propionyllysine (Kpr), butyryllysine (Kbu), and Kcr.
(B) Detection of Kcr in histones. Western blotting was carried out using the histones from HeLa cells with competition of a peptide library bearing a fixed
unmodified lysine (K) or Kcr.
(C) Dynamics of histone Kcr in response to crotonate. The histone proteins extracted from human prostate cancer cell line Du145 incubated with 0, 50, or 100 mM
crotonate for 24 hr, were western blotted with anti-Kcr pan antibody.
(D) MS/MS spectrum of PEPA KD4-crSAPAPK identified from D4-crotonate-labeled sample. The mixture of D4-, D3- and D2-crotonyl groups was used for the
identification of D4-crotonyl peptide.
(E) Kcr signals in core histones of S. cerevisiae, C. elegans, D. melanogaster (S2), M. musculus (MEF), as well as H. sapiens (HeLa) cells by western blotting
analysis with competition. See also Figure S2.
the identified mass shift of +68.0230 Da was very likely caused

by Kcr.
To further confirm Kcr in histones, we generated a pan antibody against Kcr. This pan anti-Kcr antibody specifically recognized a peptide library bearing Kcr, but not four other peptide
libraries in which the fixed lysine residue was unmodified, acetylated, propionylated, or butyrylated (Figure 4A); the specificity
of pan anti-Kac antibody was confirmed likewise (Figure S2B).
The specificity of the pan anti-Kcr antibody was further demonstrated by western blotting with three different bovine serum
albumin (BSA) derivatives, where peptide lysines were chemically modified by a crotonyl, vinylacetyl, or methacryl group,
respectively. The result showed that pan anti-Kcr antibody

recognized only the lysine crotonylated BSA, but not the unmodified, lysine vinylacetylated or lysine methacrylated BSA (Figure S2C). This pan anti-Kcr antibody was subsequently used
for western blotting and immunostaining of Kcr signal.
The antibody detected a Kcr signal among all core histone
proteins: H2A, H2B, H3, H4, and linker histone H1. In each protein, the signal could be efficiently competed away by a peptide
library bearing a Kcr, but not peptide libraries bearing an unmodified lysine (Figure 4B), metharcryllysine (Figure S2D), acetyllysine, propionyllysine, or butyryllysine (Figure S2E). The strong
Kcr signals detected in histones indicated that this modification
is present in nuclei and is associated with chromosomes.

Indeed, immunostaining using the pan antibody showed that
Kcr mainly existed in nuclei (Figure S2F; see also the analysis
below of spermatogenic cells).
Isotopic labeling is an established method to confirm in vivo
protein modifications that was previously used for the study of
histone Kac (Allis et al., 1985). We rationalized that D4-crotonate
can be converted into crotonyl-CoA in vivo, which then functions
as a lysine crotonylation donor. Consistent with this hypothesis,
the histone Kcr signal from HeLa cells increased dramatically
after cells were cultured with crotonate (Figure 4C). After tryptic
digestion of histones from D4-crotonate labeled HeLa cells, we
carried out peptide immunoprecipitation using the anti-Kcr
antibody. The enriched Kcr peptides were then subjected to
HPLC/MS/MS analysis and protein sequence alignment, which
confirmed Kcr on histone H2BK5 by D4-crotonate labeling (Figure 4D). (Note: The D4-crotonic acid was mixed with D3- and
D2-crotonic acid (Figure S2G). This characteristic isotopic distribution was used for the identification of D4-crotonyl peptide.)
Using a fluorometric assay, we also demonstrated that histone

lysine deacetylases (HDACs) 1, 2, 3, and 6 exhibit potent lysine
deacetylation activities, but have very weak or no effect on lysine
decrotonylation. For example, lysine decrotonylation activity
was not detected with HDAC6 exposure, though the enzymes
deacetylation activity was very strong in this assay (Figure S2I).
These results suggest that enzymes required for the addition
and removal of Kcr on histone proteins may be different from
those for histone Kac regulation.
Histone Kcr Is an Evolutionarily Conserved Histone PTM

To test if lysine crotonylation is present in histones from other
eukaryotic cells, we isolated histones from yeast S. cerevisiae,
C. elegans, Drosophila S2 cells, mouse embryonic fibroblast
(MEF) cells, as well as human HeLa cells. Kcr signals were detected among core histones from all five species by western
blotting (Figure 4E). Taking advantage of affinity enrichment
using the pan anti-Kcr antibody and HPLC/MS/MS, we identified
24 Kcr sites on mouse MEF cells (Figure 1E). All of the annotated
MS/MS spectra for Kcr mouse histone peptides are included
in Supplemental Information 3. Taken together, our results
revealed that Kcr is an evolutionarily conserved histone mark
appearing in eukaryotic cells from a wide range of species.
Genome-wide Mapping of Histone Kcr in Human Cells

In order to explore the in vivo function of histone Kcr, we performed ChIP-seq analysis with the pan anti-Kcr antibody to
determine the genomic distribution of histones with this modification in the human fetal lung fibroblast IMR90 cell line. As a
control, we compared the Kcr distribution with previously obtained ChIP-seq results for H3K4me3 (marking gene promoters)
and H3K4me1 (marking enhancers) distribution (Heintzman
et al., 2007).
Strikingly, we found an abundance (totaling 84,435 peaks) of
this histone modification in the human genome. Histone Kcr
was largely associated with active chromatin, including both
the transcription starting site (TSS) and regions previously
predicted to be enhancers (Hawkins et al., 2010) (Figures 5A
and 5B).
The majority (68%) of histone Kcr peaks was associated with
either promoter or predicted enhancer regions (Figure 5B).
At promoters, histone Kcr showed the strongest enrichment
flanking TSS, in contrast to the observation that H3K4me3 was
more enriched downstream of TSS (Figure 5C). At predicted
enhancers, we observed a strong enrichment of Kcr in agreement with the H3K4me1 pattern at these sites (Figure 5D).
Furthermore, we also observed a strong correlation between
gene expression and Kcr level at promoters (Figure 5E). Taken
together, these data strongly support a role for histone Kcr in
gene regulation, especially at promoters and enhancers.
In addition, we also compared Kcr sites with histone lysine
acetylation sites revealed by ChIP-seq assay using a pan antiKac antibody. The result showed that histone crotonylation
generally occupies similar locations as acetylation in IMR90 cells
as there is a significant overlap between Kac and Kcr peaks (Figure S3A). These results suggested that in resting somatic cells,
open chromatins are simultaneously marked by both histone
Kac and Kcr. Given the finding that these two marks are catalyzed by different sets of enzymes, we assumed that the regulation of Kcr would likely have different spatial and temporal
dynamics from Kac. To test this prediction, we explored the
function of Kcr in the highly dynamic spermatogenesis process.
Histone Kcr and Kac Are Mechanistically

and Functionally Different
In order to examine whether histone Kcr is derived from promiscuous histone acetyltransferase (HAT) activity, we overexpressed two HATs, CBP and p300, in 293T cells and tested for
histone crotonylation levels by western blotting. We found that
overexpression of CBP or p300 led to the enhancement of
histone Kac but did not significantly change the levels of histone
Kcr (Figure S2H).
Histone Kcr Marks Testis-Specific Genes Activated

in Postmeiotic Cells
We next investigated the genomic distribution of histone Kcr in
spermatogenic cells, where a very specific gene expression program directs key steps of differentiation (Boussouar et al., 2008;
Gaucher et al., 2010). We first examined the global dynamics of
Kcr during mouse spermatogenesis using immunohistochemistry. We observed an intense labeling of Kcr during elongating
steps 911 in spermatids (Figure 6A), which coincides with a
Histone Kcr Sites in Human Cells

To identify histone Kcr sites, we used a Mascot sequence alignment to analyze MS/MS data derived from HeLa histones, using
Kcr (+ 68.02621 Da) as a variable modification. These analyses
led to the identification of 28 Kcr sites in human histones (Figure 1E). In addition, 19 of these 28 identified sites were confirmed
by in vivo D4-crotonate labeling experiments. The Kcr peptides
and their corresponding MS/MS spectra are included in Supplemental Information 1 and Supplemental Information 2.
In summary, we have used five independent methods MS/
MS and HPLC coelution of synthetic peptides, D4-crotonate
labeling, western blotting, and immunostaining to show that
histone Kcr exists in cells, and identified 28 lysines on various
histones that are subject to this PTM.
A
Input
H3K4me3
H3K4me1
Kcr
Mean occupancy around promoters

70
ChIP-seq re
eads density (RPKM)
H3K4 1
H3K4me1
Others
15%
TSS
23%
Gene
body
17%
Enhancer
45%
60
50
H3K4me3
Kcr
Crotonylation
40
30
20
10
0
-10
Genomic distribution of
84 435 Crotonylation
84,435
Croton lation peaks
-2000
lowest 20%
H3K4me3
Kcr
Crotonylation
6
4
2
0
-2
ChIP--seq reads density (RPKM)
ChIP-seq reads density (RPK

KM)
50
H3K4me1
2000
Promoter Kcr occupancy levels are correlated with gene

activity
Mean occupancy around predicted enhancers

10
-1000
0
1000
Relative distance to TSS (bp)
2nd 20%
40
3rd 20%
4th 20%
30
top 20%
20
10
-4000
-2000
2000
4000
Relative distance to center of predicted enhancers

(bp)
-2000
-1000
1000
2000
Relative distance to TSS (bp)
Figure 5. Enrichment of Histone Kcr on Active Chromatin

(A) ChIP-seq snapshots of input, H3K4me3, H3K4me1 and Kcr in IMR90 cells.
(B) Pie chart showing the genomic distribution of all histone Kcr peaks with annotated genomic regions. TSS is defined as regions +/ 2.5kb around known
transcription starting sites in RefSeq database. Enhancers are promoter distal regions associated with H3K4me1 as predictive mark.
(C) Curves showing the mean reads density of indicated histone modification around all known TSS. Reads densities are calculated within a 100bp sliding window
and normalized by subtracting reads density in the control input ChIP-seq data. RPKM is calculated as the number of reads which map per kilobase of genomic
region per million mapped reads.
(D) Average normalized read densities of H3K4me3, H3K4me1 and histone Kcr around predicted enhancers are plotted.
(E) All RefSeq genes are divided into 5 groups based on their expression level calculated from mRNA-seq data, and the average normalized read densities of Kcr
around each group of TSS are plotted. See also Figure S3.
genome-wide histone hyperacetylation previously reported

(Hazzouri et al., 2000). Since a general transcriptional shutdown
occurs at these stages of germ-cell differentiation (Zhao et al.,
2004), we investigated the relationship between Kcr and gene

expression using a ChIP-seq experiment in earlier stages. Spermatogenic cells were fractionated to enrich in spermatocytes
Figure 6. Correlation of Histone Kcr with Gene Expression in Meiotic and Postmeiotic Male Germ Cells and in Tissues
(A) Hypercrotonylation wave in elongating spermatids. Kcr was detected on paraffin mouse testis tubule sections representing different stages of spermatogenesis by immunohistochemistry (IH) using an anti-Kcr antibody. Pre-meiotic spermatogonia (Spg) and meiotic spermatocytes (Spc) cells are present at the
periphery and middle of the tubule sections, whereas postmeiotic round (RS), elongating (ES) and condensing (CS) spermatids are near the lumen. The nuclei of
ES are positive for Kcr.
(B and C) Genes associated with higher Kcr in RS than Spc are mostly postmeiotically activated and show a predominant expression in the testis. The genes
associated with Kcr peaks were divided into three categories according to their Kcr levels in Spc and RS: (1) Spc = RS, similar Kcr levels between Spc and RS; (2)
Spc > RS, lower Kcr levels in RS than Spc (fold change > = 2); (3) Spc < RS, higher Kcr levels in RS than Spc (fold change > = 2). The expression of these genes in
male germ cells (B) and tissues (C) was then compared among the three categories. (B) Expression in male germ cells. Left panel: respective proportions of genes
(y axis) with higher expression either in Spc or RS among the three gene categories (x axis). Right panel: heatmap showing the expression of the third category of
genes (Kcr, Spc < RS) in Spc (4 samples) and in RS (4 samples). Color scale showing low expression in green to high expression in red. (C) Expression of genes in
tissues. Left panel: pie charts showing the respective proportions of genes with the highest level of expression in the indicated tissues. Genes with the highest
tissue-specific expression are those whose levels of expression, in the indicated tissue, are elevated by at least two standard deviations above the mean
expression in all tissues. Right panel: heatmap showing the expression of the third category of genes in the indicated tissues.
(Spc) and postmeiotic round spermatids (RS), which were used

for ChIP-Seq analysis. As shown in Figure S3B, the genomic
regions associated with histone Kcr from mouse spermatogenic
cells showed a pattern very similar to that of human somatic
cells: histone Kcr is enriched at the TSS of genes (Figure S3C).
Additionally, we identified many Kcr sites lacking histone Kac,
suggesting that crotonylation can occur independently of Kac
(Figure S3D).
To gain insight into the biological significance of promoter
histone Kcr, we classified the genes associated with histone
Kcr into three categories, depending on the variation of Kcr
levels during and after meiosis: (1) Genes associated with similar
levels of promoter Kcr between spermatocytes (Spc) and round
spermatids (RS); (2) genes associated with lower promoter Kcr
levels in RS than Spc; and (3) genes associated with higher
promoter Kcr levels in RS than Spc (Figure 6B). We then
analyzed expression levels among the three categories of genes.
To this end, transcriptomic data from staged mouse spermatogenic cells (available on GEO website: GSE217 and GSE4193)
were downloaded. The respective proportions of genes showing
higher expression in either Spc or RS were determined among
the three categories of genes. This analysis revealed no significant connection between histone Kcr levels and higher gene
expression in either meiotic (Spc) or postmeiotic (RS) cells for
the first two categories of genes (Kcr: Spc = RS and Spc >
RS). However, the vast majority of the genes with higher Kcr in
RS (Kcr: Spc < RS) were found to be activated after meiosis (Figure 6B, left). This result indicates a tight correlation between
postmeiotic histone Kcr and a specific haploid cell gene expression program.
To further characterize these genes, we analyzed their expression patterns in testis and several somatic tissues from adult
mice. Here again, most of the genes associated with a postmeiotic increase of Kcr (Spc < RS), but not the other two gene categories, were highly expressed in testis (Figure 6C). A parallel
ChIP-seq analysis of Kac showed that only 35% of these genes
are associated with histone Kac at the promoters (data not
shown). Therefore, the majority of the highly-expressed testis
genes seemed to be predominantly labeled by histone Kcr but
not by histone Kac.
Finally, we analyzed the genomic localizations of genes
among the three categories of Kcr-marked genes. Our data
showed that 31.4% and 1.2% of the third category of genes
(Kcr, Spc < RS) were located on the X and Y chromosomes,
respectively, while the other two categories of genes showed
no particular bias in genomic distribution (Figure 6D, left). In
contrast, three categories of Kac-marked genes showed no
bias toward a particular chromosome (Figure 6D, right). This
result therefore confirms distinct genomic distribution patterns
for histone Kcr and histone Kac.
Given the postmeiotic enrichment of Kcr on sex chromosomes, we reasoned that an in situ approach may allow us to
visualize a preferential association of Kcr with sex chromosomes. Remarkably, immunohistochemistry (IH) staining of
testis sections (without counter-staining) revealed the presence

