www.nature.com/nature
7296 | 20 May 2010
Still prime time for primates
Rats turn out to be surprisingly useful for research on cognition. But if the goal is to understand
the human brain and its many disorders, then primate studies remain essential.
R
esearchers who study cognition in non-human primates have
long had to contend with protests from animal-rights activists,
who argue that experimenting on such close human relatives
is ethically abhorrent. Now a wave of research showing that some
cognition experiments can be carried out in rodents (see page 282)
has given scientists something else to worry about. Will activists try to
exploit these developments to argue that there is no longer a scientific
justification for using primates?
Even the most enthusiastic proponents of rodent-cognition
research share this worry. They have had to argue hard to convince
some people within the scientific community that useful work can
be done in rats — and some sceptics remain unconvinced. But the
rodent researchers have never argued that rats could or should
replace primates in research that is ultimately directed at under-
standing how the human brain works — and thus what goes wrong
in neurological and psychiatric conditions.
What these scientists have discovered over the past decade or so is
that rats can do simple cognitive tasks that had often been assumed to
be beyond them and that, with appropriate training, they can indicate
what is going on in their minds by poking their noses in different direc-
tions. In retrospect, this is perhaps not so surprising. After all, rodents
in the wild have to navigate safely and successfully in constantly chang-
ing environments, just as primates do. Because the brain is the organ
that has evolved to fulfil this task, the basic mechanisms for cognitive
functions such as remembering, paying attention and discriminating
certain stimuli are likely to have been conserved across evolution.
This approach — combined with the low cost of rearing and keep-
ing rodents and the wide availability of genetic tools for studying
them — promises to help scientists to reach these basic cognitive
components with unprecedented speed and rigour. Rodent research
is also a less ethically sensitive issue than primate research, so the
Change of purpose
The United States should protect investments used
to find new uses for old drugs.
I
n 2007, a paper in the journal Cancer Cell announced that the com-
pound dichloroacetate (DCA) had been found to shrink tumours
in rats (S. Bonnet et al. Cancer Cell 11, 37–51; 2007). That news by
itself would not have created much of a stir: many compounds tested
in rodents raise hopes of their becoming potential cures, and almost
as many go on to fail in human clinical trials. But DCA had already
been tested in humans against a condition called lactic acidosis,
and so seemed to be relatively safe. Indeed, the authors of the paper
argued that DCA could swiftly find its way into late-stage clinical
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465 | Issue no.
more information that can be wrung out of rats and mice the better.
However, scientists will not be able to extrapolate directly from the
rodent brain to the human brain to work out what has gone wrong in
complex disorders such as schizophrenia. Nor will rodents do much
to help scientists develop neuroprosthetics that may one day help to
compensate for loss of brain tissue. Structurally, the brains of rodents
and humans are just too different.
The human organ has a large, highly evolved outer layer — the
cortex — that provides a processing power unequalled in the animal
world for making sense of external stimuli. Included in this layer,
situated right behind the forehead, is a pronounced prefrontal cortex
where thoughts are orchestrated and
complex plans are formulated. Rodents, “Non-human
by contrast, have a minimal cortex and primates have a level
no prefrontal cortex. Their brains of social intelligence
accordingly cannot help to illuminate
the additional levels of complexity of
that is missing in
functional brain networks in humans. rodents but highly
The brains of non-human primates relevant for humans.”
are anatomically closer to those of
humans, so their cognitive strategies are more likely to be similar.
Non-human primates also have a level of social intelligence that is
missing in rodents but is highly relevant for humans. The empathy
between primates allows them to form complex social bonds — and
perhaps underlies the human instinct to protect monkeys and apes.
Rodent studies have the potential to deliver reliable data that can
inform human, and primate, cognition research, and allow those
experiments to become even more revealing. But, if the goal is to
understand the human brain and mind, rodent and primate work
will need to be continued in parallel for the foreseeable future. It’s a
no-brainer. ■
trials against cancer — except for one problem. The drug was no
longer protected by patents, and no pharmaceutical company would
invest the millions of dollars needed to fund clinical trials.
Since then, the researchers have managed to pull together enough
funding for preliminary trials, the first of which was published last
week (E. D. Michelakis et al. Science Transl. Med. 2, 31ra34; 2010).
Supported by a mix of Canadian federal grants and donations from
philanthropic organizations and individuals, the study showed prom-
ising results when DCA was used against a form of brain cancer. But
the team was able to test the drug in only five patients. And although
the researchers have gathered enough funding for additional small
studies, they admit that the prospect of moving into the final phase
of clinical testing — which typically involves a much larger trial — is
daunting.
Over the past few years, as observers have lamented the declining
267
EDITORIALS
productivity of the pharmaceutical industry, there have been many
calls to speed up the process by ‘repurposing’ or ‘repositioning’ exist-
ing drugs. Unfortunately, that suggestion runs up against the same
stumbling block faced by DCA: when drugs have gone off patent, it is
difficult for a company to recoup the substantial investment required
to test the drug in a new population of patients. Large pharma-
ceutical companies have expressed little interest in repurposing
drugs — and at least two of the small companies that have tried have
gone bankrupt.
The fundamental difficulty is that patent systems generally reward
innovation, not development. Although it is possible to file a new
‘method of use’ patent to cover a repurposed drug, such patents are
difficult, if not impossible, to enforce if a generic copy of the drug is
already on the market.
Several solutions have been proposed, none of them perfect.
One would be for the federal government to pay for clinical
trials of repurposed drugs that — like DCA — lack intellectual-
property protection. This is an expensive proposition, however,
and taxpayers might chafe at the notion of paying for trials that
Singular vision
Reforms that could harmonize and enhance
European research deserve support.
I
f the plan set out on 11 May by Máire Geoghegan-Quinn, Europe’s
research commissioner, comes to fruition, the continent may finally
realize its long-standing goal of a single market for science and tech-
nology. Her proposals aim to break down barriers to the transfer
of knowledge and researchers, as well as to reform regulations that
hamper high-tech businesses.
The plan has three key elements. It would create a single patent
system that would grant companies protection for their inventions
across all the European Union (EU) member states at once, removing
the need to file separate patents in multiple countries. It would also
set up an EU-wide pension scheme to make it easier for researchers to
move around the continent. At present, pensions are not transferable
from one member state to the next, which discourages movement.
Finally, it would increase public procurement: directing the money
that EU agencies spend on areas such as telecommunications, energy-
efficient buildings and computer software towards EU businesses,
thereby spurring home-grown innovation rather than going after
the cheapest price abroad.
None of these ideas is new, although in the past they have strug-
gled to gain traction. The EU patent, for example, has previously
proved too controversial in too many countries to make much head-
way, and the others have never had a sufficiently vigorous political
champion.
This time things may be different. Geoghegan-Quinn was an
experienced politician in her native Ireland before assuming her
current post, and she has quickly earned a reputation for being no-
nonsense, hard-driving and determined. And, for the first time, she
is pushing an integrated plan. Instead of treating research (primarily
268
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
were previously the responsibility of the private sector.
Another possibility is to follow the path taken by the European
Union, which allows an extra year of patent exclusivity if new uses
are found for an approved drug. In the United States, a similar
approach is already being used to encourage the testing of drugs
for treating children: drug manufacturers get an extra six months
of protection if they find a new paediatric use for their product. But
this programme, although successful, comes with a trade-off: the
drugs will remain free from generic competition, and therefore more
expensive, for longer.
It is a difficult conundrum, but one that warrants serious thought
and creativity from researchers, agencies and policy-makers alike.
This is especially true as the US National Institutes of Health designs
its Cures Acceleration Network, which was mandated in the recent
health-care reform bill. Funds permitting, the network will try to
repurpose drugs that pharmaceutical companies have abandoned.
Such an initiative could cut down the time and expense needed
to find new therapies, and those who would take up the effort
deserve support. ■
in academic institutions) and innovation (primarily in the business
world) in a piecemeal fashion, as the commission has done in the
past, her plan treats them as an organic whole. The goal is to create
a smooth flow from research discoveries to products and services
on the market.
Geoghegan-Quinn says that the plan will refocus Europe’s research
efforts on a series of grand challenges facing the continent as a whole,
such as climate change and an ageing population. New partnerships
would bring together the EU, member states and public and private
researchers to work on specific aspects
“The goal is to create
of these grand challenges. For example,
for the ageing-population challenge, a smooth flow from
these partnerships could work on tack- research discoveries
ling chronic diseases or on developing to products and
technologies to allow older people to services on the
stay in their homes for longer. Existing
initiatives, such as Europe’s multibil-
market.”
lion-euro Framework programme for research, and the Joint Tech-
nology Initiative’s public–private research partnerships, will also be
integrated with the plan to avoid overlap.
Although there are some potential pitfalls in the plan — the pursuit
of direct societal benefits and high-tech industrial growth cannot
be allowed to undermine basic research, for example — it is on the
right track. Geoghegan-Quinn’s vision will come under scrutiny this
autumn, when EU heads of state meet to discuss it in detail. The task
now is to sustain political momentum, and to ensure that the neces-
sary decisions are taken at that autumn summit. European research
minsters should explicitly give their endorsement for moving this
agenda forward when they next meet on 25 May.
Geoghegan-Quinn’s reforms are especially important given Europe’s
current financial crisis. Budgetary pain is looming for everyone, so
the kind of integration and coherence that Geoghegan-Quinn has
outlined is essential for making the most effective use of the research
money that scientists do have. ■
 Vol 465|20 May 2010

RESEARCH HIGHLIGHTS

Bacterial power
Proc. Natl Acad. Sci. USA doi:10.1073/
pnas.0910426107 (2010)
Nanotechnology is still far
from producing anything as
efficient and compact as self-
propelling bacteria. But Roberto
Di Leonardo at the National
Research Council in Rome and
his colleagues have managed to
harness Escherichia coli to power
a small ratchet motor.
They made a variety of
symmetrical and asymmetrical
glass-based gears (pictured),
48 or 80 micrometres wide. The
shape of the asymmetrical gears
means that bacteria swimming
into them either slide off the end
of a cog tooth or become stuck in
a corner. Bacteria that stick exert
a force that drives the gear around
until they free themselves.
r. Di LeonarDo et al.
When the researchers placed
their cogs in a liquid bacterial
suspension, they observed about
one rotation per minute for the
asymmetrical gears, showing that
geometry can be used to convert
chaotic bacterial motion into
predictable movement. D.P.C.

NEUROSCIENCE The team devised a method to produce the ECOLOGY

Instant learning valeric acid and refined known procedures

Neuron 66, 438–448 (2010)
Many species can learn a new rule by trial and
error. But is such learning a gradual process,
or does it come all at once in a ‘eureka!’
moment?
Daniel Durstewitz at the University of
Heidelberg in Germany, Jeremy Seamans
at the University of British Columbia in
Vancouver, Canada, and their colleagues
planted electrodes in a brain region called
the prefrontal cortex in male rats. They then
observed patterns of neural activity while the
rats learned a new rule to obtain a food reward.
In many cases, an abrupt shift in neuronal
activity corresponded with a change in
the strategy an animal used. Such neural
and behavioural transition points may
correspond to moments of ‘sudden insight’,
the authors suggest. L.O.-S.
ORGANIC CHEMISTRY
Biofuel boost
Angew. Chem. Int. Edn doi:10.1002/anie.201000655
for the other steps. The authors say that
valeric biofuels, owing to their chemical
properties, may perform better than other
candidate biofuels. A blend of the valeric
fuels with regular petrol or diesel did not
cause any discernable engine damage in ten
vehicles tested. K.S.
CANCER
Melanoma’s moving target
Cell 141, 583–594 (2010)
In many cancers, tumour growth has been
attributed to a small, fixed subset of cells
known as cancer stem cells. But melanomas
are thought to follow a different path.
Researchers now report that a dynamic
group of melanoma cells is responsible for
continuous tumour growth, with individual
cells in this group transiently acquiring and
losing this ability to drive tumour growth.
Meenhard Herlyn at the Wistar Institute
in Philadelphia, Pennsylvania, and his
colleagues found that the enzyme JARID1B,
which is involved in cancer and in normal
No farm is an island
Ecol. Lett. doi:10.1111/j.1461-0248.2010.01481.x
(2010)
It seems intuitive that organic farming should
be good for biodiversity, but studies at the
single-farm scale have had mixed results.
This is probably because many species need
more space than one field or farm provides to
support their life cycle.
Doreen Gabriel at the University of Leeds,
UK, and her colleagues decided to expand
the analysis area. They compared biodiversity
levels in 100-square-kilometre areas of
British farmland that contained either high
or low amounts of organic cultivation.
The researchers sampled thousands of
organisms — including plants, birds, insects
and worms — in the farms’ fields and along
their edges (pictured), and found that the
farming practices of both a given farm
and its neighbours affect biodiversity. A
conventional farm in an organic hotspot is
likely to be as biodiverse as an organic farm
surrounded by conventional ones. E.M.
M. Tasker

(2010) stem-cell biology, is a marker for this

Woody plants could become a viable
feedstock for biofuels, thanks to research
at oil-giant Shell.
Jean-Paul Lange at the Shell Technology
Centre Amsterdam and his colleagues
have developed a technique to convert
lignocellulose — a component of plant cell
walls that is hard to break down — into
levulinic acid, and from that to valeric acid.
This can then be used to make esters that can
be burned as fuel.
270
subpopulation of melanoma cells. Silencing
the JARID1B gene in human melanoma
cells in vitro resulted in an initial increase
in cell proliferation, which then levelled off.
When transplanted into mice, the JARID1B-
knockdown cancer cells eventually stopped
dividing. They also resulted in the formation
of fewer secondary lung tumours.
Cells positive or negative for the enzyme
each gave rise to a mixed population of these
two cell types. C.L.
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465|20 May
NATURE|Vol 465|20
2010May 2010 RESEARCH HIGHLIGHTS

JOURNAL CLUB
Marc Vrakking
Max Born Institute for Nonlinear
Optics and Short Pulse
CELL BIOLOGY PHYSICS Spectroscopy, Berlin

Viral vote Not a WISP of evidence A physicist discusses how to

Cell 141, 682–691 (2010)
Why do identical cells often respond
differently to the same stimulus? Researchers
generally blame noise inherent in biological
systems, but there may, in fact, be specific
processes at play, according to Ido Golding,
now at Baylor College of Medicine in Houston,
Texas, and his co-workers.
They watched individual particles of a
bacterial virus infect single Escherichia coli
cells (pictured, top panel, green dots indicate
virus particles) — in theory subjecting the cells
to the same stimulus. Each virus particle, they
found, makes an individual
visualize a molecule changing
Phys. Lett. B doi:10.1016/j.physletb.2010.04.066 (2010) shape.
In extensions to the standard model,
which describes the fundamental particles It is the dream of many a chemist
and forces of physics, some theorists to watch a movie of a molecule
have proposed the existence of very light undergoing structural change.
subatomic particles called WISPs. These So how can we achieve this? One
could be dark matter, which keeps a spinning way is to use the relationship
galaxy from flying apart. between a molecule’s absorption
One way to detect WISPs would be to look spectrum and its structure to
for the rare conversion of light particles to deduce how the structure changes
WISPs, and later back to photons. In between over time. However, a drawback
these conversions, a WISP could zip through of this technique is its reliance on
any barrier. So Axel Lindner at DESY, the prior knowledge of the molecular
German electron synchrotron absorption spectrum.

‘decision’: to kill the host cell
or to become dormant by
integrating into the host’s
DNA. Those decisions are
then summed to determine
the cell’s ultimate fate. Only
a unanimous decision by all
virus particles to integrate
into the DNA of a particular
cell keeps that cell alive (red).
If even one particle ‘votes’ for
death, the cell bursts (bottom
panel, in green). A.K.
NEUROSCIENCE
Bright eyed
Sci. Transl. Med. 2, 31ra33 (2010)
Light is an important regulator
of the body’s circadian
rhythms. In humans, this
is thought to be mediated
by light-sensing cells in the
eye known as intrinsically
photosensitive retinal ganglion
cells (ipRGCs). Steven
Lockley at Brigham and
Women’s Hospital in Boston,
Massachusetts, and his team
have shown that the retina’s
colour-detecting cones are also involved.
They shone blue or green light of varying
intensities directly into the eyes of awake
volunteers for 6.5 hours during the night.
Cones are most sensitive to green light,
whereas ipRGCs mainly detect blue light. The
researchers also monitored changes in the
volunteers’ circadian responses.
The team found that cones contribute to
circadian responses on initial exposure to light
and in dim light conditions, but that ipRGCs
are the main photoreceptors when light is
bright and long-lasting. The authors suggest
their findings could help enhance light therapy
for conditions such as sleep disorders. C.L.
© 2010 Macmillan Publishers Limited. All rights reserved
eLsevier
in Hamburg, and his
colleagues shone green laser
light at a ‘wall’, a thick piece
of light-absorbing material,
hoping that a few photons
might pop out the other side.
They increased the chances of
a WISP conversion by using
optical resonators to boost the
power of the laser light and by
applying a strong magnetic
field. But the researchers
did not detect any emerging
photons, limiting the chance
of a WISP conversion to nearly
1 in 1025 — the most sensitive
limit yet. E.H.
MICROBIAL GENOMICS
A happy marriage
PLoS Genet. 6, e1000943 (2010)
Poplar trees may not be the
first plants to spring to mind
when thinking of biofuels,
but they are a potentially
important feedstock, partly
because they can grow on soils
unsuitable for food crops.
To see how poplar growth
could be improved, Daniel van der Lelie
at the Brookhaven National Laboratory
in Upton, New York, and his colleagues
analysed the genome and metabolites of the
microbe Enterobacter sp. 638, which lives
naturally inside poplars and increases the
plant’s growth by as much as 40%.
Among the key genes identified in the
bacterium are those encoding enzymes
involved in the production of plant growth
hormones. This hormone production
depends, in turn, on the presence of
compounds such as sucrose synthesized by
the plant, showing how much the bacterium
and tree depend on each other. L.O.-S.
Faton Krasniqi at the Max
Planck Advanced Study Group
in Hamburg, Germany, and his
co-workers present an alternative
idea: using photoelectrons ejected
from molecules excited by X-ray
free-electron lasers to determine
molecular structures that change
over time (F. Krasniqi et al. Phys.
Rev. A 81, 033411; 2010).
They explain how electrons
that are ejected and directly
detected without any further
interaction with the molecule
interfere with electrons that
scatter off the surrounding atoms
in the molecule, thereby creating
holographic patterns. These
patterns encode the molecule’s
three-dimensional structure. As an
example, the researchers present
calculations through which they
reconstruct the six-membered
phenyl ring in a chlorobenzene
molecule.
This approach of holographic
structure retrieval promises
powerful insight into time-
dependent molecular dynamics
in the next few years. It is an idea
that is well founded in earlier
experiments at synchrotrons.
Many of the required experimental
techniques — such as the ability to
position molecules in space using
moderately intense laser fields —
have recently been demonstrated.
The Linac Coherent Light Source
(LCLS), an X-ray free-electron laser
at Stanford University in California,
has been up and running since last
year. Now it’s showtime!
Discuss this paper at
go.nature.com/KzvyE9
271
 Vol 465|20 May 2010

NEWS BRIEFING
● POLICY

H. HAMbUrg/AP
Oil spill: The oil spewing into
the Gulf of Mexico was partially
stemmed this week, after BP
managed to force a siphon
tube into the leaking wellhead
pipe. But the political fallout
intensified as Congress sought
answers about the explosion
of the Deepwater Horizon rig.
US interior secretary Ken Salazar
said on 11 May that the Minerals
Management Service would be
split up, separating safety and
environmental operations from
its oil-leasing arm. For more on
the spill, see pages 274–275.

Lab fix needed: NASA’s ageing

research labs are in dire need
of an overhaul, according to a
report released on 11 May by the
National Academies. Some 80%
of the agency’s research facilities
are more than 40 years old, and
its deferred-maintenance bill
has risen from US$1.77 billion
in 2004 to $2.46 billion in 2009.
US President Barack Obama’s
proposed 2011 budget for the
agency would rectify some
problems by transferring money
from the cancelled Constellation
moon-rocket programme to long-
term technology-development
and research. But legislators in
Congress have so far fought the
President’s budget plan.
US spending hitch: The House
of Representatives unexpectedly
shot down science and education
legislation on 13 May after
Republicans raised concerns
about the federal budget
deficit and about government
New CLImate bILL aRRIves IN Us seNate
US senators John Kerry (Democrat, Massachusetts; pictured centre) and Joseph Lieberman (Independent,
Connecticut; pictured right) released their much-anticipated climate legislation on 12 May, setting the stage
for one last attempt at getting climate legislation through the current Congress before the midterm elections
in November. The bill would use a cap-and-trade scheme to curb emissions by 17% by 2020, and by some 80%
by 2050 (relative to 2005 levels). It contains mechanisms intended to stabilize carbon prices and make costs
predictable for industry (see go.nature.com/2mpfSN). A day later, the Environmental Protection Agency turned
up the pressure on Congress by releasing a rule clarifying how greenhouse-gas emission restrictions would be
phased in for major industrial polluters beginning next year — if Congress fails to enact climate legislation.
its president, Ketan Desai, was grants include $7.4 million
arrested on corruption charges to the University of Alaska in
(see Nature 464, 1251; 2010). Fairbanks for building clinical-
An ordinance signed by Indian trial facilities to study health
President Pratibha Devisingh disparities in Native Americans,
Patil establishes a board of and $15 million to the University
governors to run the council of Colorado at Boulder for a new
for one year. During that year, biotechnology facility housing
the government may also 60 faculty members and more
dissolve the MCI, which sets and than 500 graduate students,
maintains standards of medical
education and accredits Indian NUmbeR researchers and support staff.

employees, including some at the medical schools. CRUNCh Laboratory fire: A fire at a

National Science Foundation,
viewing Internet pornography
sites. The legislation would have
authorized future spending under
the 2007 America COMPETES
Act, which called for a doubling
of research funding in the
physical sciences. See go.nature.
com/aSMYlX for more.
Medical disarray: The Indian
government on 15 May took
over the Medical Council of
India (MCI), three weeks after
272
● ReseaRCh
Stimulus salvo: On 14 May, the
US National Institutes of Health
doled out the final big chunk of
new awards to be funded from its
US$10.4-billion 2009 economic
stimulus package: $1 billion to
construct, repair and renovate
research labs at extramural
institutions in 44 states,
Puerto Rico and the District of
Columbia. The 146 individual
© 2010 Macmillan Publishers Limited. All rights reserved
14.5 °C
research centre in São Paulo,
Brazil, has destroyed a leading
collection of dead snakes. The
april’s combined state-funded Butantan Institute,
which was nearly 100 years
land and sea surface old, contained around 80,000
temperature: the preserved snakes and thousands
hottest april on record, of spiders and scorpions that were
at 0.76 °C above used for biomedical research.
the average for the The curator Franciso Franco
twentieth century. has told press agencies that its
destruction on 15 May was a “loss
to humanity”. The cause of the
Source: NOAA
fire is being investigated.
 NATURE|Vol
Vol 465|13 May
465|20
2010May 2010 News bRIeFING

Phones and cancer: There is

News genetic report costing $79–249.
no clear link between mobile-
phone use and the risk of brain
maKeR
The kit is already available to
order online. the weeK
aheaD
cancer, according to a major
E. DEbEbE/UN PHOTO

study published this week (The
INTERPHONE Study Group
Int. J. Epidemiol. doi:10.1093/ije/
dyq079; 2010). The study, run by
the World Health Organization’s
International Agency for
Research on Cancer in Lyon,
France, interviewed thousands of
adults with and without cancer in
13 countries about their mobile-
phone usage. See go.nature.com/
uZph7a for more.
India defence research: India’s
largest military technology
research body is set for a
management revamp under
government measures announced
on 13 May. The 52-year-old
state-owned Defence Research
& Development Organisation
currently employs more than
5,000 scientists in 51 laboratories.
It will be split into seven
separately directed centres, such
as life sciences and materials
science, and projects will be
monitored by a new oversight
committee. A 2007 review had
recommended restructuring the
organization after criticisms that
projects such as combat aircraft
and guided missiles encountered
huge time and cost overruns.
● bUsINess
Genome shopping: The US
pharmacy chain Walgreens
postponed plans to start selling
a personal genome-testing
kit in thousands of its shops
last week, after the Food and
Drug Administration began an
investigation into whether the
Chinese solar: Jinko Solar, one
of several Chinese firms hoping
to dominate the crowded silicon 20–22 May
photovoltaic market, received a Mouse models of autism are one
cautious welcome at its 14 May of the items on the agenda at the
initial public offering on the international meeting for autism
New York Stock Exchange. The research, which convenes in
company, based in Shanghai, Philadelphia, Pennsylvania.
raised some US$64 million at ➧ www.autism-insar.org
$11 per share — a price at the
lower end of its predicted range, 22 May
and which stayed at that level This year’s International Day
throughout the day. It will use the for Biological Diversity (as
Christiana money to expand manufacturing proclaimed by the United
Figueres capacity. Nations) has been allotted
the theme of poverty and
the Costa Rican Cancer acquisition: Astellas development.
diplomat, 53, Pharma, headquartered in Tokyo, ➧ www.cbd.int/idb
is the new head Japan, said on 16 May that it had
of the United agreed to pay US$4 billion to 23–27 May
Nations Framework purchase OSI Pharmaceuticals, The american astronomical
Convention on based in New York. The US firm Society holds its 216th meeting
is best known for its anticancer in Miami, Florida.
Climate Change, drug erlotinib (Tarceva). ➧ aas.org/meetings/aas216
succeeding Yvo de
boer.
● PeOPLe Varmus return: Nobel laureate
UK science minister: David Harold Varmus is to be the next
Willetts has been named as director of the US National
Britain’s minister for universities Cancer Institute at the National
and science. He was appointed Institutes of Health (NIH),
on 12 May, after a coalition replacing John Niederhuber.
government had been formed The $5.1-billion cancer institute
between the Conservative is the largest of the NIH’s 27
and Liberal Democrat parties. institutes and centres. Varmus,
Willetts, 54, studied philosophy, whose appointment was
kits require regulatory approval. politics and economics at announced on 18 May, headed
The $30 saliva-collection kit the University of Oxford, the NIH during the Clinton
is manufactured by Pathway and has been a Member of administration, between 1993
Genomics, a direct-to-consumer Parliament since 1992. He and 1999, and then became
genetic-testing company based in was the Conservative Party’s president of the Memorial
San Diego, California. Shoppers shadow minister for innovation, Sloan-Kettering Cancer Center
would have to send their universities and skills for the in New York, before advising
DNA sample to the company’s three years leading up to the Barack Obama during his run
laboratory for a customized 6 May general election. for the presidency.

bUSINESS WATCH
SOUrCE: NEW YOrK STOCK ExCHANgE

Investors are losing confidence in Monsanto,
the agricultural biotech giant based in St Louis,
Missouri (see chart). Some farmers aren’t
seeing big yield increases with the company’s
GLOOM IN ST LOUIS
Monsanto’s share price has fallen
the North American soya and maize market by since the beginning of 2010
the world’s number-two seed company DuPont, 10
which owns plant-genetics firm Pioneer
Hi-bred International, based in Johnston, Iowa. 0
Change in share price (%)

new herbicide-tolerant soya bean line, roundup Alexander worries that Monsanto will cut prices
ready 2 Yield. They are also hesitant about to protect its share, which could potentially hurt –10
buying its forthcoming SmartStax maize (corn), its research budget. but on 5 May, Carl Casale,
which incorporates eight genes conferring Monsanto’s executive vice-president and chief –20
herbicide tolerance and insect protection. If financial officer, said research spending was not
farmers don’t switch to roundup ready 2, the in jeopardy. Meanwhile, Swiss firm Syngenta,
–30 Dow Jones
company’s profits may suffer, as the original based in basel, is flexing its muscles with the
Monsanto
roundup ready trait goes off patent in 2014. imminent release of its herbicide-tolerant,
According to Laurence Alexander, an analyst insect-resistant Viptera maize. In the longer –40
Jan Feb Mar Apr
at investment bank Jefferies, headquartered in term, the three firms will also be vying to offer Month
New York City, Monsanto is being squeezed in farmers drought-tolerant maize.
273
© 2010 Macmillan Publishers Limited. All rights reserved

NEWS
Oil cruise finds deep-sea plume
Nature reports from the research ship Pelican as scientists map the hidden extent of the Deepwater disaster.
Houma, Louisiana
The first oceanographic research expedition
into the Deepwater Horizon oil-spill zone has
uncovered evidence that a deep-sea plume —
probably made of oil, and not visible on the
surface — seems to be spreading from the rup-
tured wellhead.
Environmental concerns following the oil-
well blowout on 20 April initially focused on
the effects of spilled oil on the Gulf of Mexico
shoreline, which hosts many fishing commu-
nities and wildlife reserves. The discovery of
the plume suggests that much of the oil may
instead be lurking deep below the sea surface,
with potentially dire consequences for marine
organisms.
The find is already raising questions about
the possible impact of the widespread use of oil
dispersants to tackle the spill, and comes amid
growing dissatisfaction among researchers
about the limited efforts that have been made
so far to study the spill and accurately gauge
its size.
274
The team that found the plume is from the
National Institute for Undersea Science and
Technology (NIUST), a cooperative effort
between the University of Mississippi in Oxford
and the University of Southern Mississippi in
Hattiesburg, funded by the National Oceanic
and Atmospheric Administration (NOAA) in
Silver Spring, Maryland. The researchers had
originally been scheduled to map sea-floor for-
mations in the Gulf of Mexico, just 15 kilometres
from the Deepwater Horizon platform, and to
survey historically significant shipwrecks using
autonomous underwater vehicles launched from
the Louisiana Universities Marine Consortium’s
35-metre-long research vessel Pelican.
But when the oil-well blowout happened,
just days before the ship was scheduled to
depart, team leaders decided that the group
should divert to oil studies and set about get-
ting approval from NOAA, which is fund-
ing the expedition through a competitively
awarded grant. “We felt that the mapping that
we wanted to do could wait and that it would
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465|20 May 2010
m. Schrope
Researchers including (left to right)
Matt Lowe, Vernon Asper and Andy
Gossett have been studying the impact
of the oil spill on ocean chemistry.
be an inappropriate use of this valuable ship
time to do something that was not as urgent
as the oil study,” says NIUST oceanographer
Vernon Asper.
For the first week of the cruise, much of
the NIUST work focused on taking samples
of sediment cores throughout the region, to
develop a reliable baseline for future studies of
oil that may settle to the sea floor. After a week
of establishing a series of study sites, the team
returned to port briefly to take on additional
equipment — and a Nature journalist.
On their fourth day back at sea, on 12 May,
the scientists made an astonishing find. “You’ve
got to see this,” said Arne Diercks, rushing into
the ship’s main lab. As those aboard gathered
in the control room where data from a lowered
sampling system come through in real time,
Diercks, NIUST’s chief scientist for the cruise,
pointed out the telltale instrument readings.
At a depth of around 1,000 metres, the team
seemed to have struck oil.
The team’s hypothesis was based on three
 Vol 465|20 May
NATURE|Vol 465|20
2010May 2010
key readings. The transmissometer, which
measures the blocking of light by particles or
opaque dissolved matter in the water, showed
a serious hike in murkiness. The fluorometer,
tuned to measure fluorescence given off by
dissolved oil, was also giving readings many
times higher than normal. And oxygen levels
had dropped, suggesting heightened activity by
microbes as they consume the oil and associ-
ated organic material.
“We’ve got to home in on this,” Asper said
excitedly. “You never see signals like that in
the open ocean.” The team spent much of the
remaining time at sea mapping the boundaries
of a plume that extends about 45 kilometres
southwest from the wellhead and roughly 10 kil-
ometres wide at depths of 1,000–1,400 metres
(see ‘Oil zone’). On returning to previously
sampled sites, the team showed that the plume
was shifting, but that it generally remained at
least 100 metres above the sea floor.
Dispersant debate
Data received from NOAA while the research-
ers were still at sea confirmed that deep-water
currents at the time were flowing southwest,
further suggesting that the plume they were
measuring was oil emanating from the well.
However, the group will not be able to confirm
the plume’s composition until tests on collected
water samples are performed this week.
Aboard the Pelican, the NIUST team watched
as news of the plume spread, and eventually
began getting satellite calls from journalists. On
16 May, at a daily press briefing, officials from
energy company BP, which operates the well,
skipped over an initial request for comment on
the plume. In response to a second question,
BP spokesman Andrew Gowers said: “We have
no confirmation of that, but my observation as
a layman is that oil is lighter than water and it
tends to go up.”
Many scientists had also assumed that this
was the case, although others had predicted
that because of the depth of the leak, certain
components of the oil would separate out as
SoUrce: noaa
they rose to the surface and settle into a sub-
surface layer. Still, the magnitude of the plume
was an unpleasant surprise.
Experts including Jeffrey Short, an environ-
mental chemist with the conservation advo-
cacy group Oceana, based in Washington DC,
have suggested that oil coming straight from
the well could naturally break into small, less-
buoyant droplets that would be capable of
forming such a plume below the surface. But
underwater use of dispersants, a previously
untried technique that was approved by the
Environmental Protection Agency (EPA) on
15 May, may also play a part by shaping the oil
into smaller droplets. For several days before
© 2010 Macmillan Publishers Limited. All rights reserved
neWs
The science Of
S. Lehmann/US coaSt GUard
dispersanTs
The giant experiment in the
Gulf of Mexico.
go.nature.com/ffaaPt
that announcement, and while the NIUST oil come to the fore, accurately assessing how
team was studying the plume, BP applied more much oil is gushing from the wellhead will be
than 100,000 litres of dispersant at the sea floor even more important. The official US Coast
during EPA-sanctioned tests. Guard estimate is 795,000 litres per day. How-
Biogeochemist Samantha Joye at the Univer- ever, a number of groups have questioned the
sity of Georgia in Athens, who works with the figure, and Short says that the underwater plume
NIUST team and will be analysing the plume could be further evidence that the true flow rate
water samples, says that either possibility or a is much higher than the official figure.
combination of both could explain the plume. Whereas NOAA administrator Jane
But “it doesn’t matter if it’s dispersed oil or Lubchenco has argued that the estimate is rea-
natural crude, it’s going to have a huge impact”, sonable and that having a more accurate rate
she says. would not change the response strategy, some
Thomas Shirley, a marine biologist at Texas researchers feel that knowing the spill’s true
A&M University in Corpus size is essential to understand-
Christi, says that although lit- “We’ve certainly put ing its full biological impacts,
tle oil has washed ashore and
the harm to coastal ecosystems
a kink in the efficiency and deciding whether massive
deployment of dispersants is
has so far been minimal, off- of the system out the best option.
shore species may be at greater there, but how much “I think that knowing the
risk. Toxic compounds from volume of oil is very impor-
the floating oil could threaten
effect that will have tant,” says Shirley, “and I would
species living near the sur- and for how long, we urge BP to make all the video
face, including commercially don’t know.” they have available and work
important fish and their prey, with people to provide all the
he says. Meanwhile, toxins from the under- data possible.” BP’s ultimate legal liability for
water plume could affect deep corals and other damages could be directly tied to the size of
species, a problem that could be exacerbated the spill, adds Short: “That is a long-standing
by dispersant use, which breaks up the oil principle in these sorts of cases.” Last week, the
into smaller particles and makes it easier for administration of President Barack Obama
animals to take in. Shirley suggests that deep- sent a letter to BP asking for clarification on
dwelling organisms such as zooplankton the company’s financial redress plans and
might be hit by the low oxygen levels in the reiterating the position that BP is responsible
plume, which could take months or years to for all clean-up costs and economic damage.
recover because oxygen is slow to diffuse into A BP spokesman declined to comment on the
the deep. The plume could form a barrier that potential implications of the plume.
blocks the normal up-and-down daily migra- For now, both the gushing oil and the US
tion of numerous organisms, and could block political debate over drilling continue. On
the flow of particles of organic debris from the 16 May, BP reported that it had managed
surface to the deep where they are a critical to insert a tube into the pipe coming out of
food source. the well, which it says is capturing about
“We’ve certainly put a kink in the efficiency 320,000 litres of oil per day. And the company
of the system out there,” says Shirley, “but how is pursuing several other options to capture
much effect that will have and for how long, leaking oil or close off the well before it can
we don’t know.” finish drilling a separate relief well, a process
Shirley says that as the effects of the deeper that could take months.
On 14 May, the team aboard the Pelican all
gathered in the galley to watch a press con-
OIL ZONE ference in which President Obama said that
While surface oil continues to spread from the
Deepwater Horizon site, a previously unseen offshore drilling remains an important part
underwater plume of oil is spreading southwest. of the overall US energy policy, although any
movement towards expanding it is on hold.
Mobile
Gulfport Milton Freeport Short says that the drilling debate centres in
Pensacola
part on weighing the benefits of oil against the
Chandeleur environmental impact. “If the environmental
Sound
Breton Oil coverage
impacts are an order of magnitude larger than
Sound Light anyone dreamed of, that’s probably going to be
Medium a factor in the debate,” he says, “I suspect BP
Terrebonne Heavy
Bay Underwater has its eye on that too.” ■
plume Mark Schrope
0 100 km Read the full account of the Pelican’s mission at
http://go.nature.com/TBKWnY
275
NEWS
Space-science hopes rest on rocket test
New launch vehicle could carry next generation of NASA’s research probes.
When the Falcon 9 rocket makes its inaugural
test flight, expected later this month, it will
carry with it NASA’s hopes for a new genera-
tion of low-cost rockets to ferry cargo and peo-
ple into space.
The rocket — touted as a possible saviour
of human spaceflight — could also solve a
serious problem facing the next generation of
space probes. Satellites that observe Earth and
nearby planets, as well as space telescopes able
to look deep into the cosmos, are about to be
hit by the retirement of the Delta II rocket, a
workhorse that has launched 60% of NASA’s
science missions during the past decade.
Among NASA’s stable of rockets, the Delta II
is the right size for all but the most ambitious
science missions, and at about $50 million per
rocket the price was right too — ten years ago.
But since then, the cost of a Delta II launch has
roughly doubled, making it unaffordable. The
last science launch scheduled for the Delta II
is in 2011, for GRAIL (Gravity Recovery and
Interior Laboratory), a mission to study the
Moon’s core by mapping tiny variations in its
weak gravitational field.
“We’re almost reaching the stage of despera-
tion for launch vehicles,” says Jack Burns, a space
scientist at the University of Colorado at Boul-
der and a member of NASA’s science advisory
committee. NASA science chief Edward Weiler
adds, “If there is no replacement ever for the
Delta II, that would take away a critical capa-
bility.” He hopes that in three or four years the
Falcon 9, developed by SpaceX of Hawthorne,
California, will emerge as a low-cost replace-
ment. “Very much hoping, I might add.”
SpaceX unveiled its plans for Falcon 9 in
2005, and a year later won NASA contracts
worth $278 million to develop rockets that
MID-SIZED ROCKETS NEED A BOOST
SpaceX hopes the Falcon 9 will fill a hole left by the retirement
of the Delta II.
TAURUS II
DELTA II (7920) • First launch: 2011
• First launch: 1990
• Payload to LEO:
• Payload to LEO:
5,500 kg
6,100 kg
LEO, low Earth orbit
276
Countdown is imminent for the Falcon 9 rocket.
could carry supplies and science experiments
to the International Space Station (ISS). In
December 2008, SpaceX won a $1.6-billion
contract for 12 ISS resupply flights, up until
the end of 2015. The rocket’s potential role
expanded in February, when President Barack
Obama proposed axing the suite of NASA
rockets intended to replace the Space Shuttle
and once again send humans to the Moon.
Some $6 billion over 5 years — money that
would have gone to the Constellation rockets
— would instead be ploughed into commercial
providers such as SpaceX, in the hope that they
could transport not only cargo, but people as
well (see Nature 463, 716–717; 2010).
ATLAS V (401)
• First launch: 2002
FALCON 9 • Payload to LEO:
• First launch: 2010 9,800 kg
• Payload to LEO:
10,450 kg
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol
Vol 465|20 May 2010
This radical transition is still very much in
SpaceX
doubt, as legislators in Congress fight to protect
space-industry jobs (see Nature doi:10.1038/
news.2010.189; 2010). Yet even the most ardent
critics of Obama’s new space vision are eager to
see whether Falcon 9 can help to keep NASA
astronauts in space.
In the obsession over human space flight,
many are overlooking the role that the Falcon 9
could have in replacing the Delta II, which came
to prominence during a golden era in the 1990s
when rockets were plentiful and relatively inex-
pensive. The rocket’s biggest buyer, the US Air
Force, had to launch a constellation of global
positioning satellites, and private satellite-com-
munication companies such as Iridium were
also snapping up the rockets by the handful.
But then the communications satellite market
dried up, and the Air Force said it no longer had
an essential need for the Delta II, conducting
its final launch with one last year. Instead, the
Air Force has committed to sustaining the very
large Delta IV and Atlas V rockets in the Evolved
Expendable Launch Vehicle programme.
Price hikes
As buyers have bailed, the price of a Delta II,
including launch, has shot up. A decade ago,
they were a relative bargain, ranging from
$50 million to $80 million apiece. Now, each
one costs about $120 million — almost as
expensive as the much bigger Atlas V — with
further hikes expected. United Launch Alli-
ance, the joint venture of Lockheed Martin and
Boeing that makes the Delta II, would be happy
to continue selling it to NASA. The compo-
nents for five rockets are waiting to be assem-
bled, says William Wrobel, who directs NASA’s
launch services programme. The problem is
60 m
SHUTTLE
• First launch: 1981
• Payload to LEO:
24,400 kg 40 m
20 m
0m
Vol 465|20 May
NATURE|Vol 465|20
2010May 2010
that, without the additional buyers and big-
ger market, NASA cannot afford to pay for
the upkeep of the launch-pad infrastructure
that the Air Force had paid for. “NASA can’t
go it alone,” says Wrobel.
Wrobel says that most imminent mis-
sion launches, such as the 2011 launch of
the heavy Mars Science Laboratory and
Juno, a mission to Jupiter, needed the extra
thrust of an Atlas V anyway. But if NASA
officials are forced to buy more Atlas Vs
in the future, they will be paying extra for
unused launch capacity. “The more that the
launch vehicle costs, the less science mis-
sion you get for your money. Or fewer mis-
sions,” says Alan Stern, a planetary scientist
at Southwest Research Institute in Boulder,
NASA’s former science chief and an advo-
cate of commercial space flight.
Several missions could take advantage
of Falcon 9’s leaner launch capability and
lower price of about $50 million per launch.
These include the Soil Moisture Active and
Passive mission, an Earth-observing sat-
ellite due for launch in 2015; the Interna-
tional Lunar Network, a system of landers
designed to meas-
“The more the ure the Moon’s seis-
launch vehicle mic activity, among
other things; and
costs, the less modest-sized
science mission astrophysics and
you get for your planetary-science
missions that would
money.” launch in 2016.
All bets are not riding on the Falcon 9,
however, which will launch from Cape
Canaveral, Florida, and carry a prototype
of its cargo capsule Dragon. Orbital Sci-
ences, headquartered in Dulles, Virginia,
and another winner of ISS cargo-transport
money, is developing the Taurus II, another
medium-sized launcher that is scheduled
for first test flights in 2011 (see Mid-sized
rockets need a boost). For the planned 2012
launch of the Lunar Atmosphere and Dust
Environment Explorer (LADEE), Orbital
is putting together stockpiled ballistic mis-
siles into the Minotaur 5, which costs less
than $50 million and is just the right size for
the small Moon mission.
If Falcon 9’s test launch is successful, it
should be carrying cargo to the ISS within a
few months. But scientists will have to wait
a while longer — before a new rocket can
carry a scientific payload, NASA requires
three successful launches and a technical
certification that takes about three years.
NASA hopes to certify Falcon 9 or one of
its competitors by the end of 2013. ■ estimated US$3 billion spent on such
Eric Hand
© 2010 Macmillan Publishers Limited. All rights reserved
Neglected diseases fund touted
Despite decades of research into drugs
and vaccines for neglected diseases such as
tuberculosis and dengue fever, few products
have made it through clinical development
and into the hands of the millions who
desperately need them.
One of the biggest hurdles is the
sheer expense of running clinical trials,
compared with the small profits that
commercial companies can expect to
make from treatments for diseases that
disproportionately affect poor and
marginalized populations.
This week, alongside the World Health
Organization’s (WHO’s) annual assembly
of health ministers in Geneva, Switzerland,
a consortium of industry and non-
governmental organizations proposed
a scheme to help address the problem: a
global fund that would channel billions of
dollars a year into product development.
Plans for the fund were put forward by
the International AIDS Vaccine Initiative
(IAVI), the pharmaceutical giant Novartis
and the George Institute for International
Health in Sydney, Australia. It would
channel money from donors towards
product-development partnerships
(PDPs) — collaborative efforts between
research agencies, donors and biotech and
pharmaceutical companies to develop
drugs, diagnostics and vaccines for the
developing world. The fund aims to
encourage contributions from donors who
lack the resources or expertise to assess the
quality and progress of the various PDP
offerings.
Dozens of not-for-profit PDPs, including
the IAVI, the Medicines for Malaria
Venture, the Global Alliance for TB Drug
Development, and the Drugs for Neglected
Diseases initiative (DNDi), have been set
up during the past 15 years. Their aim is
to bridge the gap between basic research
and product development, and to prevent
promising research leads for neglected
diseases from languishing on the shelf. They
are run like businesses, but are supported
by donor funding, and have generous
intellectual-property rules to make any
products affordable to poor countries,
allowing generic manufacturers to make
cheap versions freely.
PDPs have become an attractive choice
for neglected-disease donors — of the
research in 2008, about one-fifth was
NEWS
DoctorS SEEk rEform
puNchStock
Cumbersome rules for UK
clinical trials are driving
research overseas.
go.nature.com/q2q7BT
channelled through PDPs. Although only
13 of more than 1,000 drugs developed
between 1975 and 1997 were for neglected
diseases, PDPs created over the past decade
already have 143 candidate products
in development, and have rolled out 11
products for malaria, leishmaniasis and
meningitis.
But many PDPs need a fresh cash
injection as more of their product
candidates begin to enter clinical
development, the most expensive phase
of drug and vaccine discovery. Without
substantial new funding, projects will stall
and waste much of the earlier development
work and investments, says Paul Herrling,
head of the Novartis Institutes for
Developing World Medical Research.
The proposed PDP+ Fund would seek to
raise funds from governments and other
donors, and through bond financing and
innovative taxation schemes. It would act as
a one-stop-shop for donors, coordinating
funding of projects to many different PDPs.
“Do we need a super-PDP fund? Without
a doubt,” says Jean-François Alesandrini,
a spokesman for the DNDi. However, he
cautions that the details of the scheme still
need fleshing out, particularly on the issue
of attracting new donor funding. Many
questions remain as to how the PDP+ Fund
would work, adds Alesandrini, in particular
its governance structure, and how its expert
committees would choose which projects to
fund.
Herrling says that informal discussions
he has had with donors, including the US
government, have been positive, as have
those with other companies and groups that
will be invited to come on board.
The PDP+ Fund resembles the idea for a
global research fund that was floated by a
WHO expert panel in 2002 to complement
the multibillion-dollar Global Fund to Fight
AIDS, Tuberculosis and Malaria. This was
created in the same year, but funds only
disease-control measures, not research. The
global research fund proposal never gained
traction because governments considered it
a risky venture, says Mary Moran, director
of health policy at the George Institute. The
difference now, says Herrling, is that PDPs
are widely recognized to produce drug
leads. “We have a state-of-the-art pipeline
that now needs investment to take forward,”
he says. ■
Declan Butler
277
 Vol 465|20 May
NATURE|Vol 465|20
2010May 2010 NeWs

Lizards’ darK future

F. R. M. de La CRuz
Climate change could make
one-fifth of species extinct
by 2080.
go.nature.com/ag4nBn

Pact protects Canadian forests

Huge conservation deal will benefit caribou and maybe climate.
An unlikely coalition of logging alive, they keep accumulating

G. Lenz
companies and environmental carbon,” says Luyssaert.
groups has reached an agreement The northern circumpolar
to protect more than 300,000 permafrost region, which
square kilometres of Canadian includes most of the boreal forests
boreal forest — an area larger earmarked for protection, con-
than the United Kingdom — tains “approximately 50% of the
the biggest forest-conservation estimated global belowground
deal in history. An additional organic carbon pool”, according
385,000 square kilometres will to a study co-authored by Josep
fall under strict guidelines that Canadell, director of the Global
will promote sustainable logging Carbon Project in Canberra,
and protect ecologically and Australia (C. Tarnocai et al. Glo-
culturally sensitive sites. bal Biogeochem. Cy. 23, GB2023;
Under the agreement, unveiled 2009). Canadell says that cutting
on 18 May in Toronto, Ontario, 21 down forests sometimes results
members of the Forest Products in the drying out of wetlands
natuRe ConseRvanCy

Association of Canada, which PROTECTION PLAN Boreal region and peat bogs and the release of
represents the majority of com- A new agreement will protect large swathes of Logging suspended their huge carbon stores — which
panies in the Canadian logging Canada’s boreal forest from logging. Logging restricted hold an average of 7,800 tonnes of
industry, will set aside slightly less Extant boreal forest carbon per hectare, far more than
than half of the land for which any other ecosystem.
they hold leases across seven But Werner Kurz, a senior
provinces. In exchange, nine researcher at the Canadian For-
environmental groups, includ- est Service in Victoria, British
ing Greenpeace and the Nature Columbia, isn’t sure that forest
Conservancy, have pledged to conservation is going to slow
suspend do-not-buy campaigns C A N A D A down warming. Given the speed
against the loggers’ products, of climate change, it’s not clear
which range from construction that intact forests will necessarily
lumber to toilet paper, and to U N I T E D S T A T E S 500 km be more resilient than well-man-
actively endorse them. aged ones, he says. For instance,
“We strongly believe that every improvement be off-limits to logging, which will make it the harvesting old-growth trees and replacing them
in environmental quality can translate into mar- largest area of protected forest in the world. with seeds obtained from warmer climes can
ket value for our products,” says Avrim Lazar, This is good news for caribou, whose num- produce trees that will better withstand tem-
president of the Forest Products Association. bers in Canada have been in steep decline. But perature increases, he says, and as such would
it may also slow down global warming, given be more likely to thrive and sequester carbon.
Crucial habitats mounting evidence that boreal forests are “We’re still not sure exactly how useful these
The finer details of areas to be preserved (see important carbon stores. forests are going to be in mitigating global
‘Protection plan’) will be chosen by biologists “Protecting these forests and their soil, which warming,” says Hank Margolis, a forest eco-
and logging companies. Their choices will have has enormous amounts of carbon, is a hugely system scientist at Laval University in Quebec
to be approved by the provincial governments important step forward,” says Stuart Pimm, a City, who heads the Canadian Carbon Pro-
and Native Canadians. conservation ecologist at Duke University in gram. “That’s why it makes sense to keep them
“This will allow us to protect the most intact Durham, North Carolina, who advises Pew. intact until we figure it out. And to do that,
parts of the boreal forest that are critical habi- Unlike tropical forests, which constantly we’re going to need to do a lot of science.”
tat for the caribou and other species,” says Steve recycle atmospheric carbon through phases The forests may fare better than the research
Kallick, director of the Pew Environment Group’s of growth and decay, boreal forests experi- programmes. For years, the Canadian gov-
International Boreal Conservation Campaign, ence less decay and instead tend to pool ernment has declined to renew funding for
based in Seattle, Washington, which brokered carbon in soil and peat. A recent study led the Canadian Foundation for Climate and
the deal it hails as “radically pragmatic”. by Sebastiaan Luyssaert, a biologist at the Atmospheric Sciences — its granting agency
When added to earlier commitments by University of Antwerp, Belgium, found that dedicated to climate research in universities.
Ontario, Quebec, Manitoba and the Northwest mature boreal forests remain active carbon “I’m afraid most of our research will die out by
Territories as well as the federal government, sinks rather than becoming carbon-neutral the end of the year,” says Dawn Conway, the
this week’s agreement means that 1.6 million ecosystems as they mature (S. Luyssaert et al. foundation’s executive director. ■
square kilometres of Canada’s boreal forest will Nature 455, 213–215; 2008). “As long as they’re Christopher Pala
279
© 2010 Macmillan Publishers Limited. All rights reserved
NEWS NATURE|Vol
Vol 465|20 May 2010

Malaria may not rise as world warms

Studies suggest that strategies to combat the disease are offsetting the impact of climate change.
Of the many climate-change catastro- change per se is not something that

W. Daniels/Panos
phes facing humankind, the antici- should be central to the discussion. The
pated spread of infectious tropical risks have been overstated.”
diseases is one of the most frequently Some earlier analyses painted a dire
cited — and most alarming. But a picture of a malaria-ridden future,
paper in this week’s Nature adds to the but these models often exclusively
growing voice of dissent from epide- evaluated the impact of warmer tem-
miologists who challenge the predic- peratures without taking other factors
tion that global warming will fuel a into consideration, says Paul Reiter,
worldwide increase in malaria. an entomologist at the Pasteur Insti-
On the surface, the connection tute in Paris. The latest assessment of
between malaria and climate change the Intergovernmental Panel on Cli-
is intuitive: higher temperatures are mate Change noted these concerns:
known to boost mosquito popula- “Despite the known causal links
tions and the frequency with which between climate and malaria trans-
they bite. And more mosquito bites mission dynamics, there is still much
mean more malaria. uncertainty about the potential impact
Yet when epidemiologists Peter of climate change on malaria at local
Gething and Simon Hay of the Malaria and global scales.”
Atlas Project at the University of Preventative measures such as the widespread use of bed nets Gething and colleagues’ study is the
Oxford, UK, and their colleagues com- have outweighed the effects of climate warming on malaria. first of its kind to provide a detailed sta-
piled data on the incidence of malaria tistical model of global trends over the
in 1900 and 2007 (see page 342), they found the of malaria, the impact of public-health meas- twentieth century, but it does have limitations.
opposite: despite rising temperatures during ures such as improved medications, widespread For instance, the data used to generate a global
the twentieth century, malaria has lost ground. insecticide use and bed nets have overwhelmed map of malaria in 1900 sum up all malarial
According to the models the researchers used the influence of climate change. “Malaria is still infections, including those by a malaria parasite
to tease out the factors affecting the incidence a huge problem,” says Gething. “But climate named Plasmodium vivax, whereas the data in

Pacific tuna population may crash at any time

Numbers of Pacific bluefin tuna may be about the population “does says Katsukawa.