of a single, dot-like structure exclusively found in round spermatids (Figure 7A and Figure S3E). To analyze the nature of this
structure, we used sections of seminiferous tubules containing
mostly round spermatids for the costaining of either Kac and
Kcr (Figure S3F), or Kcr and HP1g (Figure 7B). HP1g marks
the heterochromatic chromocenter, as well as adjacent mostlyinactive sex chromosome, which is consistent with previously
reported works (Namekawa et al., 2006; Turner, 2007). Kcr
concentrates at the sex chromosomes and is visible as a
DAPI-dense structure adjacent to the chromosome center,
which itself is composed of centromeric and pericentric heterochromatin and devoid of Kac (Turner, 2007). The costaining of
Kcr and HP1g confirmed these findings and supported the
ChIP-seq results of an accumulation of Kcr on the sex
chromosomes.
These data indicate that histone Kcr may play an important
role in epigenetically marking sex chromosomes in the postmeiotic stages of spermatogenesis. On these chromosomes, this
histone mark may be involved in specifically defining a subset
of genes that escape sex chromosome inactivation after the
completion of meiosis.
DISCUSSION
In this study, we described an integrated approach for the
systematic analysis of histone PTMs (Figure 1). With this unique
approach, we identified 130 PTM sites on human histones,
including 63 known and 67 novel histone marks (Figure 1C).
There are about 93 previously described histone marks based
on the UniProt database (http://www.uniprot.org; Accession
number: P16403 for histone H1.2; P0C0S8 for histone H2A;
P62807 for histone H2B; P84243 for histone H3.3; P62805 for
histone H4) and 3 histone O-GlcNAc sites that were recently
identified (Sakabe et al., 2010). Thus, our study expands the
catalog of known histone marks by around 70%, which affirms
the unprecedented quality and sensitivity of our approach. In
particular, we have identified Yoh and Kcr as two novel types
of histone PTM. Therefore, this work dramatically extends the
catalog of histone PTM sites in mammalian cells and provides
a platform for the discovery of novel mechanisms of histone
regulation.
Five lines of evidence suggest that histone Kcr and histone
Kac are different from each other and involve distinct pathways
for addition and removal. First, the two PTMs are structurally very
different. A crotonyl group has a more rigid structure, due to the
presence of a flattened C-C p-bond (Figure 2B), while the methyl
group of an acetyl group is tetrahedral and rotatable. In light of
these obvious electronic and geometric differences, it is highly
likely that Kcr uses distinct regulatory enzymes from Kac.
Second, some lysine residues appear to be modified by crotonylation but not acetylation in core histones from specific cells
of interest (e.g., H2AK118, H2AK119, H2AK125, H2BK34,
H1.2K89, and H1.2K167 in HeLa and MEF cells reported here).
(D) The X-linked genes are highly and specifically marked by histone Kcr in RS. The respective proportion (%) of genes associated with Kcr (left panel) or Kac
peaks (right panel) among chromosomes in male germ cells is shown. See also Figure S3.
Figure 7. Sex Chromosomes Are Highly Crotonylated in Round Spermatids

IH (A) on a testis tubule section and immunofluorescence (IF) (B) of male germ cells showing high crotonylation enrichment in the sex chromosome region (besides
the HP1g enriched chromocenter) in postmeiotic round spermatids. IH is shown without counterstaining for better visualization of the region detected by the
antibody. IF: pan-Kcr was detected in green, and is shown in costaining with HP1g in bright red fluorescence. Scale bar for IF: 5 mm. See also Figure S3.
Third, our results show that KATs and HDACs exert very different
effects upon Kac and Kcr. Fourth, the status of short-chain lysine
acylation may be modulated in response to CoA concentration.
In the case of crotonyl-CoA, it can be generated either from
butyryl-CoA by short chain acyl-CoA dehydrogenase, or from
glutaryl-CoA by glutaryl-CoA dehydrogenase. Once formed,
crotonyl-CoA can be converted to acetyl-CoA for the TCA cycle.
Finally, histone Kac and Kcr mark different sets of gene in some
differentiated cells, as we report here in mouse sperm cells.
Strikingly, in both human somatic and mouse male germ cell
genomes, histone Kcr specifically labels enhancers and, most
precisely, the TSS of active genes. In postmeiotic male germ
cells, a gain in histone Kcr is a consistent indicator of an X-linked
haploid cell-specific gene expression program. Indeed, the identification of a subset of genes presenting increased Kcr in round
spermatids allowed us to show that such genes are enriched on
the X chromosome and are mostly predominantly expressed in
the testis with a postmeiotic pattern of expression.
Meiosis is known to be associated with the inactivation of sex
chromosomes, called meiotic sex chromosome inactivation
(MSCI). MSCI initiates in pachytene spermatocytes, and sex
chromosome gene silencing continues after meiosis in round
spermatids (RS) until the general shutdown of transcription
(Namekawa et al., 2006; Turner, 2007). However, in round spermatids, a significant number of X-linked genes have been shown
to be specifically reactivated (Mueller et al., 2008; Namekawa
et al., 2006). Interestingly, our results showed that histone Kcr
specifically marks X-linked genes that are postmeiotically
expressed. Consistent with this observation, our in situ experiment uncovered a remarkable postmeiotic labeling of Kcr on
sex chromosomes. In addition, our result indicates that the addi1026 Cell 146, 10161028, September 16, 2011 2011 Elsevier Inc.
tion of Kcr goes beyond the specific sets of genes identified

above and also includes intergenic regions, a conclusion which
has been confirmed by analysis of ChIP-seq data (data not
shown).
Another interesting observation is the occurrence of genomewide histone Kcr in elongating spermatids following the general
transcriptional shut-down associated with histone replacement.
A genome-wide histone modification previously known to be
associated with histone removal is histone hyperacetylation
(Govin et al., 2004). The occurrence of global histone Kcr at the
same time strongly suggests that, like histone Kac, Kcr is likely
to affect chromatin structure and hence facilitate histone
replacement. Our data indicate that there is a limited overlap
between histone Kcr and Kac. Although both histone modifications are associated with gene expression and their occurrence
at a larger scale could facilitate histone removal, they may differently orient the reorganization of the marked genomic regions
after histone removal, and hence be important histone marks
in the establishment of a region-specific male epigenome
organization.
Given the fact that all the previously-studied histone lysine
PTM pathways, such as lysine acetylation and methylation,
have important biological functions, we anticipate that Kcr will
also likely play a major role in the regulation of histone protein
structure and function. Identification of histone Kcr raises
many interesting questions. What enzymes regulate the addition
or removal of Kcr? What role does histone Kcr play in the regulation of histone structure and function? Are there proteins that
bind to histone Kcr? Is lysine crotonylation also present on
nonnuclear proteins as well? Answers to these questions will
require further investigation.
In-Solution Proteolytic Digestion and Chemical Derivatization
of Histone Proteins
In-solution tryptic digestion of histone samples was carried out using a
protocol previously described (Kim et al., 2006; Luo et al., 2008). In vitro lysine
propionylation of histone extract and tryptic histone peptides was performed
as previously described (Garcia et al., 2007a). Histone extracts were in-solution digested without chemical propionylation, chemically propionylated after
in-solution digestion, or chemically propionylated before in-solution digestion.
Isoelectric Focusing Fractionation
The histone proteolytic peptides were separated using an Agilent 3100
OFFGEL Fractionator (Agilent, Santa Clara, CA) according to the manufacturers instructions. Twelve fractions were obtained from each IEF fractionation experiment.
Nano-HPLC/Mass Spectrometric Analysis
The tryptic digests were injected into a NanoLC-1D plus HPLC system
(Eksigent Technologies, Dublin, CA), and analyzed by an LTQ-Orbitrap Velos
mass spectrometer (Thermo Fisher Scientific, Waltham, MA). Full scan MS
spectra from m/z 3501400 were acquired in the Orbitrap. Twenty of the
most intense ions were isolated for MS/MS analysis.
Protein Sequencing Alignment
All MS/MS spectra were searched against the NCBInr human protein
sequence database using Mascot and PTMap software (Chen et al., 2009).
For histone samples that were generated by tryptic digestion of propionylated
histones, the specific parameters included lysine propionylmethylation
(+ 70.04187 Da) and lysine propionylation as variable modifications. For
histone samples propionylated after trypsin digestion, N-terminal propionylation was included as a fixed modification. All the identified peptides were
manually verified according to the rules described previously (Chen et al.,
2005).
Generation of Pan Anti-Kcr Antibody
The pan anti-Kcr antibody was generated and purified from rabbit with
lysine-crotonylated bovine serum albumin (BSA) as an antigen. For more
details, see Extended Experimental Procedures.
ChIP-Seq
ChIP-seq for histone Kcr or Kac was carried out as previously described
with 500 mg IMR90 chromatin (or 100 mg of fractionated germ cells chromatin)
and 5 mg pan anti-Kcr or anti-Kac antibody (Hawkins et al., 2010). ChIP-seq
libraries for sequencing were prepared following Illumina protocols (Illumina,
San Diego, CA) with minor modifications. Libraries for input samples were
generated using 20 ng corresponding input chromatin. Briefly, ChIPed DNA
was first blunted with END-IT DNA repair kit (Epicenter Biotechnology, Madison, WI) and then incubated with Klenow (exo-) (New England Biolabs, MA)
and dATP to generate single base 30 -dA overhang. Illumina sequencing
adaptor was then ligated to the resulting DNA, and followed by size selection
(180-400bp) from a 8% acrylamide gel. This size-selection step was repeated
after PCR amplification with DNA primers supplied by Illumina. Libraries were
sequenced using Illumina GAII or HiSeq machine as per manufacturers protocols. Following sequencing cluster imaging, base calling were conducted
using the Illumina pipeline. Reads were mapped to human hg18 (for IMR90
data) or mouse mm9 (for sperm cell data) genome build with a bowtie software
package. Total mapped tags were paired down to unique, monoclonal tags.
These are tags that mapped to one location in the genome and each sequence
is represented once.
For additional experimental materials and methods, see Extended Experimental Procedures, including methods for preparation of histones from
HeLa cells, in-solution proteolytic digestion and chemical derivatization of
histone proteins, HPLC/MS/MS analysis and protein sequence database
searching, verification of lysine crotonylated peptides by HPLC/MS/MS analysis, synthesis of BSA derivatives, conjugation of Kcr-immobilized agarose
beads, generation of pan anti-Kac and anti-Kcr antibodies, western blotting
with competition of a peptide library, affinity enrichment of Kcr peptides,

synthesis of Boc-Lys(crotonyl)-AMC and Boc-Lys(ac)-AMC, in vitro lysine
decrotonylation and lysine deacetylation reaction assays, transient transfection, in vivo D4-crotonate labeling of histones, immunofluorescence of HeLa
cells and germ cells using anti-Kcr antibody, immunohistochemistry of testis
tubule cross-sections using anti-Kcr antibody, mouse spermatogenic cell
fractionation, chromatin immunoprecipitation in fractionated germ cells,
expression analyses, identification of mono-, di-, and trimethylated lysine
residues, identification of N-formylated and acetylated lysine residues, identification of monomethylated arginine residues, identification of hydroxylated
tyrosine residues.
Annotated MS/MS spectra of all the modified peptides bearing a PTM other
than Kcr were included in Supplemental Information 4.
Supplemental Information includes Extended Experimental Procedures, three
figures, four Supplemental Information data files, and two tables and can be
ACKNOWLEDGMENTS
This work was supported by National Institute of Health grants to B.R., J.W.,
and Y.Z., by INCa-DHOS, ANR blanche EpiSperm, and ARC research
programs to S.K. Z.C. is a shareholder of PTM BioLab, Co., Ltd (Chicago,
IL); Y.Z. is a member of the scientific advisory board of PTM BioLab, Co.,
Ltd (Chicago, IL).
Received: February 18, 2011
REFERENCES
Allis, C.D., Chicoine, L.G., Richman, R., and Schulman, I.G. (1985). Depositionrelated histone acetylation in micronuclei of conjugating Tetrahymena. Proc.
Natl. Acad. Sci. USA 82, 80488052.
Arents, G., and Moudrianakis, E.N. (1993). Topography of the histone octamer
surface: repeating structural motifs utilized in the docking of nucleosomal
DNA. Proc. Natl. Acad. Sci. USA 90, 1048910493.
Berdasco, M., and Esteller, M. (2010). Aberrant epigenetic landscape in
cancer: how cellular identity goes awry. Dev. Cell 19, 698711.
Boussouar, F., Rousseaux, S., and Khochbin, S. (2008). A new insight into
male genome reprogramming by histone variants and histone code. Cell Cycle
7, 34993502.
Chen, Y., Chen, W., Cobb, M.H., and Zhao, Y. (2009). PTMapa sequence
alignment software for unrestricted, accurate, and full-spectrum identification
of post-translational modification sites. Proc. Natl. Acad. Sci. USA 106,
761766.
Chen, Y., Kwon, S.W., Kim, S.C., and Zhao, Y. (2005). Integrated approach
for manual evaluation of peptides identified by searching protein sequence
databases with tandem mass spectra. J. Proteome Res. 4, 9981005.
Chu, F., Nusinow, D.A., Chalkley, R.J., Plath, K., Panning, B., and Burlingame,
A.L. (2006). Mapping post-translational modifications of the histone variant
MacroH2A1 using tandem mass spectrometry. Mol. Cell. Proteomics 5,
194203.
Cosgrove, M.S., Boeke, J.D., and Wolberger, C. (2004). Regulated nucleosome mobility and the histone code. Nat. Struct. Mol. Biol. 11, 10371043.
Fullgrabe, J., Kavanagh, E., and Joseph, B. (2011). Histone onco-modifications (Oncogene).
Garcia, B.A., Mollah, S., Ueberheide, B.M., Busby, S.A., Muratore, T.L.,
Shabanowitz, J., and Hunt, D.F. (2007a). Chemical derivatization of histones
for facilitated analysis by mass spectrometry. Nat. Protoc. 2, 933938.
Garcia, B.A., Pesavento, J.J., Mizzen, C.A., and Kelleher, N.L. (2007b).
Pervasive combinatorial modification of histone H3 in human cells. Nat.
Methods 4, 487489.
Garcia, B.A., Shabanowitz, J., and Hunt, D.F. (2007c). Characterization of
histones and their post-translational modifications by mass spectrometry.
Curr. Opin. Chem. Biol. 11, 6673.
Gaucher, J., Reynoird, N., Montellier, E., Boussouar, F., Rousseaux, S., and
Khochbin, S. (2010). From meiosis to postmeiotic events: the secrets of
histone disappearance. FEBS J. 277, 599604.
Govin, J., Caron, C., Lestrat, C., Rousseaux, S., and Khochbin, S. (2004). The
role of histones in chromatin remodelling during mammalian spermiogenesis.
Eur. J. Biochem. 271, 34593469.
Hawkins, R.D., Hon, G.C., Lee, L.K., Ngo, Q., Lister, R., Pelizzola, M., Edsall,
L.E., Kuan, S., Luu, Y., Klugman, S., et al. (2010). Distinct epigenomic landscapes of pluripotent and lineage-committed human cells. Cell Stem Cell 6,
479491.
Hazzouri, M., Pivot-Pajot, C., Faure, A.K., Usson, Y., Pelletier, R., Sele, B.,
Khochbin, S., and Rousseaux, S. (2000). Regulated hyperacetylation of core
histones during mouse spermatogenesis: involvement of histone deacetylases. Eur. J. Cell Biol. 79, 950960.
Heintzman, N.D., Stuart, R.K., Hon, G., Fu, Y., Ching, C.W., Hawkins, R.D.,
Barrera, L.O., Van Calcar, S., Qu, C., Ching, K.A., et al. (2007). Distinct and
predictive chromatin signatures of transcriptional promoters and enhancers
in the human genome. Nat. Genet. 39, 311318.
Johnson, L., Mollah, S., Garcia, B.A., Muratore, T.L., Shabanowitz, J., Hunt,
D.F., and Jacobsen, S.E. (2004). Mass spectrometry analysis of Arabidopsis
histone H3 reveals distinct combinations of post-translational modifications.
Nucleic Acids Res. 32, 65116518.
Kim, S.C., Sprung, R., Chen, Y., Xu, Y., Ball, H., Pei, J., Cheng, T., Kho, Y.,
Xiao, H., Xiao, L., et al. (2006). Substrate and functional diversity of lysine
acetylation revealed by a proteomics survey. Mol. Cell 23, 607618.
Martin, C., and Zhang, Y. (2007). Mechanisms of epigenetic inheritance. Curr.

Opin. Cell Biol. 19, 266272.
McGhee, J.D., and Felsenfeld, G. (1980). Nucleosome structure. Annu. Rev.
Biochem. 49, 11151156.
Mersfelder, E.L., and Parthun, M.R. (2006). The tale beyond the tail: histone
core domain modifications and the regulation of chromatin structure. Nucleic
Acids Res. 34, 26532662.
Mueller, J.L., Mahadevaiah, S.K., Park, P.J., Warburton, P.E., Page, D.C., and
Turner, J.M. (2008). The mouse X chromosome is enriched for multicopy testis
genes showing postmeiotic expression. Nat. Genet. 40, 794799.
Namekawa, S.H., Park, P.J., Zhang, L.F., Shima, J.E., McCarrey, J.R.,
Griswold, M.D., and Lee, J.T. (2006). Postmeiotic sex chromatin in the male
germline of mice. Curr. Biol. 16, 660667.
Ruthenburg, A.J., Li, H., Patel, D.J., and Allis, C.D. (2007). Multivalent engagement of chromatin modifications by linked binding modules. Nat. Rev. Mol.
Cell Biol. 8, 983994.
Sakabe, K., Wang, Z., and Hart, G.W. (2010). Beta-N-acetylglucosamine
(O-GlcNAc) is part of the histone code. Proc. Natl. Acad. Sci. USA 107,
1991519920.
Turner, J.M. (2007). Meiotic sex chromosome inactivation. Development 134,
18231831.
Wisniewski, J.R., Zougman, A., Kruger, S., and Mann, M. (2007). Mass spectrometric mapping of linker histone H1 variants reveals multiple acetylations,
methylations, and phosphorylation as well as differences between cell culture
and tissue. Mol. Cell. Proteomics 6, 7287.
Wysocka, J., Swigut, T., Milne, T.A., Dou, Y., Zhang, X., Burlingame, A.L.,
Roeder, R.G., Brivanlou, A.H., and Allis, C.D. (2005). WDR5 associates with
histone H3 methylated at K4 and is essential for H3 K4 methylation and
vertebrate development. Cell 121, 859872.
Kouzarides, T. (2007). Chromatin modifications and their function. Cell 128,

693705.
Wysocka, J., Swigut, T., Xiao, H., Milne, T.A., Kwon, S.Y., Landry, J., Kauer,
M., Tackett, A.J., Chait, B.T., Badenhorst, P., et al. (2006). A PHD finger of
NURF couples histone H3 lysine 4 trimethylation with chromatin remodelling.
Nature 442, 8690.
Luger, K., Mader, A.W., Richmond, R.K., Sargent, D.F., and Richmond, T.J.
(1997). Crystal structure of the nucleosome core particle at 2.8 A resolution.
Nature 389, 251260.
Zee, B.M., Levin, R.S., Xu, B., LeRoy, G., Wingreen, N.S., and Garcia, B.A.
(2010). In vivo residue-specific histone methylation dynamics. J. Biol. Chem.
285, 33413350.
Luo, H., Li, Y., Mu, J.J., Zhang, J., Tonaka, T., Hamamori, Y., Jung, S.Y., Wang,
Y., and Qin, J. (2008). Regulation of intra-S phase checkpoint by ionizing
radiation (IR)-dependent and IR-independent phosphorylation of SMC3. J.
Biol. Chem. 283, 1917619183.
Zeng, L., and Zhou, M.M. (2002). Bromodomain: an acetyl-lysine binding

domain. FEBS Lett. 513, 124128.
Margueron, R., Trojer, P., and Reinberg, D. (2005). The key to development:
interpreting the histone code? Curr. Opin. Genet. Dev. 15, 163176.
Zhao, M., Shirley, C.R., Hayashi, S., Marcon, L., Mohapatra, B., Suganuma, R.,
Behringer, R.R., Boissonneault, G., Yanagimachi, R., and Meistrich, M.L.
(2004). Transition nuclear proteins are required for normal chromatin condensation and functional sperm development. Genesis 38, 200213.
Resource
Sperm Methylation Profiles Reveal

Features of Epigenetic Inheritance
and Evolution in Primates
Antoine Molaro,1,3 Emily Hodges,1,3 Fang Fang,2 Qiang Song,2 W. Richard McCombie,1 Gregory J. Hannon,1,*
and Andrew D. Smith2,*
1Howard
Hughes Medical Institute, Cold Spring Harbor Laboratory, 1 Bungtown Road, Cold Spring Harbor, NY 11724, USA
and Computational Biology, University of Southern California, Los Angeles, CA 90089, USA
*Correspondence: hannon@cshl.edu (G.J.H.), andrewds@usc.edu (A.D.S.)
DOI 10.1016/j.cell.2011.08.016
2Molecular
SUMMARY
During germ cell and preimplantation development,

mammalian cells undergo nearly complete reprogramming of DNA methylation patterns. We profiled
the methylomes of human and chimp sperm as
a basis for comparison to methylation patterns of
ESCs. Although the majority of promoters escape
methylation in both ESCs and sperm, the corresponding hypomethylated regions show substantial
structural differences. Repeat elements are heavily
methylated in both germ and somatic cells; however,
retrotransposons from several subfamilies evade
methylation more effectively during male germ cell
development, whereas other subfamilies show the
opposite trend. Comparing methylomes of human
and chimp sperm revealed a subset of differentially
methylated promoters and strikingly divergent methylation in retrotransposon subfamilies, with an evolutionary impact that is apparent in the underlying
genomic sequence. Thus, the features that determine DNA methylation patterns differ between male
germ cells and somatic cells, and elements of these
features have diverged between humans and chimpanzees.
INTRODUCTION
In mammals, proper DNA methylation is essential for both fertility
and viability of offspring (Bestor, 1998; Bourchis and Bestor,
2004; Li et al., 1992; Okano et al., 1999; Walsh et al., 1998).
DNA methylation in germ cells is required for successful meiosis
(Bourchis and Bestor, 2004), and blastocysts derived from
embryonic stem cells (ESCs) lacking DNA methyltransferases
(DNMTs) cannot survive past approximately 10 days of development (Li et al., 1992).
Mammalian germ cells are derived from somatic cells, rather
than being set-aside during the first zygotic cleavages. During
germ cell development, the genome undergoes a wave of nearly

complete demethylation and remethylation (Popp et al., 2010;
Walsh et al., 1998). This reprogramming event correlates with
re-establishment of totipotency and with the creation of sexspecific methylation patterns at imprinted loci (reviewed by
Sasaki and Matsui, 2008). Germ cell methylation patterns are
erased and reset during a second wave of epigenetic reprogramming that occurs during preimplantation development. Postfertilization, DNA methylation levels reach a nadir around the
eight-cell stage, after which methylation is rewritten, attaining
its somatic level by the blastocyst stage (Mayer et al., 2000).
Because this is completed prior to the establishment of the inner
cell mass from which cultured ESCs are derived, one can view
ESCs and mature germ cells as the terminal products of the
two landmark epigenetic reprogramming events in mammals.
Mobile genetic elements constitute roughly half of most mammalian genomes (Lander et al., 2001). Repression of transposons
relies critically on DNA methylation and is essential for the
maintenance of genomic stability in the long term and of germ
cell function in the near term (Bestor, 1998; Bourchis and Bestor,
2004; Okano et al., 1999; Walsh et al., 1998). At least in part,
silencing of repeated DNA depends upon an abundant class of
PIWI-associated small RNAs, called piRNAs (reviewed in Aravin
and Hannon, 2008). In the absence of this pathway, methylation
is lost on at least some element copies, transposons are derepressed, and germ cell development is arrested in meiosis.
CpG dinucleotides are underrepresented in mammalian
genomes, most likely because a higher rate of spontaneous
deamination of methylated cytosines exerts evolutionary pressure for CpG depletion by frequent CpG-to-TpG transitions
(Duncan and Miller, 1980; Ehrlich et al., 1990). Mammalian
genomes contain areas of relatively high CpG density, called
CpG islands (CGIs) (Gardiner-Garden and Frommer, 1987),
which have avoided CpG depletion over evolutionary time.
CGIs are frequently observed at promoters and in some cases
have been shown to exert regulatory effects. Thus, selection
against CpG depletion may reflect the importance of specific
CpG dinucleotides as sequence-based binding sites or simply
the requirement for a certain regional density of CpGs. As an
alternative, the existence of CGIs may simply be an artifact of
longstanding hypomethylation of these regions, and consequent
Table 1. Shotgun Bisulfite Sequencing of Human and Chimp Sperm Methylomes

Species
Sample
Mapped
Distinct
Mismatches
BS Conversion
Methylation
CpG Coverage
CpGs Covered
Human
sperm (1)
609,127,589
388,835,058
1.58
0.992
0.724
8.8
0.96
sperm (2)
588,920,777
316,860,245
1.84
0.983
0.674
7.3
0.94
sperm (both)
1,198,048,366
705,695,303
1.70
0.988
0.701
16.1
0.96
0.93
Chimp
ESCs
940,731,922
366,844,212
0.64
0.988
0.663
14.1
sperm (1)
459,258,834
255,193,493
1.87
0.985
0.665
6.2
0.95
sperm (2)
520,905,232
327,796,614
1.70
0.984
0.672
7.4
0.94
sperm (both)
980,164,066
582,990,107
1.78
0.985
0.669
13.6
0.96
Mapped: reads mapping optimally to a single location in the reference genome. Distinct: number of genomic locations to which a read maps; when
multiple reads map to the same position, one with the best mapping score was selected at random, and all others discarded. Mismatches: average
number of mismatches for the reads indicated in the distinct fragments column. Bisulfite (BS) conversion rate was calculated at non-CpG cytosines.
Methylation: proportion of Cs in reads mapping over CpG dinucleotides.
relief from CpG erosion, in mammalian germ cells. Under this

hypo-deamination model, selective pressure is independent of
CpG density, per se, and CGIs may instead be a secondary
consequence of protection from methylation at specific sites
combined with prevalent methylation elsewhere in the genome
(Cooper and Krawczak, 1989; Duncan and Miller, 1980; Ehrlich
et al., 1990).
Studies encompassing evolutionarily distant species have
shown that broad features of the epigenome, such as the high
methylation levels of gene bodies and repeats, are deeply
conserved (Zemach et al., 2010). In closely related species,
however, fine-scale analysis of DNA methylation state reveals
variation. The chimpanzee and human genomes share more
than 95% sequence homology but display regions of differential
methylation (Enard et al., 2004). Through focused studies, we
have gained glimpses into the characteristics of the methylome
and the evolutionary pressures that shape it. We wished to
enable genome-wide comparisons of DNA methylation states
in closely related species and to examine possible differences
between the two major waves of epigenetic remodeling that
occur during the mammalian life cycle. We therefore produced
full-genome, single-CpG resolution DNA methylation profiles in
human and chimp sperm and compared these with methylation
maps from human ESCs (Laurent et al., 2010).
comparison, we applied our analysis pipeline to a whole-genome

bisulfite dataset from human ESCs (Laurent et al., 2010). This
dataset was comparable to our own, with 93% of CpG
dinucleotides covered and an average depth of 143 on CpGs
genome-wide.
We identified contiguous domains of low methylation, termed
hypomethylated regions or HMRs, in a manner independent of
genomic annotations such as CGIs and promoters. Because
methylation levels in sperm were generally high, HMRs appeared
obvious on browser plots as valleys in which methylation dropped to very low levels. To call HMRs in a statistically principled
manner, we designed a novel computational approach, based
on a two-state hidden Markov model with Beta-Binomial emission distributions (see Extended Experimental Procedures).
This algorithm identified 79k HMRs in human sperm and
70k HMRs in chimp sperm. Only 44.5k HMRs were identified
using the human ESC dataset, despite similar sequence
coverage and overall methylation level (Laurent et al., 2010;
see Table 1 and Table S1A available online). The sizes of
HMRs also differed between germ and ESCs. In both chimp
and human sperm, the mean size of HMRs was 1.8 kb, and
the median was 1.3 kb. In ESCs, HMRs showed a mean size
of 1.2 kb with a median of 833 bp. HMRs overlapped all classes
of genomic annotation (see Table S1B).
RESULTS
Global Comparisons among Primate Sperm Methylomes

and with Human ESCs
Average methylation levels differed by a small amount among
the human donors (donor 1: 72%; donor 2: 67%) but were
more similar among chimp donors (donors 1 and 2: 67%). The
methylation status of individual CpGs of HMRs correlated very
highly between individuals, with divergence being higher in
repeats as compared to promoters (Figures 1A and 1B). High
interindividual correlations at the CpG and the HMR levels imply
that our datasets permit accurate calling of CpG methylation
genome-wide.
We also compared methylation between species at an individual nucleotide level (see Extended Experimental Procedures
for details). As expected, the correlations between human and
chimp sperm methylation are high, but the correlation remains
generally highest within species.
Methylomes of Mature Male Germ Cells in Human

and Chimp
We conducted genome-wide shotgun bisulfite sequencing of
sperm DNA samples isolated from two human and chimp donors
(see Extended Experimental Procedures for details). Basic data
analysis was conducted using a custom pipeline. We were
able to determine methylation status for 96% of genomic
CpGs in the human and chimp samples from a total of 28 million
and 27 million CpGs, respectively (Table 1). Read coverage for
CpGs on autosomes averaged 163 in human with an overall
methylation level of 70% for all CpG sites. For chimp we
sequenced to an average coverage of nearly 143 and observed
an average methylation level of 67%. We did not observe
significant methylation at non-CpG sites in either dataset. For
Promoter
Figure 1. A Global View of Sperm and ESC