to go the same way as those of their Atlantic
cousins. The Pacific population was thought
to be the endangered fish’s only remaining
stronghold. But official reports on stock sizes
have overlooked changes to fishing practices
that could mean they are heading for a crash,
according to recent assessments.
In the Atlantic Ocean and the Mediterranean
Sea, fishing has caused tuna populations to fall
below 15% of their historical levels. Even so, an
attempt this year to ban trade in the fish failed
to gain international approval at a meeting in
March of the Convention on International Trade
in Endangered Species of Wild Fauna and Flora
(see doi:10.1038/news.2010.139).
By contrast, Pacific bluefin (Thunnus orientalis)
populations had been thought to be stable, with
enough young fish maturing each year to replace
those caught. In July 2009, a working group of
the International Scientific Committee for Tuna
and Tuna-like Species in the North Pacific Ocean
(ISC) concluded that recruitment of juveniles to
280
R. HeRRmann/PHotolibRaRy.com
not appear to have been The change in fishing
adversely affected by practice has gone hand
the relatively high rate in hand with a growth in
of exploitation”. the market for juvenile
But the working tuna, says Chien-
group admitted that Chung Hsu, a fisheries
a lack of data meant biologist at the National
recruitment since 2005 Taiwan University
is highly uncertain. in Taipei. Juveniles
And the belief that the Pacific bluefin tuna are in high demand. 10–15 centimetres
stock is stable is based long fetch US$100
on a false assumption that fishing practices have each in Japanese markets, he says. By contrast,
not changed, says Toshio Katsukawa, a fisheries adult Pacific bluefins can sell for $100,000 or
expert at Mie University in Tsu City, Japan. more, and Katsukawa points out that fishermen
From speaking to fishermen, Katsukawa is most could make more profit by waiting for the fish to
concerned that, in the past few years, boats have mature. Overall, more than 70% of the Pacific
begun targeting the tuna’s spawning grounds. bluefin catch is taken by fishermen from Japan,
This tactic increases catches, simultaneously where the fish is prized for its use in sushi.
making the stock seem bigger but damaging the Data from the ISC show that more than 70%
fish’s breeding capacity. “If things go on like this, of Pacific bluefin tuna caught today are less than
the Pacific [bluefin] populations will be the one year old, compared with around 60% in the
first to collapse [before the Atlantic stock],” 1960s, although Katsukawa believes that even
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465|20 May
NATURE|Vol 465|20
2010May 2010
the 2007 map track infections by just one malaria
parasite, P. falciparum, which carries the highest
disease burden. And the analysis does not take
into account some parameters that are likely to
change as a result of global warming, such as
rainfall patterns and human migrations.
Nevertheless, the results largely match those
of several other recent studies, including one
published last year by Kevin Lafferty of the US
Geological Survey in Santa Barbara, California,
which also concluded that rising temperatures
over the last century had no net impact on the
incidence of malaria (K. D. Lafferty Ecology 90,
888–900; 2009). In 2000, models designed by
David Rogers and Sarah Randolph at the Uni-
versity of Oxford predicted that although some
parts of the world would gain malaria because
of climate change, large areas would see a drop
in disease due to reductions in rainfall and
humidity (D. J. Rogers and S. E. Randolph Sci-
ence 289, 1763–1766; 2000). The result: no net
difference.
“The complexity of malaria and the other
vector-borne diseases is astonishing,” says
Reiter. “To bring it down to just one factor —
climate change — is totally unjustifiable.”
The same could hold true for other diseases
that rely on intermediate vectors like mosqui-
toes. Dengue fever, for example, has also been
touted as an infectious disease that could get a
boost from climate change. The virus, which is
carried by mosquitoes, is already on the march,
and is spreading more rapidly into the southern
this increase is an underestimate. More than
90% of the catch is less than two years old. “The
increasing juvenile catch is increasing the risk of
population collapse,” says Hsu. “The population
is dropping for all age classes and we see signs
of serious problems if no management measures
are set immediately.”
Responding to concerns about the health of
the stock, Japan’s Fisheries Agency last week
outlined new measures to monitor and manage
the tuna populations. These specify limits on
the weight of fish that each boat can catch,
restrictions for the large boats using encircling
nets that account for most tuna fishing, along
with new requirements for other boats to report
their catches. And fish farmers, who collect an
increasing but unknown number of juveniles and
raise them in pens, will be required to register
and report their activities. The agency plans to
implement the restrictions by April 2011.
The measures also call for better information
on juvenile stocks and spawning areas, and for
the establishment of an international network of
researchers to study and manage Pacific bluefin
tuna fisheries. “To protect these spawning
regions, we have to find out where they are
© 2010 Macmillan Publishers Limited. All rights reserved
NEWS
United States. But the impact it will have there
is likely to be minimal compared with that in
less developed regions of the world because
of lifestyle differences, says Reiter. Americans
spend more time indoors, he says, and are
more likely to have window screens to keep
mosquitoes out.
Laura Harrington, who studies mosquito-
borne diseases at Cornell University in Ithaca,
New York, agrees. “In the context of vector-
borne diseases, climate change is probably
going to have a minimal role in comparison
to other factors,” she says. Even so, Gething’s
results are likely to be controversial, cautions
Richard Ostfeld, a disease ecologist at the
Cary Institute of Ecosystem Studies in Mill-
brook, New York. Although global analyses
such as Gething’s are useful, he says, they may
miss important regional trends — such as the
spread of mosquitoes from the lowlands to the
highlands of eastern Africa, which some argue
is the result of rising temperatures. “There’s a
pretty strong spatial disconnect between areas
where there is strong economic development
and increasing control of mosquitoes, versus
those areas where the risk of climate-induced
disease is highest,” he says.
Others worry that the results of the study will
be misinterpreted. “On smaller scales, climate
change certainly has big effects on disease,”
says Harrington. “This does not diminish the
importance of climate change at all.” ■
Heidi Ledford
and when the peak seasons are. Then we can
start to make restrictions on which ones can be
fished and when they can be fished,” says Yukio
Takeuchi, a researcher at the National Research
Institute of Far Seas Fisheries, an arm of the
Fisheries Agency based in Shizuoka, Japan, who
chaired the ISC working-group report.
Katsukawa says that the new measures, “if
done effectively”, could have a major impact. But
he fears they will be implemented too slowly to
head off an irrecoverable drop. “It could happen
suddenly, but we won’t see it until it happens.” ■
David cyranoski
Corrections
The News story ‘China and Taiwan strengthen
academic ties’ (Nature 465, 148–149; 2010)
incorrectly stated that Xiao-fan Wang is the first
mainland-China-born president of the Society of
Chinese Bioscientists in America. In fact, he is the
first president to be born, raised and educated
under the communist government of the People’s
Republic of China.
The News Feature ‘Life after death’ (Nature 465,
150–155; 2010) misspelt the name of Debra
Moriarity throughout.
281
 NEWS FEATURE
THE RAT PACK
Studying primates is the only way to understand human
cognition — or so neuroscientists thought. But there may be
much to learn from rats and mice, finds Alison Abbott.
A
nne Churchland had little time for
rats. In the course of 13 years’ work on
decision-making in monkeys, she had
never questioned that primate studies
were the only way to understand the neurobiol-
ogy of human cognition. Her work in the lab of
Michael Shadlen at the University of Washing-
ton, Seattle, had monkeys watch moving dots
flitting about on a screen until the animals indi-
cated, with a flick of their eyes, the direction in
which most of the dots were going. She recorded
from single brain neurons as the monkeys
slowly made sense of this ‘fuzzy’ information
— the sort of sophisticated experiment that she
did not think was possible in rodents.
“I didn’t think rats would have the right
sorts of brains to contemplate accumulating
evidence,” says Churchland. And with poor
eyesight, and heads that bob around, “I didn’t
imagine they would be able to convey to us any
decision they might be silently making”.
All that changed a year ago, when Church-
land visited Cold Spring Harbor Laboratory in
New York. Working with scientists there, she
saw that rats could also learn to gather ‘fuzzy’
sensory information — in this case to decide
whether the frequency of a rapid sequence of
tones was mostly high or low. And they could
convey their decision with a poke of the nose.
Churchland was not alone in her earlier scep-
ticism. Neurophysiological research into higher
cognitive functions such as decision-making,
attention, working memory — even risk-tak-
ing — have traditionally been carried out on
non-human primates. That seemed an obvi-
From top: A. mendonçA; A. mendonçA ; A. mendonçA ; rennA@CSHL
ous choice, given the closeness of their brain
anatomy to that of humans, the sophistication
and breadth of their behaviour and their abil-
ity to reliably report to experimenters much of
what is going on in their minds through eye,
hand or other movements. But primate work
comes with major downsides: the animals are
so expensive, and their use so highly regulated,
that a research paper typically relies on data
from just a couple of precious animals, which
have been used for multiple experiments over
their lifetime. This raises concerns that obser-
vations could be unique to those animals, rather
than a general property of the primate brain.
Mice and rats, by contrast, can be studied in
the tens or hundreds. But with brains a fraction
of the size of those of humans or non-human
282
NATURE|Vol
Vol465|20
465|13 May 2010
primates — and no prefrontal cortex, the
highly-evolved brain area where ‘higher’
cognition is thought to take place — neu-
roscientists assumed that rodents were
simply incapable of learning complex
behavioural paradigms.
That scepticism is dissolving, thanks in
large part to a ‘rodent cognition movement’
started by a small group of researchers at Cold
Spring Harbor almost a decade ago and now
spreading far beyond its grounds. Using care-
fully designed tasks, these researchers have
shown that rodents can undertake some types
of complex cognitive behaviour just like exper-
imental primates, and just like humans.
“Primate used to be the only game in town,”
says Zachary Mainen, now at the Champali-
maud Neuroscience Programme in Lisbon,
Portugal, but one of the founders of the rodent
movement when he was at Cold Spring
Harbor. “Now we are starting to appear as
a small force in cognition meetings.”
Evolutionary similarities
Mainen joined Cold Spring Harbor Labo-
ratory in 1999, the same year as his col-
league Tony Zador. Both wanted to move
beyond their backgrounds in computa-
tional neuroscience and cellular neurophys-
iology, and find out how electrical activity
in neurons — such as that stimulated by
sensory input — related to behaviours such
as decision-making. They thought that
these components of behaviour “would
likely be evolutionarily similar across mam-
mals”, says Mainen. And rats, they thought,
would move the field forwards faster than pri-
mates, particularly given the greater availability
of tools for manipulating rodent genes.
Zador sees the choice of primates for cog-
nition experiments as a “historical accident”,
naturally evolving from research begun in the
1960s to understand how vision was proc-
essed in the brain. “Using primates made
complete sense, because vision is so highly
specialized in primates for functions such as
face recognition,” he says. Then, in the 1980s,
some primate-research groups went on to
ask how visual information couples to motor
output; having seen an object, what happens
in an animal’s brain as it decides whether to A rat indicates a decision by poking its nose through a
reach for it? The interesting questions were ‘port’ when it has discriminated between two odours.
© 2010 Macmillan Publishers Limited. All rights reserved
 Vol 465|13 May
NATURE|Vol 465|20
2010May 2010 NEWS FEATURE

now about cognition. “At this point primates

B. GeddeS/CSHL
r. oCHôA/FUndAção CHAmpALImAUd

offered no unique advantage, because the tasks

that the researchers were asking monkeys to do
were so simple,” says Zador.
To study these tasks in rats, Zador and Mainen
had to decide what sensory system to use, and
establish a behavioural read-out. Rodents pri-
marily rely on senses other than vision, such as
hearing and smell, to guide their behaviour. So
Zador began to look at how rats processed audi-
tory information; Mainen focused on odours.
It took a few years, and a lot of trial and
error. But by 2003 Mainen had published his
first paper1 showing that rats could be trained
to reliably repeat behaviour, discriminating Researchers including Zachary Mainen (left) and Tony Zador founded a ‘rodent cognition movement’.
between similar smells after a single whiff.
More to the point, he showed that they could activity can be switched on or off with flashes probes the ability to make appropriate deci-
indicate their detection of an odour by poking of laser light3, allowing the role of neurons in sions in the face of stacked odds4. Researchers
their noses through a ‘port’ in the cage wall. a behaviour circuit to be dissected. Right now have also claimed that a paradigm based on the
That paper was an eye-opener for Carlos these systems work best in mice. But because prisoner’s dilemma, which explores why people
Brody, a computational neuroscientist at Cold most behavioural studies have been carried might not cooperate even when it is in their best
Spring Harbor. “I had theories that I would out using rats, cognitive scientists have mostly interests, shows that rats can understand the
have liked to test in a primate lab, but this paper chosen to start on rats in the hope that the complex pay-offs that cooperation entails5.
showed that you could do equally rigorous techniques will be quickly transferred. Zador Some primate researchers, though, remain
work with rats,” he says. Brody joined forces says it’s still not clear whether the smaller- unconvinced. “It is good to develop rodent
with Mainen and Zador and the three of them brained mouse is capable of the behaviours in models and see what they are capable of,”
persuaded the laboratory to set up the Center which the field is interested. says Shadlen. “But it still isn’t clear to me
for Neural Mechanisms of Cognition in 2006, that rodents do any serious deliberating in
devoted to rodent work. Rodent logic decision-making.” And Daniel Salzman at
In 2008, Mainen published a second water- Churchland is so taken with the experimental Columbia University in New York says that
shed paper, using electrophysiology to show possibilities afforded by rodents that she is the differences, rather than the similarities,
how rats make everyday decisions on the basis staking her career on them. This summer she in brain anatomy and circuitry are going to
of fuzzy evidence2. This time he trained rats is moving to Cold Spring Harbor, where she be decisive, such as the smaller rodent frontal
to distinguish between two odours delivered will establish her own lab to study rat deci- cortex. Rodent researchers “are quickly going
through the central port of a row of three. If a sion-making. Looking back, she wonders why to run up against a wall,” he predicts.
mixture had more of odour one, the rat had to she doubted rats’ cognitive abilities so much. Still, few on either side of the species divide
poke its nose through the left-hand port; if it “They also have to make decisions in the wild care to be too categorical. Rodent-cognition
had more of odour two, it had in order to survive, and would researchers have presented enough new data
to select the right-hand one. “The rat brain shares obviously have to accumulate at meetings to discourage dogmatism from pri-
The decision became very dif- some of the most and sift evidence to do so.” She mate loyalists. And rodent proponents empha-
ficult when the mixtures con- wants to explore why some rats size that primates will always be required to
tained nearly equal parts of fundamental design choose a strategy of decision- reality-check the theories about cognition
the two odours, but if the rats principles with those making that sacrifices accu- spawned by rodent research.
decided correctly, and waited of humans.” racy for speed. “With higher “Primates are going to be capable of some
long enough at the correct numbers, we can start to look cognitive processes that rats are simply not
port, they received a drink reward there. The at individual differences,” she says. capable of,” says Brody, who thinks both types
more confident rats were about their decision, Other committed primate researchers, such of research should run in parallel. “But the jury
Mainen found, the longer they were prepared to as Daeyeol Lee, a neuroeconomist at Yale Uni- is still very much out in terms of where the
wait. And when he took recordings from single versity School of Medicine in Connecticut, are capability border lies, and we think it is worth
neurons in the orbital frontal cortex, a brain area also exploring the use of rodents. Lee has been finding out.” ■
involved in decision-making, he found patterns working on primate decision-making for 15 Alison Abbott is Nature’s senior European
of electrical activity that correlated with the rats’ years. “The rat brain shares some of the most correspondent.
conviction. “It hadn’t been clear whether the rat fundamental design principles with those of
1. Uchida, n. & mainen, Z. F. Nature Neurosci. 6,
brain was going to be up to the task of estimating humans and other primates, such as connectiv- 1224–1229 (2003).
confidence in decisions,” says Mainen. “But we ity between the cortex and some sub-cortical 2. Kepecs, A., Uchida, n., Zariwala, H. A. & mainen, Z. F.
showed it was, and at least in this sort of task, areas,” he says. “Rats may be able to teach us a Nature 455, 227–231 (2008).
rats are as good as monkeys as subjects.” lot.” And behavioural researchers are working to 3. Buchan, L. Nature 465, 26–28 (2010).
4. Zeeb, F. d., robbins, t. W. & Winstanley, C. A.
In certain ways, rodents are better. In the see how far they can go with rodents, developing Neuropsychopharmacol. 34, 2329–2343 (2009).
past few years the development of ‘optoge- new paradigms in rats that might even mimic 5. Viana, d. S., Gordo, I., Sucena, e. & moita, m. A.
PLoS ONE 5, e8483 (2010).
netic’ tools has allowed rodent researchers some classic human psychology tests, includ-
to engineer particular neurons so that their ing a version of the Iowa gambling task, which See Editorial, page 267.
283
© 2010 Macmillan Publishers Limited. All rights reserved
 neWS FeatUre
Death anD rebirth
in the Deep
When a submarine volcano erupts, the results can be devastating —
and fascinating. Jane Qiu finds new drama in underwater biogeography.
R
ichard Lutz, a marine biologist at
Rutgers University in New Brunswick,
New Jersey, and his colleagues were
2,500 metres beneath the ocean’s surface
when they encountered the ‘blizzard’. It was
April 1991, and an underwater ridge, 900 kilo-
metres off the coast of Acapulco, Mexico, was
splitting open, introducing 1,200 °C molten rock
to 2 °C water. The results were apocalyptic.
While the researchers kept a safe distance
from the action in their submersible, the ‘snow’
of microbial debris blowing around them sig-
nalled devastation at the site of the eruption.
The bacteria had been thriving at the mouths
NOAA, NSF
of hydrothermal vents in the area, extracting
energy from the hydrogen sulphide and other
chemicals pouring out of the sea bed. In turn,
the bacteria nourished a diverse ecosystem of
organisms: clams, crabs, mussels and armies
of tubeworms. The eruption had laid waste to
nearly all of this.
But nature wasn’t permanently obliterat-
ing life at this site at latitude 9° 50ʹ N on the
East Pacific Rise — rather, it was hitting the
reset button. Lutz and his colleagues now had
an opportunity to see how life returns to the
vents. Putting monitoring equipment in place
and returning regularly, they documented the
drama of life-after-death as it unfolded. And
after 15 years of research, nature repeated the
experiment: Nine North, as the researchers call
the site, erupted again in 2006.
“It is the only deep-sea location where a full
284
eruptive cycle has been observed,” says Lauren
Mullineaux, an ecologist at Woods Hole Ocea-
nographic Institution (WHOI) in Massachu-
setts who has been working at the site. Nine
North has allowed marine biologists to iden-
tify the first species to return to a vent and the
parade of life that follows. And now, four years
on from the latest eruption, researchers have
begun to answer tougher questions: where do
these species come from, how do they travel
and how do their populations shift over time?
But more than that, says Robert Vrijen-
hoek, a molecular ecologist at the Monterey
An underwater eruption, caught on film in 2009.
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol
Vol 465|20 May 2010
Bay Aquarium Research Institute in Moss
LeFt tO right: r. A. Lutz/rutgerS uNiv.; t. ShANk; t. ShANk, W. LANge, WhOi, r. A. Lutz/rutgerS uNiv., LOW PrOductiONS
Landing, California, “habitat turnover is a
window into evolutionary pressures that con-
tribute to speciation and genetic diversity”. By
identifying more vent sites around the world,
researchers are starting to untangle the intri-
cate interconnectedness between these tiny
islands of life scattered through the abyss
and separated by vast distances and rugged
topography. These sites share similar geologi-
cal features, with mineral-spewing vents and
daunting ridges and valleys. But they don’t
always share the same species. Researchers are
now starting to understand why.
Rapid succession
The first glimpses through the window
provided by Nine North revealed a picture of
a community rebuilding quickly, says Timothy
Shank, a marine biologist at WHOI who col-
laborates with Lutz (see ‘The cycle of life’). The
microbial snowfall that piled around the volcano
after the 1991 eruption first attracted grazers —
shrimps, crabs and fish — both vent and non-
vent in origin. Within a year of the eruption,
the tubeworms returned. First came Tevnia
jerichonana, which thrives at high sulphide con-
centrations; then, as sulphide levels dwindled
in the subsequent year, dense bushes of Riftia
pachyptila, the giant tubeworm, took over.
Thousands of these creatures, with sturdy
tubes up to two-metres long, created a rich
environment for other organisms to inhabit.
 Vol 465|20 May
NATURE|Vol 465|20
2010May 2010 neWS FeatUre

Snails, and more crabs and swarming shrimps
arrived1. It was a never-before-told story of bio-
logical succession. But it was unclear whether
the organisms taking over the vents were from
local sources or came from far away. Nor was it
clear whether species composition in the region
had changed as a result of the eruption, says
Shank, because the area had not been studied
in great detail previously.
This is where Mullineaux, a specialist in
marine larvae, comes in. Although many vent
inhabitants are fused to the sea floor or to other
organisms as adults, most spend their early
lives as free-swimming larvae that can ride the
currents to new homes. She and her team have
been studying larvae at Nine North ever since
the 1991 eruption. They’ve collected them at
different depths and at the sea floor to get a
sense of abundance and species make-up.
As Mullineaux and her colleagues describe
in a paper published this year2, the species
composition of the larvae at the hydrothermal
vents changed markedly after the 2006 erup-
tion. Larvae from species common at the site
before the eruption were nowhere to be seen
afterwards — even though there were potential
sources of recolonization within a few kilome-
tres. By contrast, other species that had been
rare became abundant after the catastrophe.
The Woods Hole team also discovered larvae
of a species never seen before at Nine North, a
rock-clinging snail called Ctenopelta porifera.
The nearest hydrothermal system known to
host the species is more than 300 kilometres
to the north. “This has greatly exceeded our
N. SPeNcer
expectation of how far vent larvae could travel currents near the crest of the ridge were faster
in the ocean,” says Mullineaux. than those farther away, and that currents on
Mullineaux’s group had in 2001 calculated the eastern flank of the ridge flowed south-
that a larva with a lifespan of about a month wards, whereas those on the west side went
could travel at most up to northwards.
100 kilometres from Nine Surprisingly, the study
North and that the major-
“Habitat turnover is a shows that currents closer
ity stayed within 60 kilo- window into evolutionary to the sea floor are among
metres3. Even if larvae lived pressures that contribute the strongest, at about4
longer, they wouldn’t be able 10 centimetres per second .
to travel farther; the flow of
to speciation.” This caught the researchers
the current reverses every by surprise. “Our observa-
few weeks along the ridge axis and so limits how tions are markedly different from the much
far larvae could go. How C. porifera had made a weaker and wider currents we see in many
300-kilometre journey was a mystery. areas of the deep ocean,” says Thurnherr. It
may also explain how C. porifera larvae could
Complex flow travel so far.
The researchers had based their calculation of Diane Adams, a former graduate student
flow on measurements of the current at a single of Mullineaux’s, uncovered another surprise.
location at the crest of the ridge, and assumed She had noted a drop in larval abundance at
that the direction and speed of the currents Nine North whenever surface eddies — loops
were the same over the whole area. “This is a of rotating currents resulting from differ-
reasonable assumption in many parts of the ences in water density interacting with Earth’s
ocean,” says Andreas Thurnherr, a physical rotation — were passing by above.
oceanographer at Columbia University’s Lam- The researchers hadn’t thought that surface
ont-Doherty Earth Observatory in Palisades, eddies could reach down to the ocean floor,
New York. But the topography of Nine North but they were able to pick up signals of rotating
disrupts current flow significantly, he says. current loops a short time after surface eddies
To better estimate currents, Thurnherr, passed by. “The surface eddies could, in effect,
Mullineaux and their colleagues placed 15 blast larvae off the ridge and have an important
current-measuring devices along the crest and role in their dispersal,” says Mullineaux.
both flanks of the ridge; they also deployed Such interplay between geology and ocean
two mobile current meters that travelled currents has fascinated researchers of chemo-
between the sea floor and the top of the ridge, synthetic life for decades. “It’s important not
sampling as they went. The team found that only for recolonization after natural disasters

THE CYCLE OF LIFE Within two years, giant

After an eruption such as that witnessed at Nine North in 1991, tubeworms, Riftia pachyptila,
life returns quickly to hydrothermal vents, changing rapidly as begin to dominate vent
the concentration of hydrogen sulphide diminishes. openings.

Within one year the mats are

reduced and a small tubeworm
species, Tevnia jerichonana,
takes residence.

Microbial mats that piled

around the volcano attract
shrimps, crabs, fish and
other scavengers.

Within three years

mussels begin to appear,
and the number of
By year four, mussels
species present expands.
have colonized tubes of
Riftia pachyptila; bivalves
will begin to dominate.

Change in H2S concentration

over time (mmol kg–1) >1 0.97 0.88

© 2010 Macmillan Publishers Limited. All rights reserved

neWS FeatUre NATURE|Vol
Vol 465|20 May 2010

but for the evolutionary connectivity of marine some slight genetic differences7. Gene flow has biodiversity of the ocean. “Some strategic loca-
life,” says Robert Cowen, a marine biologist at taken place from populations in the north to tions are missing pieces of the puzzle of how
the University of Miami in Florida. their southern counterparts, the same direction things have evolved.” These include the Arctic
Researchers want to know why pockets of as the ocean currents in the region. “Tubeworm and the Antarctic, where thick ice and turbulent
life-enabling chemicals in different parts of larvae probably have a sufficiently long lifespan seas make exploration immensely challenging.
the ocean host distinct yet overlapping assem- or stay at the right parts of the water column to Others include the Chile triple junction,
blages of species. And where did these species allow them to make that jump,” he says. where three tectonic plates meet, resulting in
originate? Did they arise first at the vents or Geological changes may have separated the subduction of the Chile rise, a mid-ocean
perhaps in shallow-water cold seeps — areas species in other locations as well. For example, ridge, under the South American plate. “There
where hydrogen sulphide and hydrocarbon the Logatchev hydrothermal vent, located just you have the potential for vents and seeps
sources such as methane leak out from Earth’s east of the Caribbean, is the only place in the in close proximity,” says Tyler. “This allows
interior? And how do organisms such as those Mid-Atlantic Ridge known to host vesicomyid you to address the evolutionary relationship
living off the sulphide-laden oozes of dead clams, which are more typical of the Pacific. between vent and seep animals without the
whales contribute to the connectivity between Some researchers suspect that the animals variable of geography.”
chemosynthetic communities?
The key to the divergence and CURIOSITIES OF THE DEEP The great unknown
convergence of these marine species Researchers have focused on several hydrothermal-vent communities To many, the Caribbean, especially the
during evolution lies in the ability of around the world for the lessons they can teach in biogeography. little-explored Mid-Cayman rise near
larvae to negotiate ocean currents, geo- the island of Grand Cayman, might hold
logical barriers and changes in sea-floor The Blanco transform fault the key to some of the most perplexing
topology over millions of years. In this Northeast
Pacific questions in deep-sea biology. At almost
amount of time, the movement of only ridges 5,000 metres deep, hydrothermal vents
a small number of larvae is sufficient Mid-Cayman rise Logatchev in the region should exist at unrecorded
hydrothermal vent
to allow genetic exchange between environmental extremes and may reveal
Nine North
geographically separated populations. new species. The location is also ideal
The scale of such connectivity “can for studying the role of the isthmus of
East
be astonishing”, says Charles Fisher, a Pacific Panama in the divergence and connec-
marine biologist at Pennsylvania State Rise tivity of vent species.
University in University Park. Chile Researchers wonder whether the
rise
fauna on the Mid-Cayman rise will be
Mid-Atlantic
Long-distance taxi Ridge more closely related to that on the East
Chile triple junction
Fisher’s team has found, for example, Vent communities Pacific Rise, on the other side of the
that some species of tubeworm and Undersea ridge Panama land bridge, or to the organ-
Subduction zone
mussel living on cold seeps in the Gulf isms on the Mid-Atlantic Ridge. Chris
of Mexico are genetically related to their German, a marine geochemist at WHOI
counterparts off the west coast of Nigeria, an might have originated from the Pacific — and co-chair of the ChEss programme, led an
indication of genetic exchange between two arriving through an ancient seaway between expedition in October 2009 that detected sig-
regions that are more than 10,000 kilometres North and South America before the rise of the nals of hydrothermal plumes near the Mid-
apart5. This connectivity occurs over numer- isthmus of Panama 5 million years ago. Cayman rise. A follow-up expedition by the
ous generations through steps that are not yet Other marine biologists, including Cindy UK National Oceanography Centre managed
clear. But researchers suspect it may be aided Van Dover, a deep-sea biologist at Duke to find the vents at a depth of 5,000 metres, the
by the equatorial deep jets in the Atlantic, University Marine Laboratory in Beaufort, deepest ever recorded. As to what they found
which alternate between easterly and westerly North Carolina, say that the clam species are there, the team would say little. Nevertheless,
flow depending on depth and so could trans- probably more common on the Mid-Atlantic German warns to keep expecting surprises. “A
port larvae both ways. Ridge than it looks at present and didn’t nec- few years ago, we thought we knew everything
In other instances, larval dispersal is blocked essarily originate in the Pacific. “We simply about geological barriers and ocean currents
over quite short distances, leading to specia- don’t have enough samples to know for sure.” to predict what we are going to find, but we
tion. In the northeastern Pacific Ocean, the This highlights a limitation for this relatively have been wrong every time since.” ■
450-kilometre-long Blanco transform fault has young research field: scientists have only a Jane Qiu writes for Nature from Beijing, China.
separated the Juan de Fuca and Gorda ridge rudimentary knowledge of the distribution of 1. Shank, t. M. et al. Deep Sea Res. II 45, 465–515 (1998).
systems, off the coast of Washington State and chemosynthetic life, let alone the underlying 2. Mullineaux, L. S., Adams, d. k., Mills, S. W. & Beaulieu, S. e.
Oregon (see ‘Curiosities of the deep’). Vrijen- mechanisms of connectivity and speciation8. Proc. Natl Acad. Sci. USA 107, 7829–7834 (2010).
hoek’s team has found that similar-looking So far, only 200 or so hydrothermal vents and 3. Marsh, A. g., Mullineaux, L. S., Young, c. M. &
Manahan, d. t. Nature 411, 77–80 (2001).
snails at Juan de Fuca and Gorda are related, a few dozen cold seeps have been discovered 4. Mcgillicuddy, d. J., Lavelle, J. W., thurnherr, A. M.,
but quite different, species that diverged around the world. “Most parts of the ocean are kosnyrev, v. k. & Mullineaux, L. S. Deep-Sea Res. I (in the
press).
roughly 11 million years ago, consistent with unexplored,” says Paul Tyler, a marine biolo- 5. cordes, e. e. et al. Deep Sea Res. I 54, 637–653 (2007).
6
the time of the fault’s formation . gist at the National Oceanography Centre in 6. Johnson, S. B., Young, c. r., Jones, W. J., Waren, A. &
But, says Vrijenhoek, “the barrier doesn’t Southampton, UK, and chair of the Biogeog- vrijenhoek, r. c. Biol. Bull. 210, 140–157 (2006).
affect all animals in the same way”. The tube- raphy of Deep-Water Chemosynthetic Eco- 7. Young, c. r., Fujio, S. & vrijenhoek, r. c. Mol. Ecol. 17,
1718–1731 (2008).
worms at the two ends of the fault, for example, systems (ChEss) programme of the Census of 8. van dover, c., german, c. r., Speer, k. g., Parson, L. P. &
are the same species, even though there are Marine Life, a global initiative to document the vrijenhoek, r. c. Science 295, 1253–1257 (2002).

286
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465|20 May
NATURE|Vol 465|20
2010May 2010 COLUMN

Disaster, unmitigated biodiversity, Djoghlaf points to the designation

of 2010 as the International Year of Biodiversity
(it’s also the International Day for Biodiversity
this Saturday, 22 May). It is very hard to see such
An oil slick will not re-engage the public with environmental issues, sterile designations registering with the wider
warns Colin Macilwain, but it might lead to a saner US energy policy. public, however.
In the United States, the Senate has failed

T
he economist Paul Krugman suggested even to ratify the CBD, signed by President Bill
in his New York Times column earlier this Clinton in 1993, because public pressure on
month that the BP oil leak in the Gulf of senators to vote for it is too weak to overcome
Mexico could provide the flagging environ- the mild objections of the drug industry to its
mental movement with the renewed impetus call for bioprospectors to share patent rights
it so badly needs. with local people.
The modern movement, he pointed out,
gained a great deal of momentum from a fire Troubled waters
on the Cuyahoga River in Cleveland, Ohio, on The oil leak in the Gulf has certainly elicited a
22 June 1969. Legend holds that the sight of the sharper short-term political response than the
blazing river fuelled public support for meas- set up to confront environmental challenges. slow-burn issue of biodiversity conservation
ures including the creation of the powerful The result is that environmental issues has ever managed. Arnold Schwarzenegger,
Environmental Protection Agency by Richard consistently rank close to bottom on the list the governor of California, was soon in full
Nixon in 1970, and the passage of the Clean of voters’ priorities. In an Opinion Research Terminator mode. “All of you have seen, when
Water Act two years later. Corporation poll for CNN in March, for exam- you turn on the television, the devastation in
Environmentalists have long since lost the ple, “energy and environmental policies” were the Gulf. That will not happen here in Califor-
power, in the United States and elsewhere, to identified as “the most important issue” in con- nia,” he said, reversing his previous support for
achieve legislative success on anything like that gressional elections by just 2% of voters. fresh oil drilling off Santa Barbara.
scale. And it will take more than an environ- In retrospect, 1992’s Earth Summit in Rio de The weather, together with the success or
mental disaster on the shores of the southern Janeiro was something of a high-water mark otherwise of BP’s well-capping efforts, will
states to restore that kind of influence. for the political salience of green issues. There, determine the extent to which the Gulf spill
The river fire reflected the chronic urban world leaders sought to catalyse international will ingrain itself on the public conscious-
pollution being experienced by a great many action to meet grave environmental threats by ness. If it makes its mark, the spill could shift
people at that time — at home, at work and on agreeing to the Framework Convention on Cli- US energy policy, forcing the administration
holiday. The conundrum for environmental mate Change and the CBD. to push both renewable sources and nuclear
activists and scientists today is that the issues Instead of galvanizing public concern, this power even harder than it has already. So far,
that matter most no longer global approach has diluted President Barack Obama’s policy response has
affect voters in developed coun- “The global approach it by violating the axiom that been measured: he remains committed to off-
tries so immediately. to environmental all politics is local. Take the shore drilling, but is tightening its regulation.
Having dealt successfully CBD. On 10 May, it released its However, the spill is unlikely to reinvigor-
with the flagrant issues of
issues has violated third Global Biodiversity Out- ate an environmental movement whose inter-
filthy water and urban smog, the axiom that all look report, summarizing the ests and mode of operation remain too far
environmentalists have turned politics is local.” progress made by the parties to removed from mainstream politics to match
to global trends that pose exis- the convention. It makes grim the influence that was fleetingly enjoyed four
tential threats to our world. But the two leading reading: despite some local successes, none of decades ago.
problems — climate change and biodiversity 21 subsidiary targets to the CBD’s 2002 goal of Public indifference to environmental issues,
conservation — come across to many people achieving a “significant reduction” in the rate if left unchecked, could eventually undermine
as mere abstractions. of biodiversity loss by 2010 has been met. Ten support for scientists in the plethora of subdis-
of 15 indicators tracking biodiversity show ciplines, from ecology to atmospheric physics,
Remote control negative trends. that are now strongly oriented towards meeting
Environmental activism in the United States Ahmed Djoghlaf, who runs the CBD’s secre- global environmental threats.
has changed in other ways. By moving on from tariat in Nairobi, Kenya, hopes the findings will There has always been a lively debate (origi-
neighbourhood-based, grass-roots campaign- get world leaders to acknowledge the impor- nally in ecology, and now more widely) about
ing to a reliance on expensive court actions tance of biodiversity. “This report makes it whether a scientist can mix objectivity with
— an approach that has, admittedly, yielded clear why their response to the economic crisis advocacy. It’s not an argument that needs to be
successes — environmental groups have dis- must take on board the biodiversity agenda,” he resolved: it depends on the outlook and temper-
tanced themselves from the ordinary people says. A day of biodiversity talks is planned for ament of the scientist. However, those research-
whose interests they seek to serve. heads of state at the United Nations in Septem- ers who do feel comfortable with advocacy need
This sense of remoteness pervades not just ber, followed by the expected endorsement of to spend more time on the ground, talking to
the leadership of the main environmental a new strategic plan at the next conference of real people about why their work matters. ■
activist groups, it also clings to the scientific the 193 parties to the CBD in Nagoya, Japan, Colin Macilwain is based in the United
and semi-scientific bodies, such as the Inter- in October. Kingdom.
governmental Panel on Climate Change and But asked what the CBD is doing to build e-mail: cfmworldview@gmail.com
the Convention on Biological Diversity (CBD), public support for faster action to conserve See go.nature.com/ILx8PC for more columns.

287
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
How government
spending cuts put
lives at risk
World leaders currently making
tough economic decisions
should be guided by the
physicians’ mantra: “First, do no
harm”. Austerity programmes,
even if justifiable in terms of
promoting growth (itself highly
questionable), may exacerbate
the health risks posed by financial
crises.
Inflicting short-term pain
constitutes a massive, uncontrolled
experiment with people’s lives.
Recessions themselves pose a risk
to health, but empirical data
reveal that the decisions made by
governments crucially determine
how bad the outcome will be.
To mitigate the effects of the
Great Depression in the 1930s, US
policy-makers created a social-
welfare system and invested
in public-health programmes.
Mortality rates fell by about 10%,
mainly owing to a decrease in
infectious diseases and road-
traffic accidents (P. Fishback et al.
Rev. Econ. Stat. 89, 1–14; 2007).
But in several eastern European
countries after the economic
collapse of the early 1990s, policy-
makers massively cut social and
health budgets, while deregulating
the economy. Mortality rates rose
by about 40%, mainly through
heart attack, stroke, alcohol-
related death, suicide and
homicide. There were more than
3 million excess deaths (D. S. et al.
Lancet 373, 399–407; 2009) —
the worst peacetime mortality
crisis in the past half-century.
European Union (EU) mortality
trends during recessions in the
past three decades indicate
that member states can avoid a
rise in suicide rates by spending
US$200 per capita a year or more
on labour-market programmes,
designed to improve people’s
chances of gaining employment
(D. S. et al. Lancet 374, 315–323;
2009). In those spending less
than $70 — such as Spain and
the mainly eastern European
countries that joined from 2004
CORRESPONDENCE
onwards — a deteriorating
economy correlates with a rise
in suicide rate. But in Finland and
Sweden, which spend at least
$300, economic change has no
discernible short-term effect on
overall population health.
What can scientists do to help
protect public health in times of
economic crisis? They should
promote an evidence-based
approach to economic and public-
health recovery, analysing past
successes and failures. This will
lead to a better understanding of
why some people, households,
communities and societies are
resilient to external shocks.
David Stuckler Department of
Sociology, Oxford University,
Oxford OX1 1DP, UK
e-mail: david.stuckler@chch.ox.ac.uk
Sanjay Basu Department of Medicine,
University of California, San Francisco,
California, USA
Martin McKee London School of
Hygiene and Tropical Medicine,
London, UK
Biodiversity: linking
Singapore’s
fragmented habitats
As we celebrate the International
Day for Biological Diversity on
22 May, Singapore — a participant
in the preparatory committee of
the 1992 United Nations Earth
Summit and a signatory to the
Convention on Biological Diversity
— is making another contribution.
In addition to its proposed
index on cities’ biodiversity, to
measure urban efforts towards
conservation and sustainable
development (Nature 460, 33;
2009), Singapore is to construct
an ecological corridor known as
the Eco-Link over the Bukit Timah
expressway. With completion
expected by 2013, this hourglass-
shaped bridge will re-establish
a connection that was severed
in 1985 between the city’s Bukit
Timah and Central Catchment
nature reserves.
The corridor is intended for
tropical conservation and should
redress trade-offs made in the
© 2010 Macmillan Publishers Limited. All rights reserved
past for economic reasons. It has
been compared with Natuurbrug
Zanderij Crailo (‘Crailo sand-
quarry nature bridge’) in the
Netherlands. However, its
tropical setting is likely to pose
engineering and restoration
difficulties, owing to the diversity
of tropical habitats and inhabitants.
It will be planted up like a
forest to enable animal and
plant transfer between the
two reserves. Longer term, it
is hoped that the corridor will
restore the ecological balance
in the fragmented habitats and
rectify the loss of biodiversity. If
successful, it could be replicated
in other tropical cities to help
conserve native biodiversity
and to teach us how to restore
degraded natural landscapes.
It is instructive that a nation as
small, land-scarce, resource-poor
and highly urbanized as Singapore
is leading this promising initiative.
Kwek Yan Chong, Alex Thiam Koon
Yee, Chow Khoon Yeo
Plant Systematics Laboratory, National
University of Singapore, 14 Science
Drive 4, Singapore 117543
e-mail: kwek@nus.edu.sg
Biodiversity: need for
balanced reports of
solutions and failures
Some leading conservation
biologists deliver unnecessarily
gloomy addresses, closing with no
solutions or a few anecdotal
success stories. Non-governmental
organizations (NGOs), by contrast,
can be too optimistic in their fund-
raising lectures and magazines:
false starts and dead ends don’t
figure, because donors prefer
winners. Both groups are
misjudging their audiences.
Without the attraction of
carefully considered general
solutions, young scientists will
become disheartened or even
dismissive of the conservation
crisis. On the other hand,
without honest admission that
conservation programmes can
be messy and sometimes fail,
NGOs will not gain the confidence
OPINION
of increasingly well-educated
donors. Comparative evaluations
do exist of what works in
conservation and what doesn’t,
and they need airing.
Students and philanthropists
want to help. But we should strive
for a nuanced balance of optimism
and brutal honesty in conservation
public relations.
Tim Caro Department of Wildlife, Fish
and Conservation Biology, University
of California, 1 Shields Avenue, Davis,
California 95616, USA
e-mail: tmcaro@ucdavis.edu
Controls needed to
reduce problem of
plastic contamination
Your online News report about
assay contamination by chemicals
leaching from plastic containers
(go.nature.com/R7eAFN)
doesn’t do justice to the scale
of the problem and the effort
needed to tackle it. The cost of
biological experiments performed
in standard polypropylene test
tubes and polystyrene Petri dishes
is enormous, including the price of
millions of animals and expensive
growth factors. Worse, systematic
errors resulting from plastic
leaching are biasing conclusions
drawn from these experiments.
This could have important
implications in stem-cell culture,
for example, aggravated by the
smooth surface of the Petri dish,
which is unlike the cells’ natural
environment.
Companies should address the
problem with innovative products.
But many reagents extract
additives from plastics, and no
manufacturer or legislator can
address the whole variety of
conditions used in laboratories
around the world. Rigorously
designing appropriate controls
could be another way forward.
Andrei P. Sommer Institute of Micro
and Nanomaterials, University of Ulm,
89081 Ulm, Germany
e-mail: andrei.sommer@uni-ulm.de
Noah Lotan Department of Biomedical
Engineering, Technion,
32000 Haifa, Israel
289
 Vol 465|20 May 2010

OPINION
Scientific steps to nuclear disarmament
An advisory group and a network of international labs is needed to lay the groundwork for multilateral
disarmament and forge links between nations, say Martin Rees, Ben Koppelman and Neil Davison.

D
elegations from the 189 countries that more appropriate for verifying the presence
SuMMaRy
have signed the 1968 Nuclear Non- of highly enriched uranium. The basic scien-
Proliferation Treaty (NPT) are at the ● Scientific collaboration has helped tific understanding of these processes is well
United Nations headquarters in New York nuclear negotiations to date established. What is needed is a truly inter-
this month for the treaty’s five-yearly review. ● Researchers must now develop national approach to the design, testing and
Their discussions are focused on the treaty’s technology to support disarmament implementation of these techniques so that all
‘grand bargain’, in which the five recognized ● This requires a new international stakeholders are confident in their use for dis-
nuclear-weapons states (China, France, Rus- advisory group, and cooperating armament verification. A central challenge is
sia, the United Kingdom and the United laboratories to study verification to develop ‘information barrier’ technologies.
States) agree to pursue negotiations towards These confirm the presence of a nuclear war-
nuclear disarmament, and those states with- disarmament. A 2007 article in The Wall Street head without disclosing sensitive details about
out nuclear weapons agree to forgo acquiring Journal by a ‘gang of four’ authors including its design, the revelation of which is prohibited
or developing them. Nuclear-weapons states former secretaries of state Henry Kissinger under the NPT. Such details may also be classi-
are now showing signs of taking their part and George Shultz, called for the United States fied by national laws.
of their bargain more seriously, to ensure to take a leadership role in reversing global Other areas for research include finding
the international cooperation necessary for reliance on nuclear weapons. This catalysed suitable and verifiable ways to dispose of fissile
tighter controls on nuclear technology, to discussion in many countries, culminating in US material, and developing ways to verify that a
resolve the cases of Iran and North Korea and President Barack Obama’s speech in Prague last nation holds no clandestine nuclear weapons,
to prevent further proliferation. year setting out a vision of a world free of nuclear material or facilities. Remote detection is a key
The scientific community must now help to weapons. Political challenges remain, including: area for further study.
develop the technology to support the process questions between, and within, nuclear-weap-
of disarmament, so that the technical ground- ons states about confidence in the verification Science without borders
work is done when multilateral negotiations of disarmament; the distrust between certain As scientists around the world become more
require it. Scientists have long played a cru- countries; and how to engage countries with interested in solving these problems, they
cial part in nuclear-arms nuclear weapons outside of can look to the United Kingdom as a leader.
control — for example, the NPT. In July 2009, the UK government laid out its
developing the technolo-
“Even at the height of the The US–Russian bilateral strategy for the NPT review conference in
gies for detecting illicit cold war, Russian and US Strategic Arms Reduc- its The Road to 2010 document. This states
underground testing of scientists worked together tion Treaty, renewed in that the country has “become a disarma-
weapons.
We see two important
on verification approaches.” Prague last month, limits
the number of deployed
ment laboratory”, pointing to the verification
research programme at the Atomic Weapons
ways to aid disarmament. nuclear warheads in these Establishment (AWE) as a key player. Set up a
First, a new international scientific advisory countries. This does not require mechanisms to decade ago, AWE has been collaborating with
group should be set up to guide international verify that nuclear warheads are dismantled and laboratories in Norway, helping each country
cooperation on disarmament research. Sec- eliminated. The same is unlikely to be true for to gain a better understanding of the trans-
ond, a network of disarmament laboratories future treaties. As the Russian and US nuclear- parency needs of a nuclear-weapon state and
with international participation should be weapons stockpiles drop to sizes comparable non-nuclear-weapon state.
established. These labs could take forward the to those of the other nuclear-weapons states, More international disarmament laboratories
recommendations of the advisory group, and agreed verification mechanisms need to be in this vein should be founded. Such labs should
ensure that nations work together to create ready, to maintain progress towards multilateral also create partnerships between governments
internationally acceptable solutions. The trust negotiations and reductions. and the non-governmental community. The
built through such international cooperation Two key challenges are building confidence UK–Norway partnership is a powerful model:
would also aid wider political negotiations. The and detecting cheating during the dismantle- it has been assisted by the non-governmental
scientific community’s well-established inter- ment process, especially when inspectors will Verification Research, Training and Infor-
national networks can reach into countries need to authenticate the presence of genuine mation Centre (VERTIC) in London. This
where political links are tense or weak, ena- warheads. Passive detection of the spontaneous partnership has been developing prototype
bling collaboration even with those countries stream of γ-rays and neutrons emitted by the information-barrier technology to identify a
outside the NPT that have nuclear weapons fissile material in nuclear weapons can be used radiological source (representing a warhead)
(India, Israel and Pakistan). to verify the presence of plutonium and deter- to a specified level of confidence. It has also
Now is a good time to start these efforts. This mine its isotopic content (to tell whether it is been developing a methodology for on-site
treaty review, in contrast to others in recent weapons grade). Active methods — that detect inspections, to provide inspectors with carefully
years, has been preceded by serious and cred- radioactive signatures emitted after interrogat- managed access to highly sensitive facilities.
ible public debate about multilateral nuclear ing fissile material with radiation — may be Renewed interest in disarmament coincides
290
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010 OPINION

United States could submit classified forms of

NoRwegiaN DefeNce ReSeaRch eSTabliShmeNT

fissile material from their nuclear weapons to
the International Atomic Energy Agency for
monitoring and verification.
There are also precedents for cooperation
at the non-governmental level. In 1980, the
US National Academy of Sciences established
a Committee on International Security and
Arms Control, and a counterpart committee
was set up at the Soviet Academy of Sciences.
Ongoing dialogue between these groups has
been credited with helping to reduce tensions
and for laying the groundwork for eventual
dialogue between presidents Ronald Reagan
and Mikhail Gorbachev. This sort of scientific
cooperation can initiate and sustain dialogue
that may not be possible at the governmental
level, and can explore contentious issues that
may impede official negotiations.
There have been calls for a high-level group
of international experts to advise governments
and develop a framework for cooperation on
the scientific and technical aspects of nuclear
disarmament. For example, in 2008, Jonas
Gahr Støre, the Norwegian foreign minister,
called for an Intergovernmental Panel on
Norwegian and UK researchers are testing verification techniques on mock nuclear weapons. Nuclear Disarmament.
Given the political sensitivities around
with discussions about the role of nuclear- based on the activities of a group of scientific disarmament, an attractive alternative might
weapons laboratories, including the US labs. experts, from more than 30 countries, in the be a non-governmental initiative, perhaps
As nuclear stockpiles shrink, the technical skills development and testing of seismic monitoring facilitated by national academies of science, to
and specialized facilities of these labs will need of nuclear-test explosions from the mid-1970s bring together relevant experts from all states
to be maintained to certify the safety, security to the early 1990s, despite a difficult political cli- with nuclear weapons. Relevant research has
and reliability of the remaining weapons. Similar mate. Although scientists continue to improve been carried out in many different countries
skills will be necessary to verify the disarmament this system’s capabilities, the technical ground- but much of it is fragmented. This group could
process, as well as helping to prevent, detect and work on verification was well established and begin by reviewing what has already been done
respond to acts of nuclear proliferation. internationally accepted by the time the CTBT and identifying what gaps remain, to integrate
The best scientific minds and latest advances came to the negotiating table in the mid-1990s. efforts into a coherent research road map.
across government, industry and academia We hope that diplomats at the NPT review
must be integrated into these efforts. For Careful steps conference will support such a group, and
example, in 2007 the Royal Society organized Even at the height of the cold war, Russian and commit to setting up and funding a network of
a workshop on innovative methods to detect US scientists worked together on common collaborative and international disarmament
nuclear materials, which has since engendered verification approaches for nuclear-arms con- laboratories to take forward its recommenda-
several effective collaborations between AWE trol, thanks in part to legal agreements, such tions. This could pay significant dividends
and academic departments. Collaboration as the Warhead Safety and Security Exchange during discussions in New York and beyond,
needs to be interdisciplinary, Agreement, that clearly by providing concrete evidence that nuclear-
drawing on the natural- and “Science can help to pave articulated the areas for col- weapons states are taking seriously their obli-
social-science communities,
and other relevant commu-
the road towards a world laboration. In contrast, in the
late 1990s, scientists from
gations to pursue disarmament. Although
nuclear disarmament and non-proliferation
nities, such as policy organ- free of nuclear weapons.” Chinese and US nuclear- remain politically sensitive, science diplomacy
izations and the military. weapons labs began to col- can help to pave the road towards a world free
This will ensure that technological solutions laborate on arms control without such a careful of nuclear weapons. ■
not only work but are politically viable. framework outlining the boundaries of discus- Martin Rees is master of Trinity College,
To some, it may seem unrealistic to propose sion. That work broke down amid claims that University of Cambridge, Cambridge CB2 1TQ,
that scientists from countries between which sensitive information had been revealed. UK, and president of the Royal Society, London
political links are strained should collaborate The option of trilateral agreements, whether SW1Y 5AG, UK; Ben Koppelman and Neil
on nuclear disarmament. However, there are involving a nation or independent organisation Davison are senior policy advisers with the Royal
precedents, including the global monitoring as the third party, can also help to smooth talks Society’s Science Policy Centre.
set up to verify compliance with the Compre- between contentious political partners. The e-mail: ben.koppelman@royalsociety.org
hensive Nuclear-Test-Ban Treaty (CTBT) once Trilateral Initiative of 1996–2002, for example, Further reading accompanies this article online at
it formally comes into force. This system is developed a system by which Russia and the go.nature.com/49kGXb.

291
© 2010 Macmillan Publishers Limited. All rights reserved

OPINION
Decentralize, adapt and cooperate
Two years ago Raphael D. Sagarin and colleagues proposed that security systems should learn
from nature. Now they’ve worked with defence professionals on putting that call into practice.
H
umankind faces a wide range of threats
to its security and safety, from ter-
rorist groups and cybercriminals to
disease pandemics and climate change. All
these threats share one characteristic: they are
constantly changing. Decision-makers can
never be sure whether the next tropical
storm will be as violent as the last, or whether
Taliban insurgents will use a roadside impro-
vised explosive device or a suicide bomber
for their next attack. Therefore, many of our
security systems — those that are resistant
to change, or that try to eliminate all risk —
are doomed. Firewalls have failed to protect
computers from hackers for 40 years; screen-
ing airline passengers for liquids didn’t prevent
Umar Abdulmutallab from taking a powdered
incendiary onto a plane; and so cumbersome
is the military procurement cycle that heavy
armoured vehicles designed to repel impro-
vised explosive attacks were deployed in Iraq
a full three years after soldiers had identified
the need.
The world needs a new way to deal with
constantly shifting threats. Two years ago we
suggested in our book Natural Security: A
Darwinian Approach to a Dangerous World
(Univ. California Press, 2008) that the best
place to look for such an approach is the
natural world, because the security issues of
modern human societies are analogous to
those of many organisms. In nature, risks
are frequent, variable and uncertain. Over
3.5 billion years, organisms have evolved an
enormous variety of methods to survive, grow
and proliferate on a continually changing
planet. The key to their success is adaptability
— the capacity to change structures, behav-
iours and interactions in response to selective
pressures.
To explore how ‘natural security’ could apply
in practice we have now worked with many
people — including emergency management
coordinators, cybersecurity experts, soldiers,
police chiefs, air marshals, homeland security
officials, fire chiefs and public-health officials.
We have identified several features of natural
systems that we believe would translate well to
human security. These are common patterns
and behaviours in natural systems that have
probably evolved independently many times
and have proved successful against a range of
threats. We have analysed many human situa-
tions that would benefit from natural security,
292
and several that already have. Although other
researchers have used ecological models to
analyse patterns of violence during conflicts1,
we believe that using natural and social-
science methods to look at the broad spectrum
of human-security concerns will lead to more
adaptable and effective defences.
Embrace uncertainty
One of our key observations is that the most
adaptable and successful organisms largely
avoid centralization. They devolve powers
of detection and responses to environmental
change to several independent sensory mecha-
nisms, such as specialized organs or clusters
of nerve cells. Octopuses, for example, use
networks of pigment cells to match the col-
our of their surroundings. Clonal organisms
such as corals distribute tasks, including feed-
ing, reproduction and defence, among clones
depending on the local need.
Certain human organizations have already
recognized the advantages of this approach.
Some of the most lethal modern terrorist
groups are loosely organized, virtually lead-
erless and capable of causing huge disrup-
tion at low material cost — the 11 September
2001 attacks cost al-Qaeda about US$400,000.
Google drives the development of many of
its products by encouraging Internet users to
test them out, and some of its products feed
off user activity. Google Flu Trends analyses
search terms to provide accurate indications
Ground squirrels deter different predators with
calls, moves and tail-heating.
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465|20 May 2010
of flu outbreaks up to two weeks before the US
Centers for Disease Control and Prevention in
Atlanta, Georgia2.
Most security measures are designed by a
small number of experts and implemented
by a central authority. But some agencies have
demonstrated that decentralization is easy and
effective. In 2004, the US Defense Advanced
Research Projects Agency funded a compe-
tition to build a driverless vehicle that could
navigate an obstacle course. Of the 15 vehicles
that entered, none finished the course, yet the
groups learned from each other and modified
their designs so that the next year all but one of
the cars that started got further than the previ-
ous year’s winner, and five completed the race.
The next race was even more successful, with
competitors negotiating a more sophisticated
urban environment and responding to traffic
laws and other vehicles.
Another feature of natural systems that
governments could adopt is the capacity
to reduce uncertainty or capitalize on it. In
nature, predators can create uncertainty by
stalking prey from concealed positions. Their
prey may reduce it by signalling the presence
of predators to others. To be effective, this
signalling must be specific to particular
threats. For example, ground squirrels make
vocal signals to deter bird and mammal preda-
tors that can hear, but switch to silent tail-flag-
ging displays to deter snakes that cannot hear,
and heat their tails when confronted by snakes
such as pit vipers that can sense infrared3. By
M. Schuyl/FlPA
contrast, individuals that make constant,
ambiguous alarm calls only increase uncer-
tainty for other members of their group, who
must waste resources determining whether the
alarm is genuine4.
The US Homeland Security’s threat advi-
sory for national and international flights
ignores this principle: it has remained at level
orange (high) since August 2006. This static,
ambiguous and nonspecific system creates
uncertainty, or indifference, among the popu-
lation that it is meant to help protect. An alter-
native approach could be to screen passengers
for irregular behaviours or facial expressions
that might betray ill intentions. This could
work against different types of threat — ter-
rorists or drug-smugglers, for example — and
return control of uncertainty to the security
services because it could be conducted from
hidden vantage points or by video. Although
NATURE|Vol 465|20 May 2010 OPINION

M. AllERuzzO/AP PhOTO
failures in our own security systems.
Translating ideas from nature into usable
security solutions is complex. It requires
sensitivity to the differences and similarities
between human societies and other evolu-
tionary systems. For example, fundamental-
ist behaviours at the core of many security
problems make more sense when viewed as
deeply rooted evolutionary biases towards
strengthening group identity against outsiders,
which cannot be easily manipulated through
material negotiations7,8. We are not propos-
ing the wholesale replacement of human
security systems with biological models.
Rather, we are arguing for a series of deliberate
interventions aimed at improving a system’s
adaptability. As with any science-based approach,
many considerations will determine how it
is applied, including ethics, politics, budgets
and value-systems.
The most potent biological analogy for
human security is the immune system, which
shifts from early, generalized responses to
more adaptive responses as pathogens become
more threatening. Encouragingly, one major
General David Petraeus’s alliances with local leaders in Iraq resulted in fewer American casualties. security organization seems to be moving
towards this approach. The US Transportation
the effectiveness of behavioural profiling is because they were mandated by government Security Administration launched a blog in
questionable, it has robust evolutionary under- or international treaty, but as local, adaptive 2008 to encourage greater interaction and
pinnings in that facial and behavioural-pattern responses to the need to protect food supplies dialogue with air travellers under the slogan,
recognition is widespread in social organisms and human health from pathogens that do not “Terrorists Evolve. Threats Evolve. Security
including humans. Moreover, studies have recognize international borders. Must Stay Ahead. You Play a Part”. Hope-
shown that people can quickly The MECIDS network fully this is not a flash in the pan, but part of a
learn to recognize facial expres- encapsulates the three essentials general acceptance that societies must adapt to
sions that betray particular “The most potent of natural security: decentral- survive and be successful. ■
emotions5. biological analogy ized organization, the flexibil- Raphael D. Sagarin, Candace S. Alcorta, Scott
Just as crucial for survival for human security is ity to adapt to uncertainty and Atran, Daniel T. Blumstein, Gregory P. Dietl,
— and relevant for human symbiotic interaction. These Michael E. Hochberg, Dominic D. P. Johnson,
security — is the capacity to the immune system.” features greatly enhance the Simon Levin, Elizabeth M. P. Madin, Joshua S.
cooperate with other organ- capacity of any of the member Madin, Elizabeth M. Prescott, Richard Sosis,
isms to exploit resources and environments. states to tackle outbreaks alone. Furthermore, Terence Taylor, John Tooby & Geerat J. Vermeij
Symbiosis is ubiquitous in nature and takes although the network was not designed to Raphael D. Sagarin is at the Institute of the
many forms. For example, blue-ringed octo- address the much more complex issue of pro- Environment, University of Arizona, Tucson,
puses have toxin-producing bacteria in their moting peace in the region, it undoubtedly spurs Arizona 85716, USA.
salivary glands, making them more formi- wider cooperation by necessitating sustained e-mail: rafe@email.arizona.edu
dable predators. This cooperation lesson was dialogue between senior officials from foreign
demonstrated in Iraq in 2007, when General affairs, security, agriculture, immigration, A full list of author affiliations accompanies this
David Petraeus’s strategy to form alliances customs and other government departments6. article online at go.nature.com/ZeOpVX.
with local leaders — including those who
1. Bohorquez, J. c., Gourley, S., Dixon, A. R., Spagat, M. &
had been hostile — resulted in more tip-offs Evolve to thrive Johnson, N. F. Nature 462, 911–914 (2009).
about improvised explosive devices and fewer Nature’s approaches to security are enormously 2. Ginsberg, J. et al. Nature 457, 1012–1014 (2009).
American casualties. diverse, and therefore worthy of more scrutiny. 3. Rundus, A. S., Owings, D. h., Joshi, S. S., chinn, E. &
Giannini, N. Proc. Natl Acad. Sci. USA 104, 14372–14376
One of us (Taylor) is involved with another To help us manage a diversity of risks, we need (2007).
example of successful symbiosis: the Middle to address why certain adaptations arise in 4. Blumstein, D. T., Verneyre, l. & Daniel, J. c. Proc. R. Soc.
East Consortium on Infectious Disease Sur- nature at particular times and places: are they Lond. B 271, 1851–1857 (2004).
5. Endres, J. & laidlaw, A. BMC Med. Educ. 9, 47
veillance (MECIDS; www.mecids.org), which the result of repeated interactions, a response (2009).
promotes collaboration between Israelis, to chronic stress or a way of coping with 6. Gresham, l., Ramlawi, A., Briski, J., Richardson, M. &
Palestinians and Jordanians to prevent the constant natural variation? Studying the Taylor, T. Biosecur. Bioterror. 7, 399–404 (2009).
spread of infectious diseases and food-borne myriad examples of apparently well-adapted 7. hochberg, M. E. Proc. R. Soc. Lond. B 271, S313–S316,
(2004).
illness. These efforts have led to networks of organisms that went extinct is another area that 8. Atran, S., Axelrod, R. & Davis, R. Science 317, 1039–1040
health professionals that have emerged not can potentially inform us about catastrophic (2007).

293
© 2010 Macmillan Publishers Limited. All rights reserved
 Vol 465|20 May 2010

BOOKS & ARTS

Science, freedom and trade
Michael Shermer enjoys two books that examine economics and politics from a scientific perspective — one
explaining the experimental basis for democracy, another placing trade in an evolutionary context.
In a 1963 lecture, Nobel-prizewinning physicist equal — was added to Thomas Jefferson’s chimp is more likely to come to the aid of the
Richard Feynman opined on the nature of original draft of the declaration by Benjamin groomer in a future conflict. But human trade
politics, arguing that the US governmental Franklin. Both men were schooled in the incorporates individual values: “Give me that
system “is new, it’s modern, and it is scientific”. axioms of Euclid’s geometry, an axiom being a which I want, and you shall have this which you
Feynman reasoned that the way in which the statement that is self-evidently true. want”, Smith wrote. That aspect of exchange
system had been designed from scratch in the Ferris’s case for freedom extends to economics. brings mutual prosperity. Trade is often unequal,
eighteenth century made it flexible enough to “There are no other human rights without the but still benefits both sides, Ridley explains.
evolve as ideas were “developed and tried out right to possess property — Ridley’s long view of trade

G. Stubbe/Minneapolis star tribune/MCt/NeWSCOM

and thrown away”. The writers of the Constitu- not just land and money, begins more than 100,000
tion, he noted, knew of the value of doubt. but intellectual property”, years ago, when human
Two new books discuss how science under- he writes. Trade in a free culture itself began “to rep-
pins democracy and trade. Timothy Ferris market cannot proceed with- licate, mutate, compete,
argues in The Science of Liberty that the experi- out ownership; you cannot select and accumulate”, as
mental nature of liberal democracies is no acci- speak and write freely if the genes had been doing for
dent, and in The Rational Optimist, Matt Ridley authorities own your phone, billions of years. Once early
examines economics in the light of evolution. computer and website. hominins started exchang-
Most science historians attribute the rise Ferris discusses the sci- “In the long run, what ing things, culture suddenly
and success of the scientific enterprise to the entific approach taken by became cumulative, and
Enlightenment values of reason, empiricism economist Adam Smith in counts is how consumers “the great headlong experi-
and anti-authoritarianism. Ferris reverses the his 1776 opus on political prosper by obtaining ment of human economic
causal vector. Most of the founding fathers were economy, An Inquiry into high-quality products at ‘progress’ began”. Speciali-
serious amateur scientists who deliberately the Nature and Causes of the zation and the division of
adopted methods of data gathering, hypothesis Wealth of Nations. Smith was low prices.” labour led to time-saving
testing and theory formation. Thomas Paine, practical, being more inter- innovations that translated
for example, was an amateur astronomer who ested in understanding human affairs as they to prosperity, which in turn led to more inno-
speculated that every star is a sun like our own, are rather than in how they ought to be; quan- vation and specialization that saved more time
with orbiting planets. Assuming that science is titative, stressing analysis whenever possible; and generated more prosperity.
universal, he believed that inhabitants of other and empirical, basing his arguments on objec- Such self-catalysing feedback explains the
worlds would discover the same natural and tive observations of the real world. For instance, growth of civilization. Before 10,000 years ago,
social laws as ours. “All the great laws of soci- Smith’s book is crammed with mentions of humans lived in small groups of hunter-gather-
ety are laws of nature,” Paine wrote in his 1791 bakers, weavers, coal miners, shipbuilders and ers, with an average annual income equivalent
treatise The Rights of Man. landlords. Such an understanding of economics, to about US$100 per person and around 300
These laws are discovered through experi- Ferris asserts, has helped to bring about the great products per village. Today, the global average
ment. Paine protested against ridiculing unsuc- increases in wealth and reductions in poverty income is around $6,000 per person per year
cessful experiments because through trial and that we have witnessed throughout history. ($40,000 in Western cities) and there are about
error comes progress. Moreover, political elec- The fact that this assertion remains true ten billion products available. Even within one
tions are scientific experi- today, even after the finan- generation, people today earn three times as
ments. “I smile to myself The Science of Liberty: Democracy, cial meltdown of 2007–09, much as their parents did in the 1950s, when
when I contemplate the Reason, and the Laws of Nature motivates Ridley to exam- adjusted for inflation.
ridiculous insignificance by Timothy Ferris ine the role of trade across Ridley compiles a substantial database of
into which literature and HarperCollins: 2010. 384 pp. human evolution. In The statistics and examples and weaves them into
all the sciences would sink, $26.99 Rational Optimist, Rid- a compelling narrative. Critics of capitalism
were they made heredi- The Rational Optimist: How ley — a science writer will undoubtedly be provoked, but they will
tary,” Paine growled, “and Prosperity Evolves and former non-executive find it hard to counter his well-crafted argu-
I carry the same idea into by Matt Ridley chairman of the troubled ments. Ridley praises the supermarket, which
governments.” HarperCollins: 2010. 448 pp. UK bank Northern Rock — lands in a town and causes a drop in its com-
The 1776 US Declara- $26.99 explains that trade is more petitors’ prices, saving its customers hundreds
tion of Independence, than the exchange of of billions of dollars a year. He does not mourn
Ferris says, is steeped in the language of science. money for goods and services: it satisfies needs. the loss of smaller stores that are driven out of
Its opening reference to “the laws of nature and When Smith attributed the wealth of a nation to business, as this must be balanced against the
of nature’s God” echoes René Descartes’ and “the propensity to truck, barter, and exchange”, benefits that customers, especially the poor-
Isaac Newton’s laws of motion and nature. The he was describing an evolved trait of our species. est, gain in terms of cheaper, more varied and
assertion that there are “self-evident” certain Even chimpanzees barter to ensure reciproc- better goods. The consumer is the one who
truths — among them that all men are created ity: if chimp A grooms chimp B, the groomed benefits, even if some producers don’t like it.
294
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
In The Wealth of Nations, Smith argued
that economies should be consumer-driven,
not producer-controlled. He exposed the old
mercantilist system — in which a nation’s pros-
perity was assumed to depend on its assets
— as mere economic tribalism and industry
protectionism. “The wealth of a country con-
sists, not of its gold and silver only, but in its
lands, houses, and consumable goods of all
A flowering of pleasure and pain
The latest collaborative artwork from neuroscientist Morten Kringelbach and artist Annie Cattrell
reveals — and revels in — sensory dialogues in the brain, explains Martin Kemp.
Why is the neuroscientist Morten Kringelbach
exhibiting in an art gallery a photograph of a
dried globe amaranth (Gomphrena globosa)
unfolding in a glass teapot on a scenic balcony?
His mysterious picture accompanies a sculpture
of brain connectivity made by Kringelbach’s col-
laborator, the Scottish artist Annie Cattrell. It
seems here as if the scientist is acting like an art-
ist and the artist like a scientist. But this framing
is wrong; our terms of reference are too crude.
For an explanation, we must look at the his-
tory of the developing collaboration between
Kringelbach and Cattrell (see Nature 424, 18;
2003). During Cattrell’s highly productive
residency at the Royal Institution in 2002, she
worked with the neuroscientist’s data to model,
in three dimensions, the zone of brain activity
that corresponds to the stimulation of each of
the five senses. She cast the shape of each sen-
sory zone in resin and mounted them in five
crystalline cubes to create the artwork Sense.
The intellectual and visual traffic in their
collaboration was not all one-way. Cattrell’s
translation of data into sculptures made some
of the neurological issues appear differently to
Kringelbach. His involvement with the making
of Sense confirmed to him that sensory identifi-
cation precedes and is not modified by hedonic
(pleasure-orientated) processing. The collabo-
ration heightened his concern with questions
about the vital contextualizing activities that
occur below the threshold of brain monitoring.
What determines how judgement, subjective
experience and decision-making occur on the
basis of sensory input?
Their new sculpture Pleasure/Pain — which
I deliberately attribute to both of them — mod-
els the structural connections of a small region
in the brainstem, the periaqueductal grey, as
revealed by a method of magnetic resonance
imaging called diffusion tensor imaging. The
piece explores the links that might be activated
during sensations of pleasure and pain. The
periaqueductal grey is a pivotal and primitive
© 2010 Macmillan Publishers Limited. All rights reserved
different kinds,” Smith wrote. Ridley agrees: in
the long run, what counts is how consumers
prosper by obtaining high-quality products at
low prices. He notes that of the people desig-
nated as ‘poor’ who live in industrialized West-
ern nations today, 99% have electricity, running
water, a refrigerator and flushing toilets.
The Rational Optimist dovetails well with
The Wealth of Nations and with The Science of
area of the brainstem that functions in ways
that are incompletely understood.
Low frequencies applied to the periaqueduc-
tal grey during deep-brain stimulation relieve
chronic pain; higher frequencies enhance the
pain. The region works with other areas of the
brain through a reciprocal call-and-response
mechanism. Oscillations of neural activity
bounce back and forth, moving at different
frequencies, some initiating pain, others mod-
erating it. As Kringelbach explains, “what is key
is that the sender and recipient neurons — like
two girls rhythmically swinging a jump rope
for a third to hop over — must be in sync.”
Cattrell created the model of this oscillating
connectivity using a rapid prototyping method
called selective laser sintering. It is a kind of
sculptural photocopying in which cross-sec-
tional layers are fused to create a solid model.
Delicate traces of the successive layers can be
detected on the surface of the sculpture, like
contour lines on a map. The resultant shape
invites analogies — an incredibly complicated
OPINION
Liberty in demonstrating that where people are
free to trade in the broadest sense — including
the free exchange of everything from genes to
ideas — prosperity ineluctably follows. ■
Michael Shermer is publisher of Skeptic magazine
and adjunct professor of economics at Claremont
Graduate University, Claremont, California 91711,
USA. He is the author of The Mind of the Market.
e-mail: mshermer@skeptic.com
P. Kidd/GV Art GAllery; M. l. KriNGelbACh
Pleasure/Pain (left) and Flowering embody
sensory pleasure in different ways.
vertebra perhaps, or a wondrous fungal growth,
or even a blossom flowering posthumously in
the scalding water of a teapot.
It is within the shared openness of the sculp-
ture and the photo that our crude premises
break down. The model of the connections of
the periaqueductal grey does not record a set
structure, but rather the potential pathways of
a dialogue — one that may blossom to induce
pleasure or surge back and forth to produce
pain. There is uncertainty both in the process
and in our understanding of it. The intricate,
visually delicate and evanescent form of the
sculpture embodies hedonic potentials and plots
aspects of the fuzzily understood.
In Flowering, the image of brewing tea exudes
a promise of sensory pleasure both in itself and
in its setting. The Sun slants across a balcony
set high above blue hills. The pleasure passes
beyond the visual, evoking the exhilarating
aroma and fragrant taste of the refreshing tea.
Cattrell and Kringelbach are telling us that
art can be as rationally founded as Renaissance
theorists insisted, and that science can be as
suggestively open as the act of looking itself. ■
Martin Kemp is emeritus professor in the history
of art at the University of Oxford, Oxford, UK.
Pleasure/Pain and Flowering are part of the show
Experiments, at GV Art in London until 29 May.
295
Vol 465|20 May 2010