Methylomes
Repeat
an
u m (1)
a
C n(
hi
2
m )
p
C
(
1
hi
m )
p
H
um ( 2 )
a
H n(
u m 1)
a
C n(
hi
2
m )
p
C
(
1
hi
m )
p
H
um (2)
a
H n(
um 1)
a
C n(
hi
2
m )
p
C
(
1
hi
m )
p
(2
)
Genome-wide
ESC
Chimp (2)
Chimp (1)
Human (2)
um
0.64 0.70 0.70 0.70 0.67 0.73 0.66 0.65 0.26 0.32 0.25 0.22
0.87 0.86 0.91
0.93 0.91 0.95
0.69 0.68 0.80
0.88 0.87
0.93 0.91
0.70 0.68
0.89
0.94
0.78
Human (1)
Human (2)
Chimp (1)
Chimp (2)
ESC
CGI
0.95
0.85
0.84
0.64
0.79
0.63 0.29 0.24

0.70 0.32 0.27 1.00
0.93 0.36 0.31 0.98
0.35 0.30 0.98
0.58
0.43 0.96
0.75 0.66
0.96
um
H
C
G
Repeat
an
um (1)
a
C n(
hi
2
m )
p
C
(
1
hi
m )
p
ES ( 2
C )
C
G
I
H
um
a
H n(
u m 1)
a
C n(
hi
2
m )
p
C
(
1
hi
m )
p
ES ( 2
C )
C
G
I
(2
p
C
ES
m
hi
(1
Promoter
(2
0.82 0.62
0.68
0.81
0.80 0.90
0.64 0.58
0.79 0.75
hi
an
um
H
um
H
an
(1
Genome-wide
0.99 0.95
0.95
0.98
0.98 0.99
0.96 0.93
0.96 0.92
0.95 0.90 0.86

0.95 0.90 0.87 0.93
0.99 0.90 0.86 0.79
0.90 0.86 0.78
0.93
0.88 0.46
0.92 0.92
0.80
0.78 0.53
0.59
0.74
0.73 0.86
0.45 0.35
0.81 0.72
We also directly compared the methylomes from each of the

human and chimp donors with the human ESC methylome.
The nucleotide-level correlations between sperm methylation
of each of the four primate individuals were higher than their
correlations with ESC methylation patterns (Figure 1A). However,
the human ESC methylome did show substantially higher correlation with the human germ cell methylomes than with those of
chimp donors. Considered together these results indicate that,
although waves of reprogramming in developing germ cells
and embryos culminate in high genome-wide methylation, these
two methylomes bear substantial differences overall.
Comparison of Hypomethylated Promoters between
Sperm and ESC Methylomes
The majority of promoters are associated with HMRs in both
sperm and ESCs, indicating widespread bookmarking of
promoters during both waves of epigenetic reprogramming. A
number of promoters did show differential methylation, with
1336 showing sperm-specific HMRs but only 201 showing
ESC-specific HMRs (Figure 2A). Promoters hypomethylated in
germ cells were strongly enriched for putative binding sites of
transcription factors known to function in testis, including
NRF1, NF-Y, YY1, and CREB (see Figure S1). A similar analysis
of ESC-specific HMRs failed to yield significant results.
Only the genes with sperm-specific promoter hypomethylation revealed a strong enrichment for functional Gene Ontology
(GO) categories. These were associated with germ cell functions
(Figure 2B; Table S2) at distinct stages of gametogenesis (e.g.,
embryonic germ cell development and spermiogenesis). Thus,
genes acting at developmental stages, potentially separated
by decades, appear to maintain a permissive epigenetic state.
Of the eight genes analyzed from the piRNA metabolic process
category, seven showed promoter hypomethylation in sperm
but not in ESCs, and one was hypomethylated in both
(Figure 2B).
Retention of histones in human sperm was reported to be
extensive (Hammoud et al., 2009). Our analysis of this data revealed a strong correlation between retained histones marked
by H3K4me3 and HMRs at promoters. Among the 25.8k
promoters marked by H3K4me3 in sperm, 91% overlapped an
identified HMR. In general, these results support prior observations that the presence of H3K4me3 at promoters is often
0.54 0.11 0.03

0.61 0.13 0.03
0.89 0.13 0.04
0.13 0.04
0.35
0.09
0.72 0.65
(A) Correlations between methylomes with methylation

levels measured at individual CpG sites. Correlations are
displayed for CpGs genome-wide, within promoters, and
within repeats, and correlation coefficients are colored
blue to red to indicate low to high, respectively.
(B) Overlap between sets of HMRs from human sperm,
chimp sperm, and ESC methylomes, along with annotated
CGIs. Each cell gives the fraction of HMRs corresponding
to the row that overlaps HMRs corresponding to the
column. Colors are overlaid as in (A).
See also Table S1.
accompanied by hypomethylation (Hammoud et al., 2009; Ooi

et al., 2007).
It was previously posited that genes involved in early embryonic development had a distinct chromatin status in sperm,
being hypomethylated, histone-retained, enriched in H3K4me3
marks, and thus poised for expression (Hammoud et al., 2009).
At least with respect to DNA methylation, we do not detect a
preferential link between HMRs in sperm and developmental
regulators but instead widespread HMRs. One potential explanation for this perceived discrepancy is that our comparisons
involve sperm and ESCs, whereas prior studies used a differentiated cell type to contrast with sperm.
The genes with promoters that lack HMRs in both sperm and
ESCs (n = 5,380; Figure 2A) show strong enrichment for G
protein-coupled receptors and genes involved in neurological
functions (Tables S2C and S2D). The reason why many of these
genes, associated with highly specialized cell types, seem to
lack promoter HMRs in sperm and ESCs remains obscure.
Shared HMRs Show Distinct Characteristics in Sperm
and ESCs
Differences in average size and CpG densities suggest that
the HMRs emerging after germ cell reprogramming differ qualitatively from those emerging after zygotic reprogramming (Figure 3A; Table S1A). The majority of HMRs have CpG density
between 1% and 10%, and promoter HMRs fall almost exclusively in this range for the sperm methylomes. Those HMRs
falling below 1% CpG density lie almost exclusively in repeats.
These are overrepresented in human sperm relative to chimp
sperm and human ESCs. Promoter-associated HMRs have sizes
concentrated between 1 kb and 10 kb in human and chimp
sperm, with an overall trend to be broader than promoter-associated HMRs in ESCs (Figure 3A). A notable increase in CpG
density accompanies narrowing of HMRs and results in a significant portion of ESC HMRs with a CpG density above 10%.
To probe structural differences among HMRs in ESCs and
sperm, we plotted the average methylation around HMR-associated transcriptional start sites (TSSs), genome-wide (Figure 3B,
upper). This revealed a general principle, that a core HMR in
ESCs, referred to as a nested HMR (Figure 3B, lower), often
lies within an extended HMR in sperm. The median size of nested
ESC HMRs is 1,498, less than half the median size of 3,109 for
1.0
0.8
Sperm-specific hypo
(1,336 promoters)
ESC-specific hypo
(201 promoters)
0.6
Methylated in both
(5,380 promoters)
0.4
0.0
0.2
0.4
0.6
0.8
Figure 2. Differentially Reprogrammed Genes and

Their Functions
(A) Average methylation through promoters (1 kbp to +1
kbp) in human sperm and ESCs based on RefSeq gene
annotations. Promoters that were hypomethylated only in
sperm are shown in blue, those hypomethylated only in
ESCs in red, and promoters methylated in both are shaded
orange.
(B) Average methylation of promoters associated with GO
terms found enriched in the sperm-specific hypomethylated fraction (see A), with the addition of genes from the
embryonic development term. Individual genes involved
in the piRNA metabolic process are indicated as an
example.
See also Figure S1 and Table S2.
0.0
Average sperm methylation
Promoters considered: 24,037
0.2
1.0
Average ESC methylation
Homophilic
cell adhesion
Protein-DNA
complex assembly
0.8
0.6
0.4
0.2
0.0
Average sperm methylation
1.0
distance at the boundaries of HMRs (Figure 3E).

Because our method of identifying HMRs is
Meiosis
Sexual
agnostic to inter-CpG distance, this is not
reproduction
DNA packaging
Biological
simply an artifact of our approach. One could
adhesion
imagine increases in inter-CpG distance interEmbryonic development
rupting a processive activity, preventing the
spread of de novo methylation either directly
HENMT1
piRNA
TDRD1
or indirectly.
Metabolic
PIWIL2
process
Though we had no a priori expectation that
PIWIL1
PIWIL4
sequence features would reside at sperm or
TDRD9
ESC HMR boundaries, we searched for motifs
PLD6
that might occur at or near boundary CpGs,
MAEL
independent of CpG density. We noted a trend
0.0
0.2
0.4
0.6
0.8
1.0
toward enrichment for an ACGT motif at ESC
Average ESC methylation
boundary CpGs with a corresponding depletion
immediately outside ESC HMRs (Figure S2).
the sperm HMRs in which they reside. This phenomenon was This pattern was not significantly enriched at the boundaries of
also observed independently in a comparison of somatic and extended sperm HMRs. Building upon this observation, we
sperm HMRs, where variations in boundaries were additionally also searched for larger motifs, focusing on those containing a
correlated with tissue-specific expression (Hodges et al., central CpG core. Patterns with strong differences across
2011). Extended HMRs are reminiscent of the concept of CpG HMR boundaries tended to have the ACGT core (Table S3).
shores (Doi et al., 2009), though in comparisons of sperm and The most enriched pattern for sperm was AACGTT. For ESCs,
ESCs, we made no attempt to correlate gene expression with we saw a well-known E box pattern, CACGTG. Plotting
the widespread phenomenon of nesting that we report herein.
observed-to-expected (o/e) frequencies centered on CpGs
The observation of nested HMRs could arise either from a true around boundaries of extended and nested HMRs (Figure 3F),
expansion of the hypomethylated domain in sperm or as an there was a clear depletion just outside each boundary followed
artifact of sperm having less precise HMR boundaries than by a sharp enrichment at the boundary CpG for each pattern in
ESCs. Examining degrees of change in methylation states the appropriate cell type (Figure S2B). These results raise the
across boundary CpGs in both cell types supports the former possibility that one or more DNA-binding proteins might localize
conclusion (Figure 3C). Thus, nesting appears to represent a to HMR boundaries during waves of de novo methylation and
general phenomenon and likely reflects differences in the under- help to define transitions in methylation states.
lying mechanisms by which the boundaries of hypomethylated
regions are determined during the waves of de novo methylation Differential Repeat Methylation in Sperm and ESCs
that lead to sperm and ESCs.
Consistent with prior observations and with the known role of
As a step toward addressing such mechanisms, we asked DNA methylation in transposon silencing, most repeat elements
whether any features are associated with HMR boundaries in were highly methylated in both sperm and ESCs. However,
either cell type. Two interesting characteristics emerged. Ap- a substantial fraction of HMRs overlapped transposons in chimp
proaching the boundaries of either the extended sperm HMRs and human sperm, with all repeat classes represented (Figor the nested ESC HMRs, CpG densities dropped just prior to ure 4A; Table S1B). Fewer repeat-associated HMRs appeared
the start of the HMR and rose dramatically again thereafter, in ESCs. In sperm, HMRs collectively contained 4%5% of all
though overall densities were higher in the nested portions bases assigned to repeats, compared to 1.3% in ESCs (see
(Figure 3D). This reflects an increase in the average inter-CpG Table S1B). Overall, this suggests that different mechanisms,
Human sperm
Chimp sperm
Human ESC
CpG count
1k
100
10
10
100
1k
Promoter
10k
100k
100
1k
10k
HMR size (bp)
Repeat (non-promoter)
100k
100
Other
1k
10% CpG
10k
100k
1% CpG
C
ESC meth.
0.6
0.4
0.6
0.2
0.0
0.2
0.0
-4k
-3k
-2k
-1k
1k
2k
3k
4k
-5 -4 -3 -2 -1 +1 +2 +3 +4 +5
0.6
ESC
0.4
0.2
0.0
Schematic
Sperm
0.4
0.8
Methylation
Methylation
Methylation
0.8
Sperm meth.
0.8
-5 -4 -3 -2 -1 +1 +2 +3 +4 +5
CpG position relative to HMR boundary

Nested HMR
CpG O/E
D
0.8 Methylated
in both
0.6
Extended HMR (sperm only)
0.4
Nested HMR
(ESC+sperm)
0.2
-500
500
-500
240
Inter-CpG distance
Extended HMRs
500
180
60
180
120
E-Box O/E
1.5
Methylated
Extended HMR
0
-5 -4 -3 -2 -1 +1 +2 +3 +4 +5
CpG position relative

to HMR boundary
Nested HMR 1.5
1.0
1.0
0.5
0.5
0.0
ESC
60
Position w.r.t. boundary Position w.r.t. boundary
Sperm
120
-5 -4 -3 -2 -1 +1 +2 +3 +4 +5 -5 -4 -3 -2 -1 +1 +2 +3 +4 +5
0.0
CpG position relative to boundaries
Figure 3. Characteristics of HMRs Emerging from Germline and Somatic Reprogramming

(A) Log-scale plot depicting the sizes (in bases) and numbers of CpGs for all identified HMRs in human sperm (left), chimp sperm (middle), and human ESCs (right).
Diagonal lines indicate 10% CpG density (in green) and 1% CpG density (dashed line). HMRs are colored according to promoter overlap (red), overlap with
repeats but not promoters (blue), or overlap with neither (orange).
(B) Average methylation around all TSS overlapping HMRs in both sperm (orange) and ESCs (blue); solid lines represent data smoothed using a 20 base sliding
window. A schematic depicts the concepts of extended and nested HMRs at promoters.
(C) Average methylation at the 5 to +5 CpGs around boundaries of extended sperm HMRs and nested ESC HMRs (with the +1 CpG defined as the first inside an
HMR on either side).
(D) Ratios of observed-to-expected (o/e) CpG density for each nucleotide position relative to boundaries of extended sperm HMRs (left) and nested ESC HMRs
(right). Solid lines indicate values smoothed using a 20 base sliding window.
(E) Average inter-CpG distance for 5 to +5 CpGs around HMR boundaries of extended sperm and nested ESC HMRs. Upper and lower quartiles are reported
for each position.
(F) Ratio of o/e frequencies of the CACGTG pattern at 5 to +5 CpGs for extended sperm and nested ESC HMRs.
density /10kb
0.6
repeat hypomethylation rates
0.5
0.0 0.2 0.4 0.6 0.8 1.0
Fraction hypomethylated
Satellite
methylation (sperm)
Human sperm
0.4
0.3
p12.3
chr12
Chimp sperm
p12.1
12q12
2e07
q14.1
4e07
q15 21.1
6e07
21.2
q21.31
Human ESC
0.1
q23.3
1e08
24.31
1e08
2Mb
Sperm
HMR 1 _
Meth.
0.2
q22 q23.1
8e07
0_
N
tR
N
R
N
D
N
LI
LT
R
llit
SI
te
Sa
SV
A
0.0
ESC
HMR 1 _
Meth.
Unassembled
Centromere
0_
Transposon
Satellite
D
250
HMR overlap
CpG density
0.2
0.1
ESC
L1PA2
200
Sperm
Promoter
150
100
50
0
-1K
0
1K
5UTR
2K
3K
4K
5K
6K
0
Hypo Hyper Hypo Hyper Hypo Hyper Hypo Hyper
LINE
LTR
SINE
SVA
HMR overlap
2000
LTR12C
1500
1000
500
0
-2K
-1K
0
1K
LTR
2K
3K
4K
Figure 4. Differential Repeat Methylation during Male Germ Cell and Somatic Reprogramming
(A) For each repeat class, the proportion of elements that overlap HMRs is shown for human sperm (red), chimp sperm (orange), and ESCs (blue).
(B) Upper: Average methylation level (red) and satellite density (blue) in 10 kb sliding windows across chromosome 12. Lower: Chromosome 12 centromeric
region with HMRs (blue) and methylation level (orange) for human sperm and ESCs.
(C) CpG densities of hypomethylated repeat copies (red) and methylated repeat copies (yellow) for LINEs, LTRs, SINEs, and SVAs.
(D) HMR overlap distribution around full-length L1PA2 and LTR12 ERV9 elements for human sperm (blue) and ESCs (red).
with different stringencies, direct repeat methylation during

germ cell and preimplantation development.
pericentromeric satellites, suggesting that this is a conserved

property (Yamagata et al., 2007).
Sperm-Specific Satellite Hypomethylation

Is Concentrated at Centromeres
We noted a strong decrease in methylation of sperm DNA within
pericentromeric regions, extending several megabases outward
from the unassembled core centromeres (Figure 4B). This was
not seen in ESCs or in terminally differentiated cells (Hodges
et al., 2011). This striking pattern was attributable to spermspecific hypomethylation of 75%80% of the satellite repeats
concentrated in pericentromeric regions (Figure 4A). In ESCs,
only 16% of pericentromeric satellites were hypomethylated,
a figure in accord with the overall hypomethylation rates of
nonpericentromeric satellites in ESCs and sperm (Table S4A).
Prior studies of mouse germ cells using methylation-sensitive
restriction enzymes had noted selectively low methylation at
Retroelement Methylation Patterns Are Determined

at the Subfamily Level
Proper methylation of retrotransposons is required for transcriptional silencing of full-length and potentially active copies
(Bourchis and Bestor, 2004; Goodier and Kazazian, 2008; Walsh
et al., 1998). However, specific retroelements can be active
or unmethylated in male germ cells (e.g., AluY and AluYa5)
(Schmid, 1991). Given our read lengths, we were able to address
the methylation state of virtually all repeat families and most
individual copies (see Table S4B).
Overall, retrotransposon copies that were full length or close
to consensus showed a slight bias toward hypomethylation
(Figures S3A and S3B). However, neither of these attributes
could explain the variation observed in retrotransposon
Hypomethylated copies
A
80%
Human
specific
70%
60%
50%
Human sperm
40%
Chimp sperm
30%
Human ESC
20%
10%
0%
C
D
E
SVA Family
C
40
HMR
Chimp: 4% below 0.5

358 human/chimp
orthologous SVAs
20
10
20
15
SVA
Human (1)
CpG Methylation
number of copies
30
TLR1
Human: 36% below 0.5
10
5
0
0.0
0.2
0.4
0.6
0.8
1.0
methylation level
Human (2)
0
1
Chimp (1)
0
1
Chimp (2)
0
1
ESC
Figure 5. Divergent Methylation of SVA Elements between Human and Chimp

(A) Proportion of hypomethylated SVA copies hypomethylated according to subfamily (A to F) for human sperm (red), chimp sperm (orange), and ESCs (blue).
(B) The distribution of average methylation levels is shown for 358 human (lower) and chimp (upper) SVAs forming high-confidence orthologous pairs.
(C) An SVA insertion shared by human and chimp but with differential methylation between species.
methylation. Hypomethylated repeat copies did tend to have

greater CpG density, especially within the LTR and SVA (SINER, VNTR, and Alu) classes (Figure 4C). For long interspersed
nuclear elements (LINEs), LTR elements, and terminal repeats,
HMRs concentrated within regulatory regions, which often
show higher CpG density than their coding regions (Figures
4D; Figures S3C and S3D; Tables S4D and S4G). Short interspersed nuclear elements (SINEs) displayed a more uniform
hypomethylation (Figure S4E). Thus, similar mechanisms appear
to define HMRs in both repeat and nonrepeat portions of the
genome, as for most repeats, there is a strong association of
sperm HMRs with regulatory regions.
Among the LINEs, subfamilies of L1 were often hypomethylated in both sperm and ESCs, and these trended strongly
toward the active groups (Tables S4E and S4H). L1PA subfamilies are considered the most active in the human genome
(Khan et al., 2006), and the youngest of these (L1HS and
L1PA2) were among the very few subfamilies enriched for hypomethylation in ESCs relative to sperm. Specifically in sperm, we
noted hypomethylation of several other L1 families (e.g., L1PA416 and L1M3).
Among LTR subfamilies, sperm HMRs were enriched for ERV
elements (Table S4C). Hypomethylated copies exist either as
part of full-length provirus-like elements or as solo LTRs, with
the greatest enrichment for LTRs belonging to class I elements
(e.g., LTR12; see Tables S4D and S4G). The few LTR subfamilies
with more hypomethylated copies in ESCs than sperm are all
recently derived, human-specific ERVs (e.g., LTR5 and 13 and

HERVH LTR7).
Sperm hypomethylation has been previously reported for
primate Alu elements (Kochanek et al., 1993; Liu et al., 1994),
and our data revealed several Alu subfamilies with differential
methylation in sperm and ESCs, e,g., the AluY subfamily (Tables
S4F and S4I). The more precisely defined AluYa5 (human) and
AluYd4 (chimp) showed extreme enrichment for hypomethylation in sperm.
Species-Specific Methylation of the SVA Element
SVA elements showed strong, species-specific differences in
methylation in human and chimp sperm (Figure 4A). SVAs are
composite elements consisting of hexameric repeats, an Alulike region, a VNTR (variable number of tandem repeats) region,
and a SINE-R (Shen et al., 1994). SVA elements were active in the
most recent common ancestor of chimp and human (Mills et al.,
2006), and multiple examples of neoinsertions suggest that
they still cause genomic rearrangements and disease in human
(Ostertag et al., 2003).
Among the SVAs, the youngest subfamilies, DF (Wang et al.,
2005), showed the greatest frequency of hypomethylation in
human sperm (Figure 5A). Notably, these have a higher CpG
density than do older subfamilies. Three hundred and fifty-eight
SVA insertions can be assigned as high-confidence orthologs
between human and chimp, which remain highly similar in
sequence (see Extended Experimental Procedures). Methylation
through these element copies was distributed through the full

range from very low to very high average methylation, with two
modes near 20% and 80% methylation (Figure 5B). In human
sperm, 35% of orthologous SVAs had a methylation level below
50%. In sharp contrast, only 6% of copies fell below 50% methylation in chimp. We also annotated 921 SVA elements that
appear to represent new insertions occurring after the humanchimp divergence (Mills et al., 2006). 852 (93%) of these were
hypomethylated in sperm compared with only 62 (7%) in ESCs
(Figure 5A). Considered together, our data indicate that SVA
elements have come under different degrees of epigenetic
control in the human and chimp lineages.
Many SVA insertions occur at or around promoters (Lander
et al., 2001; Chimpanzee Sequencing and Analysis Consortium,
2005), and these elements often have a CpG content high
enough to fit the traditional definition of a CpG island. Given their
properties, SVA elements have the potential to introduce differential species- and cell type-specific methylation near genes
that may be relevant for their regulation. Figure 5C exemplifies
such a situation where, in the case of TLR1, no HMR exists
near the promoter in chimp sperm or human ESCs, but one is
contributed in human sperm by a nearby SVA element. Although
sperm are largely transcriptionally silent, similar HMRs are
expected to exist in transcriptionally active developing germ
cells (data not shown).
Signatures of Selection Accompany Differential
Methylation between Primates
CGIs are the most well known evolutionary signature of vertebrate DNA methylation. Their original definition required a CpG
o/e ratio of at least 0.6. Although the full set of HMRs in human
sperm and ESCs did not reach this empirical cut off, they did
pass the 0.4 benchmark used by Weber and colleagues (Figure 6A) (Weber et al., 2007). In general, promoter-associated
HMRs did surpass the 0.6 o/e cut off in both sperm and ESCs.
The differences in CpG density in nested and extended HMRs
(Figure 3B) imply distinct CpG depletion pressure in these
regions. Average CpG composition genome-wide is 0.2 o/e
but reaches 0.35 in extended HMRs and 0.68 in nested
HMRs. We analyzed sperm-specific and ESC-specific HMRs in
an attempt to decompose the CpG depletion pressure exerted
by the two methylomes. The ESC-specific HMRs reached only
0.35 o/e CpG composition, whereas the sperm-specific HMRs
reached a CpG composition of 0.5.
The life cycle of a germ cell can be separated into two components. The first is the time from fertilization to the time that
somatically derived primordial germ cells (PGCs) reach the
genital ridge. Second is the time during which the PGC develops
into a mature germ cell, which contributes to the zygote. The
latter period generally spans from birth to the end of the reproductive life of the animal. Our data suggest a model in which
methylation patterns present during both of these intervals shape
genomic CpG distributions but indicate a greater influence of
methylation profiles during germ cell maturation (Figure 6A).
We sought to measure the degree to which differential methylation could lead to CpG decay over the 6 million years of divergent evolution separating human and chimp. We focused on
regions that qualified as HMRs in either chimp or human, as
these regions could have either lost methylation along one

lineage or gained methylation along the other. For a given
regional methylation level, we measured CpG decay as the
proportion of regions having lost more than 5% of inferred
ancestral CpGs (using gorilla as outgroup) and plotted the relationship between average methylation and decay rate (Figure 6B). The correlation between regional methylation level and
CpG decay was extremely strong for both human and chimp.
These results indicate that CpG decay is appreciable as a function of methylation even over relatively brief evolutionary periods.
This observation predicted that we might see signatures of
selective pressure preventing erosion of some CpGs that are
maintained despite germline methylation. To address this question, we analyzed segregating sites at CpG dinucleotides using
data from the HapMap 3 project (CEU population; Altshuler
et al., 2010). CpGs were treated symmetrically, so each derived
allele at these sites can be classified as A, G, or T. As expected,
segregating sites with T as the derived allele represent the vast
majority.
We generated frequency spectra for each derived allele nucleotide with sites classified according to their methylation level in
sperm (Figure S4). As methylation levels increased, derived allele
frequencies shifted toward the low ends of the spectra (Figure 6C
and Figure S4). This shift was observed not only for derived TpG
alleles, which could be explained by an extreme bias in mutation
rate, but also for ApG and GpG derived alleles. One interpretation of these findings is that selection is on average weaker at
individual CpG sites with lower sperm methylation. Such an
interpretation is consistent with recent findings of Cohen et al.
(2011), who used sophisticated evolutionary models to posit
that selection for high CpG content is not a significant factor
contributing to maintenance of CGIs in the genome.
The strong connection between HMRs and gene promoters
suggests that the evolutionary gain or loss of HMRs may be associated with changes in selective pressure on functional regulatory
regions. To investigate this possibility, we analyzed sequence
divergence in HMRs, focusing on those that are human or
chimp specific. Because these differentially methylated regions
will have different rates of C-to-T transitions, we counted
changes from the inferred ancestor only at non-CpG sites.
Genomic intervals differing by more than 1% relative to the
inferred ancestor were counted as having divergent sequences.
Only 10% of HMRs shared between human and chimp
showed divergence from the ancestral sequence at non-CpG
sites (Figure 6D). At chimp-specific HMRs, 15% of human
sequences and 19% of chimp sequences diverged from the
inferred ancestor. At human-specific HMRs, 22% of human
sequences diverged and 18% of chimp sequences diverged.
These results indicated that changes in methylation state
between human and chimp are associated with accelerated
non-CpG sequence divergence. Interestingly, in both cases the
species with the lower methylation state had a greater rate of
divergence, which is consistent with adaptation at novel regulatory regions as a driver for these changes.
We only identified 104 promoters that are hypomethylated in
human but not in chimp sperm and only 52 genes with differential
promoter methylation in the opposite orientation. Neither set
showed significant enrichment for any ontology category.
0.6
CpG decay rate
0.4
0.2
0.0
HMRs Promoter Nested &
HMRs
extended
HMRs
Cellspecific
HMRs
0.8
0.6
r2 = 0.92
Human
r2 = 0.87
0.0
CGI
0.4
Traditional CGI cutoff
0.8
Chimp
0.2
CpG decay rate
HMM
0.8 0.0
methylated
0.6
GG&F
1.0
CpG Observed/Expected
sperm
0.4
ESC
0.2
0.0
0.2
0.4
0.6
0.8
1.0
Regional average methylation
D
0.4
Derived allele:
G
A
0.3
Fraction of sequences
Derived alleles < 5%
C
T
0.2
0.1
0.0
0.0
0.2
0.4
0.6
0.8
1.0
CpG methylation level
25%
Sequences diverged at
non-CpG sites in:
Human
Chimp
20%
15%
10%
5%
0%
Common
HMRs
Chimpspecific
HMRs
Humanspecific
HMRs
E
HMR
1
HTR3E (serotonin)
Human (1)
CpG Methylation
0
1
Human (2)
0
1
Chimp (1)
0
1
Chimp (2)
0
1
ESC
Figure 6. Sequence Features Associated with Methylome Divergence