NEWS & VIEWS

DRUG DISCOVERY

Priming the antimalarial pipeline

David A. Fidock
Emerging resistance to existing antimalarial drugs could nullify efforts to eliminate this deadly disease.
The discovery of thousands of agents active against malaria parasites offers hope for developing new drugs.
Malaria affects hundreds of millions of people
TABLE 1 | EXISTING ANTIMALARIAL DRUGS AND THEIR USE
worldwide, and claims hundreds of thousands
of lives every year. These grim statistics could Common name Chemical class Clinical use Resistance
rise even higher should resistance to the existing Artemisinins Sesquiterpene In artemisinin-based combination Possibly emerging
antimalarial drugs develop further. It is against (artemether, artesunate, lactone endoperoxide therapies (ACTs)
this backdrop that the papers of Gamo et al.1 dihydroartemisinin)
and Guiguemde et al.2 merit celebration — in Lumefantrine Arylamino alcohol Most common first-line No evidence
this issue, they describe the identification and antimalarial therapy in Africa, of high-level
chemical structures of vast numbers of com- in combination with artemether resistance
pounds, each with the potential to be developed Amodiaquine 4-Aminoquinoline In combination with artesunate Limited cross-
into tomorrow’s antimalarial drugs. in parts of Africa resistance with
For much of the twentieth century, malaria chloroquine
was treated with the fast-acting and inexpen- Piperaquine Bisquinoline In combination with Observed in China
sive drugs chloroquine and pyrimethamine– dihydroartemisinin in following single-
sulphadoxine. From the 1960s onwards, these parts of southeast Asia drug therapy
drugs progressively succumbed to the appear- Mefloquine 4-Methanolquinoline In combination with artesunate Prevalent in
ance and spread of resistance around the world. in parts of southeast Asia southeast Asia
By the early 1990s, malaria’s percentage con-
Pyronaridine Acridine-type Being registered for combined No cross-resistance
tribution to ‘all-cause mortality’ in African Mannich base use with artesunate with other drugs
infants was climbing3, in some areas account- reported
ing for nearly 50% of deaths. These facts are
Quinine/quinidine 4-Methanolquinoline Mainly for treating severe malaria, Exists at a low level
attributable almost exclusively to Plasmodium often with antibiotics
falciparum, one of the five Plasmodium species
that cause malaria in humans. Atovaquone Naphthoquinone In combination with proguanil Has been observed
(a biguanide) for treatment clinically
Currently, the only fully effective class of or prevention
antimalarial drug is the artemisinins. These are
Chloroquine 4-Aminoquinoline Former first-line treatment for Widespread
remarkably potent against the asexual blood
uncomplicated malaria
stage of the parasite’s life cycle, during which
it replicates inside human red blood cells and Pyrimethamine Diaminopyrimidine For intermittent preventive Widespread
causes disease. Artemisinin-based combina- treatment, combined with
sulphadoxine (a sulphonamide)
tion therapies (ACTs) have now been adopted
globally as the first line of treatment. Together Primaquine 8-Aminoquinoline For eliminating liver-stage Unknown
with mosquito-control measures such as insec- parasites, including dormant
forms of Plasmodium vivax
ticide-treated bednets, ACTs have driven down
malaria rates across sub-Saharan Africa, where
the disease exerts the greatest burden4. Gamo et al.1 (page 305) and Guiguemde et al.2 of the 172 compounds the researchers then
But despite intensifying elimination cam- (page 311) will soon produce drugs that will tip scrutinized more closely had novel chemical
paigns5, the prospect of resistance remains. this precarious situation in our favour. structures, and at least 80% of these seemed
Delayed rates of parasite clearance after admin- Guiguemde et al. report an extraordinary to act on parasite targets distinct from those
istration of ACTs are already evident near the multidisciplinary effort. They began by screen- affected by the current drugs.
Thai–Cambodian border6 — a hotbed of multi- ing for compounds that inhibit the growth of Strains of P. falciparum that are resistant
drug resistance. Malaria control is therefore at asexual blood-stage P. falciparum in cultured to the existing antimalarial drugs showed
a pivotal stage. Continued efficacy of ACTs and red blood cells. The team used a chemically minimal cross-resistance to the shortlisted
their sustained use, along with mosquito-vector diverse library of almost 310,000 compounds, compounds2. Furthermore, two classes of the
control programmes, could well produce a selected to maximize diversity while retaining chemicals worked in synergy with ACTs — a
remarkable global-health triumph. Alterna- clusters of related agents to explore structure– favourable attribute given the current focus
tively, resistance could wipe out artemisinins, activity relationships. More than 1,100 com- on drug combinations to reduce the risk of
and malaria could resurge to former levels of pounds met the authors’ criterion of inhibiting resistance emerging rapidly.
virulence, causing millions of deaths per year. parasite growth by 80% when tested at a con- Guiguemde and colleagues also investigated
One hopes that the compounds described by centration of 7 micromolar. Promisingly, most their compounds’ effect on other pathogenic
297
© 2010 Macmillan Publishers Limited. All rights reserved
NEWS & VIEWS
parasites (Toxoplasma, Leishmania and
trypanosomes) and on replicating human cell
lines, and found that most of the compounds
were highly selective for Plasmodium. Prelimi-
nary pharmacokinetic analyses — investiga-
tions related to how a compound is absorbed,
distributed, metabolized and eliminated by the
body — suggested that many compounds were
generally suitable for further development.
As a proof of principle, the authors showed
one compound to be efficacious in treating
malaria in a mouse model, albeit at concentra-
tions 25-fold higher than the effective dose of
chloroquine in the same animal model.
Gamo et al. 1 used GlaxoSmithKline’s
in-house chemical library to screen almost
2 million compounds against asexual blood-
stage P. falciparum. Setting a similar threshold
of greater than 80% growth inhibition, in this
case at a cut-off of 2 micromolar, the authors
identified more than 13,500 active compounds.
Of these, 8,000 were equally active against
multidrug-resistant P. falciparum parasites, and
fewer than 2,000 displayed some non-selective
activity against a human liver-cancer cell line.
Gamo and colleagues confirmed the relevance
of their compounds by identifying among
them several representatives of all existing anti-
malarial drug classes (Table 1), with the excep-
tion of the artemisinins, which were known to
be absent from the starting library. Signifi-
cantly, more than 11,000 of the new hits were
previously proprietary to GlaxoSmithKline
and are now made accessible to the general
research community for the first time.
On the basis of their in-house annotations
of candidate human or microbial targets for
more than 4,200 compounds, Gamo et al. pre-
dict that many active compounds might target
kinase enzymes of P. falciparum. If confirmed,
this would constitute an important new direc-
tion for antimalarial drug development — one
that might cross paths with researchers exploit-
ing the vast chemical repositories developed
to target kinases in other disorders, including
solid-tumour cancers, inflammation, arthritis,
diabetes and cardiovascular disease.
Gamo and colleagues1 do not go as far as
Guiguemde et al.2 in experimentally deter-
mining potential parasite targets or reporting
preliminary pharmacokinetic or pharmaco-
dynamic data — this would be a huge task
for so many compounds. As such, their study
provides only starting points to test future
hypotheses concerning drug action. Nonetheless,
it is momentous that a large pharmaceutical
company has made its antimalarial drug-
discovery data, including chemical structures,
freely available. These data are fully search-
able through the European Bioinformatics
Institute’s ChEMBL database7.
Neither group1,2 claims to have discov-
ered the next antimalarial drug. Instead,
they provide a remarkable diversity of novel
chemical structures on which to base new
antimalarial drug-discovery and development
campaigns. Comparison of these biologically
298
active compounds — and an earlier, partially
described set8 identified in a high-throughput
screen against P. falciparum — should be a first
step. Compounds that prove to be potent in
rodent models, cheap to synthesize, safe and
unaffected by existing mechanisms of resist-
ance should also be evaluated for activity
against other stages of the Plasmodium life
cycle. These include the sexual blood stage
responsible for transmission to the mosquito
vectors, and the asymptomatic liver stage that
precedes blood-stage infection. Finally, activity
must also be assessed against Plasmodium vivax
— a species that, outside Africa, causes more
cases of malaria than P. falciparum and can,
according to a clinical study in Indonesia9,
cause severe, and at times fatal, disease.
Innovative efforts by many organizations —
notably the Medicines for Malaria Venture in
Geneva, Switzerland — have in recent years
greatly accelerated the development and licens-
ing of new antimalarial drugs10. But the discov-
ery pipeline remains woefully thin, and there
are precious few alternatives to artemisinins.
BIOmatERIalS
Intelligent glue
Haeshin Lee
Spiders’ webs are coated with microscopic droplets of glue, but the
properties of this adhesive were unclear. It has now been found that the
glue’s stretchiness underpins its role in catching flies.
Man-made glues are mono-functional — their
material properties are designed to stick one
thing to another, and that’s it. But in Nature
Communications, Sahni et al.1 report that the
‘glue’ droplets that coat spiders’ webs are multi-
functional. Depending on the rate at which they
are extended, the droplets act either as a viscous
adhesive or as a rubber-like elastic solid.
Adhesive coatings are found everywhere.
Think of the paint that covers walls, cars and
ships’ hulls; metals, such as gold and silver,
plated on jewellery; and dyes that change the
colour of fabrics and hair. These coatings have
a variety of functions — paint prevents barna-
cles and mussels from attaching themselves to
ships’ hulls, reducing drag on the ship and so
improving fuel efficiency, whereas gold- and
silver-plated surfaces are anti-corrosive.
Yet adhesive materials in nature do much
more. Mussels, for example, secrete a substance
that forms into ‘byssus threads’, by which they
attach themselves to rocks, shells and even to
feathers and fish skin (Fig. 1a). The threads
must have a high degree of elasticity, or they
would snap under the physical impact of lash-
ing tides. But they need the additional prop-
erty of adhesiveness, provided by a coating of
a glue-like material, to anchor the mussels in
place. What’s more, the threads must maintain
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
These reports1,2 offer tremendous opportunities
to develop the next generation of antimalarial
drugs. They also sound a call for the academic
and pharmaceutical sectors to rise to the chal-
lenge. This should include a concerted chemical-
genomics effort to identify the most appropriate
targets in the parasite. Time is of the essence. ■
David A. Fidock is in the Departments of
Microbiology and Immunology, and of Medicine
(Division of Infectious Diseases), Columbia
University Medical Center, New York,
New York 10032, USA.
e-mail: df2260@columbia.edu
1. Gamo, F.-J. et al. Nature 465, 305–310 (2010).
2. Guiguemde, W. A. et al. Nature 465, 311–315 (2010).
3. Snow, R. W., Trape, J.-F. & Marsh, K. Trends Parasitol. 17,
593–597 (2001).
4. Eastman, R. T. & Fidock, D. A. Nature Rev. Microbiol. 7,
864–874 (2009).
5. Feachem, R. & Sabot, O. Lancet 371, 1633–1635 (2008).
6. Dondorp, A. M. et al. N. Engl. J. Med. 361, 455–467 (2009).
7. www.ebi.ac.uk/chemblntd
8. Plouffe, D. et al. Proc. Natl Acad. Sci. USA 105, 9059–9064
(2008).
9. Tjitra, E. et al. PLoS Med. 5, e128 (2008).
10. Wells, T. N. C., Alonso, P. L. & Gutteridge, W. E. Nature Rev.
Drug Discov. 8, 879–891 (2009).
their grip under water, an environment in which
most adhesives function poorly.
Sahni and colleagues’ study1 of glue droplets
on spiders’ webs suggests that the coupling of
adhesion with extension is a common design
principle of natural adhesives. The droplets
consist of a complex mixture of glycoproteins
along with a variety of viscous small molecules
and salts. The role of the components of the
droplets has been difficult to prove, not least
because it is difficult to separate their properties
from those of the underlying spider silk.
The authors overcame this problem by
immobilizing silk threads on a glass surface,
and touching glue droplets to the threads with
a tiny glass probe. They then pulled the probe
away at a constant speed, and measured the
force exerted on the probe’s tip as a function
of distance from the droplet, until contact
between the droplet and the probe was broken.
They found that the force–distance response
depended on the speed with which the probe
was pulled away: rapid extensions of the glue
caused it to become highly viscous, whereas
slow extensions turned the glue into an elastic-
solid-like material.
This behaviour correlates well with the two
functions that the droplets perform in nature:
capturing prey and then retaining it. When
NATURE|Vol 465|20 May 2010
prey — typically a flying insect — is captured in
a web, the silk rapidly extends on impact. Sahni
and colleagues’ study reveals that the glue drop-
lets would become highly viscous under these
conditions, providing maximum adhesion to
effectively capture the spider’s meal.
But after the prey has been captured, the
movement produced by its attempts to escape
causes a slow extension of the web. In this
scenario, the glue droplets turn into a rubber-
band-like material to prevent the unfortunate
prey from escaping (Fig. 1b). The glue drop-
lets secreted by spiders therefore constitute an
‘intelligent’ adhesive whose properties change
significantly depending on the extension rate
of the underlying silk. Needless to say, human
technology has not been successful in producing
such a smart adhesive material.
So what is the molecular basis of the
adhesion–extension properties of naturally
occurring glues? In the case of the adhe-
sive used by mussels, these properties derive
from the presence in the glue molecules of a
repeating pair of amino acids: lysine, one of
the common amino acids, and 3,4-dihydroxy-
l-phenylalanine (l-DOPA)2, which is more
unusual. Molecules that contain extensively
repeating l-DOPA–lysine motifs stick to a wide
range of surfaces by forming various kinds of
covalent and non-covalent bonds3. Meanwhile,
metal ions in sea water infiltrate the adhesive
threads, interconnecting scaffold proteins
covalently and/or by coordination4 (a process
in which molecules bind non-covalently to
metal ions). This intermolecular crosslinking
DEVElOPmENtal BIOlOGY
Roots respond to an inner calling
Ben Scheres
In plant roots, patterning of two types of water-conducting xylem tissue is
determined by a signalling system that involves the reciprocal dance of a
mobile transcription factor and mobile microRNAs.
Reciprocal signalling over tissue boundaries is
a well-known developmental feature in fruitfly
embryos. Surprisingly, a similar mechanism
seems to apply in the plant root, as described
by Carlsbecker and colleagues on page 316 of
this issue1.
In fruitflies, the anterior and posterior
parasegments, pre-specified by transcrip-
tion factors, communicate and stabilize their
boundaries using short-range signals carried
by Hedgehog and Wingless — small secreted
proteins originating from one parasegment
and perceived by the neighbouring one.
These signals help to consolidate their own
expression at boundaries created by earlier
patterning systems. Moreover, they create new
short-range patterns with the boundary as an
‘organizing centre’2. This strategy makes sense
© 2010 Macmillan Publishers Limited. All rights reserved
a glycoproteins are responsible for the sticki-
b
Figure 1 | Extensible and adhesive. a, Mussels
secrete underwater adhesive threads that are
also extensible to resist pummelling by the
tide. b, Sahni et al.1 report that the microscopic
droplets of ‘glue’ that coat spider silk are also
extensible and viscoelastic.
confers extensibility on the adhesive threads.
But in-depth knowledge of the molecular
structure of the glue droplets on spiders’ webs
is lacking. It has been reported that the droplets
contain small molecules such as neurotrans-
mitters, amino acids and peptides5, as well as
macromolecules such as glycoproteins6. There
in the context of groups of cells that can move
relative to one another, and that need to trans-
late positional information encoded in a cell
nucleus into cellular positional information —
as in the fly embryo.
Carlsbecker et al.1 show that reciprocal
signalling operates in the root of the model
plant Arabidopsis thaliana, in which, in con-
trast to animals, cell positions are stable. The
authors reveal how a mobile transcription
factor and mobile microRNAs perform an
intimate reciprocal dance to consolidate cell
identities at the boundary between the inner
vascular tissue of the plant and the surrounding
cell layers.
The vascular tissue of plants consists of
branched networks in leaves and separated
strands in stems. But in the root, the tissue
NEWS & VIEWS
is evidence to suggest that the sugars found in
P. KAY/PHOTOLIBRARY.COM
ness of glues secreted by other organisms. For
example, the sugar N-acetyl-d-glucosamine is
important for the adhesion of the holdfast of the
bacterium Caulobacter crescentus7. The molec-
ular components that cause spider-glue drop-
lets to form a rubbery solid remain unknown.
A series of studies investigating the molec-
ular content and supramolecular assembly
of glue droplets is therefore required.
M. LUSTBADER/SPLI
Sahni and colleagues’ studies1 are just the tip
of an enormous iceberg, as there are many other
interesting biomaterials that could be investi-
gated by curious scientists — such as the adhe-
sive flavonoid compounds found in plants, or
the viscoelastic biofilms produced by microbes.
Future investigations will no doubt reveal their
secrets, and provide inspiration for the design
of advanced synthetic materials. ■
Haeshin Lee is in the Department of Chemistry,
Graduate School of Nanoscience and Technology
(WCU), and the Molecular-level Interface Research
Center, Korea Advanced Institute of Science and
Technology, Daejeon 305-701, South Korea.
e-mail: haeshin@kaist.ac.kr
1. Sahni, V., Blackledge, T. A. & Dhinojwala, A. Nature
Commun. 1, doi:10.1038/ncomms1019 (2010).
2. Waite, J. H. & Tanzer, M. L. Science 212, 1038–1040 (1981).
3. Lee, H. et al. Science 318, 426–430 (2007).
4. Waite, J. H., Vaccaro, E., Sun, C. & Lucas, J. M. Phil. Trans. R.
Soc. Lond. B 357, 143–153 (2002).
5. Vollrath, F. et al. Nature 345, 526–527 (1990).
6. Choresh, O., Bayarmagnai, B. & Lewis R. V.
Biomacromolecules 10, 2852–2856 (2009).
7. Tsang, P. H., Li, G., Brun, Y. V., Freund, L. B. & Tang, J. X.
Proc. Natl Acad. Sci. USA 103, 5764–5768 (2006).
takes on the relatively simple organization of
one central cylinder containing water-conduct-
ing xylem elements arranged in arches, which
separate the nutrient-conducting phloem
elements. The narrow Arabidopsis root has
only a single xylem axis in which the conduct-
ing cells specialize into central metaxylem
(with pitted cell walls) and peripheral pro-
toxylem (with spiral cell walls). It was shown
previously that the transcription factor SHORT
ROOT (SHR), which is expressed in the cen-
tral cylinder but is unable to efficiently influ-
ence nuclear activity there, travels from cell
to cell, ultimately leaving the central cylinder
and ending up in the surrounding cell layer3.
There, a system that uses the physical partner
and downstream target of SHR, SCARECROW
(SCR), as well as other proteins, leads to effi-
cient nuclear uptake of SHR and activation of a
transcriptional pathway that specifies this cell
layer as endodermis4–6 (Fig. 1a).
Carlsbecker et al.1 show that this act of inside-
out signalling sets up a signalling pathway tar-
geted outside-in. SHR binds directly to and
activates a set of microRNA genes, with the aid
of its helper protein SCR, resulting in endoder-
mis-specific production of these small RNAs.
Recently, the mobility of a member of an unre-
lated class of small RNA, known as tasiRNAs,
299
NEWS & VIEWS NATURE|Vol 465|20 May 2010

embryos, the ability to respond to the signals

a SHR protein spread b miR165/6 spread c PHB distribution and action is confined to different sides of the boundary
Protoxylem
by differential regulation of the signal-response
Endodermis Metaxylem pathways, determined by pre-patterning
factors. In Arabidopsis roots, the response
pathway to SHR is constrained by a capture
mechanism that is present only in the outer
cell layers5,6. In turn, the response pathway to
the microRNAs is confined to the inner layers
because this is the expression domain of the
PHB target and its family members. Again,
pre-patterns select the domain of response to be
confined to one of two tissue compartments.
In the case of flies, the resulting distribution
of signalling molecules can lead to morpho-
genetic gradients that instruct several cell-fate
decisions. The mutant analyses presented
Figure 1 | Reciprocal signalling across a tissue boundary. This side view of concentric tissue layers of by Carlsbecker et al.1 indicate that PHB and
a root depicts a two-way signalling mechanism responsible for tissue patterning. a, SHORT ROOT related genes are not only required for a binary
(SHR) transcription factor (red) is synthesized in the central vascular tissue. It moves across the choice between xylem fates, but have a broader
boundary, depicted by thicker lines, to the next layer, where it enters the cell nuclei (red dots) and role in the specification of vascular cell identi-
specifies endodermis identity. b, c, Summary of the results of Carlsbecker and colleagues1. b, Assisted ties. Future research should nail down whether
by other proteins, the nuclear SHR activates the transcription of microRNAs (dark green), specifically
the endodermis-derived microRNAs regulate
miR165/6, and their effect spreads, as shown in light green. c, In the vascular tissue, messenger RNA of
the PHABULOSA (PHB) transcription factor is downregulated by the microRNAs in a dose-dependent other aspects of vascular differentiation by
manner: low levels of PHB specify metaxylem (dark blue), and high levels specify protoxylem (light blue). acting on PHB and related factors through a
gradient distribution. Such work may estab-
lish whether, in plants, microRNAs can act as
has been implicated in establishing polar- fashion, specifying central metaxylem at low bona fide morphogen gradients. ■
ity during leaf development, providing a levels and peripheral protoxylem at high levels. Ben Scheres is in the Molecular Genetics Group,
precedent for the involvement of at least one This is a neat demonstration of the deployment Department of Biology, Utrecht University,
class of mobile small RNA in plant pattern of a two-way signalling system for specifying 3584 CH Utrecht, the Netherlands.
formation7. cell type (Fig. 1c). e-mail: b.scheres@uu.nl
Carlsbecker et al. demonstrate that the A theoretical advantage of two-way signal-
endodermis-expressed microRNAs act in ling systems is that they allow signals to 1. Carlsbecker, A. et al. Nature 465, 316–321 (2010).
2. Gritzan, U., Hatini, V. & DiNardio, S. Development 126,
cells inside this layer to repress the messenger pattern cell types with different affinities across 4107–4115 (1999).
RNA levels of the PHABULOSA (PHB) tran- boundaries, hence creating signal gradients 3. Nakajima, K., Sena, G., Nawy, T. & Benfey, P. N. Nature 413,
scription factor (Fig. 1b). In a series of elegant with a peak at one end. This ‘broken symme- 307–311 (2001).
genetic experiments using mutant plants, they try’ around the signalling source is caused by 4. Di Laurenzio, L. et al. Cell 86, 423–433 (1996).
5. Cui, H. et al. Science 316, 421–425 (2007).
show that PHB and its family members direct the different pre-patterns around the boundary. 6. Welch, D. et al. Genes Dev. 21, 2196–2204 (2007).
xylem development in a dose-dependent In the case of Wingless and Hedgehog in fly 7. Chitwood, D. et al. Genes Dev. 23, 549–554 (2009).

EXtRaSOlaR PlaNEtS heating, a gas giant of an estimated age of a

larger than they ought to be

few billion years should have lost most of its
primordial heat, shrunk to a smaller size owing
to its own gravity — and so have become much
denser — than observations indicate2. There-
Pin-Gao Gu fore, one way to account for hot Jupiters that
The finding that some gas-giant exoplanets are much larger than theory display inflated radii, and hence lower densi-
ties, is to identify a source of internal heating
predicts has been boggling astronomers’ minds. Planetary heating caused that can retard planetary contraction, or even
by gravitational tidal interactions might be a piece of the puzzle. maintain the bloated size of the planets, against
radiative cooling. The obvious ‘usual suspect’
In the past few years, the transit method of Several of these hot Jupiters have much larger is tidal heating, which arises from the tidal dis-
planet detection has emerged as one of the radii than would be expected for planets of tortion exerted on the planets by their nearby
prevailing techniques to search for exoplanets. their age on the basis of planetary evolution host stars.
The tiny amount of stellar light blocked by a theory. Writing in the Astrophysical Journal, The idea is reminiscent of the Galilean satel-
planet when it crosses (transits) in front of its Ibgui et al.1 investigate the hypothesis of tidal lite system of Jupiter. The orbital eccentricity of
host star provides information about the size heating as an explanation for these ‘bloated’ Io, the innermost moon of Jupiter, is pumped
of the planet. A number of planetary transit planets. up by gravitational interactions with its neigh-
surveys have revealed that most of these exo- If all hot Jupiters were placed in a virtual bouring satellites. As a result, Io is subject to
planets belong to the special category of ‘hot gigantic tank filled with water, many of them tidal forces that constantly change its shape.
Jupiters’, so named because, in contrast to would be ‘light’ enough to float in the water This process dissipates Io’s orbital energy and
Jupiter, these gas-giant planets orbit their stars (Fig. 1). This is surprising, because gas giants turns it into internal heat, triggering the moon’s
just a few stellar radii away from them, and cool by radiative loss as they age. Conse- renowned volcanism — Io spews out more lava
so are scorched by the stars’ intense radiation. quently, and in the absence of any internal than any other object in the Solar System. This
300
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
hypothesis of tidal heating caused by orbital
eccentricity was first applied to the first tran-
siting hot Jupiter discovered, HD 209458b, by
Bodenheimer and colleagues3 to explain the
planet’s bloated size. Although its orbit was
later found4 to be circular, astronomers have
never abandoned the idea.
In their study, Ibgui and colleagues 1
revisited the tidal-heating hypothesis as an
explanation for the observed inflated radii
of transiting hot Jupiters. They developed
an evolutionary model that handles both the
interior structure of a hot Jupiter and its planet-
ary atmosphere — Bodenheimer et al.3 did not
consider the latter. Knowing that uncertainties
in current observations of hot Jupiters may
admit non-zero (albeit small) orbital eccentric-
ities, and so allow for a possible source of tidal
heating, the authors1 assumed that a hot Jupiter
attains thermal balance between radiative cool-
ing from the planet’s surface and tidal heating
deposited deep in the planet’s interior. They
then calculated the rate of tidal heating neces-
sary to fit the observed radii of six transiting
hot Jupiters, and explored how it depends on
unknown parameters such as orbital eccentric-
ity, tidal-heating efficiency, atmospheric opac-
ity and the planet’s core mass. They showed
that orbital eccentricity and high tidal-heating
efficiency help to enlarge the radius, whereas
atmospheric opacity serves as a thermal
blanket — it retains internal energy and slows
down planetary contraction. By contrast, a
more massive core makes the radius smaller.
Among the six planets studied, TrES-4b is
the one that is most difficult to explain by their
model, because it requires much higher tidal
efficiency and/or higher atmospheric opac-
ity than standard theoretical values. Another
planet that might be difficult to reconcile with
the authors’ model is WASP-12b, because
recent measurements5 suggest a smaller eccen-
tricity (consistent with zero) than that used by
Ibgui and colleagues. However, alternative
models have long been proposed that explain
a bloated planet even in a circular orbit2.
Because the tidal heating occurs at the cost
of making a planet’s elliptical orbit circular, a
process such as the gravitational interaction
with a planetary companion is required that
continually sustains the planet’s orbital eccen-
tricity and so keeps the rate of interior heating
constant. Ibgui et al. suggest that this hypoth-
esis could be tested with accurate dynami-
cal measurements of the orbital eccentricity.
Another possible test would be to investigate
whether there exist any variations in the dura-
tion of the planetary transit and in the time
interval between successive transit events6. A
positive result may indicate the presence of a
planetary companion.
As Ibgui et al. themselves have cautioned,
one of their model’s caveats is that the tidal-
heating-efficiency parameter remains poorly
understood. Whereas the orbits of most hot
Jupiters are consistent with their being circu-
lar, the orbits of some hot Jupiters as old as a
© 2010 Macmillan Publishers Limited. All rights reserved
NEWS & VIEWS
TrES–4b (0.2) WASP–12b (0.3) WASP–15b (0.2)
WASP–4b (0.5) WASP–6b (0.4) HAT–P–12b (0.3)
Jupiter (1.3)
Figure 1 | Planets much ‘lighter’ than water. If a gigantic water tank existed, and Jupiter and the six
‘hot Jupiters’ studied by Ibgui and colleagues1 were placed in it, Jupiter would be the only planet
that would sink: in contrast to Jupiter, all six hot Jupiters have mass densities (shown in g cm−3 ) that
are smaller than that of water (1 g cm−3). The diagram illustrates the point that many hot Jupiters are
much larger, and hence less dense, than planetary evolution theory predicts. Note, however, that not
all ‘floating’ planets are abnormally large, for example HAT-P-12b (ref. 1). The relative sizes of the
planets are drawn to scale.
few billion years were found to be elliptical7, the next few years, astronomers will continue
perhaps suggesting the diversity of the tidal- to test the tidal-heating model and will find
heating efficiency. When a gas giant is tidally out whether the size anomaly is unique to hot
disturbed, low-frequency waves can be pro- Jupiters or instead is common to all gas-giant
duced that propagate within the convective part exoplanets. ■
of the planet and dissipate into heat at a rate Pin-Gao Gu is at the Institute of Astronomy and
that depends on the planet’s interior structure Astrophysics, Academia Sinica, Taipei 10617,
and rotation rate, as well as the frequency of Taiwan.
the tidal oscillation8. It would be interesting to e-mail: gu@asiaa.sinica.edu.tw
incorporate this mechanism into Ibgui and col-
leagues’ model, as well as to relax the thermal- 1. Ibgui, L., Burrows, A. & Spiegel, D. S. Astrophys. J. 713,
751–763 (2010).
balance assumption, to study how it affects the 2. Baraffe, I., Chabrier, G. & Barman, T. Rep. Prog. Phys. 73,
relationships among unknown parameters. 1–30 (2009).
In addition to ground-based observations of 3. Bodenheimer, P., Lin, D. N. C. & Mardling, R. A. Astrophys. J.
548, 466–472 (2001).
planetary transits, the CoRoT and Kepler space 4. Deming, D., Seager, S., Richardson, L. J. & Harrington, J.
telescopes are expected to detect and measure Nature 434, 740–743 (2005).
the radii9,10 of gas-giant exoplanets that are more 5. Husnoo, N. et al. Preprint at http://arxiv.org/
abs/1004.1809 (2010).
distant from their host stars than hot Jupiters 6. Holman, M. J. & Murray, N. W. Science 307, 1288–1291
are from theirs. Tidal heating should play little (2005).
part in explaining the radii of these not-so-hot 7. http://exoplanet.eu
gas-giant exoplanets; such observations should 8. Ogilvie, G. I. & Lin, D. N. C. Astrophys. J. 610, 477–509
(2004).
therefore provide valuable complementary 9. Deeg, H. J. et al. Nature 464, 384–387 (2010).
insights into the nature of bloated planets. In 10. http://kepler.nasa.gov
DNa REPaIR
Decision at the break point
Simon J. Boulton
Many decisions affect the fate of damaged DNA — for example, how to
repair the damage, or whether to repair it at all and instead let the damaged
cell die. An intricate web of molecular interactions affects such decisions.
If repair is the desired course of action, a cell show how the protein 53BP1 affects this choice
that contains a DNA double-strand break has in cells lacking the tumour-suppressor protein
two options: repair it using either the process BRCA1. The decision in turn affects a bigger
of homologous recombination or alternative, picture: the development of cancer.
error-prone pathways. Reporting in Cell and BRCA1 and BRCA2 are tumour-suppressor
in Nature Structural and Molecular Biology, genes that are closely linked to the develop-
Bunting et al.1 and Bouwman and colleagues2 ment of hereditary breast and ovarian cancer.
301
NEWS & VIEWS NATURE|Vol 465|20 May 2010

Mutations in either gene give a lifetime end joining, predominates. This leads to
risk of around 70% of developing the chromosomal abnormalities that are the
cancers. Both proteins maintain the stabil- hallmarks of Brca1 deficiency. At DSBs,
ity of the genome by performing crucial, a Wild type BRCA1 and 53BP1 probably have opposing
yet distinct, roles in the cellular response DSB roles in promoting the choice of the repair
BRCA1
to DNA damage. BRCA2 is essential for pathway: BRCA1 promotes homologous
the accurate repair of DNA double-strand ? Processing
HR Accurate recombination by facilitating resection,
breaks (DSBs) by homologous recombina- DSB repair whereas 53BP1 promotes other pathways
tion, aiding the recruitment and loading of 53BP1 of repair by blocking resection.
the RAD51 repair protein onto processed How BRCA1 and 53BP1 exert their
DSBs. BRCA1, which is also essential for b Brca1–/– effects on resection is unclear. BRCA1 has
the repair of DSBs by homologous recom- ubiquitin ligase enzymatic activity, which
bination, is further required to signal the BRCA1 attaches a ubiquitin molecule to its target.
presence of DSBs to the cell cycle, and thus So it is intriguing that it ubiquitylates CtIP,
acts as a surveillance factor, monitoring Processing NHEJ Error-prone a protein required for DSB resection7. Per-
3 DSB repair
genome integrity . Despite considerable haps ubiquitylation stimulates CtIP-medi-
research, however, the precise function of 53BP1 ated resection. Conversely, attachment of
Chromosomal
BRCA1 in the cellular response to DSBs rearrangements the ubiquitin-related protein SUMO to
and how it suppresses tumour formation 53BP1 by PIAS proteins8,9 may promote
have remained unclear. Cancer its displacement from DSBs, releasing the
A testament to the importance of c Brca1–/– 53bp1–/– barrier to resection.
BRCA1 to genome integrity is the fact Given that 53bp1 loss significantly
that Brca1∆11/∆11 mice, which express a BRCA1 reduces tumour formation in Brca1-
truncated form of this protein, die during deficient mice6, Bunting et al.1 propose
Processing HR Accurate
embryonic development4. Deletion of a DSB repair
that 53BP1 inhibitors could be used to
single copy of the p53 tumour-suppressor treat human carriers of Brca1 muta-
53BP1
gene is sufficient to rescue Brca1∆11/∆11 Cancer tions, in the hope of suppressing tumour
5
mice from embryonic death . Neverthe- formation and enhancing repair medi-
less, the resulting mice age prematurely ated by homologous recombination. It is
and are extremely cancer prone. Dele- important to note, however, that Brca1-
tion of the p53-binding protein 53BP1, Figure 1 | Choosing the correct repair pathway for DNa deficient human tumours frequently
double-strand breaks (DSBs). a, In wild-type cells, BRCA1
which mediates DNA-damage responses, lack p53, or are incapable of activating it.
promotes active DSB resection, overcoming the inhibitory
can also prevent the embryonic death of effect of 53BP1 on this process. The outcome is accurate Inhibition of 53BP1 in this context may
Brca1∆11/∆11 mice. But, strikingly, mice repair by homologous recombination (HR). b, In therefore have the undesirable effect of
with mutant BRCA1 and lacking 53BP1 Brca1-deficient cells, 53BP1 inhibits DSB processing, restoring homologous recombination in
(Brca1∆11/∆1153bp1−/−) age relatively nor- leading to error-prone repair by pathways such as tumour cells, leading to their resistance to
mally and are not prone to developing non-homologous end joining (NHEJ), and therefore to therapies involving PARP inhibition and
cancer6. chromosomal rearrangements and a predisposition to DNA-damaging agents.
1,2
Bunting et al.1 and Bouwman et al.2 cancer. c, Two new studies show that elimination of Consistent with this notion, Bouwman
investigate how, at a molecular level, dis- 53BP1 in Brca1-deficient cells relieves the inhibition of DSB et al.2 describe a remarkable correlation
ruption of 53bp1 overcomes the effects processing. Consequently, accurate repair by homologous between the absence of 53BP1 and the
recombination is partially restored, suppressing
of Brca1 deficiency. Both groups report susceptibility to cancer. most aggressive and difficult-to-treat
that, when 53bp1 (but not p53) is deleted, ‘triple negative’ breast tumours, which lack
the sensitivity of Brca1-deficient cells to the proteins ER, PR and HER2. Further-
various DNA-damaging agents and to inhibi- recombination is restored to 10–50% of the more, 53BP1 loss is frequently associated with
tors of the DNA-repair protein PARP1 can be levels in wild-type cells. In the absence of tumours carrying BRCA1 and BRCA2 muta-
reversed. Moreover, deleting 53bp1 alleviates Brca1, therefore, deleting 53bp1 at least par- tions1. So, although many intricacies must be
damage-induced chromosomal abnormali- tially overcomes defects in homologous taken into account, at the very least these stud-
ties that arise in Brca1-deficient cells. It also recombination. ies clarify one key point: the choice of DSB-
abolishes the associated increased levels of The initiation of repair by homologous repair pathway is a decisive factor in tumour
spontaneous DNA damage in these cells, as well recombination requires DNA resection formation. ■
as the resulting ‘checkpoint response’ and cell- (processing of the DSB to produce a single- Simon J. Boulton is at the DNA Damage Response
cycle arrest. What’s more, eliminating 53bp1 in stranded DNA overhang at one end). This is Laboratory, London Research Institute,
Brca1-deficient cells restores recruitment of the compromised in Brca1-deficient cells (Fig. 1). Cancer Research UK, Clare Hall, South Mimms
homologous-recombination protein RAD51 to Bunting et al. investigated whether the absence EN6 3LD, UK.
DSBs1,2. By contrast, 53bp1 deletion in Brca2- of 53BP1 changes the situation. Indeed, they e-mail: simon.boulton@cancer.org.uk
deficient cells or in cells lacking XRCC2, found that DNA processing at DSBs was
another protein involved in homologous restored in Brca1∆11/∆1153bp1–/– cells. Moreover, 1. Bunting, S. F. et al. Cell 141, 243–254 (2010).
2. Bouwman, P. et al. Nature Struct. Mol. Biol. doi:10.1038/
recombination, does not have this effect. resection in Brca1∆11/∆1153bp1–/– cells required nsmb.1831 (2010).
Is the restoration of RAD51 recruitment two enzymes — ATM kinase and CtIP — 3. Boulton, S. J. Biochem. Soc. Trans. 34, 633–645
to DSBs sufficient to re-establish homolo- that have previously been implicated in DSB (2006).
4. Cao, L., Li, W., Kim, S., Brodie, S. G. & Deng, C.-X. Genes
gous recombination in cells deficient in both processing. Dev. 17, 201–213 (2003).
Brca1 and 53bp1? Bunting et al.1 find a signifi- So it seems that, in Brca1-deficient cells, 5. Xu, X. et al. Nature Genet. 28, 266–271 (2001).
cant increase in DSB repair by homologous 53BP1 acts as a barrier to resection at DSBs. 6. Cao, L. et al. Mol. Cell 35, 534–541 (2009).
recombination in these cells. And using the Consequently, repair by homologous recombi- 7. Yu, X., Fu, S., Lai, M., Baer, R. & Chen, J. Genes Dev. 20,
1721–1726 (2006).
technique of gene targeting, Bouwman and co- nation is compromised, and instead error-prone 8. Morris, J. R. et al. Nature 462, 886–890 (2009).
workers report that, in such cells, homologous repair, by processes such as non-homologous 9. Galanty, Y. et al. Nature 462, 935-939 (2009).
302
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010 NEWS & VIEWS

SUPERNOVaE than 1 solar mass. A possible problem is that

New explosions of old stars?

SN 2005cz occurred in an elliptical galaxy.
Ellipticals completed almost all of their star
formation thousands of millions of years ago,
so they generally do not produce core-collapse
David Branch supernovae. But Kawabata et al. cite evidence
Examples of stellar explosions have emerged that fall outside the that this elliptical may be an exception, because
it has experienced a significant amount of star
traditional types of supernova. The nature of the stars that produce formation more recently.
them and the mechanism by which they explode is far from clear. The emphasis of Perets et al.2 is on another
sub-luminous and rapidly fading event,
How many ways can stars explode? At the most explode, but a white dwarf formed in a binary SN 2005E. A serious argument against a core-
fundamental level, possibly just two: ejection system may eventually accrete mass from its collapse origin of SN 2005E is the complete
of matter following the gravitational collapse evolving, expanding companion. An accret- lack of evidence for recent star formation at
of the core of a massive star, and thermo- ing white dwarf shrinks, increasing its density its site. Furthermore, the authors show that to
nuclear disruption of a white-dwarf star. and raising its temperature. In the standard a greater extent than any of the main super-
Because the two mechanisms could hardly be model of SNe Ia, as the mass approaches 1.4 nova types, even SNe Ia, recently documented
more distinct, we might expect that observa- solar masses — the maximum mass that degen- ‘SN 2005E-likes’, including SN 2005cz, tend
tions of a supernova would reveal which way erate electrons can support — nuclear fusion to occur in galaxies dominated by old stellar
the star exploded. For most supernovae this of carbon and oxygen occurs and leads to a populations — making the hypothesis of core
does seem to be the case. However, collapse unlikely, at least for the
on page 326 of this issue, Kawa- class as a whole. The properties
bata et al.1 favour the core-collapse of SN 2005E-likes are, however,
origin for an unusual supernova, different from those of SNe Ia.
whereas Perets et al.2 (page 322) 101 From their estimate that SN 2005E
Ejected mass (solar masses)

support the hypothesis of par- SNe Ia ejected only about 0.3 solar masses
tial explosions of white dwarfs SNe Ib/c (Fig. 1), primarily composed of
for a recently recognized class of SNe II helium and calcium, Perets et al.
similar events. favour an interpretation involv-
Massive stars, formed with 100 ing only partial disruption of an
about eight or more times the mass accreting white dwarf, caused
of the Sun, develop cores of heavy by unstable nuclear burning and
elements (in most cases, iron). The SN 2005E explosive ejection of a thin helium
cores undergo catastrophic col- layer near the surface. Models
lapse, usually to neutron stars or 10–1 that predict partial disruptions of
black holes, while the outer layers −20 −19 −18 −17 −16 −15 −14 −13 accreting white dwarfs do exist, but
of the stars are ejected explosively. Absolute magnitude so far none has predicted all of the
The only event for which we have Figure 1 | Supernova-ejected mass versus absolute magnitude. The location characteristics of SN 2005E-likes.
solid proof of a core-collapse origin of supernova SN 2005E, studied by Perets et al.2, in the bottom right of the The two very different inter-
is the type II supernova SN 1987A, figure is unlike that of traditional types of supernova (SNe Ia, Ib, Ic and II). pretations offered by Kawabata
from which the burst of neutrinos The values for other ‘SN 2005E-likes’ are less certain, but they also occupy et al. and Perets et al. illustrate the
that is expected to accompany the the lower-right area. Absolute magnitude is a logarithmic measure of an present uncertainty about the ori-
collapse was detected3. But there object’s luminosity; bright on the left, faint on the right. (Figure adapted gin of these explosions and, as the
2
is a broad consensus that several from Fig. S3 of Perets and colleagues’ Supplementary Information .) authors1,2 discuss, the unusual com-
of the main types of supernova — position of the ejecta of 2005E-like
those of type II (SNe II, having strong hydro- thermonuclear instability that explodes the events will have important implications for sev-
gen lines), Ib (with prominent helium lines) entire white dwarf 5. eral areas of astrophysics. The present sample
and Ic (neither hydrogen nor helium lines are Kawabata et al.1 concentrate on SN 2005cz, of these faint events is small, but traditional
conspicuous) — result from core collapse4. whose spectrum shortly after its maximum supernova searches have been strongly biased
Because massive stars consume their nuclear brightness was like that of a SN Ib, thus impli- in favour of more luminous supernovae. Less
fuel rapidly and have short lifetimes (less than cating core collapse. Compared with a normal biased searches, some recently begun and oth-
30 million years), they die near where they were SN Ib, however, this event was subluminous; ers to come6–9, are sure to lead to much larger
born and tend to explode in regions showing the rate at which its brightness decreased samples of carefully observed 2005E-likes and
signs of recent star formation. was more rapid; and, in a spectrum obtained other kinds of non-standard supernovae. This,
For several reasons, including the observa- 6 months after its discovery, emission lines of together with increasingly detailed numerical
tion that some appear in galaxies showing no calcium were extremely strong. From theo- modelling of the explosions, will advance our
traces of recent star formation, type Ia super- retical modelling, Kawabata and colleagues understanding of the multiple ways in which
novae (SNe Ia, characterized by lines of singly suggest that core collapse in a star that had an stars can explode. ■
ionized silicon and sulphur) are thought to initial mass between 10 and 12 solar masses David Branch is in the Homer L. Dodge
have the ‘white dwarf ’ type of origin. Stars that may be able to account for the low ejected Department of Physics and Astronomy,
have an initial mass between about 0.5 and 8 mass, the subluminosity and the helium- University of Oklahoma, Norman, Oklahoma
solar masses gradually lose their outer layers and calcium-rich composition of SN 2005cz. 73019, USA.
non-explosively and become carbon–oxygen They conjecture that the star lost its hydrogen e-mail: branch@nhn.ou.edu
white dwarfs supported against gravity by the envelope by interacting with a companion,
pressure of degenerate electrons (electrons and that the remainder developed a 1.5-solar- 1. Kawabata, K. S. et al. Nature 465, 326–328 (2010).
2. Perets, H. B. et al. Nature 465, 322–325 (2010).
as crowded as permitted by Pauli’s exclusion mass iron core whose collapse to a neutron 3. Hasan, Y. & Beacom, J. F. Phys. Rev. D 76, 083007
principle). An isolated white dwarf will not star was accompanied by ejection of no more (2007).

303
© 2010 Macmillan Publishers Limited. All rights reserved
NEWS & VIEWS NATURE|Vol 465|20 May 2010

4. Smartt, S. J. Annu. Rev. Astron. Astrophys. 47, 63–106 (2009). results. They perform a detailed error analysis
7. Drake, A. J. et al. Astrophys. J. 696, 870–884 (2009).
5. Kasen, D., Röpke, F. K. & Woosley, S.E. Nature 460, 8. Young, D. R. et al. Astron. Astrophys. 489, 359–375
that helps determine the confidence in the
869–872 (2009). (2008).
6. Law, N. M. et al. Publ. Astron. Soc. Pacif. 121, 1395–1408 results and where improvements can be made.
9. Lien, A. & Fields, B. D. J. Cosmol. Astroparticle Phys. 1, 047
(2009). (2009). Their conclusions provide a valuable caution
for users of these data.
In spite of all the difficulties, Lyman et al. are
able to demonstrate a robust warming of the
GlOBal CHaNGE global upper ocean from 1993 to 2008, depicted

the ocean is warming, isn’t it?

by the red line in Figure 1, which averages
0.64 ± 0.29 watts per square metre (95% con-
fidence interval) for the Earth as a whole. This
is reasonably consistent with expectations from
Kevin E. Trenberth other indications of global warming6. Nonethe-
A reappraisal of the messy data on upper-ocean heat content for less, the results reveal that all curves flatten out
after 2003 (as seen in the black line in Fig. 1; see
1993–2008 provides clear evidence for warming. But differences among also ref. 7, for example), suggesting that ocean
various analyses and inconsistencies with other indicators merit attention. warming has stalled. However, independent
analysis8 of the full-depth Argo floats for 2003
Global atmospheric temperatures at Earth’s Ocean. The XBTs recorded temperatures, but to 2008 suggests that the 6-year heat-content
surface are often taken as an indicator of their depths were estimated on the basis of an increase is 0.77 ± 0.11 W m−2 for the global ocean
global warming. Yet the atmosphere is battered assigned drop rate, which turned out to be or 0.54 W m−2 for the entire Earth, indicating
by all sorts of natural variability associated sensitive to the exact design and character of that substantial warming may be taking place
with weather phenomena. More robust indica- the XBT probe. Recent careful comparisons below the upper 700 m (Fig. 1, blue line).
tors of a warming planet come from evidence with calibrated probes deployed from research Although Lyman and colleagues’ paper1
of increasing ocean heat content and associ- vessels have shown the need for corrections3. reinforces the overall view that the ocean has
ated sea-level rise. Yet observing systems that The severe under-sampling of the ocean been warming at a rate consistent with radia-
capitalize on these insights are in tive imbalance estimates from
their infancy. On page 334 of this 10 anthropogenic climate change,
issue, a step forward is reported by the slowdown since 2003 is at
Heat content (1022 joules)

Lyman et al.1 — they find a robust 5

odds with top-of-atmosphere
warming of the global upper radiation measurements9. This
ocean to be present in the data, in discrepancy suggests that further
spite of considerable uncertain- 0 problems may be hidden within
ties arising from the observations 0–700 m
the ocean observations and their
themselves. –5 0.64 W m–2 processing. It also highlights the
In September 2009 a meeting2 0.54 W m–2 need to do better, and the pros-
called OceanObs’09, involving –10 pects for that. Experience in the
more than 600 scientists from 36 1994 1996 1998 2000 2002 2004 2006 2008 2010 atmosphere has long highlighted
nations, took place in Venice and Year the desirability of working with
focused on progress in observing Figure 1 | Changing heat content of the global ocean, with respect to the mean ‘anomalies’ as departures from
the ocean since the Initial Ocean- of 1993 to 2008. The warming ocean is revealed by changes in heat content a well-established climatology.
observing System was mooted a from 1993 to 2008, shown by the black line with error bars, as constructed Moreover, methods of analy-
1
decade earlier at the OceanObs’99 by Lyman et al. . This analysis samples −2the ocean to 700 m depth and gives sis and interpolation of gaps
conference. The participants not an average warming trend of 0.64 W m (red line). The data available8from in space and time should take
only assessed the revolutionary Argo floats since 2003 enable an estimate to 2,000 m depth (blue line) to be account of the warming climate,
made. The differences between the black and blue plots after 2003 suggest that
advances that have taken place there has been significant warming below 700 m, and that rates of warming and care is needed not to bias
in ocean observations, but also have slowed in recent years. Processing of the two data sets is not compatible, results towards background
considered how the observations, however, so firm conclusions cannot yet be drawn by comparing them. values. As the relevant analyti-
their processing and their trans- cal methods mature, ocean heat
lation into useful information for decision- until about five years ago, along with the vari- content is likely to become a key indicator
makers should progress into the future. ety of methods used to correct for problems and of climate change. ■
The revolution has been brought about by biases, has led to many estimates of how the tem- Kevin E. Trenberth is at the National Center for
two major developments. The first is milli- peratures in the ocean have changed over time. Atmospheric Research, PO Box 3000, Boulder,
metre-precision altimetry from satellites to Of particular interest for climate is the vertically Colorado 80307, USA.
give sea level globally since 1992; the second integrated ocean heat content. The reprocess- e-mail: trenbert@ucar.edu
is the Argo profiling float system of nomi- ing of XBT and Argo observations has resolved
nally 3,000 floats that provide vertical profiles some issues highlighted in the last report of the 1. Lyman, J. M. et al. Nature 465, 334–337 (2010).
2. Harrison, D. E. & Legler, D. M. Eos 91, 23 (2010).
of temperature and salinity over the upper Intergovernmental Panel on Climate Change 3. Gouretski, V. & Koltermann, K. P. Geophys. Res. Lett. 34,
4
2,000 metres of the ocean every 10 days or so. (IPCC) , in which inflated decadal variability L01610, doi:10.1029/2006GL027834 (2007).
The latter has been built up to approach global was evidently due to changes over time in XBT 4. Science Bindoff, N. L. et al. in Climate Change 2007: The Physical
Basis (eds Solomon, S. et al.) 385−428 (Cambridge
5
coverage and the nominal numbers since about design, deployment and data analysis . But Univ. Press, 2007).
2003. Before then, the bulk of the observations there remains a surprisingly large spread among 5. Domingues, C. M. et al. Nature 453, 1090–1093 (2008).
6. Trenberth, K. E., Fasullo, J. T. & Kiehl, J. Bull. Am. Meteorol.
of the ocean were from expendable bathyther- different estimates of ocean heat content. Soc. 90, 311–323 (2009).
1
mographs (XBTs) dropped from ships along Lyman et al. delve into the origins of these 7. Levitus, S. et al. Geophys. Res. Lett. 36, L07608,
their tracks as opportunities arose. As a result, differences, and compile and reprocess a com- doi:10.1029/2008GL037155 (2009).
8. von Schuckmann, K., Gaillard, F. & Le Traon, P.-Y. J. Geophys.
coverage was spotty and irregular, and miss- mon data set for the upper 700 m of the ocean Res. 114, C09007, doi:10.1029/2008JC005237 (2009).
ing over many regions such as the Southern to isolate how various assumptions affect the 9. Trenberth, K. E. & Fasullo, J. T. Science 328, 316–317 (2010).
304
© 2010 Macmillan Publishers Limited. All rights reserved
Vol 465 | 20 May 2010 | doi:10.1038/nature09107

ARTICLES
Thousands of chemical starting points for
antimalarial lead identification
Francisco-Javier Gamo1, Laura M. Sanz1, Jaume Vidal1, Cristina de Cozar1, Emilio Alvarez1, Jose-Luis Lavandera1,
Dana E. Vanderwall2, Darren V. S. Green3, Vinod Kumar4, Samiul Hasan4, James R. Brown4, Catherine E. Peishoff5,
Lon R. Cardon6 & Jose F. Garcia-Bustos1

Malaria is a devastating infection caused by protozoa of the genus Plasmodium. Drug resistance is widespread, no new
chemical class of antimalarials has been introduced into clinical practice since 1996 and there is a recent rise of parasite
strains with reduced sensitivity to the newest drugs. We screened nearly 2 million compounds in GlaxoSmithKline’s chemical
library for inhibitors of P. falciparum, of which 13,533 were confirmed to inhibit parasite growth by at least 80% at 2 mM
concentration. More than 8,000 also showed potent activity against the multidrug resistant strain Dd2. Most (82%)
compounds originate from internal company projects and are new to the malaria community. Analyses using historic assay
data suggest several novel mechanisms of antimalarial action, such as inhibition of protein kinases and host–pathogen
interaction related targets. Chemical structures and associated data are hereby made public to encourage additional drug
lead identification efforts and further research into this disease.

With approximately 243 million cases and 863,000 attributed deaths Tres Cantos antimalarial compound set (TCAMS)
reported globally in 2009 (ref. 1), malaria is one of the most severe The 1,986,056 compounds present in GSK’s screening collection in
infectious diseases, primarily affecting the world’s most disadvantaged January 2009 were tested for inhibition of P. falciparum 3D7 at 2 mM
populations. Of the four typically recognized Plasmodium species under in vitro conditions described in Methods. 19,451 primary hits
causing disease in humans, Plasmodium falciparum causes most mor- inhibiting parasite growth by more than 80% were obtained. Fresh
tality, mainly in children below the age of 5, and Plasmodium vivax samples of these primary hits were tested in two independent experi-
most morbidity, additionally representing a reservoir of latent infec- ments and compounds displaying 80% or higher inhibition of para-
tion that hampers current control and future elimination efforts2. No site growth in at least two of the three assay runs were considered
new class of antimalarials has been introduced into clinical practice confirmed hits. 13,533 compounds were identified using this pro-
since 1996 (ref. 3), owing to the intrinsic difficulties in discovering tocol (confirmation rate . 70%). We did not detect any compounds
and developing new antimicrobials, as well as a relative lack of public in this set as non-specific inhibitors of the biochemical readout sys-
and private resource commitment towards antimalarial research. tem by testing directly for inhibition of lactate dehydrogenase (LDH)
Today, the last class of widely efficacious drugs, the artemisinins, in P. falciparum extracts (Methods). Evidence of cytotoxicity against
is being compromised by the rise of P. falciparum strains with human hepatoma HepG2 cells (a widely used in vitro marker for liver
reduced clinical response to artemisinin-containing drug combina- toxicity8), or interference with the luciferase reporter system used in
tions4–6. The genomics revolution has not yet led to new antimalarial the cytotoxicity assay (Methods), was observed in just 1,982 of the
medicines and target-based lead discovery has produced disappoint- compounds when tested at 10 mM. This relative lack of non-specific
ing results, generally for lack of whole-cell activity as documented cell toxicity is probably due in part to the low (2 mM) primary screen-
for antibacterials7. To secure that property in all chemical starting ing concentration used9. Estimation of the concentrations producing
points for new antimalarial leads, we have tested the approximately 50% inhibition of P. falciparum growth (XC50, see Methods) indi-
2 million-compound library used for high throughput screening at cated that most compounds are sub-micromolar inhibitors. The full
GlaxoSmithKline (GSK) for inhibitors of P. falciparum’s intraerythro- compound set (TCAMS) and data table (Supplementary Table 1 and
cytic cycle, the Plasmodium species causing the highest mortality and available at http://www.ebi.ac.uk/chemblntd) contains 13,533 com-
the parasite growth phase responsible for disease symptoms as well as pound entries. We have detected 139 of these as variations in salt
being amenable to in vitro culture. Here we describe 13,533 com- form or stereochemistry of 68 parent structures, which make good
pounds confirmed to inhibit parasite growth by more than 80% at internal controls for the biological assay data. They appear as differ-
2 mM concentration. Only 15% displayed some cytotoxicity in that ent compounds with the same structure. When the stereochemistry is
they inhibited proliferation of the HepG2 human hepatoma cell line resolved it shows in the SMILES structural code in Supplementary
by more than 50% at 10 mM. All of these proven plasmodial inhibitors, Table 1 and in the Chembl-NTD database (http://www.ebi.ac.uk/
of which 82% were previously proprietary and thus unknown to the chemblntd).
general research community, are hereby made public to accelerate the Representatives from all but one class of clinically used antimalarials
pace of drug development for malaria. have been recovered in the screen, providing additional validation
1
Tres Cantos Medicines Development Campus, GlaxoSmithKline, Severo Ochoa 2, 28760 Tres Cantos, Spain. 2Computational and Structural Chemistry, GlaxoSmithKline, Five Moore
Drive, Research Triangle Park, North Carolina 27709-3398, USA. 3Computational and Structural Chemistry, GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road,
Hertfordshire, Stevenage SG1 2NY, UK. 4Computational Biology, Quantitative Sciences, GlaxoSmithKline, 1250 South Collegeville Road, Collegeville, Pennsylvania 19426, USA.
5
Computational and Structural Chemistry, GlaxoSmithKline, 1250 South Collegeville Road, Collegeville, Pennsylvania 19426, USA. 6Quantitative Sciences, GlaxoSmithKline, 709
Swedeland Road, King of Prussia, Pennsylvania 19406, USA.