(A) Ratio of o/e CpG density across all HMRs, those overlapping promoters, those sperm or ESC specific, and the extended/nested HMRs. Data for sperm are
indicated in blue and for ESCs are indicated in red; orange indicates ratio immediately outside extended HMRs.
(B) Frequency of regions under CpG decay as a function of methylation for both human and chimp at locations of HMRs in the other species. Decay is presented
for chimp in the upper panel and for human in the lower panel.
(C) Frequencies of rare derived alleles at CpG dinucleotides for each derived nucleotide, grouped according to methylation level in human sperm.
(D) Proportion of sequences displaying over 1% nucleotide divergence relative to the inferred ancestor using gorilla as an out-group and counting only
non-CpG sites.
(E) The promoter of the human HTR3E (serotonin receptor) gene contains an HMR in both human donors but in neither chimp donor.
However, analysis of genes with promoters within 10 kb of

an identified human-specific sperm HMR revealed a strong
enrichment for neuronal functions (see Table S5). The HTR3E
gene, a serotonin receptor subunit, is an example of such

a gene, whose promoter is selectively hypomethylated in human
sperm (Figure 6E).
DISCUSSION
Sperm Methylation Patterns Are Conserved
Overall, sperm methylation patterns were highly similar in all our
samples. However, there were differences, even among individuals. There has been much discussion regarding the role of
germline transmission of epigenetic marks in interindividual variation (Curley et al., 2011). Changes in epigenetic state could
allow flexibility in phenotype that could be reverted over short
time spans if a trait became disadvantageous. Erosion of CpG
content provides a mechanism to allow fixation of a positive trait
in the long run. Thus, changes in DNA methylation patterns
preceding changes in DNA sequence presents an attractive
model for at least one mode of adaptation. Although evaluating
such hypotheses will require many more datasets, the work presented here builds a firm foundation for such studies.
Most Promoters Have HMRs in Sperm
Global resetting of DNA methylation patterns happens twice
during mammalian development: once during germ cell development and once early in embryogenesis. Our data permit a
genome-scale analysis of these two events. Although high
genome-wide levels of methylation are re-established during
both waves of epigenetic remodeling, some regions are protected and establish HMR boundaries that appear relevant
even in fully differentiated somatic cells (Hodges et al., 2011).
A few promoters showed selective hypomethylation in sperm,
and these are strongly enriched for annotations related to germ
cell processes. Far fewer were selectively hypomethylated in
ESCs, and these were not enriched in any particular annotation
category. Promoters of genes retaining nucleosomes have
recently been shown to be hypomethylated in human sperm
(Hammoud et al., 2009), and both of these features have
been proposed to aid rapid activation during development. We
find that gene-associated hypomethylation in sperm can be
extended to more than 70% of all annotated genes in both
human and chimp. Among these we failed to find any enrichment
for regulators of early development. Instead, it seems that
promoter regions are generally identified and bookmarked in
sperm (see Zaidi et al., 2010).
Distinct Processes of HMR Formation Shape Germ Cell
and ESC Methylomes
Genome-wide, CpG sites seem to adopt a methylated state by
default (Edwards et al., 2010). This raises the problem of
precisely how regions that become HMRs are identified as
such. Regions of hypomethylation at promoters have been
correlated with regulatory DNA in various developmental
contexts (Illingworth et al., 2008; Laurent et al., 2010; Rollins
et al., 2006; Straussman et al., 2009). Based upon analysis of
histone marks and on the proposed binding properties of
DNMT3s (Dhayalan et al., 2010; Zhang et al., 2010), active transcription and accompanying methylation of K4 on histone H3 are
thought to locally inhibit the methylation machinery. This could
enable large-scale recognition of promoter regions if widespread
transcription occurs during fetal germ cell development as
genomic methylation patters are erased and reset. It is also plausible that specific protein/DNA complexes act locally even in the
absence of active transcription, to prevent access by de novo

methyltransferases. Proteins observed to function as boundary
elements, such as CTCF and Sp1 (reviewed in Gaszner and Felsenfeld, 2006), provide candidates for such functions.
Despite overall similarity in the sets of promoters they mark,
the HMRs observed at promoters in mature male germ cells
usually extend beyond the boundaries of HMRs in ESCs when
the two overlap. These wider HMRs do not seem to reflect less
precision in HMR boundaries, as methylation differences across
HMR boundaries are similar between sperm and ESCs. Because
this nested HMR phenomenon is observed at so many
promoters, it does not seem to be associated with the regulation
of any specific genes during germ cell development. We have
observed a clear increase in CpG content through the extended
portion of these HMRs relative to the genome-wide average,
suggesting that they have to some degree avoided pressure to
decay and hence are more than a transient state. The phenomenon that we observe is similar to the concept of CpG shores
(Doi et al., 2009). Perhaps the extended HMRs in germ cells
presage the extent of shores that correlate with changes in
gene expression.
Our data suggest that HMRs emerge from de novo methylation in male germ cells with sizes that differ from those that
emerge from somatic reprogramming. Thus, despite involvement of similar methyltransferases and targeting of similar sets
of sequences, the determinants of HMR sizes likely differ
between the two reprogramming events. We have begun to
see hints to the mechanisms determining such differences by
comparing boundary-associated motifs in sperm and ESCs.
Transposon Hypomethylation in Sperm
It is thought that germ cell genomes must be closely guarded
from the activity of mobile genetic elements. Although repeats
were generally heavily methylated, we did find HMRs that overlapped repeats, and these were substantially more prevalent in
sperm. We and others have characterized a conserved, small
RNA-based silencing pathway, termed the piRNA pathway,
that is important for recognizing and silencing mobile elements
in germ cells (Aravin and Hannon, 2008). Our data indicate that
both individual element copies and broader element subfamilies
can evade piRNA-based silencing. Yet, both these element
copies and element families are often efficiently silenced during
preimplantation development. This suggests fundamental differences in the mechanisms that recognize repeats and mark them
for repression during the two major waves of epigenetic reprogramming in mammals.
Examining patterns of repeat-associated HMRs is potentially
enlightening. HMRs are more prevalent in younger transposon
subfamilies, and the hypomethylated regions themselves tend
to overlap with promoters or regulatory regions, just as they do
in genes. Thus, it may be that active elements evade default
methylation by being initially recognized as gene-like as a consequence of their binding transcription factors and possibly even
being transcribed. In these cases, we imagine that silencing of
most elements would be enforced by the piRNA pathway but
that some sites, such as those we observe herein, might still
escape. A number of examples can be cited in support of this
hypothesis. The 50 untranslated regions (UTRs) of the L1PA
subfamilies are known to carry conserved YY1-binding sites,

whereas other recent subfamilies acquired RUNX3- and SRYbinding motifs, all of which could promote transcription in developing germ cells (Khan et al., 2006; Lee et al., 2010). Similarly,
the sperm-enriched hypomethylated EVR9 LTR12 elements
have been shown to bind NF-Y, MZF1, and GATA-2 in erythroid
K562 cells (Yu et al., 2005). In each of these cases, HMRs within
these elements tend to encompass such potential transcription
factor-binding sites.
Similarly, Alu RNAs have been detected in human sperm (Kochanek et al., 1993). This suggests a potential link between Alu
HMRs and the transcriptional activity of individual repeats,
though previous studies also reported that the binding of
SABP across Alu elements in sperm prevents their methylation
(Chesnokov and Schmid, 1995). Interestingly, Alu hypomethylation is not seen in female germ cells (Liu et al., 1994) and has
been proposed as one mediator of sex-specific imprints.
Centromeric Satellite Methylation
Satellites resist methylation in sperm when localized in clusters
at centromeres but are generally methylated when located
elsewhere even if they are clustered. This is consistent with
previous observations made in mouse through the use of methylation-sensitive enzymes (Yamagata et al., 2007). Recent
reports have shown that the transient transcriptional activation
of paternal pericentromeric satellites was essential for centromeric heterochromatin formation in two-cell zygotes (Probst
et al., 2010). This could indicate that hypomethylation of satellite
repeats in male germ cell marks paternal centromeres, in a
manner similar to imprinting, allowing their rapid transcriptional
activation upon fertilization.
In addition to a characteristic location within chromocenters
in sperm, centromeres display a distinct chromatin structure
differentiating them regionally during meiosis from other chromosomal regions (reviewed by Dalal, 2009). This has prompted
suggestions that centromeric chromatin states might be critical
for proper meiosis, a hypothesis strongly supported by our
observation of selective hypomethylation of megabase domains
of centromeric satellite clusters. Prior studies have demonstrated that derepression of satellite repeats in mitotic cells
creates segregation defects due to the formation of anaphase
bridges (Frescas et al., 2008). Low methylation levels have also
been correlated with the ability to bind cohesin complexes
(Parelho et al., 2008). Considered as a whole, these observations
suggest a model in which selective hypomethylation of centromeric satellites might be critical for accurate chromosome
segregation during meiosis.
Differential Repeat Methylation between Species
The most striking example of species-specific methylation to
emerge from our analysis involved the SVA elements. These
primate-specific composite elements contain a high density of
CpGs, remain active in human and chimp, and include many
copies that are clear orthologs between human and chimp (Bantysh and Buzdin, 2009; Mills et al., 2006). Transduction of SVAs
has been implicated in human diseases and gene formation
(Damert et al., 2009; Ostertag et al., 2003). Our results indicate
that for a subset of SVA elements, the ability to methylate these
elements has either been acquired along the chimp lineage or

lost in the human lineage during the past 6 million years, despite
very little sequence change in these elements.
Mutual Canalization of the Genome and the Epigenome
It has been thought that CGIs arose as the result of protection
from methylation-associated deamination over long evolutionary
periods. This is consistent with the observed correlation
between the location of CGIs and regions that lack methylation
in both germline and somatic cells. However, recent results
have pointed to functions for CGIs that may be associated with
their high CpG density (Thomson et al., 2010), with the plausible
interpretation that selection may be acting to preserve CpG
density in CGIs. We find that although most CGIs fall within
HMRs of sperm, most HMRs extend well beyond the annotated
CGIs, even using weaker CGI definitions. Thus, hypomethylated
regions in male germ cells do not appear to require a critical CpG
density to avoid methylation. Instead, our results are consistent
with CGIs arising as a consequence of different mutational pressures rather than selection for CpG density.
In our datasets, signatures of deamination-induced CpG
depletion are clear. Yet we also observe CpG depletion from
many sperm and ESC HMRs. Several scenarios could resolve
this conundrum. For example, such regions may have been
methylated for substantial periods prior to assuming their unmethylated status. Thus, they may have decayed at some time in the
past but are now stabilized by their hypomethylated status. Such
sites could also actually be methylated during a period of germ
cell development to which our current datasets are blind (e.g.,
in fetal gonocytes or female germ cells). In accord with this
explanation, we have observed distinct CpG densities associated with sperm-specific and ESC-specific HMRs. Moreover,
at HMRs where the only central, nested portion is hypomethylated in ESCs, we observe greater CpG retention through regions
hypomethylated in both ESCs and sperm. Overall, we cannot
exclude a model in which selection acts to preserve critical
functions requiring specific local CpG densities. However, our
results lend additional support to recent conclusions of Cohen
et al. (2011), whose sophisticated evolutionary modeling showed
that CGIs can be explained without invoking selection on CpG
sites. Our results suggest a refinement of the hypo-deamination
model in which CpG retention is a function of the time spent
hypomethylated during each generation in germ cells and their
somatic precursors.
The detailed comparative analysis performed here has revealed that, over the 6 million years since the divergence of
human and chimp, most patterns of DNA methylation remain
conserved in male germ cells. We have directly related evolutionary changes in CpG methylation with loss of CpG dinucleotides and have shown that even small differences in methylation
can lead to substantial loss of CpGs over relatively short evolutionary periods. At the same time, there are many genomic
regions that are highly conserved in sequence yet show quite
different patterns of methylation. This could indicate an ability
of the genome and the epigenome to evolve independently.
However, we do find that the most drastic changes in methylation between human and chimp, where an HMR in one species
shows high levels of methylation in the other, are accompanied
by an increased sequence divergence even at non-CpG dinucleotides. One interpretation is that most species-specific HMRs
have arisen newly along one lineage with these novel functional
elements showing signs of recent adaptation. On the other hand,
if this accelerated sequence change were more a reflection of
relaxed selective pressure, we would expect species-specific
HMRs to more frequently result from loss of functional elements
along the opposite lineage. Resolution of these questions can
only come from a broadening to many more species of the
studies reported herein.
Detailed methods can be found in the Extended Experimental Procedures.
Sperm Collection
Two anonymous human donors were used and data pooled after sequencing.
Two chimp donors were used. Semen was collected at the New Iberia
Research Center (New Liberia, LA) or the Southwest National Primate
Research Center (San Antonio, TX, USA). Coagulated semen was separated
from the liquid phase manually. Both human and chimp samples were diluted
(1:1) in HBS buffer (0.01M HEPES, ph 7.4; 150 mM NaCl) and passed though
a silica-based gradient, SpermFilter (Cryobiosystems), by centrifugation
(according to manufacturers instructions).
Library Preparation
DNA from 100 million cells was extracted and sheared to a size of 150200
nt by sonication. Double-stranded DNA fragments were end repaired, A-tailed,
and ligated to methylated Illumina adaptors. Ligated fragments were bisulfite
converted using the EZ-DNA Methylation-Gold Kit (Zymo research). Following
PCR enrichment, fragments of 340 to 360 bp were size selected and
sequenced.
Computational Methods
Reads were mapped with RMAPBS (Smith et al., 2009). The accuracy of our
mapping method is discussed in the Extended Experimental Procedures.
Mapped reads were used to infer the methylation frequency at each CpG
dinucleotide. These frequencies, along with the number of reads contributing
to each frequency estimate, were supplied to a segmentation algorithm used
to identify HMRs. Ortholog mapping between human and chimp was done with
the liftOver tool available through the UCSC Genome Browser. Sequence
conservation between human, chimp, and was measured based on MULTIZ
44-way vertebrate alignments, also available through the UCSC Genome
Browser. Complete details of all computational methods are provided in the
Extended Experimental Procedures.
ACCESSION NUMBERS
Data analyzed herein have been deposited in GEO with accession GSE30340.
was supported in part by grants from the NIH (R01HG005238 and

1RC2HD064459) and by a kind gift from Kathryn W. Davis.
REFERENCES
Altshuler, D.M., Gibbs, R.A., Peltonen, L., Altshuler, D.M., Gibbs, R.A., Peltonen, L., Dermitzakis, E., Schaffner, S.F., Yu, F., Peltonen, L., et al; International
HapMap 3 Consortium. (2010). Integrating common and rare genetic variation
in diverse human populations. Nature 467, 5258.
Aravin, A.A., and Hannon, G.J. (2008). Small RNA silencing pathways in germ
and stem cells. Cold Spring Harb. Symp. Quant. Biol. 73, 283290.
Bantysh, O.B., and Buzdin, A.A. (2009). Novel family of human transposable
elements formed due to fusion of the first exon of gene MAST2 with retrotransposon SVA. Biochemistry (Mosc.) 74, 13931399.
Bestor, T.H. (1998). Cytosine methylation and the unequal developmental
potentials of the oocyte and sperm genomes. Am. J. Hum. Genet. 62, 1269
1273.
Bourchis, D., and Bestor, T.H. (2004). Meiotic catastrophe and retrotransposon reactivation in male germ cells lacking Dnmt3L. Nature 431, 9699.
Chesnokov, I.N., and Schmid, C.W. (1995). Specific Alu binding protein from
human sperm chromatin prevents DNA methylation. J. Biol. Chem. 270,
1853918542.
Chimpanzee Sequencing and Analysis Consortium. (2005). Initial sequence of
the chimpanzee genome and comparison with the human genome. Nature
437, 6987.
Cohen, N.M., Kenigsberg, E., and Tanay, A. (2011). Primate CpG islands
are maintained by heterogeneous evolutionary regimes involving minimal
selection. Cell 145, 773786.
Cooper, D.N., and Krawczak, M. (1989). Cytosine methylation and the fate of
CpG dinucleotides in vertebrate genomes. Hum. Genet. 83, 181188.
Curley, J.P., Mashoodh, R., and Champagne, F.A. (2011). Epigenetics and the
origins of paternal effects. Horm. Behav. 59, 306314.
Dalal, Y. (2009). Epigenetic specification of centromeres. Biochem. Cell Biol.
87, 273282.
Damert, A., Raiz, J., Horn, A.V., Lower, J., Wang, H., Xing, J., Batzer, M.A.,
Lower, R., and Schumann, G.G. (2009). 50 -Transducing SVA retrotransposon
groups spread efficiently throughout the human genome. Genome Res. 19,
19922008.
Dhayalan, A., Rajavelu, A., Rathert, P., Tamas, R., Jurkowska, R.Z., Ragozin,
S., and Jeltsch, A. (2010). The Dnmt3a PWWP domain reads histone 3 lysine
36 trimethylation and guides DNA methylation. J. Biol. Chem. 285, 26114
26120.

four figures, and five tables and can be found with this article online at
doi:10.1016/j.cell.2011.08.016.
Doi, A., Park, I.H., Wen, B., Murakami, P., Aryee, M.J., Irizarry, R., Herb, B.,
Ladd-Acosta, C., Rho, J., Loewer, S., et al. (2009). Differential methylation of
tissue- and cancer-specific CpG island shores distinguishes human induced
pluripotent stem cells, embryonic stem cells and fibroblasts. Nat. Genet. 41,
13501353.
ACKNOWLEDGMENTS
Duncan, B.K., and Miller, J.H. (1980). Mutagenic deamination of cytosine residues in DNA. Nature 287, 560561.
We thank Michelle Rooks, Pramod Thekkat, and Colin Malone for help with
experimental procedures and Assaf Gordon, Luigi Manna, and the CSHL
and USC High Performance Computing Centers for computational support.
We thank Babette Fontenot (New Iberia Research Center) and Jerilyn Pecotte
(Southwest National Primate Center) for help with chimp sperm collection. We
thank Sergey Nuzhdin, Ed Green, Peter Calabrese, Maren Friesen, Magnus
Norborg, and Marie-Stanislas Remigereau for helpful discussions. This work
Edwards, J.R., ODonnell, A.H., Rollins, R.A., Peckham, H.E., Lee, C., Milekic,
M.H., Chanrion, B., Fu, Y., Su, T., Hibshoosh, H., et al. (2010). Chromatin and
sequence features that define the fine and gross structure of genomic methylation patterns. Genome Res. 20, 972980.
Ehrlich, M., Zhang, X.Y., and Inamdar, N.M. (1990). Spontaneous deamination
of cytosine and 5-methylcytosine residues in DNA and replacement of 5-methylcytosine residues with cytosine residues. Mutat. Res. 238, 277286.
Enard, W., Fassbender, A., Model, F., Adorjan, P., Paabo, S., and Olek, A.
(2004). Differences in DNA methylation patterns between humans and chimpanzees. Curr. Biol. 14, R148R149.
Frescas, D., Guardavaccaro, D., Kuchay, S.M., Kato, H., Poleshko, A., Basrur,
V., Elenitoba-Johnson, K.S., Katz, R.A., and Pagano, M. (2008). KDM2A
represses transcription of centromeric satellite repeats and maintains the
heterochromatic state. Cell Cycle 7, 35393547.
Gardiner-Garden, M., and Frommer, M. (1987). CpG islands in vertebrate
genomes. J. Mol. Biol. 196, 261282.
Gaszner, M., and Felsenfeld, G. (2006). Insulators: exploiting transcriptional
and epigenetic mechanisms. Nat. Rev. Genet. 7, 703713.
Goodier, J.L., and Kazazian, H.H., Jr. (2008). Retrotransposons revisited: the
restraint and rehabilitation of parasites. Cell 135, 2335.
Hammoud, S.S., Nix, D.A., Zhang, H., Purwar, J., Carrell, D.T., and Cairns, B.R.
(2009). Distinctive chromatin in human sperm packages genes for embryo
development. Nature 460, 473478.
Hodges, E., Molaro, A., Dos Santos, C.O., Thekkat, P., Song, Q., Uren, P.,
Park, J., Butler, J., Rafii, S., McCombie, W.R., Smith, A.D., and Hannon,
G.J. (2011). Directional DNA methylation changes and complex intermediate
states accompany lineage specificity in the adult hematopoietic compartment.
Mol. Cell. Published online September 15 2011. 10.1016/j.cell.2008.06.028.
Parelho, V., Hadjur, S., Spivakov, M., Leleu, M., Sauer, S., Gregson, H.C., Jarmuz, A., Canzonetta, C., Webster, Z., Nesterova, T., et al. (2008). Cohesins
functionally associate with CTCF on mammalian chromosome arms. Cell
132, 422433.
Popp, C., Dean, W., Feng, S., Cokus, S.J., Andrews, S., Pellegrini, M., Jacobsen, S.E., and Reik, W. (2010). Genome-wide erasure of DNA methylation in
mouse primordial germ cells is affected by AID deficiency. Nature 463,
11011105.
Probst, A.V., Okamoto, I., Casanova, M., El Marjou, F., Le Baccon, P., and
Almouzni, G. (2010). A strand-specific burst in transcription of pericentric
satellites is required for chromocenter formation and early mouse development. Dev. Cell 19, 625638.
Rollins, R.A., Haghighi, F., Edwards, J.R., Das, R., Zhang, M.Q., Ju, J., and
Bestor, T.H. (2006). Large-scale structure of genomic methylation patterns.
Genome Res. 16, 157163.
Sasaki, H., and Matsui, Y. (2008). Epigenetic events in mammalian germ-cell
development: reprogramming and beyond. Nat. Rev. Genet. 9, 129140.
Schmid, C.W. (1991). Human Alu subfamilies and their methylation revealed by
blot hybridization. Nucleic Acids Res. 19, 56135617.
Illingworth, R., Kerr, A., Desousa, D., Jrgensen, H., Ellis, P., Stalker, J., Jackson, D., Clee, C., Plumb, R., Rogers, J., et al. (2008). A novel CpG island set
identifies tissue-specific methylation at developmental gene loci. PLoS Biol.
6, e22.
Shen, L., Wu, L.C., Sanlioglu, S., Chen, R., Mendoza, A.R., Dangel, A.W.,
Carroll, M.C., Zipf, W.B., and Yu, C.Y. (1994). Structure and genetics of the
partially duplicated gene RP located immediately upstream of the complement
C4A and the C4B genes in the HLA class III region. Molecular cloning, exonintron structure, composite retroposon, and breakpoint of gene duplication.
J. Biol. Chem. 269, 84668476.
Khan, H., Smit, A., and Boissinot, S. (2006). Molecular evolution and tempo of
amplification of human LINE-1 retrotransposons since the origin of primates.
Genome Res. 16, 7887.
Smith, A.D., Chung, W.Y., Hodges, E., Kendall, J., Hannon, G., Hicks, J., Xuan,
Z., and Zhang, M.Q. (2009). Updates to the RMAP short-read mapping software. Bioinformatics 25, 28412842.
Kochanek, S., Renz, D., and Doerfler, W. (1993). DNA methylation in the Alu
sequences of diploid and haploid primary human cells. EMBO J. 12, 1141
1151.
Straussman, R., Nejman, D., Roberts, D., Steinfeld, I., Blum, B., Benvenisty,
N., Simon, I., Yakhini, Z., and Cedar, H. (2009). Developmental programming
of CpG island methylation profiles in the human genome. Nat. Struct. Mol.
Biol. 16, 564571.
Lander, E.S., Linton, L.M., Birren, B., Nusbaum, C., Zody, M.C., Baldwin, J.,
Devon, K., Dewar, K., Doyle, M., FitzHugh, W., et al; International Human
Genome Sequencing Consortium. (2001). Initial sequencing and analysis of
the human genome. Nature 409, 860921.
Thomson, J.P., Skene, P.J., Selfridge, J., Clouaire, T., Guy, J., Webb, S., Kerr,
A.R., Deaton, A., Andrews, R., James, K.D., et al. (2010). CpG islands influence
chromatin structure via the CpG-binding protein Cfp1. Nature 464, 10821086.
Laurent, L., Wong, E., Li, G., Huynh, T., Tsirigos, A., Ong, C.T., Low, H.M., Kin
Sung, K.W., Rigoutsos, I., Loring, J., et al. (2010). Dynamic changes in the
human methylome during differentiation. Genome Res. 20, 320331.
Walsh, C.P., Chaillet, J.R., and Bestor, T.H. (1998). Transcription of IAP endogenous retroviruses is constrained by cytosine methylation. Nat. Genet. 20,
116117.
Lee, S.H., Cho, S.Y., Shannon, M.F., Fan, J., and Rangasamy, D. (2010). The
impact of CpG island on defining transcriptional activation of the mouse L1
retrotransposable elements. PLoS ONE 5, e11353.
Wang, H., Xing, J., Grover, D., Hedges, D.J., Han, K., Walker, J.A., and Batzer,
M.A. (2005). SVA elements: a hominid-specific retroposon family. J. Mol. Biol.
354, 9941007.
Li, E., Bestor, T.H., and Jaenisch, R. (1992). Targeted mutation of the DNA
methyltransferase gene results in embryonic lethality. Cell 69, 915926.
Weber, M., Hellmann, I., Stadler, M.B., Ramos, L., Paabo, S., Rebhan, M., and
Schubeler, D. (2007). Distribution, silencing potential and evolutionary impact
of promoter DNA methylation in the human genome. Nat. Genet. 39, 457466.
Liu, W.M., Maraia, R.J., Rubin, C.M., and Schmid, C.W. (1994). Alu transcripts:
cytoplasmic localisation and regulation by DNA methylation. Nucleic Acids
Res. 22, 10871095.
Mayer, W., Niveleau, A., Walter, J., Fundele, R., and Haaf, T. (2000). Demethylation of the zygotic paternal genome. Nature 403, 501502.
Mills, R.E., Bennett, E.A., Iskow, R.C., Luttig, C.T., Tsui, C., Pittard, W.S., and
Devine, S.E. (2006). Recently mobilized transposons in the human and chimpanzee genomes. Am. J. Hum. Genet. 78, 671679.
Okano, M., Bell, D.W., Haber, D.A., and Li, E. (1999). DNA methyltransferases
Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian
development. Cell 99, 247257.
Ooi, S.K., Qiu, C., Bernstein, E., Li, K., Jia, D., Yang, Z., Erdjument-Bromage,
H., Tempst, P., Lin, S.P., Allis, C.D., et al. (2007). DNMT3L connects unmethylated lysine 4 of histone H3 to de novo methylation of DNA. Nature 448,
714717.
Ostertag, E.M., Goodier, J.L., Zhang, Y., and Kazazian, H.H., Jr. (2003).
SVA elements are nonautonomous retrotransposons that cause disease in
humans. Am. J. Hum. Genet. 73, 14441451.
Yamagata, K., Yamazaki, T., Miki, H., Ogonuki, N., Inoue, K., Ogura, A., and
Baba, T. (2007). Centromeric DNA hypomethylation as an epigenetic signature
discriminates between germ and somatic cell lineages. Dev. Biol. 312,
419426.
Yu, X., Zhu, X., Pi, W., Ling, J., Ko, L., Takeda, Y., and Tuan, D. (2005). The long
terminal repeat (LTR) of ERV-9 human endogenous retrovirus binds to NF-Y
in the assembly of an active LTR enhancer complex NF-Y/MZF1/GATA-2.
J. Biol. Chem. 280, 3518435194.
Zaidi, S.K., Young, D.W., Montecino, M.A., Lian, J.B., van Wijnen, A.J., Stein,
J.L., and Stein, G.S. (2010). Mitotic bookmarking of genes: a novel dimension
to epigenetic control. Nat. Rev. Genet. 11, 583589.
Zemach, A., McDaniel, I.E., Silva, P., and Zilberman, D. (2010). Genome-wide
evolutionary analysis of eukaryotic DNA methylation. Science 328, 916919.
Zhang, Y., Jurkowska, R., Soeroes, S., Rajavelu, A., Dhayalan, A., Bock, I.,
Rathert, P., Brandt, O., Reinhardt, R., Fischle, W., and Jeltsch, A. (2010). Chromatin methylation activity of Dnmt3a and Dnmt3a/3L is guided by interaction
of the ADD domain with the histone H3 tail. Nucleic Acids Res. 38, 42464253.
Correction
Complete Kinetochore Tracking Reveals

Error-Prone Homologous Chromosome
Biorientation in Mammalian Oocytes
Tomoya S. Kitajima, Miho Ohsugi, and Jan Ellenberg*
*Correspondence: jan.ellenberg@embl.de
DOI 10.1016/j.cell.2011.08.028
(Cell 146, 568581; August 19, 2011)

The supplemental files that were originally published online with the above article inadvertently included the incorrect movies.
The correct movies are now available with the article online.
Stochastic State Transitions

Give Rise to Phenotypic Equilibrium
in Populations of Cancer Cells
Piyush B. Gupta,* Christine M. Fillmore, Guozhi Jiang, Sagi D. Shapira, Kai Tao, Charlotte Kuperwasser,
and Eric S. Lander*
*Correspondence: pgupta@wi.mit.edu (P.B.G.), lander@broadinstitute.org (E.S.L.)
DOI 10.1016/j.cell.2011.08.030
(Cell 146, 633644; August 19, 2011)

In the above article, mammary epithelial cells were sorted on the basis of markers, including CD24. Due to a copyediting error, the
superscript neg was inadvertently removed from CD24neg throughout the text. The corrected article is now available online.
Correction
Complete Kinetochore Tracking Reveals

Error-Prone Homologous Chromosome
Biorientation in Mammalian Oocytes
Tomoya S. Kitajima, Miho Ohsugi, and Jan Ellenberg*
*Correspondence: jan.ellenberg@embl.de
DOI 10.1016/j.cell.2011.08.028
(Cell 146, 568581; August 19, 2011)

The supplemental files that were originally published online with the above article inadvertently included the incorrect movies.
The correct movies are now available with the article online.
Stochastic State Transitions

Give Rise to Phenotypic Equilibrium
in Populations of Cancer Cells
Piyush B. Gupta,* Christine M. Fillmore, Guozhi Jiang, Sagi D. Shapira, Kai Tao, Charlotte Kuperwasser,
and Eric S. Lander*
*Correspondence: pgupta@wi.mit.edu (P.B.G.), lander@broadinstitute.org (E.S.L.)
DOI 10.1016/j.cell.2011.08.030
(Cell 146, 633644; August 19, 2011)

In the above article, mammary epithelial cells were sorted on the basis of markers, including CD24. Due to a copyediting error, the
superscript neg was inadvertently removed from CD24neg throughout the text. The corrected article is now available online.
Erratum
Synaptic PRG-1 Modulates

Excitatory Transmission
via Lipid Phosphate-Mediated Signaling
Thorsten Trimbuch, Prateep Beed, Johannes Vogt, Sebastian Schuchmann, Nikolaus Maier, Michael Kintscher,
Jorg Breustedt, Markus Schuelke, Nora Streu, Olga Kieselmann, Irene Brunk, Gregor Laube, Ulf Strauss, Arne Battefeld,
Hagen Wende, Carmen Birchmeier, Stefan Wiese, Michael Sendtner, Hiroshi Kawabe, Mika Kishimoto-Suga, Nils Brose,
Jan Baumgart, Beate Geist, Junken Aoki, Nic E. Savaskan, Anja U. Brauer, Jerold Chun, Olaf Ninnemann,
Dietmar Schmitz,* and Robert Nitsch*
*Correspondence: robert.nitsch@unimedizin-mainz.de (R.N.), dietmar.schmitz@charite.de (D.S.)
DOI 10.1016/j.cell.2011.08.029
(Cell 138, 12221235; September 18, 2009)