305
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 465 | 20 May 2010

of the selected assay. Several analogues of 4-aminoquinolines (for
example, chloroquine), 8-aminoquinolines (for example, prima-
quine), methanol-quinolines (for example, quinine), diaminopyrimi-
dines (for example, pyrimethamine), diaminotriazines (for example,
cycloguanil) and naphthoquinones (for example, atovaquone) are all
present in the hit collection. They do not represent more than 10% of
all the hits by our similarity criteria. Endoperoxides (for example,
artemisinin) were not identified, as they are not present in the screen-
ing library.
In addition to screening the reference laboratory strain 3D7, TCAMS
was subsequently screened against the multidrug resistant Dd2 strain at
the same concentration used in the primary screen (2 mM). This strain is
insensitive to several quinolines and antifolates. Results showed
approximately 8,000 compounds (,60%) inhibited Dd2 growth by
more than 50% (Supplementary Table 1). Compounds potent against
3D7 (XC50 lower than 200 nM), but less active against Dd2 (growth
inhibition less than 50% at 2 mM), are predominantly quinolines or
structures related to antifolates, as expected. But Dd2 has also been
shown to be less sensitive to unrelated chemotypes10, probably due to
amplification and mutation of the efflux pump gene pfmdr1 (ref. 11).
Specific characterization of individual hits is required before more
detailed conclusions can be drawn, but it is encouraging that more than
half of our hits retain potent activity against a multidrug resistant strain.
Structural characterization of hits
We used two methods to structurally characterize TCAMS and
estimate the number of different chemotypes present. The hits can
be described by 416 molecular frameworks12 or by fingerprint clus-
ters. Here, we use the Daylight fingerprint methods with a Tanimoto
similarity index of 0.85 (ref. 13) which yields 857 clusters and 1,978
‘singletons’ (three or fewer similar compounds) (Supplementary
Table 1). These fingerprint cluster annotations can be used to sepa-
rate distinct classes from within the broader molecular framework
categories. This is because the molecular frameworks describe the
core template of compounds well (and therefore provide a robust,
consistent categorization of compounds) whereas the fingerprint
methods will often capture subtle substituent patterns but miss the
commonalities in the core; together the two methods allow an
ordered navigation through the chemical structure space represented
by TCAMS and exemplified graphically in Fig. 1. In the absence of
specific target information, such chemical structure clustering pro-
vides a framework to enable a systematic exploration of the mode of
action for the compounds. The assumption being that compounds in
the same cluster share mode of action (similarity principle14). This
may not always be true but it is the best starting hypothesis in
the absence of additional information. Therefore, taking just a few
exemplars from each cluster can reduce the work required to identify

pXC50

50 25 150 160
100 75 170
150 125 180
175 190
200
225
8.5
8.5
Cl
N Cl Cl
OS
O N
N NN O
F N N
8
O
F O N O N
O 8
N N N
O N N O O O
Cl O N Cl
N O
7.5 O O
N N
N N N
N N 7.5
N F
S N N F N O
7 F
O N
N
N O
7
F 6.5
O
O S
O N N
N 6 6.5
O

N
Framework
number 150
6

O 160N N N
N
Cl 25
N 170
50 Cluster
N N 75S
O
number
S N N
O 100
N
S O N N
N 180 O O 125 N
N S Cl N SO N
150 N
F N N N O
Cl 175
O N Cl

Figure 1 | Three-dimensional plot of some of the novel chemical diversity (pXC50 5 2logXC50, where XC50 is in molar units and pXC50 is
present in TCAMS. Compounds are represented by coloured spheres dimensionless; XC50 5 1 mM corresponds to pXC50 5 6). Inserted structures
plotted against their assigned molecular framework number, chemical are examples of drug-like molecules not previously described to possess
fingerprint cluster number and estimated antiplasmodial potency antiplasmodial activity.
306
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 ARTICLES

specific targets for a set of molecules. Fingerprinting methods also
enable searching outside of TCAMS for the activity of related mole-
cules that can inform the biological activity of a given chemical clus-
ter. Molecular framework and fingerprint cluster numbers are
indicated for all compounds in Supplementary Table 1.
From a physicochemical point of view, our screen selected for
compounds having a larger molecular mass and a higher hydropho-
bicity index than the average for the source compound collection (446
versus 385 Da; 5 versus 3.3 clogP, respectively; Fig. 2a, b). The meaning
of this observation is unclear, but it may have to do with the physico-
chemical requirements needed to reach intracellular targets. The same
reasons may also explain the bias towards compounds made for
internal projects (82% of the hits) versus those commercially sourced
(50% of the screening collection), as the latter tend to be smaller and
more hydrophilic (commercial suppliers are listed in Supplementary
Table 1). Database searches show that most internal compounds have
not been published previously (see Supplementary Table 2 for excep-
tions), thus potentially represent highly novel chemical diversity made
available to the malaria community. Some automatic methods may
flag compounds in TCAMS with structural features a priori undesir-
able in drug leads15, but because structural alerts are not always borne
out and TCAMS compounds are not overtly cytotoxic we have
retained them, because they can at least be useful for target identifica-
tion and validation.
Modes of action hypotheses
We have used historical GSK data to develop testable hypotheses for
the antimalarial mode of action of hits in an attempt to facilitate
target identification efforts in the broader malaria community. We
found 4,205 compounds with unambiguous annotations regarding
their activity in biochemical assays against human (3,435 com-
pounds) or microbial targets (770 compounds), some of which have
been reported in the open literature (Supplementary Table 2).
identified by calculating an ‘inhibition frequency index’ (IFI) (see
Methods) that is reported for every structure (Supplementary Table 1).
These compounds were not used for analysis but remain in the data
set. We also used conservative thresholds to consider a compound a
true effector of a specific human target (that is, sub-micromolar 50%
inhibitory concentration (IC50), see Methods). Overall, activities at
413 targets (some assayed in both antagonist and agonist modes) were
analysed among the antimalarial compounds. Figure 2c shows the
distribution of known human target class activity among the hits.
Approximately 70% are G-protein coupled receptors (GPCRs) or
protein kinases. This reflects partly the relative abundance of different
ligand classes in GSK’s screening collection, and it would bias any
straightforward prediction of the antimalarial mode of action. We
therefore wanted to distinguish targets preferentially inhibited by
TCAMS compounds from those equally prevalent among the com-
pounds not present in TCAMS, under the hypothesis that the former
were more likely to be human orthologues of the lethal target in
Plasmodium. We calculated an ‘enrichment’ factor as simply the ratio
of target actives among hits, over a background defined as all target
actives among all screened compounds with data for that target (see
Methods). An enrichment factor of two indicates that inhibitors of
that target appear at twice the relative frequency among the anti-
malarial compounds than among the inactive ones, indicating that
the target is more likely to be relevant for the antimalarial mode of
action. Figure 2d shows the distribution of known target class activity
when only human targets enriched by at least twofold, plus the anti-
microbial targets, are considered. The added restriction reduced the
total number of probable targets to 146 (from 413) and interestingly
increased the relative proportion of kinases to 48%, strongly suggest-
ing that this target class, still unexploited for antimalarials, can be a
rich source of novel drug leads. Tables 1 and 2 show the number of
Mal1 fC0525 4

Pf14_0431
Pf11_0156
Pf11_0
Pf14
P _029
Inhibitors with highly promiscuous and non-specific profiles were 3P1.1 c
096
AGC

c
85

50
GC * * * * * * * *

22
* * *

CM
Pf

* *

w
* * * * * *
* * *
D0

PfL
* * * *

85
* * *
* * * * * *
M

* * * *
* * *

a b * *

I16
86

* *
al

* * *
* * * *
13

* * * *
5c

* *
Pf *

Pf
* * * *
P1

* * *
16 11 * * * * *
* *

30 TCAMS * * *
Other
.2

TCAMS _0 *
* * * *
* *
*
79

* *
24 * * * *
5c 8
14 GSK 2 *
* *
* *
*
14 25
25 GSK *
*
*
*
*
F1 _0
Pf f13
* *
12 *
*
Percentage

* * *
* *
P
MK

* * *
20

CA
Percentage

* *
* * *
10 *
*
*
*
*

MK
*
CA

15 8 *
* * * *
*
*

516
* *
* * * *
PfB *
4_0
10 6 Pf07 0815w * *
* * * *
*
*
Pf1
Pf13 _0072 * * * *
*

CK1
5 _
4 PfC0 0211 * * *
*

420w * * * * *

2 PfF0520 * *
0 w ** *
*
*
*
* *
*
45 –4 0
50 –5 0
30 –3 0

0– 00
60 –6 0
65 –650
0 0
0– 0
0– 50
0 00
0 0

80 0
0+

*
0 0
0 5
0 5

0 5
35 –35

0 0

70 –70

80

0 *
25 –2

40 –4

55 5

*
75 7

*
*
–0.5–0
0–0.5
0.5–1
1–1.5
1.5–2
2–2.5
2.5–3
3–3.5
3.5–4
4–4.5
4.5–5
5–5.5
5.5–6
6–6.5
6.5–7
7–7.5
7.5–8
–3––2.5
–2.5––2
–2––1.5
–1.5––1
–1––0.5

8+
0

* * *
20

*
Pf11_0239 *
* *
*
PfF0260w
Binned molecular mass (Da) Binned clogP PfC0385c *
*
*
* *
* *
*
*
0w
PfL228 76 *
*

1 (0) 1 (0) 1 (0)

_04
* *

c 23 81 d 6 (2)
1 (1)
Pf14 1885c *
*
* *
*
*
*

PfL 520w
* * *
24 8 (10) 0
PfB
*
* * * *
TK

* *
* *
8 (0) *
* *
*

31 158 * * * *
*
*
*
Antibacterial 5 * * *
*

08
*
* *
Antifungal _0
*
* *
11 (0) 13
* *

Pf
*
Antimalarial *
Ot

* *

36 70 (30) * * * *
Antiviral * *
he

* *
0w

* *
GPCR * *
r

*
37

* * *
* * *
L1

*
Kinase 12 (8) *
*
* * *
* *
Pf

* * * *
Nuclear receptor *
*
*
*
*
* *
*
* * *
*
*
*
* *
*
*

* * * *
Other enzyme * *
*
* * * *
* * *
* *
0C

* * * * * * *
Other receptors
* * * * *
L
TK
015

Protease
STE
PfB

132 Transporter 27 (0)

Figure 2 | Description of TCAMS and its target space. a, Relative molecular
mass distributions of the TCAMS molecules and the general GSK screening
collection. b, Relative clogP distributions of the TCAMS molecules and the
general GSK screening collection. c, Distribution of target classes affected by
the compounds in the set with potency and selectivity criteria described in
the text. The number of targets in each class is indicated. d, Target classes
remaining after applying the criterion that human targets be enriched at least
twofold among the hits relative to the screening compound library, plus
antimicrobial targets known to be relevant for Plasmodium. Number of
targets in each class is indicated, with the number of identified malarial
orthologues (BLASTP E-value # 1.0220) in parentheses.
©2010 Macmillan Publishers Limited. All rights reserved
Figure 3 | Phylogenetic tree of combined human and P. falciparum kinomes.
P. falciparum and human lineages are coloured red and black, respectively.
Major human kinase subfamilies are labelled. TK: tyrosine kinases; TKL:
tyrosine-like kinases; STE: homologues of yeast sterile 7, 11 and 20 kinases;
CK1: casein kinase 1; AGC: PKA, PKG and PKC kinases; CAMK: calcium/
calmodulin kinases; CMGC: CDK, MAPK, GSK3 and CLK kinases. Malarial
kinases that are hypothesized targets in Table 1 are labelled. This neighbour-
joining tree is based on pairwise amino acid sequence similarity for all
known human22 and P. falciparum24,27 kinase domains. Nodes supported
by $ 60% of 1,000 bootstrap replicates are indicated by ‘*’ (see Methods for
tree reconstruction and Supplementary Fig. 1).
307
ARTICLES NATURE | Vol 465 | 20 May 2010

Table 1 | P. falciparum loci encoding proteins proposed as probable targets for TCAMS compounds
Target class Target hypothesis/ Number of Example inhibitor Target class Target hypothesis/ Number of Example inhibitor
P. falciparum locus chemotypes/number structures (compound ID)* P. falciparum locus chemotypes/number structures (compound ID)*
of inhibitors of inhibitors
Antimalarial Dihydroorotate 5/20 (TCMDC-124418) Proteases Aspartic protease/ 7/41 (TCMDC-134572)
dehydrogenase/ PF14_0076
PFF0160c PF14_0075
PF14_0077
PFC0495w
Electron transport 5/122 (TCMDC-135426)
chain{/PlfaoMp3

Antibacterial Dihydrofolate 4/9 (TCMDC-139863) Cysteine protease/ 15/51 (TCMDC-135965)

reductase/ PF11_0162
PFD0805w PF11_0161
PF11_0165
DNA gyrase/ 7/15 (TCMDC-134527)
PF11_0174
PFL1120c
PF14_0553
PFL1915w
PFL2290w

Isoleucyl-tRNA 1/1 (TCMDC-131575)

Kinases Phosphatidylinositol 12/16 (TCMDC-134265)
synthetase/
3-kinase/PFE0765w
PF13_0179

Ser/Thr protein 95/264 (TCMDC-133561)

kinase/PFL2250c
PF11_0156
PF13_0085
Methionyl-tRNA 2/4 (TCMDC-139627) PFI1685w
synthetase/ PFF1145c (TCMDC-141334)
PF10_0340 PFD0865c
PF14_0476
PF14_0431
PF11_0239
PF11_0096
Phenylalanyl- 3/7 (TCMDC-140563) PFC0525c
tRNA synthetase/ (TCMDC-133396)
MAL13P1.185
PFA0480w PF13_0258
MAL13P1.279
PF14_0294
PF14_0516
PFB0150c
PFL2280w
Tyrosyl-tRNA 2/2 (TCMDC-141232) PFL1370w
synthetase/ PFC0385c
PF11_0181 PFF0260w
PFF0520w
PF13_0211
PFC0420w
PFL1885c
Ribosome 1/1 (TCMDC-123920)
PF11_0242
complex{/
PFB0815w
MAL1P3.03a
PF07_0072
PFB0520w
Other enzymes b-Keto-acid 2/2 (TCMDC-125270)
reductase/
PFI1125c

* Compound identification number as in Supplementary Table 1 (EXT_CMPD_NUMBER). Full compound assay data and structures are available at ChEMBL – Neglected Tropical Disease Archive
(http://www.ebi.ac.uk/chemblntd).
{ Multi-protein complexes.

chemotypes predicted to inhibit each target class. We have only added
target annotations to compounds for which we had experimental
evidence meeting the criteria above. Readers should use their own
judgement when extrapolating proposed modes of action to chemical
analogues.
We next searched for orthologues of the human targets in the
P. falciparum genome, using conservative criteria (see Methods) for
human–P. falciparum sequence homology to address this species’
complex and divergent evolutionary relationship to mammals16,17.
Forty-one Plasmodium proteins could be linked to human targets by
those measures (Table 1). A number of compounds annotated as active
in assays against other infectious agents, including Plasmodium, and
308
©2010 Macmillan Publishers Limited. All rights reserved
with known mode of action, were also included in the analysis, yielding
predictions of activity against organellar functions, such as protein
synthesis, DNA metabolism or electron transport. This brings the total
number of indicative Plasmodium targets to 51 (Table 1 and Sup-
plementary Table 1).
To support the mode of action hypotheses we also checked gene
expression profiles of the proposed targets using data from studies of
the P. falciparum transcriptome across all life stages18 and specifically,
the intraerythrocytic development cycle measured over 48 h (ref. 19).
Most targets proposed were expressed at various intervals in the intra-
erythrocytic malarial parasite and several were also present in other life
stages (Supplementary Fig. 1).
NATURE | Vol 465 | 20 May 2010 ARTICLES

Table 2 | Functions proposed as targets for TCAMS compounds without obvious homology to P. falciparum loci
Target class Target hypothesis Number of Example inhibitor Target class Target hypothesis Number of Example inhibitor
chemotypes/number structures* (compound ID){ chemotypes/number structures* (compound ID){
of inhibitors of inhibitors
GPCR Adrenergic 6/9 (TCMDC-137488) Ion channel Ion channel 71/108 (TCMDC-136042)
receptor inhibitor
antagonist

Cannabinoid 1/1 Nuclear Nuclear receptor 7/9 (TCMDC-125489)

receptor hormone agonist
antagonist receptor
Chemokine 4/7 (TCMDC-137146)
receptor
antagonist
Nuclear receptor 9/14 (TCMDC-131829)
antagonist

Cholinergic 1/1
receptor agonist
Free fatty acid 1/1
receptor agonist Peptide hormone 5/9 (TCMDC-138566)
G protein-coupled 9/10 (TCMDC-139980) receptor agonist
receptor agonist

Peptide hormone 19/62 (TCMDC-137435)

receptor
antagonist

G protein-coupled 1/2
receptor Other Phospholipase 3/3 (TCMDC-141264)
antagonist enzyme inhibitor
Serotonin receptor 9/27 (TCMDC-140065)
agonist

Lipid amide 3/3 (TCMDC-139221)

hydrolase
Serotonin receptor 14/23 (TCMDC-135435) inhibitor
antagonist
Serine protease 5/7 (TCMDC-137765)

Opioid receptor 1/1

agonist
Opioid receptor 4/5 (TCMDC-133483)
antagonist Toll-like receptor 1/1
agonist

* Selected compounds where two or more chemotypes were identified.

{ Compound identification number as in Supplementary Table 1 (EXT_CMPD_NUMBER). Full compound assay data and structures are available at ChEMBL – Neglected Tropical Disease Archive
(http://www.ebi.ac.uk/chemblntd).

In these data, proteases and kinases are the most prominent human
target classes with identified P. falciparum orthologues (Table 1 and
Fig. 2d). Proteases are needed for bulk degradation of haemoglobin in
the food vacuole, but also for signalling and remodelling of cellular
structures, presenting a large array of potential lethal targets for
TCAMS compounds20. But the larger proportion of hits with anno-
tated activity against ‘enriched’ targets are protein kinase inhibitors,
which provide new tools to exploit the malarial kinome for drug dis-
covery21. The human genome encodes over 450 kinases22 whereas
P. falciparum has a much reduced complement of 85–99 kinases23,24.
We reconstructed a phylogenetic tree of the entire complement of
human and P. falciparum kinases to assist in assigning putative targets
for kinase inhibitors. Figure 3 highlights the ancestral relationship of
Plasmodium kinases to multiple human homologues, allowing inhibi-
tors of several related human kinases to be mapped against a single
©2010 Macmillan Publishers Limited. All rights reserved
malarial isoform, but keeping in mind that many kinase inhibitors have
been shown to inhibit multiple, distantly related enzymes, often owing
to three-dimensional structural similarities among binding sites25.
Some compounds had historical assay activity that met our criteria
for inclusion in the analysis, but against drug targets without obvious
orthologues in the malarial genome, such as GPCRs, nuclear recep-
tors, ion channels and transporters (Table 2). The simplest explana-
tion would be that these compounds are killing Plasmodium through
interactions with unrelated targets. More interestingly, P. falciparum
could have essential proteins that are structurally and functionally
similar to the human targets yet have no significant primary amino acid
homology. This possibility is supported by recent analyses of structural
rather than sequence homology, showing at least two candidate trans-
membrane proteins with similarity to human orphan GPCRs26.
The screens in this study were conducted using Plasmodium-infected
309
ARTICLES
erythrocytes; therefore, activity against specific human red blood cell
targets is also a possible mode of action. Further exploration of this
compound collection for disruptors of host–parasite interactions could
lead to novel antimalarial therapeutic strategies, free from the selection
pressures driving drug resistance in the parasite.
The public availability of this large set of potent and drug-like
antiplasmodial compound structures enables important research
questions to be addressed in future investigations, including char-
acterizing the links between the biochemical activity of specific com-
pounds and parasite killing and the extent of activity of these
compounds in inhibiting P. vivax, which will be a critical attribute
of new antimalarial treatments.
METHODS SUMMARY
Compounds were pre-dispensed in 384-well plates, RPMI/AlbuMAX growth
media was added and P. falciparum inoculated in a biological safety level 3
laboratory, where plates were incubated for 72 h and then frozen at 270 uC over-
night. LDH activity was quantified with the modified cofactor 3-acetylpyridine
adenine dinucleotide (APAD) by measuring absorbance of the tetrazolium indi-
cator nitro blue tetrazolium (NBT) at 650 nm, using an extensively modified LDH
assay as described in Methods. Screening data were analysed using Microsoft Excel
and Spotfire DecisionSite 8.2.1 software. Chemical structures were clustered using
Daylight13. Mode of action hypotheses were generated searching archived target
activity data of hits in the internal databases and filtering results for potency,
selectivity, enrichment in the data set relative to the general screening collection
and degree of homology to P. falciparum proteins, with the cut-off values described
in the text. Plasmodium orthologues of human targets were searched and analysed
using GCG Wisconsin Package v11.0 and PHYLIP 3.67. software packages as
described in Methods.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 8 February; accepted 22 April 2010.
1. World Health Organization. World malaria report. Æhttp://www.who.int/
malaria/publications/atoz/9789241563901/en/index.htmlæ (2009).
2. Anstey, N. M., Russell, B., Yeo, T. W. & Price, R. N. The pathophysiology of vivax
malaria. Trends Parasitol. 25, 220–227 (2009).
3. Ekland, E. H. & Fidock, D. A. In vitro evaluations of antimalarial drugs and their
relevance to clinical outcomes. Int. J. Parasitol. 38, 743–747 (2008).
4. Andriantsoanirina, V. et al. Plasmodium falciparum drug resistance in Madagascar:
facing the spread of unusual pfdhfr and pfmdr-1 haplotypes and the decrease of
dihydroartemisinin susceptibility. Antimicrob. Agents Chemother. 53, 4588–4597
(2009).
5. Bonnet, M. et al. Varying efficacy of artesunate1amodiaquine and
artesunate1sulphadoxine-pyrimethamine for the treatment of uncomplicated
falciparum malaria in the Democratic Republic of Congo: a report of two in-vivo
studies. Malar. J. 8, 192 (2009).
6. Carrara, V. I. et al. Changes in the treatment responses to artesunate-mefloquine
on the northwestern border of Thailand during 13 years of continuous
deployment. PLoS ONE 4, e4551 (2009).
7. Payne, D. J., Gwynn, M. N., Holmes, D. J. & Pompliano, D. L. Drugs for bad bugs:
confronting the challenges of antibacterial discovery. Nature Rev. Drug Discov. 6,
29–40 (2007).
8. O’Brien, P. J. et al. High concordance of drug-induced human hepatotoxicity with
in vitro cytotoxicity measured in a novel cell-based model using high content
screening. Arch. Toxicol. 80, 580–604 (2006).
9. Xia, M. et al. Compound cytotoxicity profiling using quantitative high-throughput
screening. Environ. Health Perspect. 116, 284–291 (2008).
10. Yuan, J. et al. Genetic mapping of targets mediating differential chemical
phenotypes in Plasmodium falciparum. Nature Chem. Biol. 5, 765–771 (2009).
310
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
11. Reed, M. B., Saliba, K. J., Caruana, S. R., Kirk, K. & Cowman, A. F. Pgh1 modulates
sensitivity and resistance to multiple antimalarials in Plasmodium falciparum.
Nature 403, 906–909 (2000).
12. Bemis, G. W. & Murcko, M. A. The properties of known drugs. 1. Molecular
frameworks. J. Med. Chem. 39, 2887–2893 (1996).
13. Daylight Chemical Information Systems, Inc. Daylight theory manual. Æhttp://
www.daylight.com/dayhtml/doc/theory/index.htmlæ (2008).
14. Johnson, M., Lajiness, M. & Maggiora, G. M. in Qsar: Quantitative structure-activity
relationships in drug design (ed Fauchere, J. L.) 167–171 (Alan R. Liss, 1989).
15. Davis, A. M., Keeling, D. J., Steele, J., Tomkinson, N. P. & Tinker, A. C. Components
of successful lead generation. Curr. Top. Med. Chem. 5, 421–439 (2005).
16. Köhler, S. et al. A plastid of probable green algal origin in apicomplexan parasites.
Science 275, 1485–1489 (1997).
17. McFadden, G. I., Reith, M. E., Munholland, J. & Lang-Unnasch, N. Plastid in human
parasites. Nature 381, 482 (1996).
18. Le Roch, K. G. et al. Discovery of gene function by expression profiling of the
malaria parasite life cycle. Science 301, 1503–1508 (2003).
19. Bozdech, Z. et al. The transcriptome of the intraerythrocytic developmental cycle
of Plasmodium falciparum. PLoS Biol. 1, e5 (2003).
20. Wu, Y., Wang, X., Liu, X. & Wang, Y. Data-mining approaches reveal hidden
families of proteases in the genome of malaria parasite. Genome Res. 13, 601–616
(2003).
21. Leroy, D. & Doerig, C. Drugging the Plasmodium kinome: The benefits of
academia–industry synergy. Trends Pharmacol. Sci. 29, 241–249 (2008).
22. Manning, G., Whyte, D. B., Martinez, R., Hunter, T. & Sudarsanam, S. The protein
kinase complement of the human genome. Science 298, 1912–1934 (2002).
23. Anamika, S. N. & Krupa, A. A genomic perspective of protein kinases in
Plasmodium falciparum. Proteins 58, 180–189 (2005).
24. Ward, P., Equinet, L., Packer, J. & Doerig, C. Protein kinases of the human malaria
parasite Plasmodium falciparum: the kinome of a divergent eukaryote. BMC
Genomics 5, 79 (2004).
25. Karaman, M. W. et al. A quantitative analysis of kinase inhibitor selectivity. Nature
Biotechnol. 26, 127–132 (2008).
26. Madeira, L. et al. Genome-wide detection of serpentine receptor-like proteins in
malaria parasites. PLoS ONE 3, e1889 (2008).
27. Aurrecoechea, C. et al. PlasmoDB: A functional genomic database for malaria
parasites. Nucleic Acids Res. 37, D539–D543 (2009).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank S. Peregrina and S. Prats for technical assistance,
and D. Jiménez-Alfaro for supplying compounds pre-dispensed in microtitre
plates. We thank P. Vallance and R. Keenan for organising support for this work,
R. Macarron and J. Luengo for developing the IFI index and getting the chemistry
data ready for publication, together with J. M. Fiandor, S. Chakravorty and
members of GSK’s Chemistry Council. J. Lewis, A. Clow, J. Overington and
M. Davies were instrumental in the uploading and formatting of the data. We also
thank N. Cammack, P. Sanseau and J. Burrows for critically commenting on the
manuscript. The support and funding of Medicines for Malaria Venture is gratefully
acknowledged.
Author Contributions F.-J.G. and J.F.G.-B. planned and designed the work. F.-J.G.
supervised all experimental work and analysed the screening data, L.M.S., J.V.,
C.d.C. and E.A. performed the screening assays and contributed to data analysis.
J.-L.L., D.E.V., D.V.S.G. and C.E.P. performed the cheminformatic analyses and
wrote sections of the manuscript. V.K., S.H. and J.R.B. performed the bioinformatic
analyses and J.R.B. contributed the relevant sections to the manuscript. L.R.C and
J.F.G.-B integrated individual contributions and wrote the final manuscript.
Author Information Chemical structures and data described have been deposited
at EBI (http://www.ebi.ac.uk/chemblntd). Reprints and permissions information
is available at www.nature.com/reprints. The authors declare no competing
financial interests. Readers are welcome to comment on the online version of this
article at www.nature.com/nature. Correspondence and requests for materials
should be addressed to J.F.G.-B. (jose.f.garcia-bustos@gsk.com).
doi:10.1038/nature09107
METHODS
Parasites. P. falciparum strains 3D7 and Dd2 used in this study were obtained from
the Malaria Research and Reference Reagent Resource Center (MR4). Accurate
descriptions can be obtained at http://www.mr4.org. Parasite strains were cultured
using standard procedures as described28. An inoculum of parasitized red blood
cells (PRBC) at 0.25% parasitaemia and 2% haematocrit in RPMI-1640, 5%
AlbuMAX, 2% D-sucrose, 0.3% glutamine and 150 mM hypoxanthine was used
for the assay.
Growth inhibition assay. Assay plates (384-well) were prepared by dispensing
0.05 ml of compound from master plates at 1 mM in each well. Final assay volume
was 25 ml and final compound concentration was 2 mM. The sixth column was
the positive growth control and had 0.05 ml of DMSO. The eighteenth column
had 0.05 ml of a mixture of 50 mM chloroquine and 50 mM artemisinin stock
solutions as negative growth control. The parasite inoculum (25 ml) was dis-
pensed into plates containing compounds using a Multidrop Combi dispenser
(Thermo Scientific). Plates were shaken for 10 s to ensure mixing and then
incubated at 37 uC for 72 h in an atmosphere of 5% CO2, 5% O2, 95% N2.
Evaluation of parasite growth using LDH activity. After 72 h of incubation,
plates were frozen at 270 uC overnight and then thawed at room temperature for
at least 4 h. To evaluate LDH activity, 70 ml of freshly made reaction mix
(143 mM sodium L-lactate, 143 mM 3-acetyl pyridine adenine dinucleotide
(APAD), 178.75 mM Nitro Blue tetrazolium chloride (NBT), 286 mg ml21 dia-
phorase (2.83 U ml21), 0.7% Tween 20, 100 mM Tris-HCl pH 8.0) was dispensed
using a Multidrop Combi dispenser. Plates were shaken to ensure mixing and
absorbance at 650 nm was monitored in a plate reader after 10 min of incubation
at room temperature. Data were normalized to percent growth inhibition using
positive and negative controls and the equation:
  
Awell {Aneg
Percentage inhibition growth~ 1{ |100
Apos {Aneg
where Awell is the absorbance measured in the well, and Apos and Aneg are the
average absorbances measured for the positive and negative controls, respec-
tively. This method is a modification of existing ones29 that requires only a single
pipetting step after compound incubation and gave a signal to noise ratio of 10
under the conditions chosen. The approach allowed kinetic and end-point read-
outs and produced a Z9 quality factor30 higher than 0.7 in validation assays
(Supplementary Fig. 2). Potencies of standard antimalarial agents in this assay
were comparable to those determined by the current gold-standard, 96-well,
hypoxanthine incorporation assay31 (Supplementary Table 3).
At this level of miniaturization, integrity of erythrocytes and LDH activity can
be inspected visually, allowing for rapid detection of dispensing errors, inter-
ference by coloured compounds, or haemolysis, making the method very useful
for low technology settings (Supplementary Fig. 3). Proliferation of asynchronous
parasites was measured after 72 h of incubation in the presence of 2 mM com-
pound. We chose a 72 h incubation time to ensure all parasites traversed at least
once through each stage of the cell cycle and to increase the chances of identifying
slow acting and ‘delayed death phenotype’ inhibitors32,33. Some of these were
indeed detected (for example, tetracyclines), but others, presumably with a more
delayed effect, were not (for example, azythromycin). Metrics for the full screen
were of high quality, with an average Z9 of 0.7 and fewer than 3,000 wells lost to
analysis (Supplementary Table 4).
Given the large number of positives, it was operationally necessary to estimate
the concentrations producing 50% inhibition using the LDH assay above and
generating dose–response curves with fivefold dilution steps down to 3 nM
compound in an interplate design, instead of using the hypoxanthine incor-
poration assay with twofold dilution intraplate series generally considered the
standard method to calculate IC50 for antimalarials34. We denominated this
parameter XC50 to indicate that it is an estimation of the usual IC50 values.
The lowest concentration tested was 3 nM. Agreement between the two methods
was found to be within the expected limits with standard antimalarials
(Supplementary Table 3). To eliminate the possibility of retaining inhibitors
of the biochemical readout system, one set of the primary hits was assayed against
parasite LDH activity under identical screening conditions. Non-specific cyto-
toxicity was also a concern and we addressed it by assaying the hits at five times
the screening concentration against human hepatoma HepG2 cells as below.
Preparation of extracts to evaluate direct LDH inhibition by hit compounds.
P. falciparum 3D7 strain was grown as described above at 37 uC for 72 h. The
culture was frozen at 270 uC overnight. Cultures were thawed at room temper-
ature for at least 4 h and used as crude extracts (25 ml per well) to measure the
possible direct inhibition of LDH by the compounds, assayed as above.
Cytotoxicity assay. Actively growing human hepatoma HepG2 cells were
removed from a T-175 TC flask using Cell Dispersion Medium (5 ml), and dis-
persed by repeated pipetting. Cell suspensions were added to 500 ml of medium
©2010 Macmillan Publishers Limited. All rights reserved
(Eagle’s minimum essential medium with Earle’s salts and L-glutamine, 10% FBS,
1% non-essential amino acid solution, 1% penicillin and streptomycin). Twenty
five microlitres were dispensed into wells with compounds in Greiner TC-treated,
black 384-well clear bottom plates using a Multidrop dispenser. Seeding density
was adjusted to ensure that new monolayers were not more than , 50% confluent
(typically 3,000 cells per well). Cells were left at 37 uC, 5% CO2 in a humidified
incubator in the presence of 10 mM compound for 48 h. Intracellular ATP was
evaluated using a luciferase-coupled ATP quantification assay (CellTiter-Glo,
Promega) according to the manufacturer’s instructions.
Computational analysis. Molecular frameworks were calculated using an in-
house implementation of the algorithm described previously12. The frameworks
were then used to define clusters of compounds that share a particular frame-
work. To minimise the number of frameworks that describe only a small number
of molecules (,10 in our standard processing), we attempt to reclassify all such
molecules into a smaller framework (that is, the framework describes a smaller
proportion of the structure of the reclassified molecule). Therefore there will
exist some clusters that share a molecular framework, but that are not as struc-
turally homogenous as would be given by simply grouping together compounds
by unique frameworks. Compounds were also classified in chemical families
using the Daylight fingerprint methods with a Tanimoto similarity index of
0.85, following the procedures in the Daylight Information Systems manual13.
Among all compounds tested in the antimalarial assay, ,130,000 had pro-
prietary assay data against one or more human proteins from previous drug
development screening campaigns, including 3,435 of the active compounds,
which were associated to 413 human proteins with potencies matching the
inclusion criteria described in the main text. A further 770 compounds were
associated with specific programmes involving bacterial, malarial or antiviral
targets. To exclude promiscuous and non-specific compounds from the analysis,
an ‘inhibition frequency index’ (IFI) was calculated. This is the relative frequency
with which a compound has scored more than 50% inhibition in an HTS assay,
calculated with the equation:
Number of HTS assays where % inhibition§50%
IFI~ |100, excluding ki-
Total number of HTS assays
nase assays. For example, if a compound was screened in 100 HTS assays, and
showed $ 50% inhibition in 20 of those, the IFI 5 20. The IFI threshold used to
exclude compounds was based on the total number of screens in which they have
been tested, ranging from 5% for compounds tested in . 100 assays, to 20% for
compounds tested in . 25 assays. The threshold above which compound efficacy
against specific human targets was considered significant was defined as
pIC50 $ 7 for inhibition or antagonist assays, and pEC50 $ 6.5, for agonist,
activation, or modulator assays.
Activities at more than 600 target-result type combinations, (some targets are
assayed in both an antagonist and agonist mode) were analysed amongst the
antimalarial compounds, representing potential modes of action. The target
activities for the screened compounds were analysed to identify targets over-
represented amongst the antimalarial actives versus inactives. An ‘enrichment’
was calculated for each possible target-result type combination by dividing the
hit rate for that target amongst antimalarial compounds, by the hit rate at that
target for all compounds screened in the assay:
x=n
Enrichment~
X=N
where x, the number of antimalarial hits with an activity at the target above the
potency threshold; n, the number of the antimalarial hits with a measured pIC50
or pEC50 at the target; X, the number of compounds from the entire screening set
with an activity at the target above the potency threshold; N, the number of
compounds from the entire screening set with a measured pIC50 or pEC50 at the
target.
Compounds originating from programmes in infectious diseases were not
well represented in the historical target assay data. These compounds were
annotated with their originating target and therapeutic class (antimalarial, anti-
bacterial, antiviral, antifungal) and included in the counts and target classes
illustrated in Fig. 2d to provide a more complete view of potential mechanisms
represented among the TCAMS actives.
Literature compounds and target activities in Supplementary Table 2 were
retrieved from the AURQUEST database35. TCAMS actives were used to search
for similar compounds in the database, as defined by Daylight fingerprints and a
Tanimoto similarity coefficient of 0.85 or higher13. SciFinder (version 2007.2,
American Chemical Society) was used to search for compounds active at protein
targets not represented in AURQUEST. As some molecules, particularly kinase
inhibitors, have a very large number of close analogues in the public domain, the
most similar 100 compounds were kept. The biological results were restricted to
those data with a dose–response curve, and a pXC50 of 6 or higher.
doi:10.1038/nature09107
Using BLASTP36 we queried the protein complement of P. falciparum 3D7
genome with RefSeq proteins for all human, bacterial and viral targets accepting
a homology cut-off of an E-value # 1.0 3 10220. The top three hits were collected
for most proteins, except for kinases where more detailed phylogenetic analyses
were performed (see below). Human–malarial homology was confirmed by recip-
rocal BLASTP searches of identified P. falciparum homologues against the human
RefSeq protein databases. P. falciparum gene locus identifiers and annotations
were obtained from PlasmoDB27.
A phylogenetic tree of the combined P. falciparum and Homo sapiens kinomes
was reconstructed using all annotated human and P. falciparum kinases as anno-
tated by Interpro kinase queries of PlasmoDB, as well as from a previously reported
P. falciparum kinome24. Initial multiple sequence alignments were performed using
the program CLUSTALW v1.8 (ref. 37) with default settings and subsequently
refined manually using the program SEQLAB of the GCG Wisconsin Package
v11.0 software package (Accelrys). Redundant P. falciparum kinases were removed
after an initial multiple sequence alignment. Residues that could not be unambigu-
ously aligned or that contained insertions or deletions were removed from the
multiple sequence alignment before phylogenetic analysis. The final kinase multiple
sequence alignment was 160 amino acids in length and included a combined total of
557 human and malarial kinases (available as a Supplementary file).
We constructed phylogenetic trees using distance neighbour-joining (NJ)
method. NJ trees were based on pair-wise distances between amino acid sequences
using the programs NEIGHBOR and PROTDIST (Dayhoff option) of the
PHYLIP 3.67 package38. The programs SEQBOOT and CONSENSE were used
to estimate the confidence limits of branching points from 1,000 bootstrap repli-
cations. All trees were visualized using the program DENDROSCOPE39 (Fig. 3).
28. Trager, W. & Jensen, J. B. Human malaria parasites in continuous culture. Science
193, 673–675 (1976).
©2010 Macmillan Publishers Limited. All rights reserved
29. Makler, M. T. & Hinrichs, D. J. Measurement of the lactate dehydrogenase activity
of Plasmodium falciparum as an assessment of parasitemia. Am. J. Trop. Med. Hyg.
48, 205–210 (1993).
30. Zhang, J. H., Chung, T. D. Y. & Oldenburg, K. R. A simple statistical parameter for
use in evaluation and validation of high throughput screening assays. J. Biomol.
Screen. 4, 67–73 (1999).
31. Desjardins, R. E., Canfield, C. J., Haynes, J. D. & Chulay, J. D. Quantitative
assessment of antimalarial activity in vitro by a semiautomated microdilution
technique. Antimicrob. Agents Chemother. 16, 710–718 (1979).
32. Goodman, C. D., Su, V. & McFadden, G. I. The effects of anti-bacterials on the
malaria parasite Plasmodium falciparum. Mol. Biochem. Parasitol. 152, 181–191
(2007).
33. Ramya, T. N. C., Mishra, S., Karmodiya, K., Surolia, N. & Surolia, A. Inhibitors of
nonhousekeeping functions of the apicoplast defy delayed death in Plasmodium
falciparum. Antimicrob. Agents Chemother. 51, 307–316 (2007).
34. Fidock, D. A., Rosenthal, P. J., Croft, S. L., Brun, R. & Nwaka, S. Antimalarial drug
discovery: Efficacy models for compound screening. Nature Rev. Drug Discov. 3,
509–520 (2004).
35. Aureus Pharma AurSCOPE database. Æhttp://www.aureus-pharma.com/Pages/
Products/Aurscope.phpæ (2004).
36. Altschul, S. F. et al. Gapped BLAST and PSI-BLAST: a new generation of protein
database search programs. Nucleic Acids Res. 25, 3389–3402 (1997).
37. Thompson, J. D., Higgins, D. G. & Gibson, T. J. CLUSTAL W: improving the
sensitivity of progressive multiple sequence alignment through sequence
weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids
Res. 22, 4673–4680 (1994).
38. Felsentein, J. Phylip (phylogenetic inference package) v3.67. Department of
Genetics, University of Washington Æhttp://evolution.genetics.washington.edu/
phylip.htmlæ.
39. Huson, D. H. et al. Dendroscope: An interactive viewer for large phylogenetic
trees. BMC Bioinformatics 8, 460 (2007).
Vol 465 | 20 May 2010 | doi:10.1038/nature09099

ARTICLES
Chemical genetics of Plasmodium falciparum
W. Armand Guiguemde1, Anang A. Shelat1, David Bouck1, Sandra Duffy2, Gregory J. Crowther3, Paul H. Davis4,
David C. Smithson1, Michele Connelly1, Julie Clark1, Fangyi Zhu1, Marı́a B. Jiménez-Dı́az5, Marı́a S. Martinez5,
Emily B. Wilson6, Abhai K. Tripathi7, Jiri Gut8, Elizabeth R. Sharlow9, Ian Bathurst10, Farah El Mazouni11,
Joseph W. Fowble12, Isaac Forquer13, Paula L. McGinley14, Steve Castro14, Iñigo Angulo-Barturen5, Santiago Ferrer5,
Philip J. Rosenthal8, Joseph L. DeRisi6, David J. Sullivan Jr7, John S. Lazo9, David S. Roos4, Michael K. Riscoe13,
Margaret A. Phillips11, Pradipsinh K. Rathod12, Wesley C. Van Voorhis3, Vicky M. Avery2 & R. Kiplin Guy1

Malaria caused by Plasmodium falciparum is a disease that is responsible for 880,000 deaths per year worldwide. Vaccine
development has proved difficult and resistance has emerged for most antimalarial drugs. To discover new antimalarial
chemotypes, we have used a phenotypic forward chemical genetic approach to assay 309,474 chemicals. Here we disclose
structures and biological activity of the entire library—many of which showed potent in vitro activity against drug-resistant P.
falciparum strains—and detailed profiling of 172 representative candidates. A reverse chemical genetic study identified 19
new inhibitors of 4 validated drug targets and 15 novel binders among 61 malarial proteins. Phylochemogenetic profiling in
several organisms revealed similarities between Toxoplasma gondii and mammalian cell lines and dissimilarities between P.
falciparum and related protozoans. One exemplar compound displayed efficacy in a murine model. Our findings provide the
scientific community with new starting points for malaria drug discovery.

The widespread resistance of P. falciparum to many antimalarial drugs, tested against both the chloroquine-sensitive 3D7 strain and the
the dependence of all new drug combinations on artemisinins (for chloroquine-resistant K1 strain, giving 1,134 validated hits that had
which resistance may have emerged)1,2, and new efforts to eradicate saturated dose–response curves. Chemical structure analysis of vali-
malaria all drive the need to develop new, effective and affordable dated hits by topology mapping and clustering10 revealed a wide distri-
antimalarial drugs3. Although our understanding of the parasite’s bio- bution of chemotypes in the active chemical space, with several
logy has increased with the sequencing of the Plasmodium genome4 displaying promising structure–activity relationships (Fig. 1).
and the development of new technologies to study resistance Although all known antimalarial scaffolds (aminoquinolines, quino-
acquisition5–7, few new drug targets or classes of drugs have been lones, bis-amidines) present in the screening collection were identified,
clinically validated8. The lack of publicly accessible antimalarial che- providing positive controls for the screen, most of the chemotypes
motypes with differing modes of action has significantly hindered identified were new. A total of 561 of the validated hits had half-
efforts to discover and develop new drugs9. To address this urgent maximum effective concentration (EC50) values #2 mM against either
need, we have developed a forward chemical genetic approach to 3D7 or K1 and a therapeutic window $10-fold against two mammalian
identify novel antimalarials (Supplementary Fig. 1). cell lines (HepG2 and BJ). From this set, 228 structurally distinct, pure
compounds were re-purchased in powder form for further studies.
The forward chemical genetic screen Antimalarial potencies of ,75% of these compounds (172) were re-
A library containing 309,474 unique compounds, designed at the scaf- confirmed to within tenfold (Bland–Altman analysis, Supplementary
fold level to provide diverse, comprehensive coverage of bioactive Fig. 4) by three laboratories using distinct methods providing the cross-
space10,11, was screened against Plasmodium falciparum strain 3D7 at validated hit set used for all subsequent experiments.
a fixed concentration of 7 mM (Supplementary Information)12. Fidelity
of the assay was examined by receiver operator characteristic (ROC) Combination with antimalarial drugs
analysis and other metrics (Supplementary Figs 2 and 3), demonstrat- Owing to rapid resistance acquisition, the World Health Organiza-
ing good discriminatory power (area under the curve ,0.85) and tion (WHO) recommends combination therapy13. The agonistic and
indicating that a cutoff of $80% activity would retain most of the true antagonistic synergies of the cross-validated set were therefore quan-
positives. The strength of the assay was further determined by testing a tified by measuring EC50 shifts in the presence of a fixed fraction of
set of bioactive compounds including known antimalarials, all of potency (EC10) concentration of chloroquine, mefloquine, artemisinin
which were re-identified (Supplementary Table 3), demonstrating and atovaquone. Most cross-validated compounds were additive in
that the method was very likely to identify any molecule acting by effect or had minor synergies with existing drugs. Two classes demon-
a known mechanism. The primary screen gave approximately 1,300 strated strong synergies (EC50 values reduced by $10-fold): the dia-
hits with activity .80%. These compounds were serially diluted and minonaphthoquinones with artemisinin, and the dihydropyridines
1
Department of Chemical Biology and Therapeutics, St Jude Children’s Research Hospital, Memphis, Tennessee 38105, USA. 2Discovery Biology, Eskitis Institute for Cell and Molecular
Therapies, Griffith University, Brisbane 4111, Australia. 3Department of Medicine, University of Washington, Seattle, Washington 98195-7185, USA. 4Department of Biology and the
Penn Genome Frontiers Institute, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA. 5GlaxoSmithKline, Tres Cantos Medicines Development Campus, Diseases of
Developing World, Tres Cantos 28760, Spain. 6Department of Biochemistry and Biophysics, University of California, San Francisco, California 94158-2542, USA. 7W. Harry Feinstone
Department of Molecular Microbiology and Immunology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland 21205, USA. 8Department of Medicine, San
Francisco General Hospital, University of California, San Francisco, California 94143, USA. 9Department of Pharmacology and Chemical Biology, University of Pittsburgh, Pittsburgh,
Pennsylvania 15261, USA. 10Medicines for Malaria Venture, Geneva 1215, Switzerland. 11Department of Pharmacology, University of Texas Southwestern Medical Center at Dallas,
Dallas, Texas 75390-9041, USA. 12Department of Chemistry, University of Washington, Seattle, Washington 98195-7185, USA. 13Experimental Chemotherapy Lab, Portland VA
Medical Center, Portland, Oregon 97239, USA. 14Department of Chemistry and Chemical Biology Rutgers, The State University of New Jersey, Piscataway, New Jersey 08854, USA.

311
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 465 | 20 May 2010

K1 EC50 SJ000030570
OH
O
3D7 EC50 O N

>4 μM 400 nM 40 nM 4 nM NH

O
N
N O O

Molecule Scaffold Core O

Br
O
O
Core to core (Tanimoto)
O
Core to scaffold (parent/child) R1
O
NH
b
S
N N
Scaffold to scaffold (substructure) O
N
O
O
NH
Scaffold to molecule O N
N O
(parent/child) O
N
S
N O
Molecule to molecule R1
O
(common parent scaffold) O
N
NH Br
R1
SJ000101247 N F
O
O O
Cl R2 NH
O NH O O
O S O
N N
N
O NH S N O
S O
O
N N
R3 R1 N H
N
N O O
O O
O
O NH O
N
R1 R2
N

a O N
N c
O O
R1
O NH
S
O R1
R3
R2
N
R2 R4 R4
R3 N O
O R1 N H R2
N R3
S R4
R3 N
R1 H
R1 N O
O NH S R3
S N
R2
R1
O F
R2 O NH N
N S R3 N O O
R2 H
R1
O R3
O O O
O NH N R2
S R2 N
R4
H
O
R3 N O
H
N O
R3 R3
O SJ000025081

Amodiaquine
OH
I
N
HN HN
HN
N
N N
N Br
N N Cl N
N N
HN
Analogues to N N

known N N
DHOD
inhibitor

3D7 EC50
>4 μM 400 nM 40 nM 4 nM

Figure 1 | Chemical structure network graph and antimalarial potencies of full network graph (bottom half of the figure) indicates that the 1,300
the 1,300 primary screen hits. Topologically similar molecules cluster compounds are organized into clusters of clusters: cores are well sampled by
together in the branches of the network. To construct the graph, molecules multiple scaffolds, and the cores themselves are grouped into families of
were first abstracted to scaffolds and then further to cores using the Murcko related chemotypes. Previously reported antimalarial compounds are
algorithm10. Each of these structural entities is represented as a node, and highlighted in the lower centre. The top half of the figure provides greater
nodes are connected via edges according to topological relationships with detail on three potent chemotypes with well-developed structure activity
closeness being defined using the Tanimoto coefficient. Molecular nodes are relationships: a, tetrahydroisoquinoline; b, diaminonaphthoquinone;
coded to reflect potency against P. falciparum strains K1 (low, white; high, c, dihydropyridine. Data are in Supplementary Information.
blue) and 3D7 (low, small; high, large). The highly branched structure of the

with mefloquine (Fig. 2). One diaminonaphthoquinone and a cyclo- drug development, we also investigated the interaction of the cross-
guanil analogue displayed antagonism with chloroquine and meflo- validated set with 66 potential targets using enzyme inhibition assays
quine, respectively. and thermal melt shift assays (to detect binding).
Three high-priority, well characterized biological targets were
Reverse chemical genetics evaluated in activity assays (Fig. 3, left): P. falciparum dihydroorotate
The advantages of phenotypic screens for the identification of novel dehydrogenase (PfDHOD), haemozoin formation and P. falciparum
chemotypes are that no a priori assumptions are made concerning falcipain-2 (PfFP-2). PfDHOD catalyses the oxidation of dihydroor-
drug targets and that active compounds inherently have cellular bio- otate to orotate in de novo pyrimidine biosynthesis, which is essential
availability. Because insight into the mechanism of action is helpful for for parasite viability14,15. Three compounds inhibited this enzyme: two
312
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
20 nM Artemisinin
25 nM Chloroquine
Δ 3D7 log(EC50)
–2.5 –1.0
Synergy
Figure 2 | Reduced representation of the network map showing synergistic
activities with clinically relevant antimalarials. The size of the nodes
reflects the magnitude of the logarithmic difference between EC50 in the
presence and absence of EC10 of exemplar antimalarial drugs. Absolute
triazolopyrimidines, structurally related to known PfDHOD inhibi-
tors with comparable potencies14, and a dihydropyridine, structurally
related to the calcium blocker felodipine. The potency of these com-
pounds against PfDHOD strongly correlated with their antimalarial
activities (Supplementary Table 5). Furthermore, these compounds
were inactive against transgenic parasites expressing Saccharomyces
cerevisiae dihydroorotate dehydrogenase (Supplementary Table 6).
Next, haemozoin formation inhibition was investigated. The parasite
digests host haemoglobin to provide amino acids, detoxifying the
resulting haem molecules by conversion to a crystallized form known
as haemozoin. Haem detoxification is believed to be the target of many
antimalarial drugs16. Twelve compounds showed appreciable efficacy
in an in vitro haemozoin formation assay17, including analogues of
quinazoline, benzofuran, benzimidazole and carbazole as well as
amodiaquine, a known haemozoin formation inhibitor present in
our library. The correlation between enzyme inhibitory potency and
PfDHOD Haemozoin formation
PfFP-2 Inactive
Figure 3 | Reduced representation of the network map showing the
interaction of the cross-validated hits with potential biological targets.
The network map on the left displays compounds targeting well-validated
protein targets as measured in inhibition assays (EC50 # 15 mM). The map
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES
37 pM Atovaquone
20 nM Mefloquine
1.0 2.5
Antagonism
differences less than one log unit were not considered significant. Synergistic
and antagonistic compounds are uniformly colour-coded blue and red,
respectively. Highly synergistic compounds can be seen with artemisinin and
mefloquine. Data are in Supplementary Information.
antimalarial potency was similar to that displayed by the positive con-
trols quinine and amodiaquine (Supplementary Table 5). The third
enzyme assayed was PfFP-2, which has a critical role in haemoglobin
degradation18. Falcipains are redundant in P. falciparum, with four
known homologues including two (falcipain-2 and falcipain-3) that
seem to have key roles in erythrocytic stage parasites19. Three weakly
active PfFP-2 inhibitors were identified. Thus, 19 compounds (11%)
were inhibitors of validated antimalarial targets.
To expand the pool of potential targets, the compounds were tested
for binding to 61 recombinant malarial proteins (95% purity or better
after affinity and size exclusion chromatography; Supplementary
Table 1) in a thermal melt shift assay20. Fifteen compounds displayed
reproducible thermal shifts with seven malarial proteins (Fig. 3,
right; dissociation constant (Kd) values in Supplementary Table 2):
6-phosphogluconolactonase, 6-pyruvoyltetrahydropterin synthase,
choline kinase, D-ribulose-5-phosphate 3-epimerase, dUTPase, glycogen
PF11_0282 PF14_0020 PF14_0511 PF14_0545
PFC0525c PFL0960w PVX_114505 Inactive
on the right shows compounds that bind to purified malarial proteins
according to thermal melt shift experiments. The size of nodes representing
active or binding compounds is increased for clarity. Data are in
Supplementary Information.
313
ARTICLES NATURE | Vol 465 | 20 May 2010

synthase kinase 3 and thioredoxin. Two compounds bound multiple predict that chemical sensitivity should correlate with evolutionary
proteins. Two out of the seven proteins are in essential malarial path- history, due to homology between key protein targets, as is known to
ways: phosphatidylcholine synthesis21 (choline kinase) and redox be the case for many antiparasitic drugs23,24. Although a few com-
metabolism22 (thioredoxin). The remaining five protein targets poten- pounds showed activity in other parasites, most were highly selective
tially represent novel antimalarial drug targets. for Plasmodium (Fig. 4), whereas Toxoplasma exhibited a chemo-
sensitivity pattern more similar to human cell lines. Similarly,
The potential for cross-resistance the highly potent anti-leishmanial benzothiazoles were only weakly
To evaluate the potential for cross-resistance with existing drugs, the active against the related Trypanosoma. These findings indicate that
cross-validated compounds were tested against a panel of P. falciparum chemical sensitivity of pathogens is regulated by a combination of
strains with different chemosensitivities to known antimalarials, pathogen genetics, physiology and relationships to host and vector
including strains 3D7 (chloroquine sensitive), K1, W2, V1/S and species in vivo.
Dd2 (all resistant to both chloroquine and to antifolates), and SB-A6
and D10_yDHOD (both chloroquine sensitive and atovaquone resist- Early leads for drug development
ant). All strains were profiled for sensitivity to a set of antimalarial To understand the potential for development of the novel chemotypes,
drugs to normalize activity (Supplementary Table 3). A total of 58 the pharmacokinetic properties of the cross-validated set were
cross-validated compounds displayed similar potencies (EC50 assessed. The majority are reasonably drug-like, with 78% of com-
shift #3-fold) against 3D7, K1, V1/S and SB-A6, indicating that these pounds having no violations of the Lipinski rule of five, a well validated
compounds do not share mechanisms of resistance with chloroquine, predictor of oral bioavailability, and 99% having one or fewer viola-
atovaquone or sulphadoxine/pyrimethamine. A subset of the 172 com- tions25. Within the cross-validated set were embedded three chemical
pounds that were inactive against drug-resistant P. falciparum strains series that had multiple members that together gave structure–activity
with known mutations in target proteins were tested against 3D7 dihy- relationships that spanned 1,000-fold potency differences, had con-
drofolate reductase and Plasmodium yoelii cytochrome bc1 complex in sistent activity in drug-resistant strains, had very good cellular thera-
biochemical assays. Two inhibitors were identified for each protein peutic windows, and had at least one member with an EC50 more
(Supplementary Tables 5 and 6). potent than 50 nM. An exemplar compound was selected from each
series and fully profiled using standard models of in vitro and in vivo
Phylochemogenetic profiling adsorption, distribution, metabolism and toxicity (Supplementary
To understand relationships between chemical sensitivity of Plasmo- Table 7). Each possessed reasonable characteristics for developable
dium and related parasites, the cross-validated set was tested against hits; indeed, each comes close to passing Medicines for Malaria
three additional protozoan parasite species—Toxoplasma gondii, Venture (MMV) criteria for ‘late leads’. The compound from these
which belongs to the same phylum as Plasmodium (Apicomplexa), exemplars with the best pharmacokinetic profile was further evaluated
and Leishmania major and Trypanosoma brucei, which are both to measure in vivo antimalarial activity and displayed efficacy in a
Kinetoplastida, unrelated to the Apicomplexa—and an expanded murine malaria model infected with P. yoelii. A twice-daily admini-
panel of human cell lines including a Burkitt’s lymphoma line (Raji) stration of 100 mg kg21 for 3 days resulted in a 90% suppression of the
and embryonic kidney fibroblast cells (HEK293). Phylogenetic criteria parasitaemia (Supplementary Fig. 5). Although it is not suggested that
Eukaryotes
Euglenozoa

Apicomplexa

Chordata

HEK293
Toxoplasma gondii (Tg)
Leishmania major (Lm)
Trypanosoma brucei (Tb)
Plasmodium falciparum (Pf)

Homo sapiens

BJ

HepG2

Raji

Tg

Lm

Tb

V1/S

K1

3D7a

1 pM 1 μM 50 μM

Figure 4 | Phylochemogenetic profiling. Phylochemogenetic analysis of the organism and are clustered according to their chemosensitivity to the
cross-validated compounds using a two-way hierarchical clustering of compounds in the study. A phylogenetic tree of the organisms in this study is
growth inhibition against P. falciparum strains (3D7, K1, V1/S), other provided for reference. Note that despite the many known examples of
eukaryotic parasites (Toxoplasma gondii (Tg), Trypanosoma brucei (Tb), taxonomically conserved pathways, on a global level, phylogeny is a poor
Leishmania major (Lm)) and human cell lines (HEK293, BJ, Raji and predictor of chemical sensitivity profiles: Toxoplasma responses more
HepG2). Columns represent single compounds and are clustered according closely parallel human than their evolutionary siblings, Plasmodium. Data
to potency against the cell lines and organisms. Neighbouring compounds are in Supplementary Information.
share a similar potency spectrum. Rows represent a single cell line or
314
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
any of the compounds discussed herein are bone fide preclinical can-
didates, all provide reasonable starting points for drug development.
Conclusions
Drug therapy remains a key component in controlling malaria. Current
challenges of rapid acquisition of resistance, cross-resistance and
dependence on a limited number of chemical classes of antimalarials
highlight the need to enhance our understanding of the ‘chemical
space’ that can be brought to bear on malaria treatment. Solving this
problem requires understanding the relationships between the struc-
tures of compounds active against malaria parasites, and their potency,
selectivity and targets. We have identified a number of novel com-
pounds and defined these relationships. We expect that these findings
will provide novel paths for drug development and hope that making
this set of well characterized, non-proprietary lead antimalarials
publicly available to the global research community will help to re-
invigorate drug discovery for malaria.
METHODS SUMMARY
The primary screen was carried out by comparing quantities of DNA in treated
and control cultures of Plasmodium falciparum in human erythrocytes after 72 h
incubation with a fixed concentration of 7 mM of the test compounds. The
secondary potency determination was made by using the same assay in a
dose–response mode with 10 concentrations varying from 10 mM to 5 nM.
Chemical sensitivities of the human cell lines and T. brucei were determined
by measuring their ATP content (Cell Titer Glo, Promega). T. gondii parasites
expressing luciferase were cultured and drug sensitivity was determined by
luminescence; L. major promastigote drug susceptibility was tested using a meta-
bolic function assay (Alamar Blue, Promega). Chemicals were assayed for hae-
mozoin formation17 and PfDHOD15 and PfFP-226 inhibitory activities based on
previously described methods. Thermal shift assays were done at compound
concentrations of 25 mM and protein concentrations of 100 mg ml21. All data
processing and visualization, and chemical similarity and substructure analysis,
was performed using custom programs written in the Pipeline Pilot platform
(Accelrys, v.7.0.1) and the R program27. A complete description of the methods
can be found in Supplementary Information.
The Supplementary Information provides a summary of all relevant data
arising from the phenotypic screen and all secondary screens including relevant
diagnostics and details about the following: cell-based, enzyme and thermal shift
screens, data processing, Bland–Altman analysis, and the algorithm to generate
the chemical structure network graph. Chemical structures annotated with assay
data and high-resolution PDFs of the figures may be downloaded from http://
www.stjuderesearch.org/guy/data/malaria/.
Received 20 January; accepted 21 April 2010.
1. Wongsrichanalai, C. & Meshnick, S. R. Declining artesunate-mefloquine efficacy
against falciparum malaria on the Cambodia-Thailand border. Emerg. Infect. Dis.
14, 716–719 (2008).
2. Dondorp, A. M. et al. Artemisinin resistance in Plasmodium falciparum malaria. N.
Engl. J. Med. 361, 455–467 (2009).
3. Ridley, R. G. Medical need, scientific opportunity and the drive for antimalarial
drugs. Nature 415, 686–693 (2002).
4. Kissinger, J. C. et al. The Plasmodium genome database. Nature 419, 490–492 (2002).
5. Baniecki, M. L., Wirth, D. F. & Clardy, J. High-throughput Plasmodium falciparum
growth assay for malaria drug discovery. Antimicrob. Agents Chemother. 51,
716–723 (2007).
6. Plouffe, D. et al. In silico activity profiling reveals the mechanism of action of
antimalarials discovered in a high-throughput screen. Proc. Natl Acad. Sci. USA
105, 9059–9064 (2008).
7. Weisman, J. L. et al. Searching for new antimalarial therapeutics amongst known
drugs. Chem. Biol. Drug Des. 67, 409–416 (2006).
8. Wells, T. N., Alonso, P. L. & Gutteridge, W. E. New medicines to improve control and
contribute to the eradication of malaria. Nature Rev. Drug Discov. 8, 879–891 (2009).
9. Munos, B. Can open-source R&D reinvigorate drug research? Nature Rev. Drug
Discov. 5, 723–729 (2006).
10. Shelat, A. A. & Guy, R. K. Scaffold composition and biological relevance of
screening libraries. Nature Chem. Biol. 3, 442–446 (2007).
11. Shelat, A. A. & Guy, R. K. The interdependence between screening methods and
screening libraries. Curr. Opin. Chem. Biol. 11, 244–251 (2007).
12. Smilkstein, M. et al. Simple and inexpensive fluorescence-based technique for
high-throughput antimalarial drug screening. Antimicrob. Agents Chemother. 48,
1803–1806 (2004).
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES
13. Cibulskis, R. E. et al. Estimating trends in the burden of malaria at country level.
Am. J. Trop. Med. Hyg. 77 (suppl. 6), 133–135 (2007).
14. Gujjar, R. et al. Identification of a metabolically stable triazolopyrimidine-based
dihydroorotate dehydrogenase inhibitor with antimalarial activity in mice. J. Med.
Chem. 52, 1864–1872 (2009).
15. Patel, V. et al. Identification and characterization of small molecule inhibitors of
Plasmodium falciparum dihydroorotate dehydrogenase. J. Biol. Chem. 283,
35078–35085 (2008).
16. Weissbuch, I. & Leiserowitz, L. Interplay between malaria, crystalline hemozoin
formation, and antimalarial drug action and design. Chem. Rev. 108, 4899–4914
(2008).
17. Pisciotta, J. M. et al. The role of neutral lipid nanospheres in Plasmodium falciparum
haem crystallization. Biochem. J. 402, 197–204 (2007).
18. Sijwali, P. S. & Rosenthal, P. J. Gene disruption confirms a critical role for the
cysteine protease falcipain-2 in hemoglobin hydrolysis by Plasmodium falciparum.
Proc. Natl Acad. Sci. USA 101, 4384–4389 (2004).
19. Sijwali, P. S., Koo, J., Singh, N. & Rosenthal, P. J. Gene disruptions demonstrate
independent roles for the four falcipain cysteine proteases of Plasmodium
falciparum. Mol. Biochem. Parasitol. 150, 96–106 (2006).
20. Crowther, G. J. et al. Buffer optimization of thermal melt assays of Plasmodium proteins
for detection of small-molecule ligands. J. Biomol. Screen. 14, 700–707 (2009).
21. Witola, W. H. et al. Disruption of the Plasmodium falciparum PfPMT gene results in
a complete loss of phosphatidylcholine biosynthesis via the serine-
decarboxylase-phosphoethanolamine-methyltransferase pathway and severe
growth and survival defects. J. Biol. Chem. 283, 27636–27643 (2008).
22. Krnajski, Z. et al. Thioredoxin reductase is essential for the survival of Plasmodium
falciparum erythrocytic stages. J. Biol. Chem. 277, 25970–25975 (2002).
23. McFadden, G. I. & Roos, D. S. Apicomplexan plastids as drug targets. Trends
Microbiol. 7, 328–333 (1999).
24. Reynolds, M. G. & Roos, D. S. A biochemical and genetic model for parasite
resistance to antifolates. Toxoplasma gondii provides insights into pyrimethamine
and cycloguanil resistance in Plasmodium falciparum. J. Biol. Chem. 273,
3461–3469 (1998).
25. Lipinski, C. A., Lombardo, F., Dominy, B. W. & Feeney, P. J. Experimental and
computational approaches to estimate solubility and permeability in drug
discovery and development settings. Adv. Drug Deliv. Rev. 46, 3–26 (2001).
26. Shenai, B. R. et al. Structure-activity relationships for inhibition of cysteine
protease activity and development of Plasmodium falciparum by peptidyl vinyl
sulfones. Antimicrob. Agents Chemother. 47, 154–160 (2003).
27. Ritz, C. & Streibig, J. C. Bioassay analysis using R. J. Stat. Softw. 12, 22 (2005).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements This work was supported by the American Lebanese Syrian
Associated Charities (ALSAC) and St Jude Children’s Research Hospital (SJCRH,
R.K.G.), the Medicines for Malaria Venture (W.C.V.V. and V.M.A.), National
Institute of Allergy and Infectious Diseases (AI772682 (P.H.D.), AI075517
(R.K.G.), AI067921 (W.C.V.V.) and AI080625 (W.C.V.V.), AI28724 (D.S.R.),
AI53862 (J.L.D.), AI35707 (P.J.R.), AI053680 (M.A.P. and P.K.R.), AI075594
(M.A.P., P.K.R. and I.B.), AI082617 (P.K.R.) and AI045774 (D.J.S.)), the National
Cancer Institute (CA78039 (J.S.L.)), the Welch Foundation (I-1257 (M.A.P.)), the
Doris Duke Charitable Foundation (P.J.R.), and the Ellison Medical Foundation
(D.S.R.). We acknowledge A. B. Vaidya for providing the parasite strain
D10_yDHOD. We acknowledge M. Sigal for assistance in the early leads project
coordination, the SJCRH High Throughput Screening Center, particularly J. Cui; the
SJCRH Lead Discovery Informatics Center, and the SJCRH High Throughput
Analytical Chemistry Center, particularly C. Nelson and A. Lemoff; at UW,
F. Buckner, W. Hol and A. Napuli (AI067921, W. Hol); S. Wei and W. Hao in the UT
Southwestern HTS Center; and the Australian Red Cross Blood Service for the
provision of O1 erythrocytes to Griffith University.
Author Contributions W.A.G. and R.K.G. designed and coordinated the project.
A.A.S. wrote the algorithms for the data analysis and generated the figures. Assays
were conceived, performed and analysed by W.A.G. and D.B. (P. falciparum
phenotypic screen), M.C. (human cell lines), D.C.S. (T. brucei), P.H.D. and D.S.R. (T.
gondii), J.S.L. and E.R.S. (L. major), A.K.T. and D.J.S. (haemozoin inhibition), G.J.C.
and W.C.V.V. (thermal melt experiments), M.A.P., P.K.R., F.E.M. and I.B.
(PfDHOD), J.W.F. and P.K.R. (P. falciparum dihydrofolate reductase), J.G. and P.J.R.
(PfFP-2), I.F. and M.K.R. (cytochrome bc1), J.C. (P. falciparum mutant drug
sensitivity). E.B.W., S.D., J.L.D. and V.M.A. (independent antimalarial in vitro
experiments), F.Z. (in vitro pharmacokinetics), M.B.J.D., M.S.M., I.A.-B. and S.F. (in
vivo pharmacokinetics and efficacy), I.B. (coordination of technology development
and network development), S.C. and P.L.M. (re-synthesis). W.A.G., A.A.S. and
R.K.G. wrote the manuscript. All authors contributed to the design of the
experiments and the preparation of the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to R.K.G. (kip.guy@stjude.org).
315
 Vol 465 | 20 May 2010 | doi:10.1038/nature08977

ARTICLES
Cell signalling by microRNA165/6 directs
gene dose-dependent root cell fate
Annelie Carlsbecker1,2*, Ji-Young Lee3,4*, Christina J. Roberts2, Jan Dettmer1, Satu Lehesranta1, Jing Zhou3,4,
Ove Lindgren1,5, Miguel A. Moreno-Risueno6, Anne Vatén1, Siripong Thitamadee1, Ana Campilho1, Jose Sebastian3,
John L. Bowman7, Ykä Helariutta1* & Philip N. Benfey6*

A key question in developmental biology is how cells exchange positional information for proper patterning during organ
development. In plant roots the radial tissue organization is highly conserved with a central vascular cylinder in which two
water conducting cell types, protoxylem and metaxylem, are patterned centripetally. We show that this patterning occurs
through crosstalk between the vascular cylinder and the surrounding endodermis mediated by cell-to-cell movement of a
transcription factor in one direction and microRNAs in the other. SHORT ROOT, produced in the vascular cylinder, moves
into the endodermis to activate SCARECROW. Together these transcription factors activate MIR165a and MIR166b.
Endodermally produced microRNA165/6 then acts to degrade its target mRNAs encoding class III homeodomain-leucine
zipper transcription factors in the endodermis and stele periphery. The resulting differential distribution of target mRNA in
the vascular cylinder determines xylem cell types in a dosage-dependent manner.

Organ development involves extensive communication between cells one ground tissue layer12,16,17. Despite the endodermal-specific
to orchestrate tissue specification and differentiation. This communi- expression of SCR genome-wide mRNA profiling of wild-type, shr
cation is often mediated by mobile molecules such as hormones, and scr roots14,15 (Methods) indicated that nearly half of the genes co-
mRNAs, proteins and small RNAs1. In plants and animals, transcrip- regulated by SHR and SCR were expressed at the highest level in the
tion factors and/or mRNAs have been shown to move to neighbouring stele (Supplementary Fig. 2). Furthermore, in both scr and shr
cells and transfer positional information2–6. Recently, small RNAs mutants, metaxylem differentiates ectopically in the place of proto-
including microRNAs (miRNAs), small interfering RNAs (siRNAs) xylem (Fig. 1e, f, g), indicating that SHR and SCR affect stele develop-
and trans-acting siRNAs (tasiRNAs) have emerged as potential medi- ment in a non-cell autonomous manner.
ators of cell-to-cell communication. Studies on the mobility of small Because SHR is normally present in both the stele and the endo-
RNAs have established that siRNAs and tasiRNAs are mobile, but dermis, we determined where its activity is required for xylem pat-
evidence for the mobility of miRNAs has remained elusive2–7. terning. We first expressed SHR strictly in the stele of shr-2 by
The plant root is well suited to study cell-to-cell communication in introducing a construct in which a non-mobile version of SHR18
tissue patterning. Along the radial axis, epidermis, cortex and endo- containing a nuclear localization signal was driven by the stele-
dermis form around the stele, in which the pericycle surrounds the specific promoter of CRE1 (ref. 19). Recovery of root meristem size
vascular cylinder8 (Fig. 1a). Here, xylem develops in the centre and (Student’s t-test; P , 0.001, a 5 0.05) and root growth (P , 0.001,
forms symmetric arcs towards the pericycle. Phloem develops between a 5 0.05) showed that this version of SHR was functional. Although
these arcs, with procambium/cambium, the vascular stem cells sepa- we observed more immature phloem sieve cells than in shr-2 (Sup-
rating xylem and phloem. Xylem precursors in the centre differentiate plementary Fig. 3), protoxylem formation was not rescued (Fig. 1b).
into metaxylem with pitted secondary cell walls, while peripheral pre- We then expressed SHR strictly in the ground tissue by introducing
cursors differentiate as protoxylem with spiral walls. The strong evolu- UAS::SHR:YFP into shr-2 harbouring J0571, an enhancer trap line
tionary conservation of xylem patterning9,10 suggests the presence of that drives expression specifically in the ground tissue (www.plantsci.
common molecular mechanisms that constrain its organization. cam.ac.uk/Haseloff/; Fig. 1c and Supplementary Fig. 4). In addition
Here we report a novel regulatory pathway (Supplementary Fig. 1) to multiple ground tissue cell divisions, protoxylem formation was
that involves bidirectional cell signalling mediated by miRNA165/6 observed in 73% (24 of 33) of the lines (Fig. 1c), compared with 14%
(miR165/6) and the transcription factors SHORT ROOT (SHR) and in shr-2 J0571 alone. When SCR was expressed in the ground tissue in
SCARECROW (SCR) controlling xylem patterning. the absence of SHR, neither protoxylem nor endodermis was rescued
(Fig. 1d). Therefore, xylem patterning requires both SHR and SCR to
Endodermal SHR controls xylem patterning be present in the endodermis.
SHR is produced in the stele and moves into the endodermis to
activate SCR11–15. In mutants of SHR or SCR, the asymmetric cell Xylem patterning requires PHB restriction
division that forms endodermis and cortex fails to occur and the In a screen for altered vascular development (Supplementary Fig. 5)
quiescent centre is not maintained, resulting in short roots with only we identified a mutant with a short root and frequent differentiation
1
Institute of Biotechnology/Department of Biosciences, University of Helsinki, FIN-00014, Finland. 2Department of Physiological Botany, Evolutionary Biology Center, Uppsala
University, Norbyvägen 18D, SE-752 36 Uppsala, Sweden. 3Boyce Thompson Institute for Plant Research, Tower Road, Ithaca, New York 14853, USA. 4Graduate Field of Plant Biology,
Cornell University, Ithaca, New York 14853, USA. 5Institute of Technology, University of Tartu, Tartu 50411, Estonia. 6Biology Department and IGSP Center for Systems Biology, Duke
University, Durham, North Carolina 27708, USA. 7School of Biological Sciences, Monash University, Melbourne, Victoria 3800, Australia.
*These authors contributed equally to this work.

316
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 ARTICLES

a b c d
* *

CRE1::SHR∆NLELDV: UAS::SHR:YFP UAS::SCR:YFP

nlsGFP; shr in J0571; shr in J0571; shr
* *
* *
* *
*
* *
* *
Pericycle * *
Quiescent centre Protoxylem * *
* *
Endodermis/cortex initial Metaxylem * *
* Endodermis Phloem * *
Cortex Procambium
e* * f * * g * h * * i *
* * * * *
* * *
*
* * * * *
* *
* * *
* * * *
* * *
* * * * * * * *
* *

WT shr scr phb-7d phb shr

Figure 1 | Endodermal SHR and SCR control xylem patterning via PHB. stained cross-sections and confocal laser scanning micrographs of basic
a, Schematic representation of the Arabidopsis root meristem and stele. Cells fuchsin-stained xylem of wild type (WT) (e), shr-2 (f), scr-4 (g), phb-7d
in the ground tissue layer adjacent to the stele are marked with asterisks. (h), and phb-6 shr-2 (i). Filled arrowhead indicates metaxylem, and unfilled
b, CRE1::SHRDNLELDV:nlsGFP in shr-2. c, UAS::SHR:YFP in shr-2 indicates protoxylem. Scale bars, 10 mm.
harbouring J0571. d, UAS::SCR:YFP in shr-2, J0571. e–i, Toluidine blue-
of metaxylem in place of protoxylem (Fig. 1h). This mutant had a Consistent with the phenotype of this new allele, phb-7d, the strong
point mutation in the miR165/6 target site in the class III homeo- gain-of-function allele phb-1d (ref. 20) invariably formed ectopic
domain leucine zipper (HD-ZIP III) gene PHABULOSA (PHB). metaxylem (Supplementary Fig. 5e).
Comparison of vascular development using cell and tissue markers
a b reinforced the striking similarity of the phb-7d and shr-2 phenotypes
* (Supplementary Fig. 6 and Supplementary Table 1). To determine if
* * * * *
* * * *
the phb-7d phenotype was caused by reduced SHR activity, we
* * * expressed the green fluorescent protein fusion SHR::SHR:GFP13 in
* * * * * *
* * * phb-7d (Supplementary Fig. 7) but found no deviation from wild-
* * * * * type expression, indicating that this was unlikely.
*
As expected if the phb-7d mutation renders the PHB transcript
PHB as, WT PHB as, phb-7d
resistant to miRNA-mediated degradation, its mRNA levels were
c * d
*
e * elevated in the mutant (Supplementary Fig. 5d). RNA in situ hybridi-
* * *
* * * zation in wild type showed that PHB mRNA localized primarily to the
* * * * metaxylem precursors and neighbouring procambial cells, and at a
*
* residual level in protoxylem precursors (Fig. 2a). In phb-7d the PHB
* * * * mRNA domain expanded throughout as well as outside the stele
*
* * (Fig. 2b), indicating that miR165/6 normally acts to exclude PHB
* * * *
CNA as * ATHB8 as* REV as * mRNA from the stele periphery and ground tissue.
* The importance of miRNAs in xylem cell specification was sup-
f g sense
ported further by the ectopic metaxylem phenotype of the mutant of
* * * HYL1, which specifically binds the miRNA during miRNA-mediated
* * *
* * * * mRNA degradation21,22 (Supplementary Fig. 8). We did not detect
* * *
* * * * ectopic metaxylem in mutants of several well-characterized genes in
* * *
PHB::GFP PHB::GFP
* *
WT shr-2 PHB as, shr-2
Figure 2 | SHR post-transcriptionally represses PHB. a, b, In situ
h i * hybridization with a PHB mRNA specific probe on cross- and longitudinal
* sections of WT (a) and phb-7d (b) roots. c–e, In situ hybridization with CNA
* * * * * (c), ATHB8 (d) and REV (e) specific probes on cross-sections of WT roots.
* * *
f, Confocal laser scanning micrographs of transcriptional fusion of PHB to GFP
* * *
* * * in WT and shr-2. g, PHB mRNA in situ hybridization to cross-section of shr-2.
* * Inset is a section of shr-2 hybridized with a PHB sense probe. h, Expression of
translational fusion of PHB to GFP in WT and shr-2. i, Expression of
PHB::PHB:GFP, WT PHB::PHB:GFP, shr-2 PHB::phb-7d:GFP translational fusion of PHB with the phb-7d mutation to GFP. Asterisks,
endodermis position; arrowheads, protoxylem position; scale bars, 10 mm.
317
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES NATURE | Vol 465 | 20 May 2010

the si- and tasiRNA-mediated RNA degradation pathways, indi- region of eight of the MIR165/6 genes (except 165b) were analysed
cating that xylem patterning is primarily mediated by miRNAs. in wild type, shr-2 and scr-1 (Fig. 3a, b and Supplementary Fig. 14).
Only MIR165a and 166b promoters drove detectable GFP in distinct
SHR post-transcriptionally represses PHB patterns in wild-type roots. We note that another study28 suggests
Genome-wide expression profiling of shr, scr and wild-type roots that MIR166a is also expressed at low levels in roots. pMIR165a::GFP
indicated that PHB and the other four HD-ZIP III genes targeted showed endodermis-specific expression throughout the root
by miR165/6 (refs 23–25) are upregulated in the shr and scr mutant (Fig. 3a), whereas pMIR166b::GFP was expressed strongly in the
backgrounds (Supplementary Fig. 9). To characterize further the endodermis and quiescent centre and weakly in cortex and epidermis
relationship between SHR and the HD-ZIP III transcription factors of the meristems of embryonic, primary and lateral roots (Fig. 3b and
we crossed shr-2 with loss-of-function mutants of three HD-ZIP III Supplementary Fig. 14). In shr roots, GFP was not detected from
transcription factors, PHB, PHAVOLUTA (PHV) and REVOLUTA pMIR165a::GFP in any tissue, and only low-level activity was
(REV), which are closely related and functionally redundant in leaf observed in the ground tissue and epidermis from pMIR166b::GFP
development23. The phb-6 phv-5 rev-9 mutant did not show any (Fig. 3b and Supplementary Fig. 14). The constructs behaved simi-
major deviation in xylem patterning26 from wild-type roots (Sup- larly in scr, showing marked reduction in expression, with the excep-
plementary Figs 10 and 19). However, in the phb-6 phv-5 rev-9 shr-2 tion that pMIR165a::GFP was detected in the ground tissue from the
quadruple mutant, the xylem patterning defect of shr was completely late maturation zone (Fig. 3a, b and Supplementary Fig. 14). As in
rescued (Supplementary Fig. 10). This provides strong evidence that wild type, GFP driven by any of the other six promoters could not be
ectopic metaxylem formation in shr is the result of upregulation of at detected in shr roots. Consistent with the GFP data, real time PCR
least one of these HD-ZIP III transcription factors. This quadruple with reverse transcription analysis (RT–PCR) indicated significant
mutant also largely rescued the number of vascular cell files and root reduction of pri-MIR165a and pri-MIR166b in both shr and scr roots
length (Supplementary Fig. 10). Analysis of the segregating double (Supplementary Fig. 13).
mutants showed that phb shr fully recovered protoxylem (in about To determine whether SHR is a direct regulator of MIR165/6 we
80%) and root growth, whereas phv shr, rev shr and phv rev shr did not performed chromatin immunoprecipitation (ChIP) followed by
(Fig. 1i and Supplementary Figs 10 and 11). Loss-of-function mutations real-time quantitative PCR (Methods). We reproducibly found
in the two other HD-ZIP III genes, ATHB8 and CORONA/ATHB15 enrichment of fragments approximately 1 kb upstream of the tran-
(CNA) could not restore the root growth of shr-2, but in both double scription start site of MIR165a, 4.5 kb upstream of MIR166b (Fig. 3c),
mutants, athb8-11 shr-2 and cna-2 shr-2, stretches of protoxylem were and 2.5 kb upstream of MIR166a (data not shown). Taken together,
infrequently observed (Supplementary Fig. 11). Combining mutation our data indicate that SHR activates transcription of MIR165a and
in phb with scr also fully recovered protoxylem (Supplementary Fig. 11).
166b in the endodermis directly.
Hence, these genetic analyses indicate that SHR/SCR repression of PHB
is the primary pathway for xylem patterning. miR165/6 acts non-cell-autonomously
We asked whether PHB was repressed by SHR at the transcrip-
tional or post-transcriptional level. For this we analysed transcrip- We reported previously that the HD-ZIP III transcriptional domains
tional (PHB::GFP) and translational GFP fusions (PHB::PHB:GFP) are largely restricted to the vascular cylinder27. PHB is the only one
driven by the PHB promoter27. Patterns of PHB::GFP indicated that
PHB is transcribed throughout the stele and endodermis (ground a
tissue in shr) in both wild-type and shr root meristems (Fig. 2f), * * * *
* * * *
whereas PHB::PHB:GFP was restricted to the central vascular cylin- *
der in wild type (Fig. 2h). When the phb-7d mutation was introduced *
into the cDNA of the translational fusion, the GFP domain became
similar to the PHB transcriptional domain in wild type (Fig. 2i). In
shr, PHB::PHB:GFP and PHB mRNA domains expanded throughout
the stele, similar to the pattern of PHB::GFP (Fig. 2g, h). These data pMIR165a::GFP WT scr-1 scr-1
indicate that SHR restricts PHB at the post-transcriptional level. b * * * *
These results further implied that the meristematic zone is where * * * * *
* * *
PHB determines the root xylem cell types. Supporting this, we found
that a miRNA-resistant version of PHB (with a silent mutation)
expressed in the meristematic region under the stele-specific CRE1
promoter was sufficient for ectopic metaxylem to form both in wild
type and in the phb shr double mutant (Supplementary Fig. 12).
In summary, our data indicate that miR165/6 post-transcriptionally pMIR166b::GFP WT scr-1 shr-2
restricts PHB within the root meristem to the stele centre for proper c MIR165a MIR166b
xylem patterning and that SHR regulates this process by promoting 16 6
miR165/6 activity in the stele periphery and endodermis. 14 5
Fold enrichment

SHR activates miR165/6 in the endodermis
We compared levels of miR165/6 in root tips of shr, scr and wild type. In
shr, miR165/6 levels were reduced eightfold, and in scr threefold com-
pared to wild type (Supplementary Fig. 13; Methods). Comparison of
gene expression profiles of components related to small RNA pathways
in whole roots14,15 showed no statistically significant expression
changes in these mutant backgrounds (Supplementary Table 2), indi-
cating that SHR and SCR primarily control miR165/6 activity by regu-
lating MIR165/6 expression.
To determine which of the MIR165/6 genes are controlled by SHR
and SCR, expression patterns of transgenic promoter::GFP (endo-
plasmic reticulum-localized) lines with the complete intergenic
318
©2010 Macmillan Publishers Limited. All rights reserved
12
10 4
8 3
6
4
2
2 1
0 0
–0.8 –1.4 –2.7 –3.6 MGP –0.4 –1.6 –3.0 –3.6 –4.4 –4.7 –5.0 –5.7 –6.5
Distance from the transcription start site (kb)
Figure 3 | miR165/6 activated by SHR in the endodermis is active in the stele.
a, Expression of pMIR165a::GFP in WT and scr-1 meristems and in maturation
zone. b, Expression of pMIR166b::GFP in WT, scr-1 and shr-2 meristems.
c, Real-time PCR of ChIP on the upstream regulatory regions of MIR165a and
MIR166b using anti-GFP antibodies and a transgenic plant expressing
pSHR::SHR:GFP. The binding to the MAGPIE (MGP) promoter confirmed
previously21 was used as positive control. Asterisks, endodermis position.
NATURE | Vol 465 | 20 May 2010 ARTICLES

whose transcriptional domain partially overlaps with the endodermal but became progressively higher and ubiquitous throughout the root
activity domain of SHR and SCR. Hence, for miR165/6 produced in radius 30–40 mm distal from the quiescent centre. In cells close to the
the endodermis to encounter HD-ZIP III mRNA in the stele, there quiescent centre, mature miR165/6 levels were considerably higher in
must be a mobile signal. the cortical and epidermal cell layers than in the endodermis and stele
To understand the nature of this non-cell autonomous regulation, cells. Hence, the pattern seemed complementary to the domain of
we first compared the spatial distribution of miR165/6 activity in PHB transcription. A complementary pattern of a mature miRNA
wild-type and shr-2 roots using a ‘miRNA-sensor’29 (Fig. 4a; and its target is consistent with previous observations30, but rather
Methods). In this system, high miR165/6 activity is reflected by low contradictory to our results showing the highest promoter activity of
GFP expression. Without the miRNA recognition site the GFP levels MIR165a and MIR166b in the endodermis. We proposed that the
were uniform in the stele in both wild type and shr. In wild-type roots hybridization signal may reflect mainly the distribution of free
the sensor GFP with the miRNA recognition site was significantly miR165/6. Consistent with this, we detected less of a differential
lower in the ground tissue and stele periphery than in other cell types, distribution of miR165/6 in the phb-13 mutant and in several loss-
consistent with the observed reduction in the PHB mRNA and of-function HD-ZIP III mutants where target mRNA is very low or
protein domains in wild type. In shr-2, the sensor GFP expression absent. In these backgrounds the miR165/6 signal became high and
level was uniform throughout the root. ubiquitous throughout the root radius, even in tissues close to the
Second, we determined the spatial distribution of mature miR165/ quiescent centre (Fig. 4b and Supplementary Fig. 15). In shr-2, the
6 in the root meristem of wild type and shr using in situ hybridization miR165/6 pattern was similar to wild type, but at a markedly lower
with locked nucleic acid (LNA) probes (Fig. 4b and Supplementary level, probably reflecting the residual expression of MIR166b
Fig. 15; Methods). In wild-type meristems, mature miR165/6 was (Fig. 4b). Hence, the sensor and in situ hybridization results indicate
detected at a low level in the quiescent centre and surrounding cells, that the mature miRNA165/6 moves radially both outward and
inward from the endodermis.
Third, we drove MIR165a expression by a ground-tissue-specific
a U2::GFP U2::MIR165/6-GFP promoter (shr-2 J0571; UAS::MIR165a) in shr and observed the result-
* * *
ing xylem pattern (Fig. 4c and Supplementary Fig. 4). Five independent
* * * *
* * * * segregating T2 lines showed a clear recovery of protoxylem (Fig. 4c) at
* * *
* *
frequencies ranging from 33% to 88%, accompanied by suppression of
mRNA levels of all the HD-ZIP IIIs (Fig. 4d) and restriction of PHB and
CNA mRNA domains within the stele (Supplementary Fig. 16). Co-
segregation of protoxylem formation with the activator was verified by
backcrossing. Similarly, both J0571 ? MIR165a (Fig. 4c) in scr-4 and
WT shr-2 WT shr-2
MIR165a expressed under another ground-tissue-specific promoter,
b pSCR, in the scr-6 allele, rescued protoxylem formation (Supplemen-
tary Fig. 17).
*
* * * * * * * * Further support for this hypothesis is that protoxylem recovery
* * * * *
observed when SHR was specifically expressed in the ground tissue of
* *
* * * shr (Fig. 1c) was accompanied by an increase in MIR165a and 166b
* * *
*
* levels, and a decrease in mRNA levels of all five HD-ZIP III genes
WT athb8 cna phb phv rev shr-2
(Supplementary Fig. 18a), whereas SHR exclusively localized to the
stele of the shr mutant did not affect either miR165/6 or HD-ZIP III
c * * mRNA levels (Fig. 1d and Supplementary Fig. 18b). Taken together,
* * our results strongly support that miR165/6 in the endodermis
* *
* mediates the non-cell-autonomous action of SHR and SCR in xylem
* patterning, by moving into the stele to restrict the HD-ZIP III mRNA
* domains.
*
UAS::MIR165a
* *
UAS::MIR165a * UAS::MIR165a * HD-ZIP III levels determine xylem type
shr; J0571 shr; J0571 scr; J0571 Our results indicated that PHB was the primary determinant, but
*
d other HD-ZIP IIIs may have a role in metaxylem specification as well.
6 To investigate further the role of all five HD-ZIP III genes in root
ΔCt (target) – ΔCt (control)

5
vascular patterning, we analysed their expression and assessed their
4
3
loss-of-function phenotypes. Similar to PHB, CNA and ATHB8 were
expressed in xylem precursor cells (Fig. 2c, d): CNA in the metaxylem
ATHB8

2
166d
166c

CNA
PHB

PHV

REV

1 domain and neighbouring procambial cells and ATHB8 specifically

0 in the xylem axis including the protoxylem precursors. REV was
–1 broadly expressed in the vascular tissue just above the quiescent
miR165/6

165a

166a

166b

166e
Mature

–2
–3
–4
Col-WT
Figure 4 | Non-cell-autonomous action of MIR165a. a, miR165/6 GFP
sensor under the U2 promoter in WT and shr-2. b, miR166-specific LNA
probe hybridization to sections proximal to the quiescent centre of WT,
athb8-11 cna-2 phb-13 phv-11 rev-6 and shr-2. c, Protoxylem forms in shr
and scr backgrounds when UAS::MIR165a is introduced into shr-2 and scr-4
harbouring J0571. d, Real-time RT–PCR of pri-MIR165/6 and HD-ZIP III in
WT and a line with UAS::MIR165a in shr-2, J0571. n 5 4. Error bars indicate
6s.d. Asterisks, endodermis/ground tissue position; arrowheads,
protoxylem position; scale bars, 10 mm.
centre (Fig. 2e), but disappeared from the metaxylem domain farther
away from the quiescent centre. PHV mRNA was not detected. In
UAS::MIR165a in shr-2; J0571
summary, the HD-ZIP III mRNA levels in the root meristem appear
highest in the centre of the xylem axis and become lower towards the
stele periphery.
In contrast with ectopic PHB expression, which causes central
metaxylem fate in the stele periphery, the loss of combinations of
HD-ZIP III genes resulted in protoxylem differentiating in the central
metaxylem positions or even abolished xylem differentiation entirely
(Fig. 5 and Supplementary Fig. 19). This is consistent with previous
studies showing that HD-ZIP III transcription factors may direct xylem
development23,25,31–33. All single and most double mutants displayed
319
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES
a b
* *
* For anatomical and histological analyses, primary roots of vertically grown 4 to
* 5-day-old seedlings were used. Plastic sectioning, basic fuchsin staining and
*
* *
* *
* *
* synthesis system for RT–PCR (Invitrogen). For mature miRNA, a polyadenyla-
* *
* * * *
* method of ref. 39. Differences in gene expression were measured by real time
* PCR using an ABI 7900HT (Applied Biosystems).
cna phb phv rev athb8 cna phb phv rev
Figure 5 | HD-ZIP III levels determine xylem type. a, Basic fuchsin-stained
xylem and cross-section of cna-2 phb-13 phv-11 rev-6. b, Cleared root and
cross-section of athb8-11 cna-2 phb-13 phv-11 rev-6. c, d, Stained xylem of
roots in which MIR165a is expressed from the CRE1 promoter in WT and
shr-2. Asterisks, endodermis position; arrowheads, protoxylem position;
scale bars, 10 mm.
normal xylem patterning. However, the athb8-11 phb-13 and various
triple mutants had ectopic protoxylem partly replacing metaxylem.
When four of the five genes were mutated no metaxylem was observed
in any of the mutant combinations examined. Consistent with changes
in xylem cell fates, low or no expression of the metaxylem marker gene
ACL5 (ref. 34) and ectopic expression in the centre of the xylem axis of
the protoxylem marker AHP6 (ref. 35) were observed in athb8-11 cna-2
phb-13 phv-11 (Supplementary Fig. 20). This mutant also generated
more vascular cells and, in contrast to the invariably diarch vascular
arrangement in wild type, often formed a tri- or tetrarch arrangement,
indicating that the HD-ZIP III genes redundantly restrict vascular cell
proliferation (Supplementary Fig. 19). Surprisingly, loss of all five HD-
ZIP III transcription factors failed to form any xylem (Fig. 5b). Partial
failure in xylem differentiation was also detected in certain quadruple
mutants (Supplementary Fig. 19). These results show that expression
levels of HD-ZIP III transcription factors determine not only xylem cell
types, but also de novo formation of xylem.
Finally, we increased the level of miR165 throughout the stele in
wild type and shr-2. In both backgrounds an increased number of
stele cells was observed (not shown) and all xylem precursors
acquired peripheral fate differentiating exclusively as protoxylem
(Fig. 5c, d), similar to the phenotypes of several multiple loss-of-
function HD-ZIP III mutants. Thus, xylem patterning requires sup-
pression of HD-ZIP III mRNA in the stele periphery through the
activity of miR165/6.
Discussion
Our study highlights a novel regulatory pathway that integrates tran-
scriptional and post-transcriptional regulation and bidirectional cell-
to-cell communication to drive tissue patterning in the Arabidopsis
root (Supplementary Fig. 1). Formation of vascular tissue with a
surrounding endodermal layer was a key milestone in the evolution
of land plants17,18. Our study shows that its underlying regulation
involves evolutionarily conserved SHR/SCR and HD-ZIP III tran-
scription factors, and miR165/6 (refs 36, 37), implying that this
regulatory mechanism might underlie the evolutionary adaptation
to terrestrial growth.
The mobility of miR165/6 in the shoot apical meristem has been
suggested previously2–6. Our study indicates that miR165/6 is mobile
in the root and its mobility over a short distance is critical for dosage-
dependent regulation of HD-ZIP III transcription factors in xylem
patterning. A recent modelling study indicates that a mobile small
RNA can sharpen the boundary of the activity domain of its target38.
Our study shows that this may be the role for the endodermally
320
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
c produced miR165/6 as it moves into the vascular cylinder to encounter
its target mRNA, thereby communicating radial positional informa-
tion between cells of the root meristem.
METHODS SUMMARY
CRE1::MIR165a, WT confocal imaging were performed as described19.
d For quantification of mRNA and miRNA, root tips from 6- to 7-day-old
seedlings were collected and total RNA including small RNAs was extracted with
an miRNeasy Mini kit (Qiagen). To measure the expression level of pri-miRNA
and other mRNAs, cDNA was synthesized using SuperScript III first strand
tion reaction was performed before the reverse transcription following the
CRE1::MIR165a, shr
Sectioning, preparation of riboprobes and in situ hybridization were per-
formed as described19. LNA probes with complementary sequences to miR165
and miR166 were synthesized, 59 digoxigenin-labelled (Exiqon), and then hybri-
dized at 50 uC.
Most transgenic constructs were generated using the modified multisite gate-
way system27 and all were introduced into plants using the floral dip method40.
To identify in vivo binding activities of SHR to the promoters of MIR165/6,
plants expressing SHR::SHR:GFP in shr-2 were used. Roots were collected 6 days
after germination and processed for ChIP14. SHR binding affinity was compared
between the ChIP DNA treated with and without GFP antibody by measuring the
differential enrichment of DNA fragments using real-time PCR.
Received 29 May 2009; accepted 1 March 2010.
Published online 21 April 2010.
1. Du, T.-G., Schmid, M. & Jansen, R.-P. Why cells move messages: The biological
functions of mRNA localization. Semin. Cell Dev. Biol. 18, 171–177 (2007).
2. Tretter, E. M., Alvarez, J. P., Eshed, Y. & Bowman, J. L. Activity range of Arabidopsis
small RNAs derived from different biogenesis pathways. Plant Physiol. 147, 58–62
(2008).
3. Dunoyer, P., Himber, C., Ruiz-Ferrer, V., Alioua, A. & Voinnet, O. Intra- and
intercellular RNA interference in Arabidopsis thaliana requires components of the
microRNA and heterochromatic silencing pathways. Nature Genet. 39, 848–856
(2007).
4. Chitwood, D. H. et al. Pattern formation via small RNA mobility. Genes Dev. 23,
549–554 (2009).
5. Nogueira, F. T. S. et al. Regulation of small RNA accumulation in the maize shoot
apex. PLoS Genet. 5, e1000320 (2009).
6. Juarez, M. T., Kui, J. S., Thomas, J., Heller, B. A. & Timmermans, M. C. P.
microRNA-mediated repression of rolled leaf1 specifies maize leaf polarity. Nature
428, 84–88 (2004).
7. Lin, S.-I. et al. Regulatory network of microRNA399 and PHO2 by systemic
signaling. Plant Physiol. 147, 732–746 (2008).
8. Esau, K. Anatomy of Seed Plants 2nd edn (Wiley, 1977).
9. Pryer, K. M., Schneider, H. & Magallón, S. in Assembling the Tree of Life (eds
Cracraft, J. & Donoghue, M. J.) Ch. 10, 138–153 (Oxford Univ. Press, 2004).
10. Kevin Boyce, C., Holbrook, N. M. & Maciej, A. Z. in Vascular Transport in Plants
479–499 (Academic Press, 2005).
11. Gallagher, K. L., Paquette, A. J., Nakajima, K. & Benfey, P. N. Mechanisms
regulating SHORT-ROOT intercellular movement. Curr. Biol. 14, 1847–1851
(2004).
12. Helariutta, Y. et al. The SHORT-ROOT gene controls radial patterning of the
Arabidopsis root through radial signaling. Cell 101, 555–567 (2000).
13. Nakajima, K., Sena, G., Nawy, T. & Benfey, P. N. Intercellular movement of the
putative transcription factor SHR in root patterning. Nature 413, 307–311 (2001).
14. Cui, H. et al. An evolutionarily conserved mechanism delimiting SHR movement
defines a single layer of endodermis in plants. Science 316, 421–425 (2007).
15. Levesque, M. P. et al. Whole-genome analysis of the SHORT-ROOT
developmental pathway in Arabidopsis. PLoS Biol. 4, e143 (2006).
16. Di Laurenzio, L. et al. The SCARECROW gene regulates an asymmetric cell division
that is essential for generating the radial organization of the Arabidopsis root. Cell
86, 423–433 (1996).
17. Sabatini, S., Heidstra, R., Wildwater, M. & Scheres, B. SCARECROW is involved in
positioning the stem cell niche in the Arabidopsis root meristem. Genes Dev. 17,
354–358 (2003).
18. Gallagher, K. L. & Benfey, P. N. Both the conserved GRAS domain and nuclear
localization are required for SHORT-ROOT movement. Plant J. 57, 785–797
(2009).
19. Mähönen, A. P. et al. A novel two-component hybrid molecule regulates vascular
morphogenesis of the Arabidopsis root. Genes Dev. 14, 2938–2943 (2000).
20. McConnell, J. R. et al. Role of PHABULOSA and PHAVOLUTA in determining radial
patterning in shoots. Nature 411, 709–713 (2001).
NATURE | Vol 465 | 20 May 2010
21. Han, M.-H., Goud, S., Song, L. & Fedoroff, N. The Arabidopsis double-stranded
RNA-binding protein HYL1 plays a role in microRNA-mediated gene regulation.
Proc. Natl Acad. Sci. USA 101, 1093–1098 (2004).
22. Vazquez, F., Gasciolli, V., Cr, P. & Vaucheret, H. The nuclear dsRNA binding
protein HYL1 is required for microRNA accumulation and plant development, but
not posttranscriptional transgene silencing. Curr. Biol. 14, 346–351 (2004).
23. Emery, J. F. et al. Radial patterning of Arabidopsis shoots by class III HD-ZIP and
KANADI genes. Curr. Biol. 13, 1768–1774 (2003).
24. Mallory, A. C. et al. MicroRNA control of PHABULOSA in leaf development:
importance of pairing to the microRNA 59 region. EMBO J. 23, 3356–3364 (2004).
25. Zhou, G.-K., Kubo, M., Zhong, R., Demura, T. & Ye, Z.-H. Overexpression of
miR165 affects apical meristem formation, organ polarity establishment and
vascular development in Arabidopsis. Plant Cell Physiol. 48, 391–404 (2007).
26. Hawker, N. P. & Bowman, J. L. Roles for class III HD-Zip and KANADI genes in
Arabidopsis root development. Plant Physiol. 135, 2261–2270 (2004).
27. Lee, J.-Y. et al. Transcriptional and posttranscriptional regulation of transcription
factor expression in Arabidopsis roots. Proc. Natl Acad. Sci. USA 103, 6055–6060
(2006).
28. Jung, J.-H. & Park, C.-M. MIR166/165 genes exhibit dynamic expression patterns
in regulating shoot apical meristem and floral development in Arabidopsis. Planta
225, 1327–1338 (2007).
29. Parizotto, E. A., Dunoyer, P., Rahm, N., Himber, C. & Voinnet, O. In vivo investigation
of the transcription, processing, endonucleolytic activity, and functional relevance
of the spatial distribution of a plant miRNA. Genes Dev. 18, 2237–2242 (2004).
30. Stark, A., Brennecke, J., Bushati, N., Russell, R. B. & Cohen, S. M. Animal
MicroRNAs confer robustness to gene expression and have a significant impact
on 39UTR evolution. Cell 123, 1133–1146 (2005).
31. Baima, S. et al. The Arabidopsis ATHB-8 HD-zip protein acts as a differentiation-
promoting transcription factor of the vascular meristems. Plant Physiol. 126,
643–655 (2001).
32. Ohashi-Ito, K., Kubo, M., Demura, T. & Fukuda, H. Class III homeodomain leucine-
zipper proteins regulate xylem cell differentiation. Plant Cell Physiol. 46,
1646–1656 (2005).
33. Prigge, M. J. et al. Class III homeodomain-leucine zipper gene family members
have overlapping, antagonistic, and distinct roles in Arabidopsis development.
Plant Cell 17, 61–76 (2005).
34. Muñiz, L. et al. ACAULIS5 controls Arabidopsis xylem specification through the
prevention of premature cell death. Development 135, 2573–2582 (2008).
35. Mähönen, A. P. et al. Cytokinin signaling and its inhibitor AHP6 regulate cell fate
during vascular development. Science 311, 94–98 (2006).
36. Floyd, S. K. & Bowman, J. The ancestral developmental tool kit of land plants. Int. J.
Plant Sci. 168, 1–35 (2007).
37. Floyd, S. K. & Bowman, J. L. Gene regulation: ancient microRNA target sequences
in plants. Nature 428, 485–486 (2004).
38. Levine, E., McHale, P. & Levine, H. Small regulatory RNAs may sharpen spatial
expression patterns. PLoS Comp. Biol. 3, e233 (2007).
39. Shi, R. & Chiang, V. L. Facile means for quantifying microRNA expression by real-
time PCR. Biotechniques 39, 519–525 (2005).
©2010 Macmillan Publishers Limited. All rights reserved
ARTICLES
40. Clough, S. J. & Bent, A. F. Floral dip: a simplified method for Agrobacterium-
mediated transformation of Arabidopsis thaliana. Plant J. 16, 735–743 (1998).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank K. Kainulainen, M. Herpola, G.-B. Berglund and
J. Jung for technical assistance; M. Prigge, S. Clark, C. Bellini, N. Sauer, J. Colinas,
T. Vernoux, K. Gallagher, A. P. Mahonen, A. Bishopp, M. Bonke, N. Fedoroff,
J. C. Fletcher, B. J. Reinhart, I. Pekker, ABRC and NASC for materials, and E. Richards
and M. Harrison for comments on the manuscript; M. Tsiantis for sharing results
before publication. This work was supported by the Boyce Thompson Institute and
NSF IOS0818071 to J.-Y.L., Cornell Presidential Life Science Fellowship to J.Z.,
grants from the NIH (RO1-GM043778) and from the NSF ARABIDOPSIS 2010
programme to P.N.B., a fellowship from the MICINN, Spanish Government to
M.A.M.-R., grants by the Academy of Finland, Tekes and ESF to Y.H., S.L. and A.V.,
European Molecular Biology Organisation (EMBO, ALTF 450-2007) to J.D.,
Estonian funding agencies (ETF7361 and SF0180071s07) to O.L., Nilsson-Ehle
Foundation to C.J.R, and FORMAS and Carl Trygger’s Foundation for Scientific
Research to A.C. Imaging at Boyce Thompson Institute was supported by NSF (NSF
DBI-0618969) and Triad Foundation.
Author Contributions A.C. and J.-Y.L. contributed equally to this work, C.J.R., J.D.,
S.L. and J.Z. contributed equally to this work, O.L., M.A.M.-R. and A.V. contributed
equally to this work, Y.H. and P.N.B. contributed equally to this work, and the name
order was determined by raffle. A.C. designed and performed experiments to
characterize HD-ZIP III transcription factors and miR165/6 in vascular patterning,
J.-Y.L. identified and characterized the regulatory network of SHR, SCR and
miR165/6 in the xylem patterning, C.J.R. analysed the miRNA expression by in situ
hybridization and participated in HD-ZIP III mutant characterization. J.D.
participated in the analysis of expression of PHB (including mutant forms) and
other HD-ZIP III and generated pCRE1::MIR165a as well as participated in the
generation of J0571 lines to rescue scr and shr. S.L. participated in the PHB and
HD-ZIP III expression analysis, phb-7d mutant characterization and generated the
pSCR::MIR165a to rescue scr. J.Z. developed and characterized the xylem patterning
led by non-mobile SHR and PHB-m. O.L. participated in the characterization of
various HD-ZIP III mutant lines. M.A.M.-R. showed the non-cell autonomous action
of non-mobile SHR. A.V. participated in positional cloning of phb-7d and
establishment of the J0571 lines to rescue shr. S.T. identified the phb-7d mutant.
A.C. identified the scr-6 allele. J.S. characterized shr/scr and HD-ZIP III double
mutants and embryo expression patterns in GFP lines. J.L.B. shared informative
non-published materials. Y.H. and P.N.B. participated in experimental design. A.C.,
J.-Y.L., Y.H. and P.N.B. wrote the manuscript. A.C. and J.-Y.L. are co-corresponding
authors. All authors discussed the results and commented on the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to Y.H.
(yrjo.helariutta@helsinki.fi) or P.N.B. (philip.benfey@duke.edu).
321
 Vol 465 | 20 May 2010 | doi:10.1038/nature09056

LETTERS
A faint type of supernova from a white dwarf with a
helium-rich companion
H. B. Perets1,2, A. Gal-Yam3, P. A. Mazzali4,5,6, D. Arnett7, D. Kagan8, A. V. Filippenko9, W. Li9, I. Arcavi3, S. B. Cenko9,
D. B. Fox10, D. C. Leonard11, D.-S. Moon12, D. J. Sand2,13, A. M. Soderberg2, J. P. Anderson14,15, P. A. James15,
R. J. Foley2, M. Ganeshalingam9, E. O. Ofek16, L. Bildsten17,18, G. Nelemans19, K. J. Shen17, N. N. Weinberg9,
B. D. Metzger9, A. L. Piro9, E. Quataert9, M. Kiewe1 & D. Poznanski9,20

Supernovae are thought to arise from two different physical pro- showing lines of He but lacking either hydrogen or the hallmark Si
cesses. The cores of massive, short-lived stars undergo gravitational and S lines of type Ia supernovae in its photospheric spectra. The
core collapse and typically eject a few solar masses during their nebular spectrum of this event shows no emission from iron-group
explosion. These are thought to appear as type Ib/c and type II super- elements, which also characterize type Ia supernovae (Supplemen-
novae, and are associated with young stellar populations. In contrast, tary Information sections 3 and 4). Analysis of this spectrum indi-
the thermonuclear detonation of a carbon-oxygen white dwarf, cates a total ejected mass of Mej < 0.275M[ (M[, solar mass), with a
whose mass approaches the Chandrasekhar limit, is thought to pro- small fraction in radioactive nickel, consistent with the low lumi-
duce type Ia supernovae1,2. Such supernovae are observed in both nosity of this event. Such low ejecta mass for a supernova of any type
young and old stellar environments. Here we report a faint type Ib has not previously been firmly established using nebular spectral
supernova, SN 2005E, in the halo of the nearby isolated galaxy, NGC analysis (Supplementary Fig. 3 and Supplementary Information
1032. The ‘old’ environment near the supernova location, and the section 5). We also used the narrow, fast and faint light curve (Sup-
very low derived ejected mass ( 0.3 solar masses), argue strongly plementary Information, Supplementary Fig. 4) together with the
against a core-collapse origin. Spectroscopic observations and ana- measured ejecta velocity (,11,000 km s21) to infer the ejected mass
lysis reveal high ejecta velocities, dominated by helium-burning pro- (Supplementary Information section 6). We use these data to find
ducts, probably excluding this as a subluminous3,4 or a regular1 type consistent results of Mej < 0.360.1M[, assuming that some of the
Ia supernova. We conclude that it arises from a low-mass, old pro- mass is not accounted for by the nebular spectrum analysis (for
genitor, likely to have been a helium-accreting white dwarf in a example, high-velocity He layers and some slowly moving, denser
binary. The ejecta contain more calcium than observed in other types ejecta that are still hidden below the photosphere at that time).
of supernovae and probably large amounts of radioactive 44Ti. Finally, SN 2005E exhibits a remarkable amount of calcium in its
We discovered a supernova explosion (SN 2005E; Fig. 1) on 2005 ejecta, 0.135M[ (,0.49 of the total ejecta mass), 5–10 times more
January 13 (UT dates are used throughout this paper) shortly after it than typical supernovae of any variety, with a relative calcium frac-
occurred (it was not detected on 2004 December 24). Follow-up tion 25–350 times higher than any reported values for other super-
spectroscopy (Fig. 2) revealed strong lines of helium and calcium, indi- novae (Supplementary Table 1 and Supplementary Information
cating that it belongs to the previously identified group of calcium-rich section 7), while not showing evidence for sulphur (Supplementary
type Ib supernovae5. The supernova position is ,22.9 kpc (projected) Information section 4).
from the centre, and ,11.3 kpc above the disk, of its edge-on host galaxy, The remote position of SN 2005E in the outskirts (halo) of the
NGC 1032 (Fig. 1), which is itself at a distance of 34 Mpc. NGC 1032 is an galaxy, together with the isolation of NGC 1032 and its classification
isolated galaxy6 showing no signs of interaction, with the closest small as an S0/a galaxy (in which the star-formation rate is very low7), in
satellite galaxy found at a distance .120 kpc in projection. Deep follow- addition to our limits on local star formation, point to a supernova
up observations of the explosion site, sensitive to both ultraviolet light progenitor from an old stellar population (see also Supplementary
from hot young stars and emission lines from ionized hydrogen gas, Information section 2). In addition, the low ejected mass and nucleo-
put strict limits on any local star-formation activity at or near the super- synthetic output of SN 2005E are in stark contrast to those expected
nova location (Fig. 1). In addition, a radio signature, expected from from collapsing massive stars, whether formed locally or ejected from
some core-collapse supernovae, has not been observed (Supplemen- a distant location (Supplementary Information sections 8 and 9).
tary Information section 2). The low ejected mass is also inconsistent with those determined for
Our analysis of the spectra of SN 2005E indicates that it is similar to type Ia supernovae, restricted to a tight mass range of ,1–1.3M[,
type Ib supernovae (Fig. 2 and Supplementary Information section 3), regardless of their intrinsic luminosity (even the prototype faint SN
1
Department of Particle Physics and Astrophysics, Faculty of Physics, The Weizmann Institute of Science, Rehovot 76100, Israel. 2Harvard-Smithsonian Center for Astrophysics, 60
Garden Street, Cambridge, Massachusetts 02138, USA. 3Department of Particle Physics and Astrophysics, Faculty of Physics, The Weizmann Institute of Science, Rehovot 76100,
Israel. 4Max-Planck-Institut für Astrophysik, Karl-Schwarzschild-Str. 1, 85748 Garching, Germany. 5Scuola Normale Superiore, Piazza Cavalieri 7, 56127 Pisa, Italy. 6INAF – Oss.
Astron. Padova, vicolo dell’Osservatorio, 5, 35122 Padova, Italy. 7Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, Arizona 85721, USA. 8Department of
Astronomy, University of Texas at Austin, Austin, Texas 78712, USA. 9Department of Astronomy, University of California, Berkeley, California 94720-3411, USA. 10Department of
Astronomy and Astrophysics, Pennsylvania State University, University Park, Pennsylvania 16802, USA. 11Department of Astronomy, San Diego State University, San Diego, California
92182, USA. 12Department of Astronomy and Astrophysics, University of Toronto, 50 St George Street, Toronto, ON M5S 3H4, Canada. 13Las Cumbres Observatory Global Telescope
Network, 6740 Cortona Dr., Suite 102, Goleta, California 93117, USA. 14Departamento de Astronomia, Universidad de Chile, Camino El Observatorio 1515, Las Condes, Santiago, Casilla
36-D, Chile. 15Astrophysics Research Institute, Liverpool John Moores University, Twelve Quays House, Birkenhead CH41 1LD, UK. 16Department of Astronomy, 105-24, California
Institute of Technology, Pasadena, California 91125, USA. 17Kavli Institute for Theoretical Physics, Kohn Hall, University of California, Santa Barbara, California 93106, USA.
18
Department of Physics, University of California, Santa Barbara, California 93106, USA. 19Department of Astrophysics, Radboud University Nijmegen, PO Box 9010, NL-6500 GL, The
Netherlands. 20Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, California 94720, USA.

322
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
a b
50″
c d
Figure 1 | The environment of SN 2005E. a, NGC 1032, the host galaxy of
SN 2005E, as observed by the Sloan Digital Sky Survey (SDSS), before the
supernova explosion. The galaxy is an isolated, edge-on, early-type spiral
galaxy, showing no signs of star-formation activity, warping or interaction.
Its luminosity is dominated by the cumulative contribution of a multitude of
low-mass old stars (yellow light in this image). b, The LOSS29 discovery of SN
2005E on 2005 January 13 (shown in negative). Note the remote location of
the supernova (marked with a red arrow) with respect to its host, 22.9 kpc
(projected) from the galaxy nucleus and 11.3 kpc above the disk, whose edge-
on orientation is well determined (a). c, An image of NGC 1032 in the light of
the Ha emission line, emitted by interstellar gas ionized by ultraviolet
radiation, and a good tracer of recent star formation. There are no traces of
recent star-formation activity (usually appearing as irregular, compact
emission sources) near the supernova location or anywhere else in the host.
Panels a–c are 2750 3 1750; north is up, and east is to the left; scale bar in
a applies to b and c also. d, Zoom-in on the location of SN 2005E in
1991bg is found in this range)2. Furthermore, the light curve of SN
2005E (Supplementary Information section 6) shows a different beha-
viour from that of type Ia supernovae, declining much faster than even
the most subluminous (SN 1991bg-like) events observed8. These
properties, together with the observed He-rich spectra and inferred
composition, rule out SN 2005E as being either a regular or peculiar
type Ia supernova (see also discussion in Supplementary Information
section 10, regarding the very subluminous SN 2008ha and other
related peculiar supernovae4,9,10). Therefore, we conclude that SN
2005E is the first clearly identified example of a new, different type
of supernova explosion, arising from a He-rich, low-mass progenitor.
The spectroscopic signatures of SN 2005E are quite unusual, and
allow one to identify additional similar events5. Arising from lower-
mass progenitors, these events are likely to be found among both old
and young stellar populations—that is, we expect to find such peculiar
type Ib supernovae in both early- and late-type galaxies. Indeed, while
the unusual location of SN 2005E triggered the current study, several
other calcium-rich subluminous type Ib/c supernovae similar to SN
2005E have been observed (Supplementary Information section 11).
Of the group of eight subluminous calcium-rich type Ib/c supernovae
identified (seven identified by us and an additional one described in
ref. 11), four are observed in old-population environments: SN 2005E
presented here, as well as SN 2000ds, SN 2005cz and SN 2007ke,
residing in elliptical galaxies. SN 2000ds has pre- and post-explosion
Hubble Space Telescope images showing no evidence for either star-
forming regions or massive stars12 near its location. The host-galaxy
distribution of the supernovae in our sample (Fig. 3) is inconsistent
with that of any core-collapse supernova. Neither have radio signa-
tures been observed (Supplementary Information section 11). Thus,
all evidence suggests that a well-defined subset of type Ib supernovae,
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
e
10″
f g
pre-explosion SDSS r-band images. No source is detected near the supernova
location, marked with a yellow circle (radius 30; the astrometric uncertainty
in the supernova location is ,0.50). The SDSS catalogue does not list any
objects near that position (for example, putative faint dwarf satellites of
NGC 1032), down to a typical limit of r 5 22.5 mag. e, f, Deeper photometry
of the supernova location. A red image is shown in e, while an ultraviolet (u-
band) image is shown in f. At the distance of NGC 1032, the point-source
upper limits we find, Mr , 27.5(26.9) and Mu’ v{8:1ð{7:1Þ mag at
3(2)s, respectively, indicate that we would have detected faint star-forming
galaxies or star-forming regions at the supernova location, or indeed even
individual massive red supergiant or luminous blue supergiant stars.
g, Zoom-in on the location of SN 2005E in Ha light (see c for details). No
trace of star-formation activity is seen near the supernova location. Panels
d–g are 640 3 360; scale bar in d applies also to e–g. Technical details about
the observations can be found in Supplementary Information section 1.
all having Ca-rich spectra and faint peak magnitudes, comprise a
distinct physical class of explosions coming from low-mass, old pro-
genitors. This class includes all known type Ib/c events in confirmed
elliptical galaxies13,14 (Supplementary Information section 11). A dif-
ferent interpretation11, invoking the core collapse of a massive pro-
genitor, was suggested for one of these events (SN 2005cz). It is
difficult to reconcile our observations and analysis of SN 2005E with
such an interpretation, which is also inconsistent with the host-galaxy
distribution of all of the other Ca-rich type Ib supernovae in our
sample.
Calcium-rich supernovae were theoretically predicted to arise
from burning helium-rich material on a white dwarf (for example,
a helium white dwarf or a helium star accreting onto a carbon-oxygen
white dwarf), leading to the full disruption of a sub-Chandrasekhar-
mass white dwarf15,16. However, such models predicted the produc-
tion of supernovae far more luminous (and 56Fe rich) than SN 2005E.
Several theoretical models were suggested in the literature to possibly
produce subluminous supernovae, with low-mass and high-velocity
ejecta in an old stellar population. These include the accretion-induced
collapse of a white dwarf (see, for example, refs 17 and 18), and the
detonation of an accreted helium shell on a white dwarf in a binary
system (the ‘.Ia’ model19). These studies did not explore the burning of
large helium masses (.0.1M[), nor the production of calcium-rich
ejecta. Multi-dimensional simulations of a detonation in accreted He
layers20 showed (for low-mass white dwarfs; M 5 0.7M[) a trend
towards large Ca abundances and high Ca/S abundance ratios (a high
ratio is inferred for SN 2005E; Supplementary Information section 4)
and a light curve that was faster and dimmer than those of typical type
Ia supernovae (but still much more luminous than SN 2005E), as well
as a high production of 44Ti. It is possible that similar models, with less
323
LETTERS NATURE | Vol 465 | 20 May 2010

a 1
−14 He I 2005 January 16
0.9
−14.5
0.8 91bg
−15
2005 February 6 0.7

Cumulative fraction
log Fλ (erg s–1 cm–2 Å–1)

−15.5
0.6
−16 05E
0.5 Ia
−16.5 91bg
S 0.4
2005 March 11 Ia 91T
−17 Ib
0.3
Fe/Co blends Ic
−17.5 II
0.2 II
02cx
−18 02cx
[O I] [Ca II] Ca II 0.1 05E
−18.5 SN 1991bg
0
4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 8,000 8,500 9,000 E S0 Sa−Sab Sb Sbc Sc Scd−Sm Irr
12 Galaxy type
b
SN 2005E, 2005 March 11, Figure 3 | The cumulative distribution of host galaxies of supernovae from
10 photospheric light subtracted the KAIT (Katzman Automatic Imaging Telescope) supernova survey. We
Model corrected the classification of a few hosts of type Ib/c supernovae using
8 higher-quality observations from the Palomar 60-inch telescope (SN 2005ar,
Fλ (10–17 erg s–1 cm–2 Å–1)

SN 2006ab and SN 2006lc were found to be hosted by spiral galaxies rather

6
4
2 SN 2002cx-like type Ia, with half of the SN 2005E-like group (four out of
0 2005E and the other members of its group are therefore likely to belong to an
−2
Ca II
−4
4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 8,000 8,500 9,000
Rest wavelength (Å)
Figure 2 | The mass and composition of the ejecta of SN 2005E.
a, Photospheric spectra of SN 2005E. The top spectrum is obviously
photospheric and shows absorption lines of the He I series (marked with
black ticks after application of an 11,000 km s21 blueshift, at the top).
Nebular lines of intermediate-mass elements, most notably calcium, begin to
emerge in the middle spectrum. Calcium dominates the latest nebular
spectrum at the bottom, and nebular oxygen is visible as well. We note that
the typical Si lines of type Ia supernovae are absent in all spectra, while the
nebular spectrum of SN 2005E clearly rules out a type Ia identification
(comparison with the underluminous SN 1991bg is shown; note the lack of
the typical iron-group line blends in the blue side). The derived line
velocities are consistent with SN 2005E exploding within its putative host
galaxy, NGC 1032. b, The nebular spectrum of SN 2005E compared with a
model fit. From the fit we can derive elemental abundances and masses in the
ejecta of SN 2005E. We find masses of 0.1, 0.037, 0.135 and 0.003M[ for
carbon, oxygen, calcium and radioactive nickel, respectively. Both the low
total ejected mass of ,0.275M[ and the relative abundances are unique
among previously studied events. The lack of prominent C/O-burning
products such as S and Fe (typically seen in type Ia supernovae;
Supplementary Information section 4) argues against a C/O white dwarf
origin. Technical details of observations and additional references can be
found in Supplementary Information section 1.
burning of C and O to make S and Ni, may resemble SN 2005E. This
gains additional support from our nucleosynthetic analysis (Sup-
plementary Information section 12), showing that the unique com-
position of SN 2005E could be produced, in principle, as the product of
He ignition. Further studies in these directions are in progress.
We conclude that SN 2005E appears to be the first observed mani-
festation of the helium detonation process. This event most probably
occurred in an interacting double white-dwarf system with a helium
white-dwarf mass donor. Additional characteristics of these explosions,
including their old population origin, He-rich spectra, subluminosity
324
than elliptical galaxies). After correcting the classification, we find that all
type Ib/c supernovae found in early-type galaxies are faint Ca-rich
supernovae similar to SN 2005E. Note that the SN 2005E-like supernova
host distribution is very different from that of other type Ib/c supernovae, as
well as that of type II (known to have young massive progenitors) and that of
eight) observed in early-type (elliptical or S0) galaxies. The progenitors of SN
old, low-mass stellar population. The total numbers of host galaxies included
in this figure are 244, 25, 8, 257, 30, 63, 14, and 8 for supernovae of types Ia,
SN 1991bg (91bg), SN 1991T (91T), II, Ib, Ic, SN 2002cx (02cx) and SN
Na I [O I] [Ca II] Ca II [C I] 2005E (05E), respectively.
and low ejected mass, are broadly consistent with the predictions of
some theoretical models (.Ia19; accretion-induced collapse18; helium
detonation21), variants of which may produce the appropriate condi-
tions for such helium detonations. Alternatively, these explosions may
require a new mechanism.
Our discovery has numerous astrophysical implications. It seems
highly likely that we identified explosions arising from very close
white dwarf/white dwarf systems, the rates of which (Supplemen-
tary Information section 11) might be useful for predicting the rates
of white dwarf/white dwarf inspirals observable as gravitational wave
sources. The unique nucleosynthetic production of large masses of
calcium and radioactive 44Ti per explosion could solve puzzles related
to the source of calcium (especially 44Ca) in the primitive Solar
System22,23 and in old, metal-poor halo stars24, and the enrichment
patterns of the interstellar and intracluster medium25. Production of
most of the Galactic 44Ti and its progeny, 44Ca, in a few rare, prolific
explosions, can also explain the origins of Galactic 44Ca, given the null
detection of 44Ti traces in most nearby supernova remnants23,26.
Finally, inverse b decay of 44Ti may significantly contribute to the
Galactic production of positrons27. Assuming our estimated rates
(,10% of the type Ia supernova rate; Supplementary Information
section 11) and our 44Ti yield (0.014–0.14M[; Supplementary Informa-
tion section 12), Galactic supernovae of the type we describe here will
provide a significant contribution to the Galactic bulge component of
the positron annihilation 511 keV c-ray line, at least comparable to that
of type Ia supernovae. In fact, within the current uncertainties on the
44
Ti yield and supernova rates, these events may come within a factor of
a few of producing all of the observed positrons28.
Received 17 May 2009; accepted 23 March 2010.
1. Filippenko, A. V. Optical spectra of supernovae. Annu. Rev. Astron. Astrophys. 35,
309–355 (1997).
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 LETTERS

2. Mazzali, P. A. et al. A common explosion mechanism for type Ia supernovae. 28. Knödlseder, J. et al. The all-sky distribution of 511 keV electron-positron
Science 315, 825–828 (2007). annihilation emission. Astron. Astrophys. 441, 513–532 (2005).
3. Filippenko, A. V. et al. The subluminous, spectroscopically peculiar type IA 29. Filippenko, A. V. et al. in Small Telescope Astronomy on Global Scales (eds
supernova 1991bg in the elliptical galaxy NGC 4374. Astron. J. 104, 1543–1556 Paczynski, B., Chen, W.-P. & Lemme, C.) 121–130 (ASP Conf. Ser. Vol. 246,
(1992). Astronomical Society of the Pacific, 2001).
4. Li, W. et al. SN 2002cx: the most peculiar known type Ia supernova. Publ. Astron.
Soc. Pacif. 115, 453–473 (2003). Supplementary Information is linked to the online version of the paper at
5. Filippenko, A. V. et al. Supernovae 2001co, 2003H, 2003dg, and 2003dr. IAU Circ. www.nature.com/nature.
8159, 2 (2003).
Acknowledgements We thank P. Podsiadlowski, E. Nakar and D. Maoz for
6. Prada, F. et al. Observing the dark matter density profile of isolated galaxies.
comments. We acknowledge observations with the Liverpool Telescope, and
Astrophys. J. 598, 260–271 (2003).
various telescopes at the Lick, Palomar and Keck Observatories. We are grateful to
7. Kennicutt, R. C. Jr. Star formation in galaxies along the Hubble sequence. Annu.
the staffs of these observatories, as well as to the institutions, agencies and
Rev. Astron. Astrophys. 36, 189–232 (1998).
companies funding these facilities. This research also made use of the NASA/IPAC
8. Kasliwal, M. M. et al. SN 2007ax: an extremely faint type Ia supernova. Astrophys.
Extragalactic Database (NED). H.B.P. acknowledges the ISF/FIRST and Ilan
J. 683, L29–L32 (2008).
Ramon-Fulbright Fellowships, and is a Harvard-Smithsonian Center for
9. Foley, R. J. et al. SN 2008ha: an extremely low luminosity and exceptionally low
Astrophysics Fellow. The collaborative work of A.G.-Y. and P.A.M. is supported by
energy supernova. Astron. J. 138, 376–391 (2009).
a Weizmann-Minerva grant. A.G.-Y. acknowledges further support by the Israeli
10. Valenti, S. et al. A low-energy core-collapse supernova without a hydrogen
Science Foundation, an EU Seventh Framework Programme Marie Curie IRG
envelope. Nature 459, 674–677 (2009).
Fellowship, the Benoziyo Center for Astrophysics, and the Peter and Patricia
11. Kawabata, K. S. et al. A massive star origin for an unusual helium-rich supernova in
Gruber Awards. A.V.F. is grateful for the support of the US National Science
an elliptical galaxy. Nature doi:10.1038/nature09055 (this issue).
Foundation, the US Department of Energy, Gary and Cynthia Bengier, the Richard
12. Maund, J. R. & Smartt, S. J. Hubble Space Telescope imaging of the progenitor
and Rhoda Goldman Fund, the Sylvia & Jim Katzman Foundation, and the
sites of six nearby core-collapse supernovae. Mon. Not. R. Astron. Soc. 360,
TABASGO Foundation. R.J.F. is a Clay Fellow.
288–304 (2005).
13. van den Bergh, S., Li, W. & Filippenko, A. V. Classifications of the host galaxies of Author Contributions H.B.P. led the project, performed the calculations related to
supernovae, set III. Publ. Astron. Soc. Pacif. 117, 773–782 (2005). hyper-velocity stars, examined other putative SN 2005E-like events, collected and
14. Hakobyan, A. A. et al. Early-type galaxies with core collapse supernovae. Astron. analysed archival data concerning supernova properties and their hosts, and wrote
Astrophys. 488, 523–531 (2008). the manuscript. A.G.-Y. is the Principal Investigator of the CCCP programme and
15. Woosley, S. E., Taam, R. E. & Weaver, T. A. Models for type I supernova. I — initiated the project, collected and analysed photometric and spectroscopic data,
Detonations in white dwarfs. Astrophys. J. 301, 601–623 (1986). coordinated further observational and theoretical work, and managed the project.
16. Woosley, S. E. & Weaver, T. A. Sub-Chandrasekhar mass models for type Ia P.A.M. conducted the nebular spectral analysis and its interpretation, and
supernovae. Astrophys. J. 423, 371–379 (1994). determined the elemental abundances in the ejecta. D.A. determined that the
17. Nomoto, K. & Kondo, Y. Conditions for accretion-induced collapse of white measured composition requires He burning and performed nucleosynthesis
dwarfs. Astrophys. J. 367, L19–L22 (1991). calculations to confirm this. D.K. investigated local star-formation tracers at the
18. Metzger, B. D., Piro, A. L. & Quataert, E. Nickel-rich outflows from accretion disks location of SN 2005E. A.V.F. and W.L. contributed spectroscopic and photometric
formed by the accretion-induced collapse of white dwarfs. Mon. Not. R. Astron. observations and reductions of SN 2005E and of similar Ca-rich objects, a class
Soc. 396, 1659–1664 (2009). they originally identified, and provided most of the data on supernova host galaxies.
19. Bildsten, L. et al. Faint thermonuclear supernovae from AM Canum Venaticorum A.V.F. also edited the paper. I.A. analysed the CCCP photometry of SN 2005E and
binaries. Astrophys. J. 662, L95–L98 (2007). cross-calibrated it with other data. S.B.C., D.B.F., D.C.L., D.-S.M., D.J.S. and A.M.S.
20. Livne, E. & Arnett, D. Explosions of sub-Chandrasekhar mass white dwarfs in two are members of the CCCP and contributed to initial observations of SN 2005E.
dimensions. Astrophys. J. 452, 62–74 (1995). J.P.A. and P.A.J. obtained and analysed narrow-band images of NGC 1032 and the
21. Iben, I. J. et al. On interacting helium star-white dwarf pairs as supernova location of SN 2005E. R.J.F. and M.G. contributed to spectroscopic observations
precursors. Astrophys. J. 317, 717–723 (1987). and reductions. E.O.O. obtained deep photometric observations of the location of
22. Woosley, S. E., Arnett, W. D. & Clayton, D. D. The explosive burning of oxygen and SN 2005E. L.B., G.N., K.J.S. and N.N.W. investigated the relation of SN 2005E to .Ia
silicon. Astrophys. J. Suppl. Ser. 26, 231–312 (1973). models and contributed to the text. B.D.M., A.L.P. and E.Q. investigated the relation
23. Timmes, F. X. et al. The production of 44Ti and 60Co in supernovae. Astrophys. J. of SN 2005E to accretion-induced collapse models and contributed to the text.
464, 332–341 (1996). M.K. performed custom reductions of CCCP spectra. D.P. carried out synthetic
24. Lai, D. K. et al. A unique star in the outer halo of the Milky Way. Astrophys. J. 697, photometry analysis.
L63–L67 (2009).
25. de Plaa, J. et al. Constraining supernova models using the hot gas in clusters of Author Information Reprints and permissions information is available at
galaxies. Astron. Astrophys. 465, 345–355 (2007). www.nature.com/reprints. The authors declare no competing financial interests.
26. The, L.-S. et al. Are 44Ti-producing supernovae exceptional? Astron. Astrophys. Readers are welcome to comment on the online version of this article at
450, 1037–1050 (2006). www.nature.com/nature. Correspondence and requests for materials should be
27. Chan, K.-W. & Lingenfelter, R. E. Positrons from supernovae. Astrophys. J. 405, addressed to A.G.-Y. (avishay.gal-yam@weizmann.ac.il) and H.B.P.
614–636 (1993). (hperets@cfa.harvard.edu).