It has been brought to our attention that, in Figure 2C of the article above, the in vivo recordings for P22 PRG-1 KO mice are identical.
Upon re-examination of the original recordings, we found that the recording from the right hemisphere was mistakenly used to also
represent the recording from the left hemisphere. All in vivo recordings and related figures were made by Sebastian Schuchmann.
The corrected figure with the appropriate recording for the left hemisphere is now presented below. This error was exclusive to P22
of Figure 2C and does not affect the article beyond Figure 2C, neither the original data underlying Figure 2C nor the description in
the Results section or the conclusions resulting from these data. We apologize for the mistake and for any inconvenience caused
to the readers, and we thank the alert reader who discovered the error.
SnapShot: HighThroughput
Sequencing Applications
Hong Han,1 Razvan Nutiu,1 Jason Moffat,1 and Benjamin J. Blencowe1
Banting and Best Department of Medical Research, University of Toronto, Toronto, ON M5S 3E1, Canada
Genome Sequencing/Resequencing
Genome-wide polymorphism and mutation mapping
Genome assembly
CpG
METHYLATION
Chromatin Immunoprecipitation Sequencing

(ChIP-Seq)
G
3
Nucleosome component, Transcription factor (TF),

RNA polymerase II (Pol II) occupancy
Histone methylation or acetylation

TF
Pol ymer as e II
com pl ex
en
ic
Methyl-Seq/Bisulfite-Seq (DNA methylation status)

DNase-Seq (DNase hypersensitivity)
TF
A SNP
RNA Pol II/TF
Relative enrichment
Chromosome
H3K36me3
TF
TF
H3K36me3
H2K27ac
Exon
5 UTR
Polymerase II
Pol II
TF
RBP
Transcription
factors
Constitutively
spliced exon
TF
RBP
TF
Nucleosome
3 UTR
Cap
RBP
TRANSCRIPTION
RB
Intron Alt.
Exon
RBP
Crosslinking Immunoprecipitation Sequencing

(CLIP-Seq/HITS-CLIP)
N as cen t R N A
Transcriptome-wide RNA-binding protein (RBP) maps
SPLICING
Intron
Alternatively
spliced exon
Modified Clip For Site-specific Crosslinking

Photoactivatable ribonucleoside-enhanced CLIP (PAR-CLIP)
RBP
RNA-binding
proteins
Transcriptome Sequencing/RNA-Seq
Individual nucleotide resolution CLIP (iCLIP)
Total RNA, total RNA minus rRNA, poly(A)-selected RNA

Gene expression profiling
Long noncoding RNA profiling
Alternative splicing and trans-splicing profiling
Alternative polyadenylation profiling
Mapping transcription initiation sites
Mapping RNA editing sites (coupled with DNA-Seq)
Targeted RNA-Seq, Direct RNA-Seq, Strand-specific

RNA-Seq, Nascent RNA-Seq
AAAAAAAA
AAAA
Eg. Nova RNA-binding regulatory map
Poly(A) tail
ACTIVE TRANSLATION
Protein
m R N A EX P O R T
AAA
AAAAAAA
RNA-induced silencing complex
(RISC)
microRNA
60S
60S
Ribosome
40S
Ribosome Profiling
60S
5 UTR
AGO
mRNA
DEGRADATION
40S
40S
Reads mapped to junctions
Reads mapped to exons
Sequencing ribosome-protected mRNA fragments

Mapping ribosome footprints within
transcripts at nucleotide level resolution
TRANSLATION
SUPPRESSION
Small RNA Sequencing

e.g., microRNA/Piwi-interacting RNA profiling
Argonaute (Ago) HITS-CLIP

5 UTR
Mapping interactions between microRNAs and mRNAs
1044
Cell 146, September 16, 2011 2011 Elsevier Inc.
DOI 10.1016/j.cell.2011.09.002
Start
codon
CDS
CDS
Stop
codon 3 UTR
See online version for legend and references.
Table 1. High-Throughput Next Generation Sequencing (NGS) Platforms

Platforms
Amplification; Sequencing
Chemistries
Detection
Read lengtha
SE-single end; PE-paired end)
Reads per lane/No. of

lanes (X2 for dual flow
cell)
Run timeb
Illumina HiSeq 2000

http://www.illumina.com
Bridge Amplification; Synthesis
Fluorescence
100 bp (PE)
50 106 / 8 (X2)
11 days
Roche/454s GS FLX Titanium

http://www.454.com
Emulsion PCR; Synthesis
Luminescence
400 bp (SE)
1 106
10 hr
Life/APGs SOLiD 3
www.appliedbiosystems.com
Emulsion PCR; Ligation
Fluorescence
50 bp (PE)
40 106 / 8 (X2)
14 days
Polonator G.007
http://www.polonator.org
Emulsion PCR; Ligation
Fluorescence
13 bp (PE)
10 106 / 8 (X2)
4 days
Helicos BioSciences HeliScope

http://www.helicosbio.com
No amplification; Synthesis
Fluorescence
35 bp (SE)
20 106 / 25 (X2)
8 days
Pacific Biosciences
http://www.pacificbiosciences.com
No amplification; Synthesis
Fluorescence
>1000 bp (SE)
150,000 per SMRT cell
N/A
Ion Torrent
http://www.iontorrent.com
Emulsion PCR; Synthesis
Change in pH
200 bp (SE)
Variable
<2 hr
Complete Genomics Analysis Platform

http://www.completegenomics.com
DNA nanoballs; Ligation
Fluorescence
Complete genomic analysis service at 40 human genome coverage;

>90% of the full genome resolved (both alleles)
400 human genomes

per month
Modified and updated from Metzker (2010).

Average read length. bRun time for full sequencing experiment.
Sequencing technologies in development: Nanopore sequencing (Oxford Nanopore: http://www.nanoporetech.com; Nabsys: http://www.nabsys.com). Electron Microscopy base sequencing (Halcyon
Molecular: http://halcyonmolecular.com; ZS Genetics: http://ww.zsgenetics.com).
High-throughput, next-generation sequencing (NGS) technologies have revolutionized genomics, epigenomics and transcriptomics studies by allowing massively parallel
sequencing at a relatively low cost. In this SnapShot, we highlight the increasingly diverse applications of NGS, including genome sequencing/resequencing, transcriptome
sequencing, small RNA sequencing, analysis of DNA/RNA-protein interactions, and ribosome profiling. In addition, we provide a quick guide (Table 1) to the currently available
NGS platforms, together with their underlying methodologies and unique features.
Genome Sequencing/Resequencing
Whole-genome sequencing/resequencing and targeted genome resequencing have been used extensively for sequence polymorphism discovery and mutation mapping. These
applications are rapidly advancing our understanding of human health and disease and are also facilitating the de novo assembly of uncharacterized genomes.
DNA-Protein Interactions and Epigenome Sequencing
Chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-Seq) is a powerful technique for genome-wide profiling of DNA-protein interactions and epigenetic marks. It has facilitated a wide range of biological studies, including transcription factor binding, RNA polymerase occupancy, nucleosome positioning and histone
modifications. Complementary methods being used to study chromatin structure and composition are Methyl-Seq and DNase-Seq for profiling DNA methylation and DNasehypersensitive sites, respectively.
Transcriptome Sequencing/RNA-Seq
The introduction of transcriptome sequencing/RNA-Seq has provided a new approach for characterizing and quantifying transcripts. In general, total RNA, rRNA-depleted total
RNA, or poly(A)-selected RNA are converted to double-stranded cDNA fragments that are then subjected to high-throughput sequencing. This strategy has been applied for
profiling mRNA and noncoding RNA expression, alternative splicing, trans-splicing, and alternative polyadenylation and for mapping transcription initiation, termination, and RNA
editing sites. Related applications include targeted RNA-Seq, direct RNA-Seq, strand-specific RNA-Seq, and nascent RNA-Seq (e.g., global run-on sequencing, GRO-Seq, and
native elongating transcript sequencing, NET-Seq).
RNA-Protein Interactions
CLIP-Seq, also known as HITS-CLIP, is a method employing in vivo crosslinking of RNA to protein followed by immunoprecipitation and high-throughput RNA sequencing to
generate transcriptome-wide RNA-protein interaction maps. Modified CLIP-Seq technologies, such as PAR-CLIP (photoactivatable ribonucleoside-enhanced CLIP) and iCLIP
(individual nucleotide resolution CLIP), have been applied to increase crosslinking efficiency and resolution.
Small RNA Sequencing
Similar to RNA-Seq, sequencing of size-selected short RNA provides insight into small RNA populations in different organisms, tissue and cell types, developmental stages,
and disease states. It has greatly contributed to our understanding of the functions and regulatory mechanisms of different classes of small RNAs, such as microRNAs (miRNAs)
and Piwi-interacting RNAs (piwiRNAs). With the recent development of Argonaute (Ago) HITS-CLIP, it is possible to simultaneously detect Ago-bound microRNAs and mRNA
segments, which enables the large-scale mapping of in vivo miRNA-mRNA interactions.
Ribosome Profiling
In addition to the profound impact of NGS on transcriptomic studies, the development of methods enabling high-throughput sequencing of ribosome-protected mRNA fragments
has provided a powerful tool for the analysis of translationally engaged mRNA on a genome-wide scale.
Additional Applications and Future Directions
High-throughput sequencing is a rapidly evolving technology and will likely continue to change the face of omics studies in the years to come. Although NGS technologies
power a wide spectrum of current research applications, new innovations are continually being developed. These include barcode sequencing strategies for multiplexing the
analysis of samples, metagenomic analyses, protein-protein interactome mapping (Stitch-Seq), and high-definition measurement of DNA-affinity landscapes (HiTS-FLIP).
Future technical advances and applications are expected to further revolutionize our understanding of evolutionary biology and genotype-phenotype relationships and ultimately
to bring personalized medicine into the clinic.
1044.e1 Cell 146, September 16, 2011 2011 Elsevier Inc. DOI 10.1016/j.cell.2011.09.002
References
Chi, S.W., Zang, J.B., Mele, A., and Darnell, R.B. (2009). Argonaute HITS-CLIP decodes microRNA-mRNA interaction maps. Nature 460, 479486.
Core, L.J., Waterfall, J.J., and Lis, J.T. (2008). Nascent RNA sequencing reveals widespread pausing and divergent initiation at human promoters. Science 322, 18451848.
Ingolia, N.T., Ghaemmaghami, S., Newman, J.R., and Weissman, J.S. (2009). Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science
324, 218223.
Licatalosi, D.D., Mele, A., Fak, J.J., Ule, J., Kayikci, M., Chi, S.W., Clark, T.A., Schweitzer, A.C., Blume, J.E., Wang, X., et al. (2008). HITS-CLIP yields genome-wide insights into brain
alternative RNA processing. Nature 456, 464469.
Metzker, M.L. (2010). Sequencing technologies - the next generation. Nat. Rev. Genet. 11, 3146.
Nutiu, R., Friedman, R.C., Luo, S., Khrebtukova, I., Silva, D., Li, R., Zhang, L., Schroth, G.P., and Burge, C.B. (2011). Direct measurement of DNA affinity landscapes on a high-throughput sequencing instrument. Nat. Biotechnol. 29, 659664.
Pan, Q., Shai, O., Lee, L.J., Frey, B.J., and Blencowe, B.J. (2008). Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing.
Nat. Genet. 40, 14131415.
Park, P.J. (2009). ChIP-seq: advantages and challenges of a maturing technology. Nat. Rev. Genet. 10, 669680.
Smith, A.M., Heisler, L.E., St Onge, R.P., Farias-Hesson, E., Wallace, I.M., Bodeau, J., Harris, A.N., Perry, K.M., Giaever, G., Pourmand, N., and Nislow, C. (2010). Highly-multiplexed
barcode sequencing: an efficient method for parallel analysis of pooled samples. Nucleic Acids Res. 38, e142.
Yu, H., Tardivo, L., Tam, S., Weiner, E., Gebreab, F., Fan, C., Svrzikapa, N., Hirozane-Kishikawa, T., Rietman, E., Yang, X., et al. (2011). Next-generation sequencing to generate interactome datasets. Nat. Methods 8, 478480.
1044.e2 Cell 146, September 16, 2011 2011 Elsevier Inc. DOI 10.1016/j.cell.2011.09.002

Cell Sept 2011

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cell Sept 2011

Uploaded by

Copyright:

Available Formats

Leading Edge

Chromothripsis, a phenomenon in which numerous genomic rearrangements

Blocking Myc in Myeloma

Despite the centrality of Myc in the pathogenesis of cancer, conventional

Positive ID on a Tumor Target

Helicase Merges onto the Single-Stranded Highway

Microtubules Master Transcription

Acetylation Winds the Aging Clock

Lu et al. find that yeast lifespan is regulated by an intrinsic timing mechanism

A New Recipe for Making Gradients

Endothelium at the Eye of the Storm

Cytokine storms are excessive and potentially fatal immune responses to

A Neuron in Any Other Place Would Smell as Sweet

Histone Vocabulary Explodes

Going Ape Over the Sperm Methylome

Gene Hunter Gatherers

Emerging evidence suggests that

Ancient modern humans interbred with archaic humans including Neanderthals,

Squeezing More out a Genome

Both Parents Are to Blame

Cell 146, September 16, 2011 2011 Elsevier Inc. 849

pathway and molecular mechanisms involved in this process. Together, the

common knowledge that most complex

Cell 146, September 16, 2011 2011 Elsevier Inc. 851

demonstrate that BiP, an Hsp70 ATPase

et al., 1986) were well known to be cylindrical particles based on oligomeric

GroES binding. This change results in

Cell 146, September 16, 2011 2011 Elsevier Inc. 853

Fenton, W.A., Kashi, Y., Furtak, K., and Horwich,

Hemmingsen, S.M., Woolford, C., van der Vies,

Martin, J., Mayhew, M., Langer, T., and Hartl, F.U.

Langer, T., Lu, C., Echols, H., Flanagan, J., Hayer,

Rye, H.S., Burston, S.G., Fenton, W.A., Beechem,

Langer, T., Pfeifer, G., Martin, J., Baumeister, W.,

Xu, Z.H., Horwich, A.L., and Sigler, P.B. (1997).

854 Cell 146, September 16, 2011 2011 Elsevier Inc.

established), the projects long-term goal

recipes of Chinese traditional herbs,

Cell 146, September 16, 2011 2011 Elsevier Inc. 855

on March 8, 1972, providing

these transitions, deaths (particularly of

increasing with time. Our hope is that

of control methods, but this reduction was

Cell 146, September 16, 2011 2011 Elsevier Inc. 857

Qinghaosu Antimalarial Coordinating Research

Published online August 17 2011. 10.1016/

858 Cell 146, September 16, 2011 2011 Elsevier Inc.

Wootton, J.C., Feng, X., Ferdig, M.T., Cooper,

Salk Institute for Biological Studies, La Jolla, CA 92037, USA

tively, and one of three alternative

The mechanism by which the ticking of

Cell 146, September 16, 2011 2011 Elsevier Inc. 859

aging field, inhibition of S6K

860 Cell 146, September 16, 2011 2011 Elsevier Inc.

ceptors S1P1, 35. Its mode of action is

To address the mechanism of action,

Cell 146, September 16, 2011 2011 Elsevier Inc. 861

targets specifically endotheduring influenza infection, as

862 Cell 146, September 16, 2011 2011 Elsevier Inc.

ceptor-mediated calcium release from

recognized that such a physiological

Cell 146, September 16, 2011 2011 Elsevier Inc. 863

Figure 1. AgRP Neurons Respond to Changing Metabolic States

(Andrews et al., 2008; Kohno et al., 2008;