325
©2010 Macmillan Publishers Limited. All rights reserved
 Vol 465 | 20 May 2010 | doi:10.1038/nature09055

LETTERS
A massive star origin for an unusual helium-rich
supernova in an elliptical galaxy
K. S. Kawabata1, K. Maeda2, K. Nomoto2, S. Taubenberger3, M. Tanaka2,4, J. Deng5, E. Pian6, T. Hattori7 & K. Itagaki8

The unusual helium-rich (type Ib) supernova SN 2005E is distin-

guished from all supernovae hitherto observed by its faint and
rapidly fading light curve, prominent calcium lines in late-phase
spectra and lack of any mark of recent star formation near the 1.2 × 10–15
supernova location. These properties are claimed1 to be explained 2000H (Ib) +8
by a helium detonation in a thin surface layer of an accreting white 2000H +29
dwarf. Here we report that the observed properties of SN 2005cz, 1.0 × 10–15

Relative flux + const. (erg cm–2 s–1 Å–1)

which appeared in an elliptical galaxy, resemble those of SN 2005E.
We argue that these properties are best explained by a core-collapse
supernova at the low-mass end (8–12 solar masses) of the range of
massive stars that explode2. Such a low-mass progenitor lost its
hydrogen-rich envelope through binary interaction, had very thin
oxygen-rich and silicon-rich layers above the collapsing core, and
accordingly ejected a very small amount of radioactive 56Ni and
oxygen. Although the host galaxy NGC 4589 is an elliptical, some
studies have revealed evidence of recent star-formation activity3,
consistent with the core-collapse model.
We discovered SN 2005cz on 2005 July 17.5 UT in the elliptical
galaxy NGC 4589. The spectrum of SN 2005cz taken on July 28 is
consistent with post-maximum spectra of type Ib supernovae4. Thus,
SN 2005cz would have originated from a core-collapse of an envelope-
stripped massive star. We tentatively assume that the epoch of our first
spectrum is at t 5 126 days, where t is time after the maximum bright-
ness (Fig. 1; Supplementary Information section 1).
The late-time spectrum of SN 2005cz at t 5 1179 days is unique;
unlike most other type Ibc/IIb supernovae, SN 2005cz shows much
stronger [Ca II] at wavelengths of 7,291 Å and 7,323 Å than [O I] at
6,300 Å and 6,364 Å (Fig. 2; see refs 5, 6 for other supernovae with
large Ca/O ratios, and Supplementary Information section 3 for
comparative discussion). Oxygen is ejected mostly from the oxygen
layer formed during the hydrostatic burning phase; its mass depends
sensitively on the progenitor mass and is smaller for lower-mass
progenitors. On the other hand, Ca is synthesized during the explo-
sion. Theoretical models predict that the stars having main-sequence
masses of Mms 5 13M[ and 18M[ (M[, solar mass) produce
0.2M[ and 0.8M[ of O, and 0.005M[ and 0.004M[ of Ca, respec-
tively (see, for example, ref. 7). Therefore, the Ca/O ratio in the
supernova ejecta is sensitive to the progenitor mass8,9. To produce
the extremely large Ca/O ratio, the mass of the progenitor star of SN
2005cz should be smaller than that of the progenitor of any other type
Ib supernova reported to date. For both SN 1993J and SN 1994I,
which show weaker [Ca II] than [O I] (Fig. 2), the mass of the pro-
genitor is estimated to be ,12–15M[ (refs 10, 11), which is the
smallest among well-studied samples with [Ca II] , [O I] (see, for
example, refs 9, 12, 13; see also Supplementary Fig. 3). Thus, the
progenitor mass of SN 2005cz is likely to have been #12–15M[.
1
Hiroshima Astrophysical Science Center, Hiroshima University, 1-3-1 Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan. 2Institute for the Physics and Mathematics of the
Universe (IPMU), University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8583, Japan. 3Max-Planck-Institut für Astrophysik, Karl-Schwarzschild-Straße 1, 85741 Garching,
Germany. 4Department of Astronomy, School of Science, University of Tokyo, Bunkyo-ku, Tokyo 113-0033, Japan. 5National Astronomical Observatories, CAS, 20A Datun Road,
Chaoyang District, Beijing 100012, China. 6INAF Osservatorio Astronomico di Trieste, Via Tiepolo 11, I-3413 Trieste, Italy. 7Subaru Telescope, National Astronomical Observatory of
Japan, Hilo, Hawaii 96720, USA. 8Itagaki Astronomical Observatory, Teppo-cho, Yamagata 990-2492, Japan.
He I
8.0 × 10–16
1993J (IIb) +8
6.0 × 10–16 1993J +24
1994l (Ic) +7
4.0 × 10–16
2.0 × 10–16 1994l +26
0 2005cz
4,000 5,000 6,000 7,000 8,000 9,000 10,000
Rest wavelength (Å)
Figure 1 | Early-time spectra of stripped-envelope core-collapse
supernovae. Red, spectrum of type Ib SN 2005cz taken on 2005 July 28 with
the Keck Telescope. Also shown are the spectra of type Ib SN 2000H at
t 5 18 days (black) and t 5 129 days (grey)18, type IIb SN 1993J at
t 5 18 days (blue) and t 5 124 days (cyan)19, and type Ic SN 1994I at
t 5 17 days (magenta) and t 5 126 days (green)20,21. The type Ib is
characterized by strong helium lines and weak silicon lines, while in the type
Ic both helium and silicon lines are weak. The type IIb shows a type II-like
spectrum characterized by the strong hydrogen features at early times, and
becomes type Ib/c-like at late times. All these supernovae are thought to have
partly or fully stripped off their outer layers of hydrogen and helium before
the explosions. The epochs are also shown in the figure. The overall
appearance of spectral features in SN 2005cz is quite similar to those of the
type Ib SN 2000H at t 5 129 days, the type IIb SN 1993J at t 5 124 days
(despite its stronger H lines), and also the typical type Ic SN 1994I at
t 5 126 days (despite its lack of the strong He lines). The spectra are
corrected for the host redshift and the reddening. We adopted a total (Milky
Way 1 host) reddening of E(B 2 V) 5 0.13 (0.03 1 0.1) mag in SN 2005cz,
0.23 (0.23 1 0.0) mag in SN 2000H, 0.45 mag in SN 1994I, and 0.3 mag in SN
1993J. The flux is on an absolute scale for SN 2005cz, calibrated with the
Calar Alto photometry obtained four nights later. For the comparison
supernovae, the fluxes are on an arbitrary scale and constants are added for
presentation. The positions of the prominent He I lines are shown by the
dashed lines. The spectrum of SN 2005cz is consistent with the post-
maximum spectra of type Ib supernovae.

326
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
2.0 × 10–16
SN Ib 2004dk, +392*
Relative flux + const. (erg cm–2 s–1 Å–1)
1.6 × 10–16
SN IIb 1993J, +203
1.2 × 10–16
SN Ic 1994I, +147
8.0 × 10–17
SN Ia-pec 2005hk, +232
SN I(?) 2008ha, +65
4.0 × 10–17
SN Ib 2005cz, +179
0
4,000 5,000
Rest wavelength (Å)
Figure 2 | Late-time spectra of stripped-envelope core-collapse
supernovae and faint supernovae. Red, calcium-rich late-time spectrum of
type Ib SN 2005cz taken on 2006 December 27 (t 5 1179 days). Also shown
are type Ib SN 2004dk at t < 392 days (black)22, type IIb SN 1993J at
t 5 1203 days (blue)19, type Ic SN 1994I at t 5 1147 days (magenta)21,
peculiar (pec.) type Ia SN 2005hk at t 5 1232 days (green)23, and peculiar
(possibly type I) SN 2008ha at t 5 165 days (yellow)24. As time goes by, the
ejecta become transparent to optical light, following the expansion and
density decrease. Late-time spectra of type Ib/c supernovae are thus
characterized by various emission lines, mostly of forbidden transitions. The
spectrum of SN 2004dk is typical for type Ib/c supernovae at late times (for
example, see figure 2 of ref. 22). The spectra are corrected for the host
redshift, but not for reddening. The flux is on an approximate absolute scale
for SN 2005cz, calibrated with the spectroscopic standard star (but not with
photometry), whereas it is on an arbitrary scale for the comparison
supernovae. The asterisk on SN 2004dk indicates days since its discovery
(not maximum light). It is unique that SN 2005cz shows only weak [O I] lines
at 6,300 Å and 6,364 Å, and much stronger [Ca II] lines at 7,291 Å and 7,323 Å
than [O I]. The relatively weak Ca II infrared triplet compared with other
supernovae might suggest lower density ejecta for SN 2005cz. It is interesting
that the [Ca II] line is considerably narrow (half-width at half-maximum,
0.005c, which is probably the typical expansion velocity of the inner core
emitting the [Ca II] line) compared with the blueshift of the absorption in the
Ca II infrared triplet in the early-time spectrum (,0.04c), which is attributed
to the expansion velocity of the outer layer.
SN 2005cz is intrinsically fainter than the well-studied type Ic SN
1994I by DR < 1.5 mag (Fig. 3). In the pseudobolometric light curve,
the decline

. rate
 from the intermediate to the late phase is consistent with
2 56
Mej,8 E51 ƒ1, and the luminosity requires that M( Ni) # 0.005–
0.02M[ (Fig. 4). (Here Mej,[ is the mass of ejecta in units of solar mass,
and E51 is the kinetic energy of ejecta in units of 1051 erg.) Additionally,
(Mej,[/E51) < 1 is suggested from the line velocity (Fig. 4 legend). We
thus estimate Mej,[ # 1 and E51 # 1, indicating a small progenitor mass
(#12M[; refs 2, 14).
To explain the above peculiarities, we suggest a star with Mms 5 10–
12M[ as the most likely origin of SN 2005cz. If such a star had been
single, its mass (and thus its mass loss rate) would have been too small
to lose most of its H-rich envelope. Thus this star must have been in a
close binary system. Then it became a He star of ,2.5M[ after under-
going Roche lobe overflow. This He star formed a C1O core of
,1.5M[, which marginally exceeded the lower mass limit to form a
Fe core15,16. The overlying He layer had a mass of ,1M[. Eventually,
the He star underwent Fe core-collapse to explode as a type Ib super-
nova, leaving a ,1.5M[ neutron star behind. The ejecta had mass of
,1M[, consistent with the observed constraint. The ejecta consisted
mostly of unburned material in the He layer and a small amount of
LETTERS
–20
–17
[O I]
[Ca II] –16
–18 1993J (IIb)
–15
Absolute R-band magnitude
1994I (Ic) –14
–16
2007Y (Ib) –13
0 20 40 60
–14 2005cz (Ib)
1994I + 1.5 (lc)
–12
2008S (IIn)
–10
2008ha (I?)
–8
6,000 7,000 8,000 9,000 0 60 120 180 240 300 360
Time since the explosion (days)
Figure 3 | Absolute R-band light curves of relevant supernovae. Red
circles, light curve of the rapidly fading type Ib SN 2005cz. It is compared
with those of type IIb SN 1993J (cyan triangles), type Ic SN 1994I (blue
stars), type Ib SN 2007Y (green squares), type IIn SN 2008S (black open
circles), and the possibly type I SN 2008ha (orange open squares). Also
shown is the light curve of SN 1994I, but dimmed by 1.5 mag (magenta open
stars), which is magnified in the inset for convenience of comparison with
that of SN 2005cz. For SN 2005cz, the first three points denote unfiltered
magnitudes, which are approximately R-band magnitudes. The two points
with downward arrows are 3s upper limits. The distance moduli, m, and total
reddening values, E(B 2 V), both in units of magnitude, are taken as follows:
[m, E(B 2 V)] 5 (32.23, 0.13) for SN 2005cz (Supplementary Information
section 1), (27.8, 0.3) for SN 1993J, (29.6, 0.45) for SN 1994I, (31.43, 0.112)
for SN 2007Y, (31.55, 0.076) for SN 2008ha, and (28.78, 0.687) for SN 2008S.
We assume RV 5 3.1 to convert the colour excess to the R-band extinction.
The data points, as well as the distance and the reddening, are from the
literature19,24–27. The putative explosion date for SN 2005cz is assumed to be
2005 June 17, 30 days before the discovery and 15 days before maximum
brightness (Supplementary Information section 1). The tail of the light curve
of SN 2005cz is similar to those of type IIn SN 2008S and type Ic SN 1994I
(dimmed by 1.5 mag). From this, we estimate the mass of 56Ni as 0:018M 8 ,
assuming a nickel mass of M(56Ni) 5 0.07M[ produced in the typical type Ic
SN 1994I28.
explosively synthesized elements. The explosive burning products con-
tain some Fe, Ca, S and Si, but not much oxygen. Also, the ejected part
of the unburned oxygen-rich layer is extremely small. This scenario can
explain the peculiar nebular spectrum with large [Ca II]/[O I] ratio, as
well as the low luminosity and its relatively rapid decrease.
An alternative candidate for the progenitor is a star with
Mms < 8–10M[ in a close binary system. Such a star forms an electron-
degenerate O1Ne1Mg core and undergo electron-capture-induced
collapse16. The most likely scenario to realize a type Ib supernova
would be the merging of an O1Ne1Mg white dwarf and a He white
dwarf. The delay time between the star formation and the merging
could be long enough to explain the origin of both SN 2005cz and the
recently reported SN 2005E1 with this scenario.
As for the host galaxy problem, the ,10M[ star model is found to
be consistent with the properties recently inferred for the host galaxy
of SN 2005cz. It is still a genuine E2 galaxy17, but has a relatively
young stellar population with life times of ,107–108 years (ref. 3)
and the type Ib SN 2005cz is likely to have been the end product of
one of these young stars (see Supplementary Information section 2).
The mass range of 10–12M[ has not been theoretically investigated
in much detail so far, but, as the data from SN 2005cz suggest, the
supernovae resulting from these stars may have a very special abundance
pattern in their ejecta and play an important role in the chemical evolu-
tion of galaxies (see Supplementary Information section 3).
327
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
1994l
1042
Full deposition
Bolometric luminosity (erg s–1)
1041
2005cz
1040
1039
0 60 120 180
Time since the explosion (days)
Figure 4 | Bolometric light curve. Filled red circles, pseudobolometric light
curve of type Ib SN 2005cz. It is compared with a simple . c-ray and positron
deposition model with M(56Ni) 5 0.02M[ and Mej,8 2
E51 ~1 (red line),
where E51 is the kinetic energy EK expressed in units of 1051 erg. We also plot
the bolometric light curve of type Ic SN 1994I (open black squares)25 and a
simple deposition model with M(56Ni) 5 0.07M[ (black line) for
comparison. Except for the last point (upper limit), we simply assume the
bolometric correction BC ; Mbol – MR 5 0.5, derived from SN 1998bw, SN
2002ap and SN 2008D at similar epochs14,29,30. As this is a very crude
estimate, we adopt an error bar of 60.5 mag for the bolometric luminosity.
The deposition models adopt the c-ray opacity for the Compton scattering
(tc ! Mej2 EK {1 t {2 ) and assume the full deposition of positrons. The decline

from
rate . the intermediate to the late phase is consistent with
2
Mej,8 E51 ƒ1. Combining this expression with (Mej,[/E51) < 1 as
indicated by the similarity in the absorption velocity seen in SN 2005cz and
those in SN 1993J and SN 1994I (Fig. 1, Supplementary Fig. 1), we estimate
Mej,[ # 1 and E51 # 1. The luminosity requires that M(56Ni) # 0.02M[.
Note that the estimate for M(56Ni) is sensitively affected by the explosion
date. The upper limit to M(56Ni) is only M(56Ni) # 0.005M[, if the
explosion date is as late as 2005 July 15 (just a few days before the discovery).
Received 2 June 2009; accepted 24 March 2010.
1. Perets, H. B. et al. A faint type of supernova from a white dwarf with a helium-rich
companion. Nature doi:10.1038/nature09056 (this issue).
2. Smartt, S. J. Progenitors of core-collapse supernovae. Annu. Rev. Astron. Astrophys.
47, 63–106 (2009).
3. Zhang, Y., Gu, Q.-S. & Ho, L. C. Stellar and dust properties of local elliptical galaxies:
clues to the onset of nuclear activity. Astron. Astrophys. 487, 177–183 (2008).
4. Leonard, D. C. Supernova 2005cz in NGC 4589. IAU Circ. 8579 (2005).
5. Filippenko, A. V. et al. Supernovae 2001co, 2003H, 2003dg, and 2003dr. IAU Circ.
8159 (2003).
6. Foley, R. J. et al. SN 2008ha: an extremely low luminosity and exceptionally low
energy supernova. Astron. J. 138, 376–391 (2009).
7. Nomoto, K., Tominaga, N., Umeda, H., Kobayashi, C. & Maeda, K.
Nucleosynthesis yields of core-collapse supernovae and hypernovae, and galactic
chemical evolution. Nucl. Phys. A 777, 424–458 (2006).
8. Fransson, C. & Chevalier, R. A. Late emission from supernovae: a window on
stellar nucleosynthesis. Astrophys. J. 343, 323–342 (1989).
9. Maeda,K.etal.SN2006ajassociatedwithXRF060218atlatephases:nucleosynthesis
signature of a neutron-driven explosion. Astrophys. J. 658, L5–L8 (2007).
10. Shigeyama, T. et al. Theoretical light curves of type IIb supernova 1993J.
Astrophys. J. 420, 341–347 (1994).
328
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
11. Nomoto, K. et al. A carbon-oxygen star as progenitor of the type-Ic supernova
1994I. Nature 371, 227–229 (1994).
12. Mazzali, P. A., Nomoto, K., Patat, F. & Maeda, K. The nebular spectra of the hypernova
SN 1998bw and evidence for asymmetry. Astrophys. J. 559, 1047–1053 (2001).
13. Mazzali, P. A. et al. The aspherical properties of the energetic type Ic SN 2002ap
as inferred from its nebular spectra. Astrophys. J. 670, 592–599 (2007).
14. Tanaka, M. et al. Nebular phase observations of the type Ib supernova 2008D/X-
ray transient 080109: side-viewed bipolar explosion. Astrophys. J. 700,
1680–1685 (2009).
15. Nomoto, K. & Hashimoto, M. Presupernova evolution of massive stars. Phys. Rep.
163, 13–36 (1988).
16. Nomoto, K. Evolution of 8–10 solar mass stars toward electron capture
supernovae. I — Formation of electron-degenerate O 1 Ne 1 Mg cores.
Astrophys. J. 277, 791–805 (1984).
17. Hakobyan, A. A. et al. Early-type galaxies with core collapse supernovae. Astron.
Astrophys. 488, 523–531 (2008).
18. Branch, D. et al. Direct analysis of spectra of type Ib supernovae. Astrophys. J. 566,
1005–1017 (2002).
19. Barbon, R. et al. SN 1993J in M81: one year of observations at Asiago. Astron.
Astrophys. 110 (Suppl.), 513–519 (1995).
20. Filippenko, A. V. et al. The type Ic supernova 1994I in M51: detection of helium and
spectral evolution. Astrophys. J. 450, L11–L15 (1995).
21. Sauer, D. N. et al. The properties of the ‘standard’ type Ic supernova 1994I from
spectral models. Mon. Not. R. Astron. Soc. 369, 1939–1948 (2006).
22. Maeda, K. et al. Asphericity in supernova explosions from late-time spectroscopy.
Science 319, 1220–1223 (2008).
23. Phillips, M. M. et al. The peculiar SN 2005hk: do some type Ia supernovae explode
as deflagrations? Publ. Astron. Soc. Pacif. 119, 360–387 (2007).
24. Valenti, S. et al. A low energy core-collapse supernova without a hydrogen
envelope. Nature 459, 674–677 (2009).
25. Richmond, M. W. et al. UBV RI photometry of the type Ic SN 1994I in M51.
Astrophys. J. 111, 327–339 (1996).
26. Stritzinger, M. et al. The He-rich core-collapse supernova 2007Y: observation
from X-ray to radio wavelengths. Astrophys. J. 696, 713–728 (2009).
27. Botticella, M. T. et al. SN 2008S: an electron capture SN from a super-AGB
progenitor? Mon. Not. R. Astron. Soc. (in the press); preprint at Æhttp://arXiv.org/
abs/0903.1286æ (2009).
28. Iwamoto, K. et al. Theoretical light curves for the type Ic supernova SN 1994I.
Astrophys. J. 437, L115–L118 (1994).
29. Patat, F. et al. The metamorphosis of SN 1998bw. Astrophys. J. 555, 900–917 (2001).
30. Yoshii, Y. et al. The optical/near-infrared light curves of SN 2002ap for the first
140 days after discovery. Astrophys. J. 592, 467–474 (2003).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank D. Leonard for permission to publish his early-time
spectrum of SN 2005cz, and S. Valenti for permission to use his spectra of
SN 2008ha. We acknowledge advice and help from P. A. Mazzali. This work is based
on observations collected at the 2.2-m telescope at the Calar Alto Observatory
(Sierra de Los Filabres, Spain), at the Keck Telescope, and at the Subaru Telescope
(operated by the National Astronomical Observatory of Japan). We are grateful to
the staff members at the observatories for their assistance, especially to T. Sasaki,
K. Aoki, G. Kosugi, T. Takata and M. Iye. This research is supported by the World
Premier International Research Center Initiative (WPI Initiative), MEXT, Japan, and
by the Grant-in-Aid for Scientific Research of the Japan Society for the Promotion of
Science (JSPS) and MEXT. M.T. is supported through the JSPS Research Fellowship
for Young Scientists. J.D. is supported by the NSFC and by the 973 Program of China.
Author Contributions K.S.K., K.M., K.N., J.D. and E.P. organized the observations
and discussions; K.M., K.N and K.S.K. wrote the manuscript; K.S.K., S.T. and K.I.
were responsible for data acquisition and reduction; J.D. and E.P. were the Principal
Investigators of the relevant Subaru programmes, S05B-132 and S05B-054,
respectively; M.T. and S.T. contributed to discussions; and T.H. provided expertise
on data acquisition at the Subaru Telescope.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to K.S.K. (kawabtkj@hiroshima-u.ac.jp).
Vol 465 | 20 May 2010 | doi:10.1038/nature09054
GaAs photovoltaics and optoelectronics using
releasable multilayer epitaxial assemblies
Jongseung Yoon1*, Sungjin Jo1,4*, Ik Su Chun2, Inhwa Jung1, Hoon-Sik Kim1, Matthew Meitl3, Etienne Menard3,
Xiuling Li2, James J. Coleman2, Ungyu Paik4 & John A. Rogers1,2
Compound semiconductors like gallium arsenide (GaAs) provide
advantages over silicon for many applications, owing to their direct
bandgaps and high electron mobilities. Examples range from effi-
cient photovoltaic devices1,2 to radio-frequency electronics3,4 and
most forms of optoelectronics5,6. However, growing large, high
quality wafers of these materials, and intimately integrating them
on silicon or amorphous substrates (such as glass or plastic) is
expensive, which restricts their use. Here we describe materials
and fabrication concepts that address many of these challenges,
through the use of films of GaAs or AlGaAs grown in thick, multi-
layer epitaxial assemblies, then separated from each other and
distributed on foreign substrates by printing. This method yields
large quantities of high quality semiconductor material capable of
device integration in large area formats, in a manner that also
allows the wafer to be reused for additional growths. We demon-
strate some capabilities of this approach with three different appli-
cations: GaAs-based metal semiconductor field effect transistors
and logic gates on plates of glass, near-infrared imaging devices on
wafers of silicon, and photovoltaic modules on sheets of plastic.
These results illustrate the implementation of compound semi-
conductors such as GaAs in applications whose cost structures,
formats, area coverages or modes of use are incompatible with
conventional growth or integration strategies.
Some of the most daunting technical challenges associated with
established methods for using GaAs occur in terrestrial photovoltaics,
where the extremely high efficiencies of GaAs solar cells7,8 might other-
wise offer compelling opportunities. Here, the demands on cost and the
necessity for integration over large areas on glass or other foreign sub-
strates lead to requirements that lie well outside current capabilities.
Previously known techniques, most notably epitaxial lift-off9–12, sepa-
rate GaAs solar cells from their growth substrates to reduce their cost or
weight10,11, but difficulties in handling the lifted materials12 limit their
use. Despite continuing progress11,12, photovoltaic epitaxial lift-off has
remained as a research topic for over 30 years, with no commercial
implementation. More tractable, yet still difficult, problems appear in
advanced electronics and optoelectronics where, as examples, device-
level integration of compound semiconductors with silicon electronics
can improve high speed operation in radio-frequency systems13, add
capabilities for light emission in devices for lightwave communica-
tions14 or enable efficient photodetection for night-vision cameras15.
Wafer bonding16, heteroepitaxy17, pick-and-place techniques18 and
chip-scale packaging methods19 can meet certain requirements of these
and other applications, but not without some combination of disad-
vantages that include restricted scales or modes for integration, limited
substrate compatibility and poor cost effectiveness.
1
Department of Materials Science and Engineering, Beckman Institute for Advanced Science and Technology, and Frederick Seitz Materials Research Laboratory, University of Illinois
at Urbana-Champaign, Urbana, Illinois 61801, USA. 2Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801, USA.
3
Semprius, Inc., Durham, North Carolina 27713, USA. 4Division of Materials Science Engineering, WCU Department of Energy Engineering, Hanyang University, Seoul 133-791, South
Korea.
*These authors contributed equally to this work.
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
Alternative methods that avoid these drawbacks and, at the same
time, provide realistic means of distributing large quantities of material
over large areas on amorphous substrates could have significant
practical value. Recent work by us and others demonstrates potential
for compound semiconductor nanomaterials and unusual guided or
deterministic assembly methods to address some of these chal-
lenges20–22. Limitations in the amounts and formats of these materials,
difficulties in their scalable integration into devices and/or uncertain
compositions and properties frustrate their use for many important
applications. We present here concepts that overcome these issues by
exploiting device-quality compound semiconductor films epitaxially
grown in thick, multilayer formats specially designed for separation
from one another, release from the wafer and subsequent integration in
diverse layouts on various substrate types. This approach greatly
reduces the need for wafer refinishing, compared to analogous single-
layer lift-off approaches, and eliminates cycles of loading and unloading
of wafers into the growth chamber. In favourable cases, the net effect
can be an increase in throughput and a decrease in cost of more than an
order of magnitude (Supplementary Information). The outcomes are
applicable to every area of use for compound semiconductors, from
electronics to optoelectronics and photovoltaics, as illustrated in the
following.
The basic strategy appears in Fig. 1. Metal organic chemical vapour
deposition (MOCVD) forms multiple layers of GaAs and/or AlGaAs
or other related materials with compositions and layouts that match
device requirements, with separating films of AlAs (Al0.95Ga0.05As).
Figure 1a presents a simple case of 14 such layers, similar to designs
used in the electronics examples described next. Figure 1b shows
secondary ion mass spectrometry (SIMS) depth profiles of the com-
position of a structure with the layout of Fig. 1a, illustrating well
defined layers with abrupt interfaces. Many more layers are possible,
limited by layer morphologies, doping profiles and differential stresses
and, ultimately for certain systems (for example, Fig. 1e), by practical
constraints on growth times (Supplementary Information). Patterned
vertical etching through the entire stack exposes the sidewalls of the
layers and delineates the material into separated blocks with desired
lateral dimensions. Immersion in hydrofluoric acid (HF) selectively
eliminates the AlAs layers, thereby releasing large collections of pieces
of GaAs (and/or multilayers with AlGaAs or other materials), with
sizes that can range from micrometres to centimetres, and thicknesses
from several nanometres to micrometres.
Figure 1c shows a cross-sectional scanning electron microscope
(SEM) image of the system schematically illustrated in Fig. 1a after
vertical etching and brief immersion in HF. The etching rate for
AlxGa12xAs over GaAs depends on the aluminium fraction23.
329
LETTERS
a Release; transfer
GaAs
AIAs
GaAs
Etch in HF Regrow
substrate
b c
As AlAs GaAs
Counts (a.u.)
Al
Ga
0 1 2 3
Depth (µm)
d
200 µm
2 mm
e f
GaAs
AIAs
1 µm 500 nm
Figure 1 | Schematic illustration, optical and SEM images, and SIMS profile of
GaAs/AlAs multilayers. a, Schematic illustration of a multilayer stack of GaAs/
AlAs and schemes for release through selective etching of the layers of AlAs.
b, Corresponding SIMS profile of this stack. c, Cross-sectional SEM image after
partial etching of the AlAs layers. d, Optical image of a large collection of GaAs
solar cells formed by release from a three-layer stack, and then solution casting
onto another substrate. Inset, high magnification view. e, Cross-sectional SEM
image of a 40 repeat multilayer stack of GaAs (200 nm)/AlAs with ultrathin
(20 nm) AlAs sacrificial layers. Inset, high-magnification view. f, Cross-sectional
SEM image (coloured) of a heterogeneous multilayer stack composed of three
layers for MESFETs (green), one layer for an NIR detector (yellow) and one
layer for a single junction solar cell (blue). The substrate is purple. Each of the
device layers is separated by 20 nm AlAs (red). Details of stack design and
corresponding SIMS profile are in Supplementary Information.
Increasing the fraction (for x . 0.7) results in highly enhanced rates
of etching in HF. For example, etching of GaAs occurs at a rate 106
times lower than that of AlAs (ref. 24). The GaAs component can
incorporate other layers of epitaxial materials, with the constraint
that their etching rate is sufficiently low in HF and that thermally
driven diffusion of any mobile species during growth does not lead to
adverse effects. In the following, we exploit this flexibility to grow
multilayer stacks for metal semiconductor field effect transistors
(MESFETs), Schottky diodes, near-infrared (NIR) photodetectors
and single-junction photovoltaic cells. Figure 1d shows a macro-
scopic pile of such cells (square platelets ,200 mm on a side, and
,4.5 mm thick), formed from an AlAs-separated three-layer assembly
330
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
of GaAs stacks that each consist of six Zn and Si doped layers of GaAs
and AlGaAs; details appear below. The collective assembly on the
growth wafer incorporates 22 epitaxial layers, with a total thickness
of ,17 mm. In this case, and others, the wafer surface can be re-
finished after complete lift-off to prepare it for additional growths.
These concepts can be extended to extremely large multilayer stacks
(for example, 40-layer repeats; Fig. 1e) and to heterogeneous collec-
tions of devices (for example, solar cells, transistors and photodetec-
tors; Fig. 1f).
To demonstrate the simplest example first, we formed multilayers
(seven repeats) of n-type GaAs (200 nm; Si-doped to 4 3 1017 cm23)
and AlAs (300 nm) on a (100)GaAs substrate, for building MESFETs.
Complete etching and release of GaAs active elements, followed by
solution assembly on a target substrate, represents one conceivable
route to device integration. Here we use extensions of deterministic
assembly procedures described elsewhere25,26, in which each of the
individual GaAs layers in the stack is released and then transfer
printed one at a time, in a step and repeat fashion, onto a glass
1 µm substrate coated with a thin layer of polyimide. Printing tools, with
yields .99.5%, submicrometre placement accuracy and throughputs
corresponding to millions of dies per hour, are under commercial
development. Details, including cost estimates, appear in Sup-
plementary Information. Fabricating MESFETs from the printed
GaAs membranes involves patterned etching to define the channel
regions, followed by additional processing to create ohmic source and
drain contacts (Pd/Ge/Au) and Schottky gates (Ti/Au). Figure 2a and
b shows schematic illustrations and optical images, respectively.
Figure 2c and d presents current/voltage (I/V) characteristics and
transfer curves of representative devices (with channel widths of
130 mm, channel lengths of 25 mm and gate lengths of 3 mm) from each
of the top five layers in the stack. In all cases the layer-to-layer varia-
tions in properties are random, with magnitudes comparable to
device-to-device variations observed in any given single layer.
The electrode layouts enable probing of high frequency response
with a vector network analyser. Representative results appear in
Solar cell Fig. 2e, which plots the current gain (H21) and maximum available
gain (MAG) as a function of frequency for the fourth layer device,
NIR indicating that the unity current gain frequency (fT) and unity power
detector
gain frequency (fmax) are around ,2 and ,6 GHz, respectively, even
MESFET
for this relatively coarse channel dimension, as expected on the basis of
the high electron mobility in GaAs (ref. 27). This response, as well as the
maximum transconductances (,3.4 mS), on/off current ratios (,106),
maximum current outputs per channel width (,1.1 3 1025 A mm21)
and other parameters inferred from electrical testing, is consistent with
expectation based on devices with similar designs formed on a GaAs
wafer. Integration of multiple MESFETs yields logic gates, as a step
towards integrated circuits. Figure 2f shows optical images of NAND
and NOR gates that each consist of three interconnected devices. The
load and switching transistors have channel lengths of 55 and 25 mm,
respectively, both with gate lengths of 3 mm and channel widths of
130 mm. Figure 2g and h presents output–input characteristics of
NAND and NOR gates, respectively. In both cases, the logic ‘0’ and
‘1’ input signals correspond to 23 V and 1 V, respectively, for Vdd (the
drain voltage of the load transistor) equal to 5 V. The logic ‘0’ and ‘1’
outputs of the NAND gate are 0.5–0.9 V and 5 V, respectively; the
corresponding outputs of the NOR gate are 0.3–0.6 V and 5 V.
NIR imagers composed of interconnected arrays of GaAs photo-
diodes and blocking diodes provide a second, more advanced and
different means of exploiting the multilayer concepts. Figure 3a pro-
vides a schematic illustration of the device layout in this case. The
epitaxial stacks include bilayers of undoped GaAs (500 nm thick) and
n-type GaAs (50 nm thick; Si-doped to 8 3 1016 cm23), in a four-
layer assembly separated by AlAs (300 nm thick). To demonstrate a
processing option different from that for the MESFETs, we built fully
functional, interconnected systems using each active layer on the
growth wafer, followed, in sequential fashion, by transfer to a target
substrate. We chose silicon in this case owing to its established use for
NATURE | Vol 465 | 20 May 2010 LETTERS

100 µm n-GaAs SD 150 µm

a b a b
S D S
D GaAs MSM
S G PU
n-GaAs
PI G

Glass Silicon
500 µm
250 µm c d
c 1.6 d 1.5
First First 1.2 First
Second 10–4 Second
Third Third
Second
1.2

IDS1/2 (mA1/2)
1.0 0.9 Third
Fourth Fourth
IDS (mA)

IDS (A)

I (µA)
Fifth 10–6 Fifth Fourth
0.8 0.6
0.5
0.4 10–8 0.3

0.0 0.0
0.0 10–10 –4 –2 0 2 4
0.0 0.5 1.0 1.5 –1.5 –1.0 –0.5 0.0 0.5
VDS (V) VGS (V) V (V)
5 mm
e f VO Vdd NOR GND
NAND e f
30 VB
H21
H21, MAG (dB)

MAG
20
VB
VA
10 –20dB per
decade
GND VA
0
0.1 1 fT fmax 200 µm Vdd VO
f (GHz)
g 6 h 6
(0,1)1 (0,0)1 (1,0)1

4 VO (first) VA
4 VA (0,0)1 VO (first) Figure 3 | Multilayer GaAs NIR imagers. a, Schematic illustration of a GaAs
VO (second) VB VB VO (second) metal–semiconductor–metal (MSM) NIR detector on a Si wafer coated with
VO (third)
V (V)
V (V)

2 VO (third) 2 a photocurable polyurethane (PU). Inset, Schottky blocking diode (SD).

0 (1,1)0
–2
Time
Figure 2 | Multilayer GaAs MESFETs and logic circuits. a, Schematic
illustration of a GaAs MESFET on a polyimide (PI) coated glass substrate.
b, Optical image of arrays of MESFETs on glass substrate. Inset, a single
MESFET with source (S), drain (D) and gate (G) metal layers. c, VDS
(drain–source voltage) versus IDS (drain–source current) curves of
MESFETs fabricated from first, second, third, fourth and fifth layers at
gate–source voltages (VGS) of 0.4, 0.2, 0, 20.2, 20.4 and 20.6 V, from top to
bottom. d, IDS versus VGS transfer curves of MESFETs fabricated from first,
second, third, fourth and fifth layers, at VDS 5 1.5 V. e, Amplitude plots for
current gain (H21) and MAG measured from the fourth layer device. The
unity current gain frequency (fT) and unity power gain frequency (fmax) are
,2 and ,6 GHz, respectively. f, Optical image of NAND and NOR gates on
polyimide (VA, VB, input voltages for switching transistors; Vdd, drain
voltage for load transistor; VO, output voltage). g, Output–input
characteristics of NAND gates using devices from the first, second and third
layers. h, As g but for NOR gates.
image processing in conventional infrared cameras formed by solder
bump bonding28. Forming Schottky contacts (Ti/Au) to the top,
undoped GaAs layers yields metal–semiconductor–metal diodes.
Such contacts combined with ohmic contacts (Pd/Ge/Au) to the Si-
doped GaAs provide Schottky diodes. The metal–semiconductor–
metal devices enable NIR detection, while the Schottky diodes block
the reverse bias current to prevent crosstalk for passive matrix readout
of the array.
Figure 3b shows an image of the layout of a photodiode/blocking
diode pair, in the geometry used for the NIR imagers. Figure 3c
presents representative I/V characteristics and photoresponse for
typical devices formed using the first, second, third and fourth layers
b, Optical image of a NIR imager consisting of a 16 3 16 array of detectors
0 (0,1)0 (1,1)0 (1,0)0 (12 pixels are shown). Inset, a unit cell before interconnect metallization.
c, I/V characteristics of a cell formed with material from the first, second,
–2
third and fourth layers. Open circles and lines correspond to responses in the
Time dark and under NIR illumination (wavelength 830 nm), respectively.
d, Photograph of an NIR imager mounted on a printed circuit board.
e, Image acquired with an NIR imager formed with material from the second
layer. f, As e but using material from the fourth layer. Insets in e and
f correspond to objects that were imaged.
in the stack. The open circles and lines correspond to responses in the
dark and under NIR illumination (830 nm), respectively. The data
indicate comparable and reproducible behaviour for all layers. Ratios
of currents in the on and off states are ,200, sufficient for passive
matrix readout. A spin cast, photodefinable epoxy serves as a mech-
anical support and substrate for metal interconnection lines between
256 such devices, to yield independently addressable arrays (16 3 16
elements) for imaging. Transfer printing delivers the completed sys-
tem to a silicon wafer that mounts on a printed circuit board interface
to a computer for image acquisition (Fig. 3d). The yields of properly
functioning pixels were better than ,96% in individual pixels and
,70% in interconnected arrays, limited mainly by breaks in inter-
connection metals. Figure 3e and f shows images of pictures of owls
collected with NIR illumination and two separate cameras formed
with the second and fourth layers, respectively. Image collection
involved a scanning mode29, to achieve an effective level of resolu-
tion much higher than that defined by the numbers of pixels and
to provide oversampling for minimizing adverse effects of non-
uniformities and malfunctioning pixels.
The third demonstration is in photovoltaics, where six-layer stacks
of Zn and Si doped GaAs and AlGaAs form single-junction solar cells,
epitaxially grown in trilayer assemblies separated by layers of AlAs
(1 mm thick). Figure 4a illustrates the layout of the materials and the
331
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

100 µm
a b g
n-contact n-GaAs

p-contact n-AlGaAs n-contact

n-GaAs
p-GaAs
p-contact
p-AlGaAs
p-GaAs

Epoxy PET

500 µm
c –20
d 1.2
–20 First
Second 1.0
J (mA cm–2)

–15
Jsc (mA cm–2)

–16 Third

Voc (V), FF
0.8
–10 First –12 0.6
Second
–8 0.4 2 mm
–5 Third
–4 0.2
0
0.0 0.2 0.4 0.6 0.8 1.0 0 0.0
V (V) Jsc Voc (V) FF
h
e 1.0 f –400 0.4 i –40 0.4
20 30
0.8
Jsc (mA cm–2) –300 10 cells 0.3 –30 10 cells 0.3
IQE, EQE

15

P (mW)
P (mW)
0.6
I (µA)

I (µA)
in parallel in series
η (%)

IQE (first) –200 0.2 –20 0.2

0.4 10 η (first) 20
EQE (first)
η (second) 0.1 0.1
IQE (second)
5 –100 Current –10 Current
0.2 Jsc (first)
EQE (second)
Jsc (second) Power Power
0.0 0 10 0 0.0 0 0.0
300 500 700 900 No ARC With DLARC 0.0 0.2 0.4 0.6 0.8 1.0 0 2 4 6 8 10
Wavelength (nm) V (V) V (V)

Figure 4 | Multilayer GaAs single-junction solar cells. a, Schematic
illustration of GaAs single-junction solar cell on a PET substrate coated with
a photodefinable epoxy. b, Optical image of arrays of such devices formed on
the source wafer. Inset, magnified view of top (n-type) and bottom (p-type)
contacts. c, Representative light current density (J)-voltage (V) curves for
Zn-doped solar cells formed from the first (top), second (middle) and third
(bottom) layers, under AM1.5D illumination measured on the source wafer
with a single-layer ARC of Si3N4. d, Short-circuit current density (Jsc), fill
factor (FF) and open circuit voltage (Voc) of first, second and third layer
geometry of the contacts. Each GaAs and AlGaAs epitaxial construct
consists of an n-type top contact layer (0.2 mm thick GaAs; Si-doped),
an n-type window layer (40 nm thick Al0.4Ga0.6As; Si-doped), an
n-type emitter (0.1 mm thick GaAs; Si-doped), a p-type base (2 mm
thick GaAs; Zn-doped), a p-type back surface field layer (0.1 mm thick
Al0.3Ga0.7As; Zn-doped) and a p-type bottom contact layer (2 mm
thick GaAs; Zn-doped). Vertical etching delineates arrays of cells
with lateral dimensions of ,500 3 500 mm on the source wafer.
Controlled etching of selected regions (35 3 500 mm) near the edges
of these cells exposes the bottom p-contact layer. This process, fol-
lowed by removal of all but a small part of the top n-contact layer
(30 3 330 mm) prepares the wafer for deposition and patterning of
contact metals. The n and p contacts consist of Pd/Ge/Au and Pt/Ti/
Pt/Au, respectively (Fig. 4b). Current density/voltage (J/V) curves
and extracted characteristics of individual cells from each layer mea-
sured under simulated AM1.5D (air mass 1.5 direct) illumination
(1,000 W m22) at room temperature with a single-layer antireflection
coating (ARC; Si3N4) appear in Fig. 4c and d, respectively, where the
current densities correspond to the active areas of the cells. AM1.5D
illumination provides measurements that are relevant to terrestrial
concentrator photovoltaics.
Moderate decreases in J from the top to the bottom layer are
systematic, and can be attributed to the diffusion of Zn in the p-type
GaAs layers, as revealed by SIMS analysis (Supplementary Informa-
tion). This behaviour might be eliminated by the use of alternative
dopants, such as carbon, that have negligible diffusion coefficients.
Figure 4e shows external and internal quantum efficiencies (EQE and
IQE, respectively) with an ARC for top and middle layer devices. The
calculated efficiencies based on these measurements and AM1.5D
332
©2010 Macmillan Publishers Limited. All rights reserved
devices. Error bars, max./min. e, IQE and EQE of first and second layer
devices. f, Projected efficiencies (g) and Jsc values without ARC and with
double-layer ARCs (DLARC; MgF2/ZnS) for devices formed using material
from the first and second layers. g, Photograph of a solar module consisting
of a 10 3 10 array of GaAs solar cells (each ,500 mm 3 500 mm) on a PET
substrate. h, Light current–voltage (I/V) and power–voltage (P/V) curves for
such a module with parallel interconnection of 10 cells. i, As h but with series
interconnection. Both measurements were made in a flat configuration.
reference spectrum without an ARC are ,14.5% and ,13.5% for
top and middle layer devices, respectively (Fig. 4f). A double-layer
ARC (such as MgF2/ZnS; Fig. 4f) can increase the efficiency to
,20.5% (,19.5% for the middle layer; see Supplementary Informa-
tion), which approaches the performance of some of the best
reported devices designed for ultrahigh concentration (,20.6%)30.
The performance can be further improved by the use of optimized
cell designs and ARCs. These ultrathin cells provide other interesting
features, such as the ability to integrate onto thin (,50 mm) sheets of
polyethylene terephthalate (PET) for mechanically flexible modules
capable of bending to radii of curvature of ,5 mm without degra-
dation in performance. Figure 4g presents an image of such a device,
consisting of 100 interconnected cells (10 3 10 array), wrapped onto
a cylindrical support. The cells can be combined in series and/or
parallel configurations to yield output power at high or low voltages
(or anything in between), thereby highlighting an important advant-
age of the use of large collections of small cells. Figure 4h and i
presents output characteristics of devices that use parallel and series
interconnects for output voltages of 0.93 V and 8.9 V, respectively,
with similar total maximum output power (0.23 mW and 0.25 mW,
respectively). Similar processing steps can be used with large (.1 cm)
devices, as described in the Supplementary Information. Edge recom-
bination, even at 100 mm cell sizes, reduces the efficiency by less than
1% (Supplementary Information).
In conclusion, approaches reported here provide advantages in
integration possibilities, designs and cost, to create opportunities
that are inaccessible to established technologies. The remaining tech-
nical barriers to commercialization are specific to the application.
For example, solar cells that incorporate tunnel junctions may
NATURE | Vol 465 | 20 May 2010
require special growth strategies for production in thick, multilayer
configurations. Exploring other material systems, such as gallium
nitride and indium phosphide, and other devices, such as light emit-
ting diodes, lasers and sensors, represent additional promising direc-
tions for future work.
METHODS SUMMARY
Multilayer structures for MESFETs, NIR imagers and single-junction solar cells
were grown on (100)GaAs substrates by MOCVD. Etching through the top GaAs
layer (or GaAs/AlGaAs stack) with a mixture of citric acid and hydrogen peroxide
created vertical trenches. Partially etching into the underlying, exposed AlAs layer
with HF created a slight undercut around the periphery of the trenches. Spin
coating a layer of photoresist and patterning it by photolithography formed
structures that, after complete etching of the AlAs, tethered fully released GaAs
membranes to the underlying wafer. Techniques of transfer printing were used to
integrate the released GaAs/AlGaAs structures onto substrates of interest.
Repetitive application of this sequence of steps separated each of the layers in
the original stack, and integrated them onto target substrates. For MESFETs and
logic gates, n-type GaAs membranes were transfer printed to a glass substrate
coated with a partially cured polyimide. A sequence of photolithography, electron
beam evaporation and annealing steps yielded ohmic source and drain electrodes
and Schottky gate electrodes to form the MESFETs and logic gates. For NIR
imagers, fully interconnected device arrays were formed on the source wafer.
After undercut etching, the arrays were transfer printed to a Si wafer coated with
a photocurable polyurethane, and connected to a computer for image acquisition.
For single-junction solar cells, individual cells were formed through bottom con-
tact exposure, top contact etch, isolation and ohmic contact formation. Si3N4
served as an ARC. Arrays of the resulting cells were released by undercut-etching
and then were transfer printed to a PET substrate coated with a photodefinable
epoxy, followed by fabrication of metal interconnects and encapsulation layers.
Current–voltage measurements were conducted using a direct-current (d.c.)
source meter and a full-spectrum solar simulator. External and internal quantum
efficiencies were measured using a spectroradiometer system.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 9 December 2009; accepted 25 March 2010.
1. Bett, A. W., Dimroth, F., Stollwerck, G. & Sulima, O. V. III–V compounds for solar
cell applications. Appl. Phys. A 69, 119–129 (1999).
2. Bosi, M. & Pelosi, C. The potential of III–V semiconductors as terrestrial
photovoltaic devices. Prog. Photovolt. 15, 51–68 (2007).
3. Schwierz, F. & Liou, J. J. RF transistors: recent developments and roadmap toward
terahertz applications. Solid-State Electron. 51, 1079–1091 (2007).
4. Chang, K. Y. & Kai, F. GaAs High-Speed Devices (Wiley & Sons, 1994).
5. Wada, O. Optoelectronic integration based on GaAs material. Opt. Quantum
Electron. 20, 441–474 (1988).
6. Mokkapati, S. & Jagadish, C. III–V compound semiconductors for optoelectronic
devices. Mater. Today 12, 22–32 (2009).
7. Cotal, H. et al. III–V multijunction solar cells for concentrating photovoltaics.
Energy Environ. Sci. 2, 174–192 (2009).
8. Yamaguchi, M., Takamoto, T., Araki, K. & Ekins-Daukes, N. Multi-junction III–V
solar cells: current status and future potential. Sol. Energy 79, 78–85 (2005).
9. Konagai, M., Sugimoto, M. & Takahashi, K. High-efficiency GaAs thin-film solar-
cells by peeled film technology. J. Cryst. Growth 45, 277–280 (1978).
10. van Geelen, A. et al. Epitaxial lift-off GaAs solar cell from a reusable GaAs
substrate. Mater. Sci. Eng. B 45, 162–171 (1997).
11. Schermer, J. J. et al. Thin-film GaAs epitaxial life-off solar cells for space
applications. Prog. Photovolt. 13, 587–596 (2005).
12. van Deelen, J. et al. On the development of high-efficiency thin-film GaAs and
GaInP2 cells. J. Cryst. Growth 298, 772–776 (2007).
13. Laskar, J., Jokerst, N. M., Evers, N. & Chun, C. Thin-film integration for nanoscale
and high-frequency electronics on Si. Proc. SPIE 3212, 97–105 (1997).
14. Roelkens, G. et al. III–V/Si photonics by die to wafer bonding. Mater. Today 10,
36–43 (2007).
15. Rogalski, A. Infrared detectors: an overview. Infrared Phys. Technol. 43, 187–210
(2002).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
16. Tong, Q. Y. & Gosele, U. Semiconductor wafer bonding — recent developments.
Mater. Chem. Phys. 37, 101–127 (1994).
17. Bolkhovityanov, Y. B. & Pchelyakov, O. P. GaAs epitaxy on Si substrates: modern
status of research and engineering. Phys. Usp. 51, 437–456 (2008).
18. Demeester, P., Pollentier, I., Dedobbelaere, P., Brys, C. & Vandaele, P. Epitaxial lift-
off and its applications. Semicond. Sci. Technol. 8, 1124–1135 (1993).
19. Garrou, P. Wafer level chip scale packaging (WL-CSP): an overview. IEEE Trans.
Adv. Packag. 23, 198–205 (2000).
20. Baca, A. J. et al. Semiconductor wires and ribbons for high-performance flexible
electronics. Angew. Chem. Int. Edn 47, 5524–5542 (2008).
21. Yu, G., Cao, A. & Lieber, C. M. Large-area blown bubble films of aligned nanowires
and carbon nanotubes. Nature Nanotechnol. 2, 372–377 (2007).
22. Fan, Z. et al. Wafer-scale assembly of highly ordered semiconductor nanowire
arrays by contact printing. Nano Lett. 8, 20–25 (2008).
23. Yablonovitch, E., Gmitter, T., Harbison, J. P. & Bhat, R. Extreme selectivity in the
lift-off of epitaxial GaAs films. Appl. Phys. Lett. 51, 2222–2224 (1987).
24. Voncken, M. M. A. J., Schermer, J. J., Bauhuis, G. J., Mulder, P. & Larsen, P. K.
Multiple release layer study of the intrinsic lateral etch rate of the epitaxial lift-off
process. Appl. Phys. A 79, 1801–1807 (2004).
25. Meitl, M. A. et al. Transfer printing by kinetic control of adhesion to an elastomeric
stamp. Nature Mater. 5, 33–38 (2006).
26. Yoon, J. et al. Ultrathin silicon solar microcells for semitransparent, mechanically
flexible and microconcentrator module designs. Nature Mater. 7, 907–915
(2008).
27. Sun, Y. G. et al. Gigahertz operation in flexible transistors on plastic substrates.
Appl. Phys. Lett. 88, 183509 (2006).
28. Hayashi, T. An innovative bonding technique for optical chips using solder bumps
that eliminate chip positioning adjustments. IEEE Trans. Components Hybrids
Manufact. Technol. 15, 225–230 (1992).
29. Ko, H. C. et al. A hemispherical electronic eye camera based on compressible
silicon optoelectronics. Nature 454, 748–753 (2008).
30. Algora, C. et al. A GaAs solar cell with an efficiency of 26.2% at 1000 suns and
25.0% at 2000 suns. IEEE Trans. Electron. Dev. 48, 840–844 (2001).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank T. Banks, J. Soares and T. Spila for help using
facilities at the Frederick Seitz MRL; D. Stevenson, S. Robinson and D. Sievers for
help with photography, SEM and four point probe measurements, respectively; and
A. Gray for guidance in calculating the cell conversion efficiency and the calibration
of the solar simulator. J.Y. thanks D. Shir for help with optics simulation. This paper
is based partly on work supported by a National Security Science and Engineering
Faculty Fellowship (NIR cameras and MESFETs) and by the US Department of
Energy, Division of Materials Sciences (DEFG02-91ER45439), through the
Frederick Seitz MRL and Center for Microanalysis of Materials at the University of
Illinois at Urbana-Champaign (materials, growth aspects) and the Division of
Energy Efficiency and Renewable Energy (DE-FG36-08GO18021) (solar cells). We
also acknowledge the Center for Nanoscale Chemical Electrical Mechanical
Manufacturing Systems in the University of Illinois, which is funded by the National
Science Foundation (DMI-0328162) (printing-based manufacturing) and separate
funding from the National Research Foundation of Korea through a grant
(K2070400000307A050000310, Global Research Laboratory Program)
provided by the Korean Ministry of Education, Science and Technology. J.J.C.
acknowledges support by the Center for Energy Nanoscience, an Energy Frontier
Research Center funded by the US Department of Energy, Office of Science, Office
of Basic Energy Sciences (DE-SC0001013). J.Y. acknowledges support from a
Beckman Institute postdoctoral fellowship. S.J. thanks the Korea Research
Foundation for a postdoctoral fellowship (KRF-2008-357-D00134).
Author Contributions J.Y., S.J., I.S.C., I.J., H.-S.K., M.M., E.M., X.L., J.J.C., U.P. and
J.A.R. designed the experiments; J.Y., S.J., I.S.C., I.J., H.-S.K., M.M., E.M., X.L., J.J.C.,
U.P. and J.A.R. performed experiments and analyses; and J.Y., S.J., I.J. and J.A.R.
wrote the paper.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare competing financial interests:
details accompany the full-text HTML version of the paper at www.nature.com/
nature. Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to J.A.R. (jrogers@uiuc.edu, for any aspect of this work), U.P.
(upaik@hanyang.ac.kr, for MESFET processing) or M.M.
(matt.meitl@semprius.com, for design and characterization of solar cells).
333
doi:10.1038/nature09054
METHODS
Epitaxial multilayer growth. Multilayer structures for MESFETs, NIR imagers
and heterogeneous devices (MESFETs, NIR photodetectors and solar cells) were
grown on an n-type Si-doped GaAs(100) substrate using a Thomas Swan atmo-
spheric pressure MOCVD reactor. In all cases, a GaAs buffer layer with thickness of
200 nm was formed first, at 800 uC. Trimethyl gallium, trimethyl aluminium and
arsine (AsH3) served as precursors for Ga, Al and As, respectively. Disilane (Si2H6)
and dimethylzinc were used for Si (n-type) and Zn (p-type) doping, respectively.
Purified hydrogen served as the carrier gas. The growth temperature in all cases was
800 uC. Epitaxial multilayer structures for the MESFETs consisted of either seven or
forty bilayers of n-type GaAs (200 nm, 4 3 1017 cm23, Si-doped) and Al0.95Ga0.05As
(300 nm for 7 bilayers, 20 nm for 40 bilayers). For the NIR imagers, the structure
included four repeats of n-type GaAs (50 nm, 8 3 1016 cm23, Si-doped), undoped
GaAs (500 nm) and Al0.95Ga0.05As (300 nm). Details on heterogeneous stacks appear
in the Supplementary Information. Multilayer structures for n-on-p type single-
junction solar cells were grown by Epiworks Inc. with its commercial EpiSure solar
cell growth processes in a low-pressure MOCVD production system (modified
Aixtron 2600G3). The materials were deposited on (100)GaAs substrates (2u off
towards the nearest [111] plane) using standard precursors including trimethyl
gallium, AsH3, trimethyl aluminium, dimethylzinc and Si2H6. The multilayer struc-
tures for single-junction solar cells were composed of three repeats of n-type GaAs
(200 nm, (4–5) 3 1018 cm23, Si-doped), n-type Al0.4Ga0.6As (40 nm, 2 3 1018 cm23,
Si-doped), n-type GaAs (100 nm, 2 3 1018 cm23, Si-doped), p-type GaAs (2 mm,
3 3 1017 cm23, Zn-doped), p-type Al0.3Ga0.7As (100 nm, 1 3 1019 cm23, Zn-
doped), p-type GaAs (2 mm, 4 3 1019 cm23, Zn-doped) and Al0.95Ga0.05As (1.0 mm).
Multilayer release and assembly. Vertical etching through the entire thickness
of a multilayer stack in a mixture of citric acid (100 g of citric acid monohydrate
(Sigma-Aldrich); 83 ml of deionized water) and hydrogen peroxide (H2O2, 30%,
Fisher Scientific) at a 10:1 volume ratio, followed by removal of the AlAs in HF,
yielded solution suspensions of released GaAs structures. Subsequent assembly
onto a target substrate provided one route to devices. For work reported here,
vertical etching of trenches through the top GaAs layer (or GaAs, AlGaAs stack)
and partially into the underlying AlAs layer by brief immersion in HF created a
slight undercut around the periphery of the trenches. A layer of photoresist
formed structures that tethered fully released pieces of GaAs to the underlying
wafer after complete etching of the AlAs. Techniques of transfer printing were
used to integrate the released GaAs structures onto substrates of interest.
Repetitive application of this sequence of steps completed the process. The
Supplementary Information provides additional details.
MESFETs and logic circuits. Etching according to procedures described above
prepared the GaAs source wafer (Fig. 1a) for delivery of organized collections of
GaAs membranes to a glass substrate coated with a partially cured layer of
polyimide. Removing residual photoresist with acetone, and fully curing the
©2010 Macmillan Publishers Limited. All rights reserved
polyimide by baking at 250 uC for 2 h completed the process. A sequence of
photolithography, electron beam evaporation and annealing steps yielded ohmic
source and drain electrodes (Pd/Ge/Au (5/35/80 nm); 175 uC for 60 min) and
Schottky gate electrodes (Ti/Au (10/100 nm)) to form the MESFETs. Similar
procedures formed interconnects for the logic gates. d.c. measurements of
MESFETs were carried out using a semiconductor parameter analyser
(Agilent, 4155C). High frequency measurement used a network analyser
(Agilent, 8720) calibrated with a standard-open-load-through technique. The
Supplementary Information provides additional details.
NIR imagers. Ohmic and Schottky contacts used Pd/Ge/Au (5/35/80 nm,
annealed at 175 uC for 60 min) and Ti/Au (30/100 nm), respectively. Wet etching
with a mixture of citric acid and hydrogen peroxide at a 10:1 volume ratio removed
the n-type GaAs layer. A photodefinable epoxy (SU8, Microchem) served as a
support structure for interconnecting metal lines and as an encapsulation layer.
After etching the AlAs in concentrated HF, the entire NIR imager array was
transferred to a silicon wafer coated with a photocurable polyurethane as an
adhesive (NOA 61, Norland Products). Electrical measurements of individual
pixels were conducted using a semiconductor parameter analyser (Agilent,
4155C). The detector was mounted in a printed circuit board and connected to
a computer for image acquisition. The Supplementary Information provides addi-
tional details.
Solar cells and flexible modules. The p-type bottom contact layer was exposed
by etching with mixtures of citric acid and hydrogen peroxide at different volume
ratios (citric acid:H2O2 5 10:1 and 4:1) in a sequential manner. Next, most of the
top contact layer except the area for the ohmic contact was selectively etched away
with a 4:1 mixture of citric acid and hydrogen peroxide. Etching with a 10:1
mixture of citric acid and hydrogen peroxide provided isolation of individual
cells. Ohmic contacts for both top and bottom contact layers were then formed by
electron beam evaporation of Pd/Ge/Au (5/35/80 nm) and Pt/Ti/Pt/Au (10/40/
10/80 nm) for n-type and p-type contacts, respectively. Annealing at 175 uC under
N2 atmosphere was used for the n-type ohmic contact. Before the I/V measure-
ments on the source wafer, Si3N4 (thickness: ,80 nm, n < 1.76) was deposited by
plasma enhanced chemical vapour deposition (PlasmaTherm SLR) as a single-
layer ARC. Arrays of the resulting cells were undercut-etched in HF and transfer
printed to PET substrates using a photodefinable epoxy (SU8, Microchem) as an
adhesive. Patterning holes through the SU8 followed by lift-off of metals (Cr/Au:
10/300 nm) deposited by sputtering formed interconnects. Encapsulating with
SU8 completed the entire process. I/V measurements were carried out using a d.c.
source meter (Keithley, 2400) and a 1,000 W full-spectrum solar simulator (Oriel,
91192). External and internal quantum efficiencies of GaAs solar cells on the
source wafer were measured using an automated spectroradiometer system
(Optronic Lab., OL-750) in the wavelength range of 280–1,100 nm with 10 nm
resolution. The Supplementary Information provides additional details.
 Vol 465 | 20 May 2010 | doi:10.1038/nature09043

LETTERS
Robust warming of the global upper ocean
John M. Lyman1,2, Simon A. Good3, Viktor V. Gouretski4, Masayoshi Ishii5,6, Gregory C. Johnson2,
Matthew D. Palmer3, Doug M. Smith3 & Josh K. Willis7

A large ( 1023 J) multi-decadal globally averaged warming signal
in the upper 300 m of the world’s oceans was reported roughly a
decade ago1 and is attributed to warming associated with anthro-
pogenic greenhouse gases2,3. The majority of the Earth’s total
energy uptake during recent decades has occurred in the upper
ocean3, but the underlying uncertainties in ocean warming are
unclear, limiting our ability to assess closure of sea-level
budgets4–7, the global radiation imbalance8 and climate models5.
For example, several teams have recently produced different
multi-year estimates of the annually averaged global integral of
upper-ocean heat content anomalies (hereafter OHCA curves) or,
equivalently, the thermosteric sea-level rise5,9–16. Patterns of inter-
annual variability, in particular, differ among methods. Here we
examine several sources of uncertainty that contribute to differ-
ences among OHCA curves from 1993 to 2008, focusing on the
difficulties of correcting biases in expendable bathythermograph
(XBT) data. XBT data constitute the majority of the in situ mea-
surements of upper-ocean heat content from 1967 to 2002, and we
find that the uncertainty due to choice of XBT bias correction
dominates among-method variability in OHCA curves during
our 1993–2008 study period. Accounting for multiple sources of
uncertainty, a composite of several OHCA curves using different
XBT bias corrections still yields a statistically significant linear
warming trend for 1993–2008 of 0.64 W m22 (calculated for the
The individual OHCA curves all flatten out after around 2003,
with some variability among curves in the year in which this levelling
occurs. The causes of this flattening are unclear, but sea surface
temperatures have been roughly constant since 200017. Although
sea level has continued to rise steadily during this period, an increase
in the amount of water added to the ocean by melting continental ice
in recent years may account for most of this rise even with very little
change in ocean heat content6,7,18. However, this resolution of the sea-
level budget leaves the global energy budget with a large residual for
this time period, because it takes less energy to melt ice than to warm
the ocean for the equivalent sea-level rise18,19.
The flattening of OHCA curves also occurs around the time (2004)
that the Argo array of autonomous profiling floats first achieved near-
global coverage20 and became the primary source of OHCA data. The
Argo array affords year-round sampling of the temperature and salin-
ity of the ice-free oceans over the 0–2,000-m layer, with a nominal
separation of 3u in latitude and longitude. The transition from an
ocean temperature record consisting primarily of ship-based XBT
data to one dominated by high-quality conductivity–temperature–
depth (CTD) instrument data from Argo floats occurred between
10 Depth5,22
Depth12,22
0–700-m heat content anomaly (1022 J)

8 Depth13,22
Earth’s entire surface area), with a 90-per-cent confidence interval Depth-dependent
of 0.53–0.75 W m22. 6 temperature10
A host of choices must be made when computing OHCA curves. Depth9
4
Depth with SSH11,22
These choices include how to quality-control the data, which mapping Depth plus
2
technique to use, which baseline mean climatology to use, which annual temperature11,15
cycle to remove, how to treat unsampled or undersampled areas, and 0
how to correct biases in data from XBTs and other instruments. The −2
several teams working on the problem around the world make their
own choices and have produced apparently different OHCA curves16. −4
We assess the differences arising from these choices by overlaying −6
the curves produced by each team (Fig. 1). For this gross comparison, −8
the curves are aligned by removing their individual means for 1993–
2006, the time period over which most of the curves overlap. The −10
1994 1996 1998 2000 2002 2004 2006 2008 2010
curves show significant warming of the global upper ocean for the
Year
past 16 yr. Most of their warming rates (Table 1) are consistent,
agreeing within their published uncertainties. Figure 1 | OHCA curves using published methods. Globally integrated
However, there are differences in interannual variability among annual average OHCA curves from 0 to 700 m, estimated using methods
the curves. For example, from 1997 to 1998 (during a strong El Niño) published in papers cited in the key. All OHCA curves are estimated using
different baseline climatologies, mapping methods and XBT corrections
some curves appear to show cooling, some seem to show warming
(first reference). Types of XBT bias corrections used include depth, depth-
and others seem to show no change. Other years show similar varia- dependent temperature and depth with sea surface height (SSH; second
tions among curves. Offsets among the curves may originate from reference, if different from first). Each curve has had its 1993–2006 mean
differences in the reference period from which the heat content removed to aid comparison, except for the depth5,22 curve, which has been
anomalies are computed, making it difficult to assess differences aligned with the 1993–2002 mean of the other curves. Error bars, 1 s.e.m.
among the curves. (Supplementary Information).
1
Joint Institute for Marine and Atmospheric Research, University of Hawaii at Manoa, Honolulu, Hawaii 96822, USA. 2NOAA/Pacific Marine Environmental Laboratory, Seattle,
Washington 98115-6349, USA. 3Met Office Hadley Centre, Exeter EX1 3PB, UK. 4KlimaCampus, University of Hamburg, Grindelberg 5, 20144 Hamburg, Germany. 5Climate Research
Department, Meteorological Research Institute, 1-1 Nagamine, Tsukuba, Ibaraki 305-0052, Japan. 6Japan Agency for Marine-Earth Science and Technology, 3173-25 Showa-machi,
Kanazawa-ku, Yokohama 236-0001, Japan. 7Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California 91109, USA.

334
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
Table 1 | Warming rate bounds at 90% confidence
Type of XBT bias correction Warming rate normalized Effective degrees
to area of the Earth (W m22) of freedom
XBT depth plus 0.89–0.95 6.4
temperature11,15
XBT depth-dependent 0.52–0.58 5.4
temperature10
XBT depth9 0.35–0.45 7.2
XBT depth with SSH11,22 0.78–0.90 4.9
XBT depth12,22 0.26–0.42 5.0
XBT depth13,22 0.36–0.50 4.3
Confidence intervals are estimated from slopes of uncertainty-weighted lines fitted to the
OHCA curves in Fig. 1 for the years 1993–2006 (Supplementary Information). If no uncertainties
are plotted in Fig. 1 then the uncertainties of the pink curve were used in the weighted least-
squares fit. The first reference in the first column is for the mapping method and the second (if
different from the first) is for the XBT correction. For details on the effective degrees of freedom,
see Supplementary Information. SSH, sea surface height.
roughly 2000 and 20059, and marked a revolution in ocean observing.
Argo has dramatically increased the sampling in the Southern Ocean
since 200411. However, we find that this region continues to warm
from 2004 to 2008 (not shown) and apparently did not contribute to
the flattening of the OHCA curves. The fact that this transition
occurred at the same time as the flattening could be coincidental,
but also raises the possibility of a yet-undiscovered bias in the observ-
ing system.
The XBT was designed primarily to estimate ocean sound speed for
submarine warfare. As XBT data began to be used for more sensitive
climate research of the sort discussed here, partly correctable tem-
poral and spatial biases in both XBT temperature and XBT depth
were discovered9,10,15,21,22.
We compare the effects of five different methods of correcting
these XBT biases in OHCA curves. All the methods attempt to elimi-
nate the XBT biases by comparing XBT data with higher-accuracy
data from different platforms, mainly CTD and bottle data from
ships and Argo floats; one method in addition uses satellite-derived
sea surface height22. The corrections vary in time and in the types of
bias they are designed to remove. Biases can result from a variety of
problems with XBTs and how they are deployed, and they can affect
both the depths and the temperatures that make up a profile23. One
example is estimating how the rate at which the XBTs fall through the
water has changed over time as the instruments have subtly changed,
as this fall rate is used to infer the depths of the temperature mea-
surements. Choices that have been made are to focus on the fall-rate
issue and obtain corrections to the observation depths9,22, to correct
both temperature and depth biases using depth-dependent adjust-
ments to the temperature measurements10, and to obtain adjust-
ments for both the depths and the temperatures15.
A further difficulty in estimating XBT bias adjustments is that
there are multiple types and manufacturers of XBTs, each with its
own bias history. However, the XBT metadata are incomplete. For
example, XBT type and manufacturer were not recorded for about
half the data archive9,24. These partial metadata make consistent
application of the different XBT corrections difficult.
We acquired five differently assembled, quality-controlled XBT
data sets, each used to produce a different OHCA curve. Each research
team supplied their own bias-corrected XBT data for this study, ensur-
ing that the XBT corrections are applied in the manner intended.
Additionally, we apply four of the XBT corrections to the same EN3
(version 2a; http://www.metoffice.gov.uk/hadobs/en3) XBT data
set12,13,25 using published corrections (Fig. 2, dotted lines). Each of
these nine corrected XBT data sets are combined with other non-
XBT in situ data and mapped using the same method (Fig. 2, solid
lines; Supplementary Information), relative to the same 1993–2002
reference period. This procedure ensures that remaining differences
are entirely due to differences in XBT bias corrections and XBT quality
control. These differences are used to help determine the uncertainties
in OHCA estimates (Supplementary Information).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
10
0–700-m heat content anomaly (1022 J)
8
6
4
2
0
–2 Mean
Argo only
–4 Depth22
Depth plus
temperature15
–6 Depth-dependent
temperature10
–8 Depth9
Depth with SSH22
–10
1994 1996 1998 2000 2002 2004 2006 2008 2010
Year
Figure 2 | OHCA curves produced using the same mapping technique.
Solid lines are OHCA curves with a single 1993–2002 climatology and
variously corrected XBT data provided by individual research teams. Dashed
and dotted lines show the same thing as the solid lines, but using a single
2005–2008 climatology (dashed) or applying different published XBT
corrections to the identical EN3 (version 2a) XBT data set (dotted). The key
describes the type of XBT bias correction, with references. The trend
estimate of the mean curve (black line; red error bars, 90% confidence
intervals) is 0.64 6 0.11 W m22 and has 5.7 effective degrees of freedom
(Supplementary Information); red error bars show the overall uncertainty,
determined from combining all of the individual uncertainties in Fig. 3
assuming they are uncorrelated. Black error bars show XBT correction and
XBT quality-control uncertainty from Fig. 3. The difference between the
global means of the two climatologies has been added to the dashed lines.
A mean (Supplementary Information) of the nine resulting 16-yr
OHCA curves (Fig. 2, black line) has the same general shape as the
curves in Fig. 1: a decadal warming with a flattening after 2003.
Unlike Fig. 1, these curves can be compared directly and net offsets
among the different curves reflect biases due to differences in XBT
bias correction and quality control.
Argo has greatly improved the spatial and temporal sampling of
the global ocean11. This impact is most striking in the Southern
Ocean, where there have historically been few observations, especially
during the austral winter, owing to the remote location and harsh
environment. The anthropogenic-warming signal is thought to be
large in the Southern Ocean because of the role of varying air–sea
fluxes in the formation regions of bottom and intermediate
waters26–29. However, the poor historical sampling there means that
it is difficult to define an adequate baseline climatology for the
Southern Ocean. Recent modelling results suggest that increased
sampling in the Southern Ocean could lead to an artificial reduction
in OHCA30. To assess the sensitivity of the time series to the reference
period, the OHCA curves are also computed using a climatology for
2005–2008 (Fig. 2, dashed lines), a period when the Southern Ocean
was fairly well sampled by Argo floats. Because the global ocean was
warmer for this period than for 1993–2002, for comparison the
curves relative to the 2005–2008 climatology are shifted here by
7.7 3 1022 J, which is the difference in global heat content between
the two climatologies. The remaining differences among the OHCA
curves produced using the two different climatologies are negligible
and therefore add little to the overall uncertainty.
We also estimate uncertainty associated with mapping methodo-
logy, effects of irregular or sparse geographical sampling, and the use
of different climatologies (Fig. 3; Supplementary Information).
These uncertainty estimates are combined, assuming that they are
uncorrelated, to produce an overall uncertainty in the OHCA curve
(Fig. 3, red line). From 1995 to 2004, the XBT uncertainty dominates,
increasing from 1994 to 1999 as the number of coincident XBT–CTD
measurement pairs decreases10,15,22, and reaching a peak of about
335
LETTERS
3.5
0–700-m heat content s.e.m. (1022 J)
3
2.5
2 We focus on the period from 1993 onwards because the sampling uncertainty
1.5
1
0.5
0 2.
1994 1996 1998 2000 2002
Year
Figure 3 | Uncertainties in OHCA. Estimates of uncertainties (described in
Supplementary Information) arising from the choice of climatology
(magenta line), the method of XBT correction and XBT quality control
(black line), the choice of mapping methodology (blue line) and the effects of
irregular and sparse sampling (green line). All uncertainties are displayed as
1 s.e.m. The overall uncertainty, calculated by combining the individual
uncertainties assuming they are uncorrelated, is also shown (red line).
2.4 3 1022 J lasting from 1999 to 2000, with smaller contributions to
the overall uncertainty from the other terms. The XBT uncertainty
begins to decrease after 2000, as the Argo array of profiling floats
reaches maturity, decreasing the relative contribution of XBT data to
the OHCA estimates.
We fit a line using weighted least squares (Supplementary
Information) to the mean OHCA curve (Fig. 2, black line), using
the overall uncertainty (Fig. 2, red error bars) for each year in the fit.
These uncertainties are large enough that interannual variations,
such as the 2003–2008 flattening, are statistically meaningless. We
estimate a warming rate of 0.63 6 0.28 W m22 (uncertainties at the
90% confidence level) for 1993–2003, which is slightly (but not sig-
nificantly) higher than the value of 0.5 6 0.18 W m22 stated in the
Fourth Assessment Report of the Intergovernmental Panel on
Climate Change. The fit to the entire 16-yr record, including the
well-sampled Argo years, yields a more robust warming rate of
0.64 6 0.11 W m22. The large uncertainties in OHCA introduced
by the XBTs would undoubtedly have a similar effect on trends in
thermosteric sea level (not shown).
Since the Fourth Assessment Report, the discovery of a time-varying
bias in XBT data has prompted re-evaluations of the rate of upper-
ocean warming. We have carried out an intercomparison of these
estimates of ocean warming and made a comprehensive estimate of
the total uncertainty. We find that uncertainties in XBT bias correc-
tions are the dominant error source over the period 1993–2008, which
limits our ability to resolve interannual changes in ocean heat content.
However, despite these uncertainties, we still find a robust warming
over the 16-yr record. We are optimistic that with more work the
uncertainty associated with XBT bias corrections may be reduced in
future, possibly leaving the mapping methodology, which is also
improvable, as the largest uncertainty.
METHODS SUMMARY
We obtained Argo data from the US Argo Global Data Assembly Center (http://
www.usgodae.org/argo/argo.html) and non-Argo and non-XBT data from the
World Ocean Database 2005 (WOD05)24. We downloaded both Argo float and
WOD05 data on 14 September 2009. We estimate anomalies by taking differ-
ences both from a mean all-year climatology and from a seasonal cycle defined by
the quarterly annual cycle from the World Ocean Database 2001 (WOD01).
(WOD05 and WOD01 can be found at http://www.nodc.noaa.gov/OC5.) We
estimate baseline climatologies for different time periods using corrected XBT
data10 and by mapping all available data using an objective mapping technique11.
336
NATURE | Vol 465 | 20 May 2010
Unless otherwise stated, we map and integrate anomalies using a weighting that
assumes unsampled areas have the same OHCA as the mean of the sampled
Climatology areas11, which slightly increases our trend estimates relative to previously
XBT reported values.
Mapping The 90% confidence intervals given for the OHCA trend estimates in Fig. 2
Sampling and Table 1 are computed from weighted least-squares fits (Supplementary
Overall Information).
during this time period is well defined and relatively small11; there are sufficiently
few profiles from mechanical bathythermographs, which have their own biases9;
and the XBT correction that uses satellite-derived sea surface height is available
only from 199322.
Received 8 December 2009; accepted 22 March 2010.
1. Levitus, S., Antonov, J., Boyer, T. P. & Stephens, C. Warming of the world ocean.
Science 287, 2225–2229 (2000).
Barnett, T. P. et al. Penetration of human-induced warming into the world’s
2004 2006 2008 2010 oceans. Science 309, 284–287 (2005).
3. Levitus, S. et al. Anthropogenic warming of Earth’s climate system. Science 292,
267–270 (2001).
4. Willis, J. K., Chambers, D. P. & Nerem, R. S. Assessing the globally averaged sea
level budget on seasonal to interannual timescales. J. Geophys. Res. 113, C06015
(2008).
5. Domingues, C. M. et al. Improved estimates of upper-ocean warming and multi-
decadal sea-level rise. Nature 453, 1090–1093 (2008).
6. Cazenave, A. et al. Sea level budget over 2003–2008: a reevaluation from GRACE
space gravimetry, satellite altimetry and Argo. Global Planet. Change 65, 83–88
(2009).
7. Leuliette, E. W. & Miller, L. Closing the sea level rise budget with altimetry, Argo,
and GRACE. Geophys. Res. Lett. 36, L04608 (2009).
8. Murphy, D. M. et al. An observationally based energy balance for the Earth since
1950. J. Geophys. Res. 114, D17107 (2009).
9. Ishii, M. & Kimoto, M. Revaluation of historical ocean heat content variations with
time-varying XBT and MBT depth bias corrections. J. Oceanogr. 65, 287–299
(2009).
10. Levitus, S. et al. Global ocean heat content 1955–2007 in light of recently revealed
instrumentation problems. Geophys. Res. Lett. 36, L07608 (2009).
11. Lyman, J. M. & Johnson, G. C. Estimating annual global upper-ocean heat content
anomalies despite irregular in situ ocean sampling. J. Clim. 21, 5629–5641 (2008).
12. Palmer, M. D., Haines, K., Tett, S. F. B. & Ansell, T. J. Isolating the signal of ocean
global warming. Geophys. Res. Lett. 34, L23610 (2007).
13. Smith, D. M. & Murphy, J. M. An objective ocean temperature and salinity analysis
using covariances from a global climate model. J. Geophys. Res. 112, C02022
(2007).
14. Willis, J. K., Roemmich, D. & Cornuelle, B. Interannual variability in upper ocean
heat content, temperature, and thermosteric expansion on global scales.
J. Geophys. Res. 109, C12036 (2004).
15. Gouretski, V. & Reseghetti, F. On depth and temperature biases in
bathythermograph data: development of a new correction scheme based on the
analysis of a global ocean database. Deep-Sea Res. I doi:10.1016/j.dsr.2010.03.011
(in the press).
16. Palmer, M. et al. in Proc. OceanObs’09: Sustained Ocean Observations Inf. Soc. Vol. 2
(eds Hall, J., Harrison D. E. & Stammer, D.) (European Space Agency, 2010).
17. Knight, J. et al. Global oceans: do global temperature trends over the last decade
falsify climate predictions? Bull. Am. Meteorol. Soc. 90, S56–S57 (2009).
18. Trenberth, K. E. An imperative for climate change planning: tracking Earth’s global
energy. Curr. Opin. Environ. Sustainability 1, 19–27 (2009).
19. Trenberth, K. E. et al. in Proc. OceanObs’09: Sustained Ocean Observations Inf. Soc.
Vol. 1 (eds Hall, J., Harrison D. E. & Stammer, D.) (European Space Agency, 2010).
20. Roemmich, D. et al. Argo: the challenge of continuing 10 years of progress.
Oceanography 22, 46–55 (2009).
21. Gouretski, V. & Koltermann, K. P. How much is the ocean really warming?
Geophys. Res. Lett. 34, L01610 (2007).
22. Wijffels, S. E. et al. Changing expendable bathythermograph fall rates and their
impact on estimates of thermosteric sea level rise. J. Clim. 21, 5657–5672 (2008).
23. Reseghetti, F., Borghini, M. & Manzella, G. M. R. Factors affecting the quality of
XBT data – results of analyses on profiles from the Western Mediterranean Sea.
Ocean Sci. 3, 59–75 (2007).
24. Boyer, T. P. et al. World Ocean Database 2005 (ed. Levitus, S.) Ch. 1 (US
Government Printing Office, 2006).
25. Ingleby, B. & Huddleston, M. Quality control of ocean temperature and salinity
profiles – historical and real-time data. J. Mar. Syst. 65, 158–175 (2007).
26. Gille, S. T. Warming of the Southern Ocean since the 1950s. Science 295,
1275–1277 (2002).
27. Johnson, G. C. & Doney, S. C. Recent western South Atlantic bottom water
warming. Geophys. Res. Lett. 33, L14614 (2006).
28. Johnson, G. C., Purkey, S. G. & Bullister, J. L. Warming and freshening in the
abyssal southeastern Indian Ocean. J. Clim. 21, 5351–5363 (2008).
29. Johnson, G. C., Mecking, S., Sloyan, B. M. & Wijffels, S. E. Recent bottom water
warming in the Pacific Ocean. J. Clim. 20, 5365–5375 (2007).
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
30. AchutaRao, K. M, et al. Simulated and observed variability in ocean temperature
and heat content. Proc. Natl Acad. Sci. USA 104, 10768–10773 (2007).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements J.M.L. and G.C.J. were funded by the US National Oceanic and
Atmospheric Administration (NOAA) Climate Program Office and NOAA
Research. S.A.G., M.D.P. and D.M.S. were supported by the Joint DECC and Defra
Integrated Climate Programme DECC/Defra (GA01101). C. Domingues, S. Levitus,
T. Boyer, M. Ferrante and D. Trossman provided comments. C. Domingues,
S. Levitus, and T. Boyer also provided corrected XBT profiles. This is Pacific Marine
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
Environment Laboratory contribution number 3476 and Joint Institute for Marine
and Atmospheric Research contribution number 09-372.
Author Contributions J.M.L. led the writing and analysis, with writing contributions
from G.C.J., J.K.W., M.D.P. and S.A.G., and analysis contributions from S.A.G.,
V.V.G., M.I., M.D.P., D.M.S. and J.K.W.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to J.M.L. (john.lyman@noaa.gov).
337

LETTERS
Reconciling surface plate motions with rapid
three-dimensional mantle flow around a slab edge
Margarete A. Jadamec1{ & Magali I. Billen1
The direction of tectonic plate motion at the Earth’s surface and the
flow field of the mantle inferred from seismic anisotropy are well
correlated globally, suggesting large-scale coupling between the
mantle and the surface plates1,2. The fit is typically poor at subduction
zones, however, where regional observations of seismic anisotropy
suggest that the direction of mantle flow is not parallel to3–7 and may
be several times faster than6 plate motions. Here we present three-
dimensional numerical models of buoyancy-driven deformation
with realistic slab geometry for the Alaska subduction–transform
system and use them to determine the origin of this regional decoup-
ling of flow. We find that near a subduction zone edge, mantle flow
velocities can have magnitudes of more than ten times the surface
plate motions, whereas surface plate velocities are consistent with
plate motions8 and the complex mantle flow field is consistent with
observations from seismic anisotropy5. The seismic anisotropy
observations constrain the shape of the eastern slab edge and require
non-Newtonian mantle rheology. The incorporation of the non-
Newtonian viscosity9,10 results in mantle viscosities of 1017 to
1018 Pa s in regions of high strain rate (10212 s21), and this low vis-
cosity enables the mantle flow field to decouple partially from the
motion of the surface plates. These results imply local rapid transport
of geochemical signatures through subduction zones and that the
internal deformation of slabs decreases the slab-pull force available
to drive subducting plates.
In olivine grains, the dislocation creep deformation mechanism
leads to lattice-preferred orientation (LPO)10 and a non-Newtonian
rheology9, that is, a power-law dependence between strain rate and
stress. In the mantle, LPO can lead to anisotropy in seismic velocit-
ies11,12. Measurements of seismic anisotropy near subduction zones
show trench-parallel fast directions beneath slabs3,7 and above
slabs4,6, as well as fast directions that are perpendicular to the edges
of slabs13. If the fast seismic direction tracks mantle flow10,12, then
there is spatially variable flow at subduction zones and the motion of
surface plates is partially decoupled from mantle flow3.
Numerical models and laboratory experiments of subduction can
provide a fluid dynamics explanation for the patterns of mantle flow
inferred from observations of seismic anisotropy. Three-dimensional
(3D) models of subduction–transform systems, commonly in the
context of slab rollback, indicate a significant component of toroidal
flow around the slab edge and a small component of trench-parallel
flow in the mantle wedge and beneath the slab14–19. In addition, trench-
parallel seismic fast directions match stretching directions generated
by along-strike changes in slab geometry (dip and curvature) in models
with a non-Newtonian rheology20. These models provide valuable
insights into the dynamics governing mantle flow in subduction zones;
however, they are limited because they prescribe a velocity boundary
condition for the subducting plate (kinematically driven), do not
include an overriding plate or use a Newtonian rheology.
1
Department of Geology, University of California, Davis, California 95616, USA. {Present address: School of Mathematical Sciences & School of Geosciences, Monash University,
Clayton, Victoria 3800, Australia.
338
©2010 Macmillan Publishers Limited. All rights reserved
Vol 465 | 20 May 2010 | doi:10.1038/nature09053
The southern Alaska subduction zone is a good location to study to
understand how the 3D structure of a slab drives plate motion and
couples to mantle flow, because sufficient data exist to characterize
the slab shape, there are seismic observations constraining the pattern
of mantle flow and it is isolated from other subduction zones (Sup-
plementary Information). We construct 3D models of buoyancy-
driven flow for this subduction–transform plate boundary system
and compare model results with observations of Pacific plate motion
(PPM) and SKS fast-axis seismic directions5 (SKS waves are S waves
converted to compressional waves where they pass through the outer
core) (Fig. 1). These models test how mantle rheology, the slab geo-
metry, yield strength (sy) and the viscosity of the plate boundary
shear zone (PBSZ) influence the surface plate motions and underlying
mantle flow (Supplementary Information). For the mantle rheology,
we use either Newtonian-only viscosity or a composite (Newtonian
and non-Newtonian) viscosity (Methods). The slab geometry, con-
strained by seismicity and tomography, varies along the length of the
subduction zone, forming a flat slab beneath south-central Alaska and
steepening close to the transform boundary. Because seismic observa-
tions do not constrain the exact depth of the slab edge, we construct
two slab shapes, slabE325 and slabE115, where the subscript refers to the
depth of the slab in kilometres at the eastern slab edge (Fig. 1).
PBSZ
SMSZ NAM
185
PAC A B′ ºE
72º N
Slab
B
?
JdFR A′
0 km
ºC
T=0
E
240º
45º
1,500 km
,400 º
C
N
T=1
Figure 1 | Schematic of full model domain and slab geometry. Outline of
the high-resolution mesh region (dashed grey line); the portion of the slab
geometry that is varied (short-dashed black line); the PBSZ and southern
mesh boundary shear zone (SMSZ) (thick dark-grey lines); Juan de Fuca
Ridge (JdFR, double black line); and the locations of the cross-sections
shown in Fig. 3 (AA9 and BB9, black). NAM, North American plate; PAC,
Pacific plate; T, temperature. See Methods and Supplementary Information.
NATURE | Vol 465 | 20 May 2010
We first compare PPM predicted from the models with the
observed PPM, to verify that models producing high flow velocities
in the mantle also produce surface plate motions consistent with
first-order observations. Although the models do not include the full
Pacific plate or its slabs, they are designed such that the region of the
Pacific plate is free to move in response to the local driving forces, and
this motion should be generally consistent with PPM (Supplemen-
tary Information).
Off the shore of south-central Alaska, the observed PPM with
respect to North America is approximately N 16u W at 5.2 cm yr21
(ref. 8). In models with a Newtonian viscosity and slabE115, the pre-
dicted PPM is ,N 27u W at 1.9 cm yr21 (Fig. 2). The predicted PPM
is more northerly and slightly faster (,N 20u W at 4.4 cm yr21) in
models that use the composite rheology, owing to the reduced viscous
support of the slab and to weakening and widening of the plate inter-
face. A deeper slab along the eastern boundary (slabE325) increases the
slab-pull driving force, resulting in a more northerly and slightly faster
PPM (,N 14u W at 5.3 cm yr21). For both slabE115 and slabE325,
increasing the PBSZ viscosity from 1020 to 1021 Pa s decreases the speed
of the Pacific plate by ,2.5 cm yr21. Therefore, models with a com-
posite viscosity, a PBSZ viscosity of 1020 Pa s and either slab shape
provide good agreement with observed PPM. More importantly, none
of the composite-viscosity models, which have fast mantle flow, pro-
duce plate motions that are inconsistent with surface observations.
Subduction of the Pacific plate beneath southern Alaska induces a
spatially variable mantle flow and localized flow rates that can be more
than ten times the speed of surface plate motions (Fig. 2). In the
composite-viscosity flow models, with sy 5 500 MPa, mantle speeds
close to the slab can be up to 90 cm yr21. These high flow rates occur
where strain rates are high (10213–10212 s21), and viscosities are
therefore low (1017–1018 Pa s) owing to the non-Newtonian rheology.
For Newtonian viscosity models, mantle flow rates near the slab are
1.0–5.0 cm yr21. In all the models, far from the slab edge and the
driving force of the sinking slab, mantle velocities decrease and are
comparable to plate motions.
a SlabE115 (composite viscosity) b SlabE115 (Newtonian viscosity)
65º
5 cm yr–1
NAM
60º
PAC
55º
Latitude
210º 220º
d e
65º
60º
55º
20 cm yr–1
200º 200º
210º 230º
220º
a, b, c: Viscosity, log10 [ η (Pa s)]
17 18 19 20 21 22 23
d, e, f: Vertical velocity (cm yr–1)
–30 –20 –10 0 10
Figure 2 | Maps of flow field. a–c, Surface velocity field and viscosity
(colour scale) for three models (sy 5 500 MPa; PBSZ viscosity, 1020 Pa s).
The observed NUVEL-1A Pacific motion vector, assuming North America to
be fixed8, is indicated using a white arrow. AVO, Alaska Volcano
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
The pattern of flow in the mantle depends strongly on the slab shape,
yield strength and mantle viscosity. The sinking of the dense slab draws
flow into the mantle wedge towards the slab, resulting in a poloidal
component of flow (Fig. 3). The velocity vectors in the slab and mantle
dip more steeply than the slab, which is indicative of steepening of the
slab with time or retrograde slab motion21. Steepening of the slab
pushes material from underneath the slab around the slab edge and
into the overlying mantle wedge, forming an anticlockwise toroidal
component of flow (Fig. 3). A stronger slab (sy 5 1,000 MPa) steepens
more slowly, generating a proportionally smaller toroidal component
of flow. The locus of toroidal flow depends on the depth of the slab
edge and the mantle rheology (Fig. 2). The strength of the toroidal
component is significantly greater for models with a composite vis-
cosity. In general, the highest velocities occur in those regions of the
mantle wedge that are affected simultaneously by poloidal and toroidal
flow.
To determine which mantle rheology and slab shape are consistent
with observations, infinite-strain axes2,12 (ISAs) are calculated from
the predicted mantle flow and compared with SKS fast-axis direc-
tions indicative of seismic anisotropy5 (Fig. 4). The pattern of seismic
anisotropy is consistent with toroidal flow around the slabE115 edge
beneath south-central Alaska and a component of trench-parallel
flow beneath western Alaska, assuming A-type LPO fabric in the
mantle wedge and B-type fabric in the wedge nose (Supplementary
Information). None of the slabE325 models, in which the toroidal flow
is located further to the east, match the anisotropy north of the slab
nose; this provides a strong constraint on the slab shape. West of
longitude 210u, where slabE115 and slabE325 have the same shape along
the mantle wedge, the flow is dominantly trench-perpendicular.
However, for models with composite viscosity, the ISAs have regions
of trench-parallel orientation that match the observed trench-parallel
SKS fast-axis. In this region, stretching occurs as a result of shearing
(that is, velocity gradients) in the trench-parallel component of flow
due to pressure gradients formed by the westward steepening of the
slab. None of the models using the Newtonian viscosity or a higher
c SlabE325 (composite viscosity)
5 cm yr–1 5 cm yr–1
NAM NAM
PAC
PAC
210º 220º 210º 220º
f
5 cm yr–1 20 cm yr–1
200º
210º 220º 230º 210º 220º 230º
Longitude
Plots show subset of model domain
a, b, c: Slab surface (contours
24 Velocity every 40 km)
Plate boundary Neogene faults
Holocene volcanoes (AVO) d, e, f: 1,000 ºC temperature contour
20
Observatory (Supplementary Information). d–f, Mantle velocity field at
100-km depth and vertical velocity magnitude (colour scale). The implied
strong vertical gradient in velocity illustrates the significant decoupling of
the overriding plate from the mantle flow.
339
LETTERS NATURE | Vol 465 | 20 May 2010

a B A
200º E A A′ B B′
55º N 60º N 60º N
Slab 65º N 65º N
N

120 km

A′ B′ 370 km
0 km
72º N
240º E Log10 [ η (Pa s)]
550 km 17 18 19 20 21 22 23 24

200º E
b 55º N
N B
B′
Slab

N
0 km

550 km
Velocity (cm yr–1)
N
72º N 240º E 0 20 40 60 80

Figure 3 | 3D mantle flow field and viscosity structure. Calculated using the
model with slabE115 and composite rheology (sy 5 500 MPa; PBSZ viscosity,
1020 Pa s). Figures show subset of modelled domain. a, Isosurface and cross-
sections (AA9 and BB9) through composite viscosity show the strong slab
and the low-viscosity regions in the mantle wedge and beneath the slab.
Low-viscosity regions correlate with regions of high strain rate. b, Isosurface
of viscosity showing an oblique, cross-sectional, radial slice through the
velocity field. The cross-section BB9 shows poloidal flow and along-strike
flow. The plan view shows anticlockwise toroidal flow and an upward
component of flow east of the slab edge.

yield strength (1,000 MPa) are able to match the overall pattern of a yield strength of 500 MPa and a PBSZ viscosity of 1020 Pa s. The
seismic anisotropy. difference between the surface motion and the sinking rate of the
The broad spatial distribution of seismic anisotropy observations subducting plate requires strain within the plate. This is primarily
provides strong constraints on the slab shape, mantle rheology and accommodated by yielding within the slab hinge, which can reduce
slab yield strength, and comparison with the surface plate motions the effective viscosity from 1024 to 1022 Pa s and is more significant in
demonstrates that rapid mantle flow, which is strongly decoupled models with a composite viscosity, because the non-Newtonian vis-
from the surface plates, is consistent with surface observations. We cosity weakens the outer portions of the slab and the surrounding
find that for the Alaska subduction zone, both observations are best mantle, requiring more of the slab weight to be supported by the slab
fitted by a single model using slabE115 and a composite viscosity with strength.
The depth of the Alaskan slab and its predicted steepening are
consistent both with the transient state of slab evolution found before
a SlabE115 , σy = 500 MPa, b SlabE115 , σy = 500 MPa, c SlabE325 , σy = 500 MPa,
composite viscosity Newtonian viscosity composite viscosity
a slab reaches the more viscous lower mantle in time-dependent sub-
65º duction models16–18,22 and with global observations, which show that
slab dip increases with slab length for slabs in the early stages of sub-
64º
duction22,23. Geochemical constraints on mantle flow above the short
slab in Costa Rica suggest a trench-parallel flow component 50%
63º
greater than the convergence rate6. Thus, although localized high
62º
velocities may not occur in all subduction zones, short slabs may be
in a transient state characterized by slab steepening, localized fast
50 cm yr–1 5 cm yr–1 50 cm yr–1 mantle flow rates and a minimum in viscous resistance to sinking.
61º
Latitude

d 65º e f Rheologically controlled decoupling of plate motions from rapid

mantle flow in subduction zones has implications for a range of
processes and for developing a more dynamic understanding of plate
64º
tectonics. Rapid transport through the mantle wedge implies that
63º
material spends little time in the active melt zone beneath the arc
and therefore experiences only small degrees of melting. Locally, high
62º strain rates and complex 3D flow at the slab edge should form regions
of efficient mixing of geochemical signatures, and the significant
61º component of upwelling along slab edges may contribute to adakitic
208º 210º 212º 214º 208º 210º 212º 214º 208º 210º 212º 214º
Longitude volcanism (for example, the Wrangell volcanics in southern
Lag parameter (Π )
Alaska24). Because the mantle flow is fast (for example, 90 cm yr21),
Model-predicted ISA directions (colour scale)
Observed SKS fast direction even a small slab-parallel component of flow (10–20%) can result in
0.0 0.5 1.0 1.5 2.0 2.5 1,000 ºC temperature contour (slab outline) apparently rapid (9–18 cm yr21) transport of geochemical signatures
along the arc25,26. High strain rates around the slab can lead to more
Figure 4 | Velocity and ISA orientations at 100-km depth. a, Velocity field rapid development of LPO12, allowing changes in mantle flow on
shows a trench-parallel component of flow near the slab nose for slabE115
shorter length scales to be tracked2. Finally, partial decoupling of
with composite viscosity (sy 5 500 MPa; PBSZ viscosity, 1020 Pa s). b, The
Newtonian viscosity has a damping effect on the toroidal component of flow. the mantle flow and slab-pull force from surface plates as a result of
c, There is no toroidal flow above the slab nose for slabE325. d–f, ISA non-Newtonian viscosity and localized internal slab deformation may
orientations coloured by the lag parameter (P, colour scale). Superimposed explain why 50% of present-day global plate motions can be accounted
are SKS fast-axis directions (blue) back-projected along the ray path to 100- for without a direct contribution to slab pull from upper-mantle
km depth5. ISA orientation provides a good estimate of LPO for P , 1.0. slabs27.
340
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
METHODS SUMMARY
Surface plate motions and solid-state creep within the models are driven by
thermally induced density anomalies prescribed by the initial thermal structure.
The geometry and thermal structure of the Pacific plate, the Pacific slab and the
North American plate are constrained by geological and geophysical observables,
thereby reflecting the observed variability in plate age and slab dip (Supplemen-
tary Information). We use the finite-element code CitcomCU28,29 to solve for the
instantaneous solid-state flow of the lithosphere and mantle. The flow is governed
by the conservation of mass and momentum for an incompressible, infinite
Prandtl fluid, and a viscoplastic constitutive equation, which links the density
and viscosity through the thermal field. We use a strain-rate-dependent viscosity
assuming the experimentally determined composite rheology for olivine9 (Sup-
plementary Information). Diffusion creep is assumed for the lower mantle. To
resolve the variable slab dip and overriding plate as well as solve accurately for the
large gradients in viscosity and velocity, the models contain 100,413,929 mesh
nodes with mesh resolution of up to ,2.35 km. The highest resolution coincides
with the corner of the plate boundary, which is located far from the model
boundaries. We calculate the ISA orientations2,12, which provide a good approxi-
mation to the olivine LPO throughout much of the subduction-zone region of the
models, as indicated by lag parameter magnitudes less than one. See Supplemen-
tary Information for further details.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 20 October 2009; accepted 26 March 2010.
1. Becker, T. W., Kellogg, J. B., Ekstrom, G. & O’Connell, R. J. Comparison of
azimuthal seismic anisotropy from surface waves and finite strain from global
mantle-circulation models. Geophys. J. Int. 155, 696–714 (2003).
2. Conrad, C. P., Behn, M. D. & Silver, P. G. Global mantle flow and the development
of seismic anisotropy: differences between the oceanic and continental upper
mantle. J. Geophys. Res. 112, B07317 (2007).
3. Russo, R. M. & Silver, P. G. Trench-parallel flow beneath the Nazca Plate from
seismic anisotropy. Science 263, 1105–1111 (1994).
4. Smith, G. P. et al. A complex pattern of mantle flow in the Lau backarc. Science 292,
713–716 (2001).
5. Christensen, D. H. & Abers, G. A. Seismic anisotropy under central Alaska from
SKS splitting observations. J. Geophys. Res. 115, BO4315 (2010).
6. Hoernle, K. et al. Arc-parallel flow in the mantle wedge beneath Costa Rica and
Nicaragua. Nature 451, 1094–1097 (2008).
7. Long, M. & Silver, P. G. The subduction zone flow field from seismic anisotropy: a
global view. Science 319, 315–318 (2008).
8. DeMets, C. & Dixon, T. H. New kinematic models for Pacific-North American
motion from 3 Ma to present: evidence for steady state motion and biases in the
NUVEL-1A model. Geophys. Res. Lett. 26, 1921–1924 (1999).
9. Hirth, G. & Kohlstedt, D. in Inside the Subduction Factory (ed. Eiler, J.) 83–105
(American Geophysical Union, 2003).
10. Karato, S., Jung, H., Katayama, I. & Skemer, P. Geodynamic significance of seismic
anisotropy of the upper mantle: new insights from laboratory studies. Annu. Rev.
Earth Planet. Sci. 36, 59–95 (2008).
11. Savage, M. K. Seismic anisotropy and mantle deformation: what have we learned
from shear wave splitting? Rev. Geophys. 374, 65–106 (1999).
12. Kaminiski, É. & Ribe, N. M. Timescales for the evolution of seismic anisotropy in
mantle flow. Geochem. Geophys. Geosyst. 3, 1051 (2002).
13. Peyton, V. et al. Mantle flow at a slab edge: seismic anisotropy in the Kamchatka
region. Geophys. Res. Lett. 28, 379–382 (2001).
14. Zhong, S. & Gurnis, M. Interaction of weak faults and non-Newtonian rheology
produces plate tectonics in a 3D model of mantle flow. Nature 383, 245–247
(1996).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
15. Zhong, S., Gurnis, M. & Moresi, L. Role of faults, nonlinear rheology, and viscosity
structure in generating plates from instantaneous mantle flow models. J. Geophys.
Res. 103, 15255–15268 (1998).
16. Schellart, W. P. Kinematics of subduction and subduction-induced flow in the
upper mantle. J. Geophys. Res. 109, B07401 (2004).
17. Funiciello, F. et al. Mapping mantle flow during retreating subduction: laboratory
models analyzed by feature tracking. J. Geophys. Res. 111, B03402 (2006).
18. Piromallo, C., Becker, T. W., Funiciello, F. & Faccenna, C. Three-dimensional
instantaneous mantle flow induced by subduction. Geophys. Res. Lett. 33, L08304
(2006).
19. Stegman, D. R., Freeman, J., Schellart, W. P., Moresi, L. & May, D. Influence of
trench width on subduction hinge retreat rates in 3-D models of slab rollback.
Geochem. Geophys. Geosyst. 7, Q03012 (2006).
20. Kneller, E. A. & van Keken, P. E. Trench-parallel flow and seismic anisotropy in the
Mariana and Andean subduction systems. Nature 450, 1222–1225 (2007).
21. Garfunkel, Z., Anderson, C. A. & Schubert, G. Mantle circulation and the lateral
migration of subducted slabs. J. Geophys. Res. 91, 7205–7223 (1986).
22. Billen, M. I. & Hirth, G. Rheologic controls on slab dynamics. Geochem. Geophys.
Geosyst. 8, Q08012 (2007).
23. Lallemand, S. E., Heuret, A. & Boutelier, D. On the relationships between slab dip,
back-arc stress, upper plate absolute motion and crustal nature in subduction
zones. Geochem. Geophys. Geosyst. 6, Q09006 (2005).
24. Preece, S. J. & Hart, W. K. Geochemical variations in the 5 Ma Wrangell Volcanic
Field, Alaska: implications for the magmatic and tectonic development of a
complex continental arc system. Tectonophysics 392, 165–191 (2004).
25. Turner, S., Bourdon, B. & Gill, J. Insights into magma genesis at convergent
margins from U-series isotopes. Rev. Mineral. Geochem. 52, 255–315 (2003).
26. Conder, J. A., Wiens, D. A. & Morries, J. On the decompression melting structure
at volcanic arcs and back-arc spreading centers. Geophys. Res. Lett. 29,
doi:10.1029/2002GL015390 (2002).
27. Conrad, C. P. & Lithgow-Bertelloni, C. How mantle slabs drive plate tectonics.
Science 298, 207–209 (2002).
28. Zhong, S. Constraints on thermochemical convection of the mantle from plume
heat flux, plume excess temperature and upper mantle temperature. J. Geophys.
Res. 111, B04409 (2006).
29. Moresi, L. & Gurnis, M. Constraints on the lateral strength of slabs from three-
dimensional dynamic flow models. Earth Planet. Sci. Lett. 138, 15–28 (1996).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements This work was supported by US National Science Foundation
grant EAR-0537995. High-resolution models were run on the TeraGrid cluster
Lonestar at the Texas Advanced Computing Center, through grant
TG-EAR080015N. We thank Computational Infrastructure for Geodynamics for
the CitcomCU source code and C. Conrad and M. Behn for the source code used to
calculate the ISAs. 3D data were visualized at the W. M. Keck Center for Active
Visualization in the Earth Sciences at the University of California, Davis. We thank
D. Christensen and G. Abers (shear-wave splitting data) and N. Ruppert
(earthquake hypocentral data). We thank D. Turcotte, L. Kellogg, O. Kreylos,
D. Eberhart-Phillips, S. M. Roeske, J. Dewey, T. Taylor, L. Moresi and G. Hirth for
comments and discussions.
Author Contributions Both authors contributed equally to the overall development
of the project, model design considerations, analysis and interpretations. M.A.J.
performed all of the numerical modelling, except for the ISA calculations, which
were done by M.I.B.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to M.A.J. (majadamec@ucdavis.edu).
341
doi:10.1038/nature09053
METHODS
28,29
Model set-up. We use the finite-element code CitcomCU to solve for the
instantaneous solid-state flow of the lithosphere and mantle. Models of the
subduction–transform plate boundary in southern Alaska contain an overriding
plate (the North American plate), a subducting plate (the Pacific plate) and the
underlying mantle. A 3D low-viscosity shear zone separates the two plates (see
below). We assume that the Wadati–Benioff zone represents the shape of the sub-
ducting lithosphere. The thermal structure of the subducting plate is defined using a
half-space cooling model and lithosphere age30,31. For the subducted portion of the
Pacific plate, the age of the plate at the trench is projected down-dip. To account for
warming of the slab with depth due to subduction into the mantle, the half-space
cooling thermal field is adjusted, using a length-scale diffusion analysis. For the
overriding plate, effective plate ages are used. For simplicity, we do not include an
additional density anomaly for the subducting Yakutat terrane. Perpendicular dis-
tances between the model grid points and the slab surface are calculated for each
node in the finite-element mesh, allowing the initial 3D thermal and weak-zone
fields to be smoothly mapped onto the finite-element grid. The mesh spans
961 3 649 3 161 nodes in the longitudinal (185–240u), latitudinal (45–72u N)
and radial (0–1,500 km) directions, respectively; mesh resolution ranges from
0.04u to 0.255u, 0.0211u to 0.18u and 2.35 km to 25 km in the respective directions.
The highest resolution coincides with the corner of the plate boundary, which is
located far from the model boundaries. All regions of all model boundaries are free-
slip. The models were run on 360 processors. See Supplementary Information.
Rheology. The composite rheology includes the diffusion (df) and dislocation
(ds) creep deformation mechanisms for olivine
g g
gcom ~ df ds
gdf zgds
where we assume that both deformation mechanisms accommodate deformation at
a constant stress and that the total strain rate is a linear combination of the con-
tributions from the two mechanisms9,22. The component viscosities are defined by
 p 1=n  
d EzPV
gdf,ds ~ r e_ (1{n)=n exp
ACOH nRT
where P is the lithostatic pressure, R is the universal gas constant, T is the temper-
ature, e_ is the strain rate, d is the grain size, COH is the water content, and A, n, p, r, E,
and V are experimentally constrained flow-law parameters9. The grain size in the
upper mantle is set to 10 mm, such that the respective viscosities predicted by the
two mechanisms are equal for a strain rate of 10215 s21, with a value of 1020 Pa s at
250 km for a water content of 1,000 p.p.m.-H/Si. In the lower mantle, the grain size
is set to 40 mm, giving a tenfold increase in viscosity between the upper and lower
mantle. At low temperatures and pressures, plastic deformation occurs32; the effec-
tive viscosity, geff, is therefore limited by a depth-dependent yield stress, sy. If
s . sy then geff 5 sy/_eII , and if s , sy then geff 5 gcom, where s is the stress and
e_ II is the second invariant of the strain rate tensor. For a gradient of 15 MPa km21,
sy increases linearly from 0.1 MPa at the surface to a maximum value of either
500 MPa or 1,000 MPa. The models allow for viscosity variations of up to seven
orders of magnitude (1017–1024 Pa s).
Plate boundary shear zone. We model the boundary between the plates as a thin
viscous layer at least 40 km thick and extending to a depth of approximately
90 km (Supplementary Information). The viscosity in the thin viscous layer, gwk,
is smoothly blended into the background viscosity, geff, using
gwk ~10(1{Awk )log10 (gef f =go ) go
©2010 Macmillan Publishers Limited. All rights reserved
where Awk is a scalar weak-zone field defined a priori with values ranging from 0
to 1 (corresponding to unweakened and fully weakened regions, respectively)
and go is the reference viscosity, equal to 1020 Pa s. The gwk value is overwritten if
the viscosity calculated for geff is lower, such that the final form of the viscosity,
gf, becomes
gf ~ min (gef f , gwk )
Within the multi-resolution finite-element mesh, the thin viscous layer spans a
specified minimum number of elements rather than a specific width. This allows
us to constrain the viscosity jump across the elements, which has been shown to
be important for convergence in models with large viscosity variations33,34. We
found good convergence using eight elements to span four orders of magnitude
change in viscosity. This implementation was also tested on 3D models with a
simplified subduction zone geometry35.
ISA orientations. Although mantle velocity fields from geodynamic models have
been compared with the azimuth of the fast seismic velocity36, this is strictly valid
only for simple shear deformation and after a sufficient amount of strain has
accumulated to align the axis of maximum finite strain with the shear plane12.
However, in regions with rapidly changing flow, stretching axis orientations2 or
full tracking of LPO development is required12,37.
We calculate the lag parameter, P, which is the ratio of VISA, the rate at which
the LPO forms along the ISA direction, to Vflow, the rate at which the ISA rotates
in response to the local flow pattern2,12 (Supplementary Information). The para-
meter Vflow is the sum of a spatial contribution related to the instantaneous flow
and a time-dependent contribution related to non-steady-state flow. Because the
models are instantaneous, we account for the time-dependent contribution to
Vflow by assuming that it is equal in magnitude to the spatial contribution, as has
been shown to be true for density-driven mantle flow models2 (Supplementary
Information). For lag parameter values of less than one, the ISA gives a robust
indication of the LPO12. We expect that for lag parameter values of less than two,
the calculated ISA still provides a better indication of the fabric orientation than
does mantle flow direction. In comparing the ISAs with SKS fast-axis directions,
we assume that A-type fabric develops in the mantle below and near the slab and
in most of the mantle wedge. However, within the cold, innermost corner of
the mantle wedge, trench-parallel fast seismic directions may be due to B-type
fabric20.
30. Müller, R. D., Roest, W. R., Royer, J. Y., Gahagan, L. M. & Sclater, J. G. Digital
isochrons of the world’s ocean floor. J. Geophys. Res. 102, 3211–3214 (1997).
31. Turcotte, D. L. & Schubert, G. Geodynamics 2nd edn, 153–161 (Cambridge Univ.
Press, 2002).
32. Kohlstedt, D. L., Evans, B. & Mackwell, S. J. Strength of the lithosphere: constraints
imposed by laboratory experiments. J. Geophys. Res. 100, 17587–17602 (1995).
33. Moresi, L. N. & Solomatov, V. S. Numerical investigations of two-dimensional
convection with extremely large viscosity variations. Phys. Fluids 9, 2142–2162
(1995).
34. Moresi, L., Zhong, S. & Gurnis, M. The accuracy of finite element solutions of
Stokes’ flow with strongly varying viscosity. Earth Planet. Sci. Lett. 97, 83–94
(1996).
35. Jadamec, M. A. Three-Dimensional Lithosphere and Mantle Dynamics: Models of the
Subduction-Transform Plate Boundary System in Southern Alaska. PhD thesis, Univ.
California, Davis (2009).
36. Zandt, G. & Humphreys, E. Toroidal mantle flow through the western U.S. slab
window. Geology 36, 295–298 (2008).
37. Lassak, T. M., Fouch, M. J. & Kaminiski, É. Seismic characterization of mantle flow
in subduction systems: can we resolve a hydrated mantle wedge? Earth Planet. Sci.
Lett. 243, 632–649 (2006).
 Vol 465 | 20 May 2010 | doi:10.1038/nature09098

LETTERS
Climate change and the global malaria recession
Peter W. Gething1, David L. Smith2,3, Anand P. Patil1, Andrew J. Tatem2,4, Robert W. Snow5,6 & Simon I. Hay1

The current and potential future impact of climate change on climate change and observed under existing public health interven-
malaria is of major public health interest1,2. The proposed effects tions. We use these perspectives to reassess the rationale of existing
of rising global temperatures on the future spread and intensifica- modelling approaches to impact assessments and the threat posed by
tion of the disease3–5, and on existing malaria morbidity and mor- future climatic changes to regional malaria control.
tality rates3, substantively influence global health policy6,7. The The only global map of pre-intervention malaria endemicity dates
contemporary spatial limits of Plasmodium falciparum malaria from a 1968 study13 (Fig. 1a) in which a major synthesis of historical
and its endemicity within this range8, when compared with com- records, documents and maps of a variety of malariometric indices for
parable historical maps, offer unique insights into the changing all four Plasmodium species was used to map parasite rate (PR—the
global epidemiology of malaria over the last century. It has long
been known that the range of malaria has contracted through a
century of economic development and disease control9. Here, for a
the first time, we quantify this contraction and the global
decreases in malaria endemicity since approximately 1900. We
compare the magnitude of these changes to the size of effects on
malaria endemicity proposed under future climate scenarios and
Risk free
associated with widely used public health interventions. Our find-
Epidemic
ings have two key and often ignored implications with respect to
Hypoendemic
climate change and malaria. First, widespread claims that rising Mesoendemic
mean temperatures have already led to increases in worldwide Hyperendemic
malaria morbidity and mortality are largely at odds with observed Holoendemic
decreasing global trends in both its endemicity and geographic
extent. Second, the proposed future effects of rising temperatures b
on endemicity are at least one order of magnitude smaller than
changes observed since about 1900 and up to two orders of mag-
nitude smaller than those that can be achieved by the effective
scale-up of key control measures. Predictions of an intensification
of malaria in a warmer world, based on extrapolated empirical Risk free
relationships or biological mechanisms, must be set against a con- Unstable
text of a century of warming that has seen marked global declines Hypoendemic
in the disease and a substantial weakening of the global correlation Mesoendemic
between malaria endemicity and climate. Hyperendemic

A resurgence in funding for malaria control10, the existing efficacy of Holoendemic

affordable interventions, and a growing body of nationally or sub-

c
nationally reported declines in endemicity or clinical burden11 have
engendered renewed optimism among the international malaria
research and control community. In marked contrast, however, are
model predictions, reported widely in global climate policy debates3,6,7, –5
–4
that climate change is adding to the present-day burden of malaria and
–3
will increase both the future range and intensity of the disease. In policy –2
arenas, such predictions can support scenario analysis or serve as a call –1
to action, but the modelling approaches used and the accuracy of their 0
predictions have not always been challenged. +1
The recent publication of an evidence-based map of contemporary +2
malaria endemicity8 allows an audit of changes in the global epidemi-
Figure 1 | Changing global malaria endemicity since 1900. a, Pre-
ology of malaria since the start of the twentieth century; a period of intervention endemicity (approximately 1900) as defined in ref. 13.
undoubted climatic change12. We compare this modern-day map b, Contemporary endemicity for 2007 based on a recent global project to
with the most reliable equivalent for the pre-intervention era, around define the limits and intensity of current P. falciparum transmission8.
1900 (ref. 13), and compare the magnitude of observed changes in c, Change in endemicity class between 1900 and 2007. Negative values
range and endemicity to those proposed to occur in response to denote a reduction in endemicity, positive values an increase.
1
Spatial Ecology and Epidemiology Group, Tinbergen Building, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK. 2Emerging Pathogens Institute,
University of Florida, Gainesville, Florida 32610, USA. 3Department of Biology, University of Florida, Gainesville, Florida 32610, USA. 4Department of Geography, University of Florida,
Gainesville, Florida 32611, USA. 5Malaria Public Health and Epidemiology Group, Centre for Geographic Medicine, KEMRI – University of Oxford – Wellcome Trust Collaborative
Programme, Kenyatta National Hospital Grounds (behind NASCOP), P.O. Box 43640-00100, Nairobi, Kenya. 6Centre for Tropical Medicine, Nuffield Department of Clinical Medicine,
University of Oxford, CCVTM, Oxford OX3 7LJ, UK.

342
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 LETTERS

proportion of individuals with malaria parasites in their peripheral
blood) stratified into four endemic classes (hypoendemic, PR , 10%;
mesoendemic, PR $ 10% and , 50%; hyperendemic, PR $ 50% and
, 75%; holoendemic, PR $ 75%). This map is the only reconstruc-
tion of historical malaria at its assumed historical peak around the
start of the twentieth century and triangulates well with the plethora of
national level malaria maps published throughout the last century (see
ref. 14 for a systematic review).
We have used recently completed work that defines the 2007 limits of
stable P. falciparum transmission and its endemicity within this range8
to generate a comparable contemporary map of P. falciparum malaria
distribution. This model has been described in detail elsewhere8 and its
output allowed for a continuous estimate of parasite prevalence for the
year 2007 (Fig. 1b) stratified into the same endemicity classes used in
the historical map13 (see Methods). Comparison of the historical and
compared formally with observed or predicted effects from non-
climatic influences. The P. falciparum basic reproductive number,
PfR0, quantifies the expected number of secondary cases in a non-
immune population resulting from a single new infection and is the
appropriate metric for comparing the relative magnitude of different
effects on the underlying intensity of transmission20. Linked climate–
biological models that simulate changes to PfR0 under different
climate-change scenarios have proposed effect sizes of up to three
(Table 1), that is, a tripling of the reproductive number16,18,19. To
Table 1 | Comparison of magnitude of changes (effect size) in the basic
reproductive number in relevant observational or predictive studies
Effect size (relative Study
change in R0)

contemporary maps revealed that endemic/stable malaria is likely to Estimated changes 1900–2007
have covered 58% of the world’s land surface around 1900 (see Proportion of 1900 endemic world by
area*
Supplementary Table 1) but only 30% by 2007 when P. falciparum 13% 4 # 1{ This study
malaria has become restricted largely to the tropics. Even more marked 12% 4 1–10 This study
has been the decrease in prevalence within this greatly reduced range, 18% 4 10–100 This study
with endemicity falling by one or more classes in over two-thirds (67%) 57% 4 .100 This study
Changes predicted under climate change
of the current range of stable transmission (Fig. 1c). The contemporary Spatially aggregated mean change by Ref. 16
map indicates that holoendemic P. falciparum malaria is now rare and ,2050{1
limited to patches in West Africa totalling around 140,000 km2. The P. falciparum 3 1.27 (1.16–1.74)
Americas are entirely hypoendemic for P. falciparum, as are very large P. vivax 3 1.23 (1.15–1.39)
sections of central and southeast Asia and substantial swathes of Africa. Range of local changes by ,20501I" 3 0–2 Ref. 16
Range of local changes by ,20801I 3 0–2# Ref. 19
When empirical relationships linking the current spatial distribu- Range of local changes by ,20501I" 3 0–3 Ref. 18
tions of surface temperature and malaria are extrapolated to predict Example effect sizes for interventions
future changes in disease range or intensity under scenarios of rising predicted via biological modelling
global temperatures, a necessary assumption is that all other factors ACTs (compared to failing pre-ACT 4 1.1–1.8q Ref. 21
treatment)
either remain constant or have a relatively negligible effect. During a ITNs (at 40–60% effective coverage) 4 5–15** Ref. 22
century in which global temperature increases have been unequi- ITNs 1 LCI (both at ‘moderate’ coverage 4 15–25{{ Ref. 23
vocal12,15, we have documented a marked, global decrease in the range levels)
and intensity of malaria transmission. This suggests that such an Example effect sizes observed
concurrently with increased intervention
assumption would not have been valid for predicting the response coverage (primary interventions in use)
of malaria to the warming climate of the last 100 years. A second Western Kenya{{ (ITNs, ITNs 1 LCI) 4 7.2{{, 4 26.4{{ Ref. 25
assumption is that the nature of the link between climate and the São Tomé and Principe11 (IRS 1 ITNs 1 4 5.1II Ref. 26
global distribution and intensity of malaria is effectively immutable: ACT 1 IPTp)
Bioko Island"" (IRS 1 ACT) 4 2.9## Ref. 27
an empirical climate–malaria relationship observed at one time period Southern Mozambiqueqq (IRS 1 ACT) 4 78.8*** Ref. 28
will be preserved even under changing climate and disease scenarios. Zanzibar{{{ (ITNs 1 ACT); 0–5 yr, 4 1.6{{{, 4 1.8{{{ Ref. 29
However, a comparison of global-scale climate patterns with the his- 6–14 yr
torical and contemporary patterns of malaria endemicity presented ACT, Artemisinin-based combination therapy; EIR, entomological inoculation rate; IPTp,
here indicated a decoupling of the geographical climate–malaria rela- intermittent preventive treatment in pregnancy; ITN, insecticide-treated bed net; LCI, larval
control intervention.
tionship over the twentieth century (see Supplementary Information * Percentages do not sum to 100 because of rounding. {Consists of 11% of land area with no
for additional explanation and results of this analysis), indicating that evidence of change, and 2% with evidence of increased transmission. {The available results
relate to the combined Africa, South-East Asia, and Central and South America regions. Given
non-climatic factors have profoundly confounded this relationship are central modelled values and uncertainty intervals associated with plausible ranges of input
over time. Contemporary endemicity maps therefore provide a poor biological parameters. 1Results obtained using the MIASMA linked climate-biological model,
baseline for empirically-based predictions of future climate effects. under three future climate scenarios, and only the largest predicted changes are shown here. IIn
the original study, results were not presented for areas with very low baseline potential to avoid
Linking increasing temperatures to changes in malaria epidemi- using infinitesimal values as denominators in the comparison. "Results apply to both P.
ology is justified theoretically by known biological effects on different falciparum and P. vivax. #Range of predicted changes included a ’.2’ category but no further
details provided. qStudy reported predicted changes in parasite rate from five real-world
life-cycle stages of the Anopheles vector and Plasmodium parasite16. baseline endemicity settings under failing treatment regimes based on pre-ACT monotherapies
Such effects do not act in isolation, however, and empirical predic- (Table 4 in ref. 21). We converted transitions in parasite rate into transitions in R0 to estimate
tions are only credible if the role and relative influence of non- effect size. **Study modelled effect size as a continuous function of ITN effective coverage.
Values presented based on Fig. 1 in ref. 22. {{Study reported effect sizes in terms of EIR, which
climatic factors is considered. A simple interpretation of the observed were interpreted directly as R0 effect sizes. {{Results from a control trial. 11Integrated malaria
global recession in malaria since 1900 is that non-climatic factors, control effort involving mass intervention coverage complemented with health system
strengthening. Reported decline relates to period 2005–2007. IIStudy reported reduction in
primarily direct disease control and the indirect effects of a century of community parasite rate from 30.5% to 2.1% following expanded control efforts. We converted
urbanization and economic development, although spatially and this into transitions in R0 to estimate effect size. ""Integrated malaria control effort involving
temporally variable, have exerted a substantially greater influence mass IRS administration, improved case management with ACT and strengthened health
system surveillance and diagnostics. Reported decline relates to period 2004–2005. ##Study
on the geographic extent and intensity of malaria worldwide during reported reduction in community parasite rate from 46% to 31% following expanded control
the twentieth century than have climatic factors. This simple infer- efforts. We converted this into transitions in R0 to estimate effect size. qqIntegrated malaria
control effort involving IRS administration, improved case management with ACT and
ence is consistent with other studies that have reviewed the historical strengthened health systems. Reported decline relates to period 1999–2005. ***Study
climatic and anthropogenic forces acting on malaria17 and has reported reduction in community parasite rate from 65% to 4% in study zone with longest
important implications for the debate on the importance of climate running intervention coverage (Zone 1, values taken from Table 1 of ref. 28). We converted this
into transitions in R0 to estimate effect size. {{{Integrated malaria control effort involving scale
change in determining future malaria scenarios. up of long-lasting ITNs from 10% to 90% coverage of children, switch from chloroquine to ACT
Biological modelling provides an alternative to empirical ap- as first and second line therapeutic and strengthened surveillance and health systems. Reported
decline relates to period 2003–2005. {{{Study reported reduction in parasite rate in children
proaches, enabling the magnitude of potential disease responses to 0–5 yr from 9% to 0.3% and in children 6–14 yr from 12.9% to 1.7% following expanded control
future climate scenarios to be estimated directly16,18,19 and then efforts. We converted this into transitions in R0 to estimate effect size.

343
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
place the proposed magnitude of these effects in context, we com-
pared them to the observed reductions in global endemicity over the
past century. To allow such a comparison, a simple P. falciparum
transmission model20 was used to translate the historic and contem-
porary estimated endemicity classes into approximate values of the
reproductive number (see Methods). Using this model, the hypoen-
demic class (central PR value 5 5%) was interpreted as representing
an approximate R0 value of 1.3; holoendemic (central PR 5 30%) as
5.5; mesoendemic (central PR 5 63%) as 87.7; and hyperendemic
(central PR 5 88%) as 175.6. These conversions allowed the observed
changes between historical and contemporary endemicity to be recast
in terms of proportional changes (effect sizes) in the reproductive
number and summarized simply into orders of magnitude to reflect
their likely precision. In this way we found that, of the 66 million km2
of the Earth’s surface thought to have sustained stable/endemic
malaria in 1900, 12%, 18% and 57% had exhibited proportional
decreases in the reproductive number of up to one, between one
and two, and greater than two orders of magnitude, respectively;
11% had shown no evidence of change; and 2% had shown evidence
of an increase in the reproductive number by 2007 (see Supplemen-
tary Information for additional details of methods and results).
Although imperfect, this simple comparison illustrates that despite
warming global temperatures12, the combined natural and anthro-
pogenic forces acting on the disease throughout the twentieth
century have resulted in the great majority of locations undergoing
a net reduction in transmission between one and three orders of
magnitude larger than the maximum future increases proposed
under temperature-based climate change scenarios (Table 1).
If the effects of climate change are to reverse the observed declining
trend in malaria endemicity, they must exceed the collective counter-
acting effects of continuing economic development and increasing
control efforts. Whereas the former is hard to quantify, the potency
of currently available control and intervention tools has been assessed
theoretically using biological models21–23 and also evaluated in practice
when changes in local transmission have been measured concurrently
with aggressive malaria control initiatives24–29. We translated examples
of predicted and observed changes in endemicity caused by control
efforts into PfR0 effect sizes and compared these directly with effect
sizes proposed under climate change. This indicated that suites of
interventions at high coverage can achieve effect sizes exceeding two
orders of magnitude (Table 1). When compared to the substantially
smaller proposed magnitude of climate-induced effects, an important
and simple inference is that the latter can be offset by moderate
increases in coverage levels of currently available interventions.
Various sources of uncertainty exist in the inputs and methodo-
logies used in this study. The proposed levels of historical endemicity13
are plausible when triangulated against other values reported from the
pre-intervention era (for example, see refs 28, 30 and 31), but the
NATURE | Vol 465 | 20 May 2010
recent decades, while a growing body of evidence points to recent
regional declines in malaria morbidity or mortality in areas achieving
sustained intervention coverage11,26–29. We have interpreted the results
presented in this study at a global scale. Any effects of rising tempera-
tures on malaria transmission would most likely be extremely geo-
graphically variable, with most modelling studies suggesting that the
fringes of the current disease extent would be most sensitive to warm-
ing5,16,19. However, these more local predictions are themselves largely
undermined by current observations; the marginal transmission zones
are the places from which the largest declines in malaria transmission
or morbidity are being reported, concurrent with improved interven-
tion coverage26–29 and consistent with biological modelling studies
that indicate a greater theoretical impact on endemicity in areas of
lower baseline transmission32.
In an era when the international community has been emboldened
to provide guidelines for malaria elimination it is necessary to main-
tain the correct perspective on the future impact of climate change on
malaria epidemiology and by implication its malaria public health
importance. The quantification of a global recession in the range and
intensity of malaria over the twentieth century has allowed us to
review the rationale underpinning high-profile predictions of a cur-
rent and future worsening of the disease in a warming climate. It
suggests that the success or failure of our efforts against the parasite in
the coming century are likely to be determined by factors other than
climate change.
METHODS SUMMARY
The historical malaria endemicity map13 was scanned from the original publica-
tion, digitized on-screen and rasterized to a 5 3 5 km grid. The map of con-
temporary malaria endemicity was generated from a recently defined model8 of
age-standardised P. falciparum parasite rate, Pf PR2-10. Using a model-based geo-
statistical framework, the underlying value of Pf PR2-10 at each location was
modelled for the year 2007 as a transformation of a space-time Gaussian process
(GP), with the number of P. falciparum-positive individuals in each survey
modelled as a binomial variate given the unobserved age-standardised prevalence
surface. The GP was parameterised by a mean component (a linear function of
time and urban–peri-urban–rural status) and a space-time covariance function
which was spatially anisotropic, used great-circle distance to incorporate the
curvature of the earth, and included a periodic temporal component to capture
seasonality. Bayesian inference was implemented using Markov chain Monte
Carlo and direct simulation to generate posterior predictive samples of the
2007 annual mean prevalence surface and to assign each pixel to the endemicity
class with the highest posterior probability of membership.
Predicted Pf PR2-10 values at each pixel were converted into approximate
values of the P. falciparum basic reproductive number, PfR0, using a model that
assumes new infections are acquired and clear independently at a constant rate
and that biting is heterogeneously distributed in a population such that relative
biting rates follow a Gamma distribution. Pf PR was empirically related to the
Pf EIR using a log-linear relationship based on 91 paired observations. Pf EIR can
be inferred from Pf PR (X) by inverting the formula20,22. It follows that:
relatively crude categorization of all-cause malaria endemicity strata 

and the cartographic approach used preclude a more formal quan-
tification of the precision in the global P. falciparum endemicity
declines we report, and the subsequent conversions into P. falciparum
reproductive number effect sizes are approximate. However, the
key comparisons we present between observed declines, proposed
temperature-driven increases, and the impact of available counter-
measures rely on our estimates being accurate only to within one order
of magnitude. We have compared two snapshots of estimated global
endemicity separated by a period of 100 years during which changes in
global temperatures have been accelerating, with the largest observed
warming occurring in recent decades12. The progression of the reces-
sion in global malaria is less well measured, and several periods of
decline and resurgence are known to have occurred associated with,
for example, periods of coordinated large-scale control efforts, the
introduction and subsequent failure of successive therapeutics, or
breakdowns in public health infrastructure associated with conflict
or political upheaval. There is, however, no evidence of a systematic
upturn in malaria endemicity concurrent with the warming trend of
344
©2010 Macmillan Publishers Limited. All rights reserved
 ð1
XÞ
a
1  cð1 þ SkÞð1 þ aÞ
R0 ¼
a k
where k is the net infectiousness of humans, that is, the probability that a
mosquito will become infected after biting a human, c is the probability that a
mosquito will become infected after biting a non-immune infectious human,
and S is the stability index.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 3 February; accepted 16 April 2010.
1. Patz, J. A., Campbell-Lendrum, D., Holloway, T. & Foley, J. A. Impact of regional
climate change on human health. Nature 438, 310–317 (2005).
2. Lafferty, K. D. The ecology of climate change and infectious diseases. Ecology 90,
888–900 (2009).
3. McMichael, A. J. et al. in Comparative Quantification of Health Risks: Global and
Regional Burden of Disease due to Selected Major Risk Factors (eds Ezzati, M., Lopez,
A. D., Rodgers, A. & Murray, C. J. L.) 1543–1649 (World Health Organization,
2004).
4. Tanser, F. C., Sharp, B. & Le Sueur, D. Potential effect of climate change on malaria
transmission in Africa. Lancet 362, 1792–1798 (2003).
NATURE | Vol 465 | 20 May 2010
5. van Lieshout, M., Kovats, R. S., Livermore, M. T. J. & Martens, P. Climate change
and malaria: analysis of the SRES climate and socio-economic scenarios. Glob.
Environ. Change 14, 87–99 (2004).
6. Intergovernmental Panel on Climate Change. Climate Change 2007: Impacts,
Adaptation and Vulnerability. Contribution of Working Group II to the Fourth
Assessment Report of the Intergovernmental Panel on Climate Change (eds Parry, M.
L., Canziani, O. F., Palutikof, J. P., van der Linden, P. J. & Hanson, C. E.) (Cambridge
Univ. Press, 2007).
7. US Environmental Protection Agency. Endangerment and Cause or Contribute
Findings for Greenhouse Gases Under Section 202(a) of the Clean Air Act (Technical
Support Document) (US Environmental Protection Agency, 2010).
8. Hay, S. I. et al. A world malaria map: Plasmodium falciparum endemicity in 2007.
PLoS Med. 6, e1000048 (2009).
9. Hay, S. I., Guerra, C. A., Tatem, A. J., Noor, A. M. & Snow, R. W. The global
distribution and population at risk of malaria: past, present, and future. Lancet
Infect. Dis. 4, 327–336 (2004).
10. Snow, R. W., Guerra, C. A., Mutheu, J. J. & Hay, S. I. International funding for
malaria control in relation to populations at risk of stable Plasmodium falciparum
transmission. PLoS Med. 5, e142 (2008).
11. World Health Organization. World malaria report 2009 (World Health
Organization, 2009).
12. Intergovernmental Panel on Climate Change. Climate Change 2007: The Physical
Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change (eds Solomon, S. et al.) (Cambridge
Univ. Press, 2007).
13. Lysenko, A. J. & Semashko, I. N. [in Russian] in Itogi Nauki: Medicinskaja Geografija
(ed. Lebedew, A. W.) 25–146 (Academy of Sciences, Moscow, 1968).
14. Mouchet, J. et al. Biodiversité du paludisme dans le monde [in French] (John Libbey
Eurotext, 2004).
15. Small, J., Goetz, S. J. & Hay, S. I. Climatic suitability for malaria transmission in
Africa, 1911–1995. Proc. Natl Acad. Sci. USA 100, 15341–15345 (2003).
16. Martens, W. J. M., Jetten, T. H. & Focks, D. A. Sensitivity of malaria,
schistosomiasis and dengue to global warming. Clim. Change 35, 145–156 (1997).
17. Reiter, P. Global warming and malaria: knowing the horse before hitching the cart.
Malaria J. 7 (Suppl. 1), S3 (2008).
18. Lindsay, S. W. & Martens, W. J. M. Malaria in the African highlands: past, present
and future. Bull. World Health Organ. 76, 33–45 (1998).
19. Martens, P. et al. Climate change and future populations at risk of malaria. Glob.
Environ. Change 9, S89–S107 (1999).
20. Smith, D. L., McKenzie, F. E., Snow, R. W. & Hay, S. I. Revisiting the basic
reproductive number for malaria and its implications for malaria control. PLoS Biol.
5, e42 (2007).
21. Okell, L. C., Drakeley, C. J., Bousema, T., Whitty, C. J. & Ghani, A. C. Modelling the
impact of artemisinin combination therapy and long-acting treatments on malaria
transmission intensity. PLoS Med. 5, e226 (2008).
22. Smith, D. L., Hay, S. I., Noor, A. M. & Snow, R. W. Predicting changing malaria risk
after expanded insecticide-treated net coverage in Africa. Trends Parasitol. 25,
511–516 (2009).
23. Killeen, G. F. et al. The potential impact of integrated malaria transmission control
on entomologic inoculation rate in highly endemic areas. Am. J. Trop. Med. Hyg. 62,
545–551 (2000).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
24. Kouznetsov, R. L. Malaria control by application of indoor spraying of residual
insecticides in tropical Africa and its impact on community health. Trop. Doct. 7,
81–91 (1977).
25. Fillinger, U., Ndenga, B., Githeko, A. & Lindsay, S. W. Integrated malaria vector
control with microbial larvicides and insecticide-treated nets in western Kenya: a
controlled trial. Bull. World Health Organ. 87, 655–665 (2009).
26. Teklehaimanot, H. D., Teklehaimanot, A., Kiszewski, A., Rampao, H. S. & Sachs, J.
D. Malaria in São Tomé and Principe: on the brink of elimination after three years
of effective antimalarial measures. Am. J. Trop. Med. Hyg. 80, 133–140 (2009).
27. Kleinschmidt, I. et al. Reduction in infection with Plasmodium falciparum one year
after the introduction of malaria control interventions on Bioko Island, Equatorial
Guinea. Am. J. Trop. Med. Hyg. 74, 972–978 (2006).
28. Sharp, B. L. et al. Seven years of regional malaria control collaboration -
Mozambique, South Africa, and Swaziland. Am. J. Trop. Med. Hyg. 76, 42–47
(2007).
29. Bhattarai, A. et al. Impact of artemisinin-based combination therapy and
insecticide-treated nets on malaria burden in Zanzibar. PLoS Med. 4, e309
(2007).
30. Smith, D. L., Dushoff, J., Snow, R. W. & Hay, S. I. The entomological inoculation
rate and Plasmodium falciparum infection in African children. Nature 438,
492–495 (2005).
31. Smith, D. L., Guerra, C. A., Snow, R. W. & Hay, S. I. Standardizing estimates of the
Plasmodium falciparum parasite rate. Malar. J. 6, 131 (2007).
32. Smith, D. L. & Hay, S. I. Endemicity response timelines for Plasmodium falciparum
elimination. Malar. J. 8, 87 (2009).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank A. Bibby, H. C. J. Godfray, G. D. Shanks and
G. R. W. Wint for comments on the manuscript. S.I.H. is funded by a Senior
Research Fellowship from the Wellcome Trust (#079091) that also supports
P.W.G. and previously A.J.T. R.W.S. is funded by a Principal Research Fellowship
from the Wellcome Trust (#079080) that also supports A.P.P. D.L.S. and A.J.T. are
supported by a grant from the Bill and Melinda Gates Foundation (#49446). D.L.S.
and S.I.H. also acknowledge funding support from the RAPIDD program of the
Science & Technology Directorate, Department of Homeland Security, and the
Fogarty International Center, National Institutes of Health. This work forms part of
the output of the Malaria Atlas Project (MAP, http://www.map.ox.ac.uk),
principally funded by the Wellcome Trust, UK.
Author Contributions S.I.H. conceived the research. P.W.G. and S.I.H. drafted the
manuscript. P.W.G. led, and A.P.P., D.L.S., A.J.T. and R.W.S. contributed to, the
analyses. All authors discussed the results and contributed to the revision of the
final manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to S.I.H. (simon.hay@zoo.ox.ac.uk) or P.W.G
(peter.gething@zoo.ox.ac.uk).
345
doi:10.1038/nature09098
METHODS
Generating comparable historical and contemporary endemicity maps. The
historical malaria endemicity map13 was scanned from the original publication
and geo-referenced using ERDAS Imagine 8.5 (Leica Geosystems GIS &
Mapping). The map was then digitised on-screen with MapInfo Professional
7.0 (MapInfo), and rasterized to a 5 3 5 km grid. The map of contemporary
malaria endemicity was generated from a recently defined model8 of P. falciparum
infection prevalence within the previously defined limits of stable transmission33.
The underlying value of P. falciparum parasite rate in the 2–10-year age cohort,
PfPR2
10 ðxi Þ, at each location xi was modelled for the year 2007 as a transforma-
tion gð:Þ of a space-time Gaussian process f ðxi ; ti Þ with mean m and covariance C
superimposed with additional aspatial (random) variation eðxi Þ, represented as
Gaussian with zero mean and variance V . The number of P. falciparum positive
individuals, Niþ , from the total sample of Ni in each survey was modelled as a
conditionally independent binomial variate given the unobserved underlying age-
standardized PfPR2
10 value34. The space-time mean component m was modelled
as a linear function of time, t, and urban–peri-urban–rural status (denoted by the
endemicity class with the highest posterior probability of membership, identified
as the class containing the largest proportion of posterior samples. Pixels modelled
as being at unstable or no risk were assigned directly because these classifications
were deterministic. The age-standardization (2–10-year-olds) matched that used in
the historical map13 except for the holoendemic class (Pf PR.70%) which the
authors defined as relating to prevalence in one-year-olds.
Translating predicted endemicity into approximate values of PfR0. We have
developed a function to estimate the Pf R0 from the Pf PR on the basis of a model
that assumes that new infections are acquired and clear independently at a
constant rate and that biting is heterogeneously distributed in a population such
that relative biting rates follow a Gamma distribution with mean 1 and variance
a20,22,30. The transformation was developed with 91 paired estimates of the Pf EIR
and the Pf PR in African children, standardized following an algorithm described
in ref. 31. Pf PR is empirically related to the Pf EIR in these 91 paired estimates by
a log-linear relationship38, or by a simple formula that describes the steady state
of a malaria transmission model20. Pf EIR can be inferred from Pf PR ðXÞ by
inverting the formula20,22. It follows that:
indicator variables 1u1 (x), 1p1 (x)). The mean component was therefore defined 

as:
m ¼ bx þ bt t þ bu1 1u1 ðxÞ þ bp1 1p1 ðxÞ
where bx denotes the intercept. Each parasite rate survey was referenced temporally
using the mid-point (in decimal years) between the recorded start and end months.
Covariance between spatial and temporal locations was modelled using a spatially
anisotropic space-time covariance function C with a periodic component (wave-
length 5 12 months) added to the time-marginal covariance model to capture
seasonality35:
ðDxÞcðDtÞ KcðDtÞ ðDxÞ
Cðxi ; ti ; xj ; tj Þ ¼ t2 cð0Þ ;
2cðDtÞ
1 CðcðDtÞ þ 1Þ
1 infectiousness, we assume c ¼ 0:1 consistent with ref. 39, and we use the formula
cðDtÞ ¼ ; described elsewhere20,22, assuming no immunity. We also assume that S ¼ 1,
2r þ 2ð1
rÞ½ð1
uÞe
jDtj=wt þ u cosð2pDtÞ
Dt ¼ jti
tj j
where Kc is the modified Bessel function of the second kind of order c, and C is the
gamma function36,37. The effect of the curvature of the earth on point-to-point
separations, and a mechanism for spatial anisotropy, were incorporated by com-
puting the spatial distance between a pair of points xi and xj as great-circle distance
DGC ðxi ; xj Þ on a flexible ellipsoid. Bayesian inference was implemented using
Markov Chain Monte Carlo to generate samples from the posterior distribution
of the Gaussian field f ðxi ; ti Þ at each data location and of the unobserved parameters
of the mean, covariance function and Gaussian random noise component and
direct simulation was then used to generate samples from the 2007 annual mean
of the posterior distribution of f ðxi ; ti Þ at each prediction location across the same
template 5 3 5 km grid as that used for the historical map. Model output therefore
consisted of samples from the posterior predictive distribution of the 2007 annual
mean PfPR2-10 at each grid location. We assigned each pixel to the historical
©2010 Macmillan Publishers Limited. All rights reserved
 ð1
XÞ
a
1  cð1 þ SkÞð1 þ aÞ
R0 ¼
a k
where k is the net infectiousness of humans, the probability that a mosquito will
become infected after biting a human, c is the probability that a mosquito will
become infected after biting a non-immune infectious human, and S is the
stability index.
The transformation from Pf PR to PfEIR gives unsatisfactory estimates of
Pf EIR when Pf PR exceeds approximately 65%. At these high values of Pf EIR
and Pf PR, it is generally possible to find a value such that the function fits exactly.
These values are statistically significantly and negatively correlated with Pf PR, by
the relationship a ¼ 9:292
8:035X . To make the transformation, we take
a ¼ minð4:2; 9:292
8:035XÞ. To make the transformation from Pf PR to net
although k , 1, and S is generally lower than approximately 5, so 1 , 1 1 Sk , 1.5.
33. Guerra, C. A. et al. The limits and intensity of Plasmodium falciparum transmission:
implications for malaria control and elimination worldwide. PLoS Med. 5, e38
(2008).
34. Diggle, P. & Ribeiro, P. J. Model-based Geostatistics (Springer, 2007).
35. Stein, M. L. Space-time covariance functions. J. Am. Stat. Assoc. 100, 310–321
(2005).
36. Antosiewicz, H. A. in Handbook of Mathematical Functions (eds Abramowitz, M. &
Stegun, I. A.) 435–479 (Dover Publications, 1964).
37. Davis, G. M. in Handbook of Mathematical Function (eds Abramowitz, M. & Stegun,
I. A.) 253–295 (Dover Publications, 1964).
38. Hay, S. I., Guerra, C. A., Tatem, A. J., Atkinson, P. M. & Snow, R. W. Urbanization,
malaria transmission and disease burden in Africa. Nature Rev. Microbiol. 3, 81–90
(2005).
39. Killeen, G. F., Ross, A. & Smith, T. Infectiousness of malaria-endemic human
populations to vectors. Am. J. Trop. Med. Hyg. 75, 38–45 (2006).
 Vol 465 | 20 May 2010 | doi:10.1038/nature09074

LETTERS
Staphylococcus epidermidis Esp inhibits Staphylococcus
aureus biofilm formation and nasal colonization
Tadayuki Iwase1, Yoshio Uehara2, Hitomi Shinji1, Akiko Tajima1, Hiromi Seo2, Koji Takada3, Toshihiko Agata4
& Yoshimitsu Mizunoe1

Commensal bacteria are known to inhibit pathogen colonization; co-cultured with S. aureus, 428 of 960 isolates from volunteers’ nasal
however, complex host–microbe and microbe–microbe interac- cavities inhibited biofilm formation by S. aureus in a dose-dependent
tions have made it difficult to gain a detailed understanding of manner, whereas the remaining 532 strains did not. This finding
the mechanisms involved in the inhibition of colonization1. Here indicated that S. epidermidis can be divided into two types: one that
we show that the serine protease Esp2,3 secreted by a subset of inhibits biofilm formation by S. aureus (inhibitory type) and one that
Staphylococcus epidermidis, a commensal bacterium, inhibits biofilm does not (non-inhibitory type) (Fig. 1a and b). Furthermore, the
formation and nasal colonization by Staphylococcus aureus, a culture supernatants of inhibitory S. epidermidis destroyed pre-existing
human pathogen4. Epidemiological studies have demonstrated that S. aureus biofilms, whereas those of non-inhibitory S. epidermidis did
the presence of Esp-secreting S. epidermidis in the nasal cavities of not (Fig. 1c).
human volunteers correlates with the absence of S. aureus. Purified On the basis of these findings, we conducted an epidemiological
Esp inhibits biofilm formation and destroys pre-existing S. aureus study investigating the relationship between the nasal colonization of
biofilms. Furthermore, Esp enhances the susceptibility of S. aureus inhibitory S. epidermidis and that of S. aureus. Univariate logistic
in biofilms to immune system components. In vivo studies have regression analysis showed that S. aureus detection rate was signifi-
shown that Esp-secreting S. epidermidis eliminates S. aureus nasal cantly lower in the presence of inhibitory S. epidermidis (odds ratio
colonization. These findings indicate that Esp hinders S. aureus (OR), 0.29; 95% confidence interval (CI), 0.11–0.75; Table 1) than in
colonization in vivo through a novel mechanism of bacterial inter-
ference, which could lead to the development of novel therapeutics Table 1 | Odds ratios for S. aureus colonization in 88 volunteers (univariate
to prevent S. aureus colonization and infection. analysis)
Staphylococcus aureus is an important human bacterial pathogen S. aureus colonization
responsible for a wide variety of conditions, ranging from subclinical
Measurement Yes (n 5 28) No (n 5 60) Odds ratio (95% CI) P-value
inflammation to severe infections causing pneumonia, endocarditis and
septicaemia4. Recently, emergence of multi-drug-resistant S. aureus, Age (years) 22.2* 21.7* 1.10 (0.78–1.55) 0.58
such as methicillin-resistant S. aureus (MRSA) and vancomycin- (1.2){ (1.3){
Sex 2.61 (0.93–7.40) 0.07
resistant S. aureus (VRSA), has made treatment difficult and increased Male 22 35
the rate of mortality5,6. The primary reservoir for S. aureus is the nasal Female 6 25
cavity7,8. One-third of the population in the United States of America, Active smoking 0.33 (0.04–2.91) 0.32
Yes 1 6
the United Kingdom, Japan, and other countries is colonized with this No 27 54
pathogen asymptomatically; however, the colonization is known to be Passive smoking 0.69 (0.26–1.83) 0.46
a risk factor for staphylococcal infections9–13. Intriguingly, there are Yes 8 22
people capable of evading S. aureus colonization. Elucidating this No 20 38
Allergy in nose 1.21 (0.49–3.00) 0.69
evasion mechanism may lead to a better understanding of bacterial Yes 12 23
ecology, including commensal–pathogen interactions, and provide No 16 37
information to develop effective drugs against S. aureus. Allergy in eye 1.08 (0.30–3.95) 0.90
As a first step to explain the evasion mechanism, we focused on the Yes 4 8
No 24 52
possible inhibitory roles of Staphylococcus epidermidis in S. aureus Allergy in skin 1.49 (0.51–4.36) 0.47
colonization because S. epidermidis is the dominant commensal bac- Yes 7 11
terium found in the human nasal cavity. We isolated these bacteria No 21 49
from the nasal cavities of 88 volunteers. Bacteriological analysis Antibiotic treatment in last month 0.69 (0.13–3.67) 0.67
Yes 2 6
showed that S. aureus had colonized 28 (31.8%) subjects, of whom No 26 54
26 (92.9%) were also colonized by S. epidermidis (Table 1). Of the 60 Colonization of S. epidermidis 0.45 (0.06–3.36) 0.43
non-carriers of S. aureus, 58 (96.7%) were colonized by S. epidermidis Yes 26 58
(Table 1). The prevalence of S. epidermidis in nasal carriers and non- No 2 2
Colonization of inhibitory S. epidermidis 0.29 (0.11–0.76) 0.01
carriers of S. aureus did not differ significantly (Table 1). However, 0.30 (0.11–0.80){
further investigation indicated that some properties of S. epidermidis Yes 8 35
isolated from carriers of S. aureus might differ from those in non- No 20 25
carriers. When S. epidermidis cells or their culture supernatants were * Mean, {standard deviation, {multivariate analysis (adjusted for age, sex or smoking habits).

1
Department of Bacteriology, The Jikei University School of Medicine, Tokyo, 105-8461 Japan. 2Department of General Medicine, Kochi Medical School, Nankoku, 783-8505 Japan.
3
Department of Biochemistry, The Jikei University School of Medicine, Tokyo, 105-8461 Japan. 4Department of Environmental Health, The Jikei University School of Medicine, Tokyo,
105-8461 Japan.

346
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 LETTERS

Destruction of cavity could be a potential determining factor for the absence of

Inhibition of S. aureus biofilm formation S. aureus biofilms S. aureus colonization.
a NS
b NS
c NS
To identify the inhibitory factor, we collected the culture super-

S. aureus biofilms (A600)

S. aureus biofilms (A600)

natants of inhibitory S. epidermidis after 16 h culture, because the

S. aureus biofilms (%)

P < 0.001 P < 0.001 1.2 P < 0.001
1.2
100
biofilm-destruction activity in culture supernatants of inhibitory S.
0.9 80 0.9
epidermidis increased with bacterial growth and reached its maximal
60
0.6 0.6 level at 8–16 h (Fig. 2a). Following collection, we purified the factor
40
0.3 0.3 from the culture supernatants by a series of fractionation procedures
20
including salt precipitation, gel filtration and ion-exchange chromato-
0 0 0
0 0.01 0.1 1 0 103 104 105 0 0.01 0.1 1 graphy. The isolated protein had a single component with an apparent
Concentration of Cell concentration Concentration of molecular mass of approximately 27 kDa (Fig. 2b). Amino acid
S. epidermidis culture of S. epidermidis S. epidermidis culture sequence analysis showed that the protein factor was S. epidermidis
supernatants (%) (c.f.u. ml–1) supernatants (%)
serine protease (Esp2,3). When culture supernatants of non-inhibitory
Figure 1 | Inhibition of S. aureus biofilm formation and destruction of S. S. epidermidis were subjected to the same purification procedures, we
aureus biofilms by S. epidermidis. a, b, The inhibitory effect of S. epidermidis could not find the protease (data not shown). In addition, immuno-
culture supernatants (a) or cells (b) on S. aureus biofilm formation. c, The blotting with anti-Esp antibody showed that Esp protein was present in
destructive effect of S. epidermidis culture supernatants on S. aureus the culture supernatant of the inhibitory S. epidermidis JK16 strain,
biofilms. After these treatments by S. epidermidis culture supernatants or
but not in that of the non-inhibitory S. epidermidis JK11 strain (Sup-
cells on S. aureus, the amount of S. aureus biofilms was measured. Filled
bars, effect of inhibitory S. epidermidis (JK16 strain); open bars, effect of plementary Fig. 1).
non-inhibitory S. epidermidis (JK11 strain). Bars show the mean value of the To confirm that Esp was responsible for the destruction of
experiments (n 5 3). Error bars show standard deviation (s.d.). S. aureus biofilms, we generated an isogenic esp-deficient mutant from
the inhibitory S. epidermidis strain, JK16. The culture supernatant of
its absence. This difference remained significant (OR, 0.30; 95% CI, the mutant did not show any biofilm-destruction activity; however,
0.11–0.80; Table 1) when the multivariate logistic regression model this activity was restored by complementation with the esp gene
was adjusted for age, sex and smoking habit. These findings indicated (Fig. 2c and Supplementary Fig. 1). Because Esp shows serine-protease
that in humans, the presence of inhibitory S. epidermidis in the nasal activity2, we used the serine-protease inhibitor, amidinophenyl

a b M
g h i
Absorbance (A600)

2 100
80
Activity (%)

1.5
60 Esp (–)
1.0
40
0.5 20
j k l
0 // 0
0 2 4 6 8 16
Time (h)
Esp (+)
c P < 0.001 d P < 0.001
S. aureus biofilm (A600)

S. aureus biofilm (A600)

1.2 1.2

0.9 0.9

0.6 0.6
m P < 0.001
0.3
(percentage of viable cells)

0.3
100
Bacterial survival

0 0
Esp
l

pe

d
t

– 80
tro

+ +
an

te
ty

ut
on

en

APMSF – – +
ild

em
C

60
W

pl
om

e f 40
C
S. aureus biofilm (A600)

1.2
S. aureus biofilm (A600)

1.2 20
**
0.9
0.9 * 0
** **
0.6 hBD2 – –
0.6 ** + + + +
Esp – – – –
0.3 ** **
** 0.3
**
0 0 0.01 0.1 1 10 0 **
0 1 2 3 4 5 6
Esp (nM) Time (h)

Figure 2 | Isolation and characterization of the serine protease Esp, the f, Esp destroyed S. aureus biofilms in a time-dependent manner. The
factor responsible for the biofilm-destruction activity, secreted by biofilms were incubated in the presence (closed circles) or absence (open
inhibitory S. epidermidis. a, Growth curve (open circles) of inhibitory S. circles) of Esp for the indicated times. g–l, Microscopic observation of S.
epidermidis (JK16 strain) and the biofilm-destruction activity (closed aureus biofilms incubated for 6 h in the presence (j–l) or absence (g–i) of Esp.
circles) of the culture supernatants of the same strain. The activity of the Gram staining (g and j) and scanning electron micrographs (h, i, k, and l) of
supernatants at 8 h is shown as 100%. b, A protein having the biofilm- the biofilms with scale bars (10 mm in g, h, j, and k, and 1 mm in i and l). The
destruction activity was purified from the culture media of inhibitory S. arrows in j indicate the intercellular matrix. m, Esp enhanced the
epidermidis (arrow). M, molecular mass markers. c, Effects of the culture susceptibility of S. aureus in biofilms to hBD2. Viability of S. aureus cells in
supernatants of inhibitory S. epidermidis (JK16, wild-type strain), an biofilms, which was incubated in the presence (1 or triangles) or absence
isogenic esp-deficient strain and a complemented strain on S. aureus (2) of Esp (1 mM) and hBD2 (1, 5 and 10 mM, from left to right in each
biofilms. d, The biofilm-destruction activity of purified Esp was blocked by triangle) for 6 h. The cell viability without Esp and hBD2 was set as 100%.
APMSF. (1) or (2) indicate the presence or absence of Esp and APMSF. Plots and columns show mean and s.d. (n 5 3). Statistical significance is
e, Esp inhibited S. aureus biofilm formation in a dose-dependent manner. indicated as *P , 0.05 and **P , 0.01.
347
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
methansulphonyl fluoride (APMSF) to test whether this protease
activity was involved in biofilm destruction. The results of these tests
showed that treatment with APMSF blocked biofilm destruction by
Esp (Fig. 2d). Purified Esp inhibited S. aureus biofilm formation dose-
dependently (Fig. 2e) and destroyed S. aureus biofilms (Fig. 2f). To
investigate the destructive effect of inhibitory S. epidermidis on bio-
films formed by various S. aureus strains, including MRSA and
vancomycin-intermediate S. aureus (VISA) (Supplementary Table 1),
they were subjected to the biofilm-destruction test. Biofilms formed
by almost all strains were efficiently destroyed by Esp (Supplementary
Fig. 2). We then investigated whether S. aureus could become resistant
to the biofilm-destruction activity of Esp. After S. aureus was co-
cultured with Esp for 1 year, the sensitivity of S. aureus to Esp was
unchanged, indicating that no resistance developed during this time
(data not shown).
When analysed by microscopy, Esp changed S. aureus from the
sessile to the planktonic form (Fig. 2g and j). This same effect was also
observed in MRSA and VISA biofilms (Supplementary Fig. 3). Electron
microscopic analysis indicated that the intercellular matrix in S. aureus
biofilms disappeared after Esp treatment (Fig. 2h, i, k, and l).
We further investigated the characteristics of Esp, including its effect
on the host’s immune system. Human beta-defensin 2 (hBD2)14 is an
a Day 0 Day 1 Day 2
b 104
NS
S. aureus c.f.u. per swab
103
* at 37 uC for 16 h, and to evaluate their destructive effect on biofilms they were
* added to the culture dishes containing S. aureus biofilms and incubated for 6 h.
* *
102
101
DL
0 1 2 3
c 1.0
Carrier rate of S. aureus
0.8
0.6
0.4
esp-deficient strain
0.2 Wild-type (inhibitory-type) strain
0
0 1 2 3
Time (days)
Administration
Figure 3 | Elimination effect of inhibitory S. epidermidis cells on S. aureus
nasal colonization. a, Representative culture images of samples from test
persons after administration of inhibitory S. epidermidis (JK16, wild-type
strain). The nasal swabs from the volunteers were cultured on mannitol salt
agar with egg yolk. b, The bacterial counts of S. aureus from the nasal swabs
after the administration of S. epidermidis cells. JK16 strain (closed circles;
n 5 7) or an isogenic esp-deficient strain (open circles; n 5 6) were placed
into the nasal cavities of the volunteers. Plots show mean and s.d. Statistical
significance is indicated at *P , 0.05. NS, not significant; DL, detection
limit. c, Carrier rate of S. aureus after the administration of S. epidermidis
wild type cells (closed circles; n 5 7) and esp-deficient cells (open circles;
n 5 6). Statistical significance according to Kaplan–Meier method and log-
rank test is at *P , 0.05.
348
NATURE | Vol 465 | 20 May 2010
antimicrobial peptide component of the human innate immune sys-
tem and is secreted by keratinocytes. Esp demonstrated no bactericidal
activity, and hBD2 alone demonstrated a low bactericidal activity
towards S. aureus in biofilms (Fig. 2m). However, when combined
with Esp, hBD2 effectively killed S. aureus in biofilms (Fig. 2m).
These findings indicate that the effects of Esp secreted by inhibi-
tory S. epidermidis against S. aureus are due to a novel mechanism
distinct from all known bacterial interference mechanisms such as
growth inhibition, bactericidal activity, or competition for specific
attachment molecules15–19.
To investigate the effect of the bacterial interference on S. aureus nasal
colonization directly, S. epidermidis cells were introduced into the nasal
cavities of volunteers who were S. aureus carriers. Although the wild-
type S. epidermidis eliminated S. aureus colonization, its isogenic esp
mutant did not (Fig. 3 and Supplementary Fig. 4). Moreover, purified
Esp also cleared S. aureus (Supplementary Fig. 5). Non-inhibitory
S. epidermidis (wild-type, inherent Esp-negative strain) did not show
this eliminating effect (data not shown).
Previously, the role of S. epidermidis in S. aureus colonization has
been unclear15,20–24. Here we demonstrate that a subset of S. epidermidis
can inhibit S. aureus colonization by secreting Esp. The mechanisms
regulating Esp expression in S. epidermidis remain to be determined.
In recent years, human microbiome projects have been conducted
Day 5
worldwide at the species, genus, or phylum level25–30. Further char-
acterization of the microbes at the strain level will help increase our
understanding of microbial ecology, including the relationships
between commensal bacteria and infectious pathogens, or host–
microbe interactions.
METHODS SUMMARY
Eighty-eight healthy adult volunteers participated in the study, and 960 bacterial
strains were identified from their nasal cavities. The isolates of S. epidermidis or
their culture supernatants were used as the samples to assess the effects on
biofilms formed by S. aureus. To evaluate this inhibitory effect on biofilm forma-
tion, these S. epidermidis samples were simultaneously incubated with S. aureus
* After these treatments, the amount of S. aureus biofilm were measured spectro-
photometrically. A protein factor responsible for the biofilm destructive activity
was isolated from the culture supernatants of inhibitory S. epidermidis using a
series of chromatographic steps, and a partial amino acid sequence was deter-
mined. The relationship between colonization rates of S. aureus and inhibitory S.
4 5 epidermidis in the nasal cavity was epidemiologically investigated. The inhibitory
S. epidermidis or the protein factor obtained from it and identified as S. epidermidis
serine protease (Esp) was placed into the nasal cavity of S. aureus carriers, and their
effects on colonization of S. aureus in each cavity were analysed. Epidemiological
and in vivo studies were approved by the ethics committees of The Jikei University
P < 0.001 School of Medicine and Kochi Medical School, respectively, and informed consent
was obtained from all participants.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 28 December 2009; accepted 8 April 2010.
4 5
1. Wertheim, H. F. et al. The role of nasal carriage in Staphylococcus aureus infections.
Lancet Infect. Dis. 5, 751–762 (2005).
2. Moon, J. L., Banbula, A., Oleksy, A., Mayo, J. A. & Travis, J. Isolation and
characterization of a highly specific serine endopeptidase from an oral strain of
Staphylococcus epidermidis. Biol. Chem. 382, 1095–1099 (2001).
3. Dubin, G. et al. Molecular cloning and biochemical characterisation of proteases
from Staphylococcus epidermidis. Biol. Chem. 382, 1575–1582 (2001).
4. Lowy, F. D. Staphylococcus aureus infections. N. Engl. J. Med. 339, 520–532 (1998).
5. Klein, E., Smith, D. L. & Laxminarayan, R. Hospitalizations and deaths caused by
methicillin-resistant Staphylococcus aureus, United States, 1999-2005. Emerg.
Infect. Dis. 13, 1840–1846 (2007).
6. Srinivasan, A., Dick, J. D. & Perl, T. M. Vancomycin resistance in staphylococci.
Clin. Microbiol. Rev. 15, 430–438 (2002).
7. Kluytmans, J., van Belkum, A. & Verbrugh, H. Nasal carriage of Staphylococcus
aureus: epidemiology, underlying mechanisms, and associated risks. Clin.
Microbiol. Rev. 10, 505–520 (1997).
8. Peacock, S. J., de Silva, I. & Lowy, F. D. What determines nasal carriage of
Staphylococcus aureus? Trends Microbiol. 9, 605–610 (2001).
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
9. Graham, P. L. III, Lin, S. X. & Larson, E. L. A U.S. population-based survey of
Staphylococcus aureus colonization. Ann. Intern. Med. 144, 318–325 (2006).
10. Kuehnert, M. J. et al. Prevalence of Staphylococcus aureus nasal colonization in the
United States, 2001–2002. J. Infect. Dis. 193, 172–179 (2006).
11. Shopsin, B. et al. Prevalence of methicillin-resistant and methicillin-susceptible
Staphylococcus aureus in the community. J. Infect. Dis. 182, 359–362 (2000).
12. Perl, T. M. et al. Intranasal mupirocin to prevent postoperative Staphylococcus
aureus infections. N. Engl. J. Med. 346, 1871–1877 (2002).
13. von Eiff, C., Becker, K., Machka, K., Stammer, H. & Peters, G. Nasal carriage as a
source of Staphylococcus aureus bacteremia. Study Group. N. Engl. J. Med. 344,
11–16 (2001).
14. Lehrer, R. I. Primate defensins. Nature Rev. Microbiol. 2, 727–738 (2004).
15. Otto, M. Staphylococcus aureus and Staphylococcus epidermidis peptide
pheromones produced by the accessory gene regulator agr system. Peptides 22,
1603–1608 (2001).
16. Mackowiak, P. A. The normal microbial flora. N. Engl. J. Med. 307, 83–93 (1982).
17. Brook, I. Bacterial interference. Crit. Rev. Microbiol. 25, 155–172 (1999).
18. Falagas, M. E., Rafailidis, P. I. & Makris, G. C. Bacterial interference for the
prevention and treatment of infections. Int. J. Antimicrob. Agents 31, 518–522
(2008).
19. Uehara, Y. et al. H2O2 produced by viridans group streptococci may contribute to
inhibition of methicillin-resistant Staphylococcus aureus colonization of oral
cavities in newborns. Clin. Infect. Dis. 32, 1408–1413 (2001).
20. Speck, W. T., Driscoll, J. M., Polin, R. A. & Rosenkranz, H. S. Effect of bacterial flora
on staphylococcal colonisation of the newborn. J. Clin. Pathol. 31, 153–155 (1978).
21. Poutrel, B. & Lerondelle, C. Protective effect in the lactating bovine mammary
gland induced by coagulase-negative staphylococci against experimental
Staphylococcus aureus infections. Ann. Rech. Vet. 11, 327–332 (1980).
22. Lina, G. et al. Bacterial competition for human nasal cavity colonization: role of
staphylococcal agr alleles. Appl. Environ. Microbiol. 69, 18–23 (2003).
23. Peacock, S. J. et al. Determinants of acquisition and carriage of Staphylococcus
aureus in infancy. J. Clin. Microbiol. 41, 5718–5725 (2003).
24. Nicoll, T. R. & Jensen, M. M. Preliminary studies on bacterial interference of
staphylococcosis of chickens. Avian Dis. 31, 140–144 (1987).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
25. Gill, S. R. et al. Metagenomic analysis of the human distal gut microbiome. Science
312, 1355–1359 (2006).
26. Turnbaugh, P. J. et al. The human microbiome project. Nature 449, 804–810 (2007).
27. Gao, Z., Tseng, C. H., Pei, Z. & Blaser, M. J. Molecular analysis of human forearm
superficial skin bacterial biota. Proc. Natl Acad. Sci. USA 104, 2927–2932 (2007).
28. Hooper, L. V. et al. Molecular analysis of commensal host-microbial relationships
in the intestine. Science 291, 881–884 (2001).
29. Eckburg, P. B. et al. Diversity of the human intestinal microbial flora. Science 308,
1635–1638 (2005).
30. Bik, E. M. et al. Molecular analysis of the bacterial microbiota in the human
stomach. Proc. Natl Acad. Sci. USA 103, 732–737 (2006).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank K. Hiramatsu and T. Bae for providing materials and
M. Sekiguchi, T. Bae, S. N. Wai, B. E. Uhlin, L. Cui, T. Ito, M. Yoneda, M. Urashima
and S. Masuda for discussions, critical comments and advice. Thanks also go to
S. Kuramoto, K. Seki, F. Sato, S. Hoshina, T. Ohashi, H. Ikeshima-Kataoka,
Y. Yoshizawa, M. Murai and M. Kono for their comments on the study, and to
J. Fitzpatrick and M. Okazaki for their comments on the manuscript, and to our
colleagues for their assistance. Finally, we thank all persons involved in the study. A
part of the study was supported by The Jikei University Research Fund and by The
Jikei University Graduate Research Fund.
Author Contributions T.I. and Y.M. designed the research and wrote the
manuscript. All authors contributed the experiments; T.I., H.S., A.T., K.T. and Y.M.
for in vitro study and epidemiological study, Y.U. and H.S. for in vivo study, T.A. for
statistics. All authors discussed the results and commented on the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to T.I. (iwase.tadayuki@jikei.ac.jp).
349
doi:10.1038/nature09074
METHODS
Epidemiological study. In January 2006, 88 healthy college students were
recruited for our study. The study was approved by the ethics committees of
the Jikei University School of Medicine, and informed consent was obtained
from all participants.
Each participant answered a questionnaire that included questions about
demographics (age and sex), antibiotic treatment received within 1 month
before beginning the study, passive exposure to tobacco smoke, active smoking
habits, and allergic symptoms (Table 1). Sterile 5 mm-diameter cotton swabs
(Eiken Kagaku) soaked in sterile physiological saline were used to obtain nasal
samples from one nostril of each subject. The swabs were streaked onto mannitol
salt agar plates (Merck KGaA). S. aureus was identified using Gram staining and
tests for DNase activity, mannitol metabolism and coagulase production. MRSA
was identified using the PCR method31 and MRSA Screen Agar (Nippon Becton
Dickinson). Coagulase-negative staphylococci were identified using the PCR
method32,33, API-Staph (Japan bioMérieux), and the N-ID-Test SP-18 (Nissui
Pharmaceutical).
Bacterial strains and culture conditions. The following strains were used in this
study: 960 S. epidermidis strains isolated from the subjects (at least five from each
subject, including JK16 strain and JK11 strain), S. aureus JCM 2874 (a methicillin-
sensitive S. aureus (MSSA) strain), 12 laboratory-isolated MSSA strains; six
clinically-isolated MRSA strains; and Mu50 (one clinically isolated VISA strain)34,
an isogenic esp–deficient mutant (TI1) derived from JK 16 strain; and a TI1 strain
complemented by the esp gene (TI1/pTI1) (Supplementary Table 1). The bacteria
were grown in tryptic soy broth (TSB) (Difco), on TSB agar (Difco), or on
mannitol salt agar with egg yolk (Merck) at 37 uC under aerobic conditions.
Preparation of S. epidermidis culture supernatants. After isolating S. epidermidis
from volunteers, the culture supernatants were immediately prepared from these
fresh isolates. Culture supernatants were prepared by centrifuging cultures that
had been grown overnight in TSB with shaking at 37 uC. Samples were sterilized by
filtration across a 0.2-mm diameter membrane filter (Millipore).
Biofilm formation of S. aureus. Overnight TSB cultures of S. aureus were used to
inoculate 35 mm-diameter polystyrene dishes (Asahi Techno Glass) filled with
2.5 ml of TSB with 0.2% glucose at cell concentrations of 103–105 colony forming
units (c.f.u.) per millilitre and incubated at 37 uC for 16 h under static conditions.
Effects of culture supernatants and cells of S. epidermidis on S. aureus
biofilms. To determine their effect on the formation of S. aureus biofilms, culture
supernatants of S. epidermidis at various concentrations (from 0.01–10%) were
simultaneously added with the S. aureus inoculant (105 c.f.u. ml21) to the test
dishes. The culture media were incubated for 16 h at 37 uC under static conditions.
After the incubation, the dishes were gently shaken to remove the deposited
bacteria, and the culture media were discarded. The dishes were gently rinsed
with phosphate-buffered saline (PBS). The biofilms were scraped out and sus-
pended in PBS with vigorous vortex mixing for 1 min, and then the absorbance at
600 nm (A600) of each suspension was measured. In addition, the S. aureus cell
suspension at a concentration of 103 c.f.u. ml21 was co-incubated with various
concentrations of S. epidermidis cells (103, 104 and 105 c.f.u. ml21) for 16 h at
37 uC under static conditions. After the incubation, the dishes were gently shaken
to remove the deposited bacteria, and the culture media were discarded. The
dishes were gently rinsed with phosphate-buffered saline (PBS). The biofilms
were scraped out and suspended in PBS with vigorous vortex mixing for 1 min.
The number of S. aureus cells in the biofilms was counted by a conventional
culture method using mannitol salt agar with egg yolk. To evaluate the effect of
S. epidermidis on the stability of S. aureus biofilms, the culture supernatants of
S. epidermidis at various concentrations were added to culture dishes after 16 h of
S. aureus inoculation, and then incubated for an additional 6 h. The detached
bacteria were removed, and turbidities of the biofilm suspensions were deter-
mined as described above.
Isolation and characterization of Esp, the factor responsible for biofilm
destruction activity. The factor responsible for the biofilm destruction activity
was purified as a protein from the culture supernatants of inhibitory S. epidermidis,
and identified as Esp. The factor was fractionated and isolated using an indicator of
the biofilm destruction activity. The 16-h culture supernatants of inhibitory S.
epidermidis were sterilized by filtration, and salted out with 65% ammonium
sulphate, the precipitates were then dissolved in 1 ml H2O (the pH was adjusted
to 7.4 with phosphate buffer). This crude material was fractionated by using a
Sephadex G-200 gel filtration column (GE Healthcare UK), and by the protein
purification kits comprised cation and anion exchangers (Vivapure, Vivascience)
using 25 mM acetate buffer, pH 5.5, and 25 mM Tris-HCl, pH 8.0, for each equi-
libration, respectively. Further procedures were performed according to the
manufacturer’s instructions. The purified protein fraction having the activity
was then subjected to native-PAGE using a 10–20% gradient acrylamide gel
(Atto) and visualized with a negative staining kit (Atto); a major distinct band
©2010 Macmillan Publishers Limited. All rights reserved
was spliced out and eluted by sample buffer, and its activity was confirmed. The
isolated protein with biofilm destruction activity was also analysed with SDS–
PAGE using a 15%-acrylamide gel and stained with a silver staining kit (Atto).
In the other experiment, the isolated protein was subjected to SDS–PAGE followed
by transfer to a polyvinylidene fluoride membrane (Immobilon-PSQ, Millipore),
and visualized with Coomassie blue staining. The corresponding band was cut out
and subjected to automated Edman degradation with a protein sequencer (model
492, Applied Biosystems). An amino-terminal sequence of 15 amino acids of the
obtained protein factor was determined to be VILPNNNRHQIFNTT, which cor-
responded to that of the S. epidermidis serine proteinase, Esp, which was isolated
and cloned in 2001 (refs 2, 3) and characterized in detail35. To evaluate the effect of
the Esp on the formation of S. aureus biofilm, Esp (1 mM) and S. aureus
(105 c.f.u. ml21) were simultaneously added to the dishes filled with the test media,
and biofilms were collected from the dishes 16 h after the inoculation. The effect of
Esp on the stability of S. aureus biofilm was assessed by adding Esp to the culture
dishes 16 h after the S. aureus inoculation, and the biofilms were then collected in a
time-dependent manner. Absorbance at 600 nm was used as a quantitative mea-
surement of the biofilms. In addition, for microscopic evaluation, the S. aureus
biofilms were incubated with 1 mM Esp for 6 h, and the bacteria were collected and
subjected to Gram staining or microscopic observations.
Effect of culture supernatant and Esp on growth of S. aureus. S. aureus was
cultured in media containing the culture supernatant of S. epidermidis or Esp at
various concentrations with shaking at 37 uC for 16 h. The A600 of culture media
was used as a quantitative measurement of the bacteria.
Genetic evidence for the Esp activity on S. aureus biofilms. An esp-mutant
derivative of the S. epidermidis JK16 strain was constructed by an allelic exchange
method36,37. Briefly, the sequences flanking esp were PCR amplified with primers
Esp_attB_F1 (GGGGACAAGTTTGTACAAAAAAGCAGGCTAAAGCCAGCA
GATAGTGCAG) and Esp_R (CCTGATATAAATATTCAGTAATTAAATTTA
GTT) for the upstream region and primers Esp_F (ACATATAGATAAAAAT
CTCTTTTTCATATGATA) and Esp_attB_R2 (GGGGACCACTTTGTACAAG
AAAGCTGGGTCGCATCGGATTGTGGTT) for the downstream region. The
PCR fragments were then ligated in vitro and recombined with pKOR1. The
pKOR1 recombinant was then introduced into the S. epidermidis JK16 strain by
electroporation38. The successful deletion of the esp gene was verified by PCR
with primers Esp_F (TTTGGAGGTTATCATATGAAAAAGAG) and Esp_R
(CTGAATATTTATATCAGGTATATTGTTTC) and by the biofilm destruction
activity test. The esp-deficient mutant of JK16 was designated TI1.
In addition, to generate a complemented strain, a DNA fragment containing
the entire esp gene was obtained by PCR amplification with genomic DNA of
strain JK16 as a template and cloned downstream of the tetL gene of shuttle
vector pYT336,39,40. Briefly, the sequences flanking esp were PCR-amplified with
primers Esp1BamHI_F (TATGGATCCATATTTTTGGAGGTTATCATATGA
AAAAGAG) and Esp1HindIII_R (CCCAAGCTTTTAGTGATGGTGATGGT
GATGCTGAATATTTATATCAGGTATATTGTTTC). Esp1BamHI_F primer
includes putative Shine-Dalgarno sequence and Esp1HindIII_R primer
includes a His-Tag sequence. The resulting plasmid, pTI1, was introduced into
the esp-deficient mutant strain, TI1, by electroporation38.
The recombinant Esp protein. To obtain recombinant Esp protein with a His-
Tag, we cultured the esp-deficient mutant strain TI1 harbouring pTI1 in TSB
containing tetracycline (10 mg l21) for 16 h at 30 uC. The protein was collected
using the TALON purification kit (Clontech, Takara Bio). To evaluate the activity
of the recombinant protein, 1 mM of both the recombinant and native protein was
used in the study.
Western blotting. Polyclonal antisera against Esp were generated in rabbits
immunized with recombinant Esp protein at Operon Biotechnologies. For western
blotting, protein samples were denatured and separated on 15% SDS–PAGE as
described previously41. Briefly, blots were first incubated with antiserum against
Esp followed by the secondary horseradish-peroxidase-conjugated antibody, both
at 1:500 final dilution. Immunoreactive bands were visualized by ECL western
blotting detection reagents (GE Healthcare UK).
Interaction between Esp and innate host immune systems. The cooperative
effect of Esp and hBD2 (Peptide Institute and PeproTech) in killing S. aureus in
biofilms was evaluated. S. aureus biofilms were treated with 1 mM Esp with or
without hBD2 at various concentrations (1, 5 and 10 mM) in 10 mM sodium
phosphate buffer for 6 h42,43. Bacteria were then collected and washed with
physiological saline three times, and the cell viability of S. aureus was evaluated
with a conventional culture method using TSB agar plates.
Effect of S. epidermidis and Esp on S. aureus nasal colonization. To investigate
the effects of S. epidermidis and Esp on S. aureus nasal colonization, we intro-
duced inhibitory S. epidermidis and Esp in the subjects’ nasal cavities. Prior to
this study, persistent colonization of S. aureus without inhibitory S. epidermidis
was confirmed by at least three separate nasal cultures from the carriers for over
three months. The 19 test subjects were randomly divided into four groups: (1)
doi:10.1038/nature09074
seven subjects, including one MRSA carrier, who received daily inoculations of
the inhibitory S. epidermidis (wild type, JK16 strain) in the anterior nares using
cotton swabs, (2) six subjects who received the esp-deficient mutant bacteria
derived from the JK16 strain, (3) three subjects who received the non-inhibitory
S. epidermidis (JK11 strain), and (4) three subjects who received purified Esp.
The sample sizes in the groups (1) and (2) were determined by the results of a
pilot study according to the accepted methods for sample-size determina-
tion44,45, which gave a statistical power of 80% with an alpha value of 0.05. For
liquid formulations, the saline solutions (0.5 ml in each) containing approxi-
mately 1 3 109 c.f.u. of each bacteria46, 500 pmol Esp or none (as control group)
were administered into the nasal cavity with swabs (Eiken Kagaku).
The bacteria in the nasal cavities were collected daily with swabs by a clinical
microbiologist, and the number of S. aureus bacteria was counted by conven-
tional culture methods using mannitol salt agar with egg yolk for selection of S.
aureus. Briefly, nasal swabs were vigorously vortexed for 1 min in 500 ml physio-
logical saline, and samples were appropriately diluted. Then, the diluted samples
were spread using spreaders on mannitol salt agar with egg yolk. After 24 h
incubation, S. aureus colonies were counted. Colonies were identified as S. aureus
if a yellow colony exhibited positive egg-yolk reaction, and if the medium
surrounding the S. aureus colony turned yellow. Colonies were identified as
S. epidermidis if they appeared either pink or white, and if the medium around
the S. epidermidis colonies turned red or purple. We could not analyse samples of
3 of the 19 volunteers immediately. Therefore, the number of S. aureus bacteria
was quantified by real-time PCR47 instead of the conventional colony-counting
method to avoid the effects of a time delay until the analysis. The samples for
real-time PCR were obtained from nasal swabs kept in Tris-EDTA (TE) buffer.
Although both conventional culture and real-time PCR methods have been
applied to quantify bacteria46,47, we further checked their validity and consistency
by comparing the results of both methods (see Supplementary Fig. 6). Briefly,
nasal swabs obtained from the additional study (n 5 6) were vigorously vortex-
mixed in 500 ml physiological saline for 1 min, and then each sample was divided
into two groups; one for real-time PCR and another for the conventional culture
method. Estimation of c.f.u. values obtained by real-time PCR were calibrated by
a standard curve, which was constructed with c.f.u. values of the corresponding
samples counted by the conventional culture method, according to previous
studies32,33,46,47.
To confirm that the administered strains had colonized the participants’ nasal
cavities and to distinguish among the JK16 strain (wild type), the isogenic
mutant, and pre-colonizing S. epidermidis strains, we tested for biofilm destruc-
tion activity of the culture supernatants of these bacteria and performed PCR
using both primers designed in this study and reported previously48 for the esp
gene.
Before the study, we consulted several clinical microbiologists, bacteriologists,
immunologists, physicians and surgeons working in the field of infection con-
trol, and statisticians regarding the design of this experiment. After considering
their recommendations, we submitted our proposed experimental protocol to
the ethics committee of Kochi Medical School. We were very careful to conduct
the experiment according to the approved protocol after obtaining the ethics
committee’s approval. Nineteen healthy subjects affiliated with the university
were enrolled during March 2007–July 2009. Informed consent was obtained
from all participants.
Statistical analysis. Colonization by S. aureus was considered a dependent vari-
able. Odds ratios (ORs) and confidence intervals (CIs) were calculated, and
univariate logistic regression analysis was performed to assess the factors that
might affect colonization by S. aureus: age, sex, passive exposure to tobacco
smoke, active smoking, antibiotic treatment one month before beginning the
study, allergic symptoms, presence of inhibitory S. epidermidis, and S. epidermidis
bacterial count. To confirm the independence of these factors, multivariate
logistic regression analysis was performed using the stepwise selection method
at a significance level of 0.20. The log likelihood was used to assess goodness of
©2010 Macmillan Publishers Limited. All rights reserved
fit. In the in vitro studies evaluating the effects on formation and stability of the
S. aureus biofilms and on viability of the S. aureus cells, Student’s t test and
Dunnett’s test were used. In the in vivo studies evaluating the effects on the
S. aureus nasal colonization, Student’s t test was used for each sampling day
and the Kaplan–Meier method and log-rank test were used to compare the elimi-
nation effect on S. aureus nasal colonization by administration of the wild-type
strain, JK16, and the esp-deficient mutant strain, TI1. A value of P , 0.05 was
considered to indicate statistical significance. Calculations were performed using
the SAS statistical program, version 9.1 (SAS Institute) and Excel software
(Microsoft).
31. Carroll, K. C., Leonard, R. B., Newcomb-Gayman, P. L. & Hillyard, D. R. Rapid
detection of the staphylococcal mecA gene from BACTEC blood culture bottles by
the polymerase chain reaction. Am. J. Clin. Pathol. 106, 600–605 (1996).
32. Iwase, T., Seki, K., Shinji, H., Mizunoe, Y. & Masuda, S. Development of a real-time
PCR assay for the detection and identification of Staphylococcus capitis,
Staphylococcus haemolyticus and Staphylococcus warneri. J. Med. Microbiol. 56,
1346–1349 (2007).
33. Iwase, T. et al. Rapid identification and specific quantification of Staphylococcus
epidermidis by 59 nuclease real-time polymerase chain reaction with a minor
groove binder probe. Diagn. Microbiol. Infect. Dis. 60, 217–219 (2008).
34. Hiramatsu, K. et al. Dissemination in Japanese hospitals of strains of
Staphylococcus aureus heterogeneously resistant to vancomycin. Lancet 350,
1670–1673 (1997).
35. Ohara-Nemoto, Y. et al. Characterization and molecular cloning of a glutamyl
endopeptidase from Staphylococcus epidermidis. Microb. Pathog. 33, 33–41
(2002).
36. Cui, L., Lian, J. Q., Neoh, H. M., Reyes, E. & Hiramatsu, K. DNA microarray-based
identification of genes associated with glycopeptide resistance in Staphylococcus
aureus. Antimicrob. Agents Chemother. 49, 3404–3413 (2005).
37. Bae, T. & Schneewind, O. Allelic replacement in Staphylococcus aureus with
inducible counter-selection. Plasmid 55, 58–63 (2006).
38. Augustin, J. & Gotz, F. Transformation of Staphylococcus epidermidis and other
staphylococcal species with plasmid DNA by electroporation. FEMS Microbiol.
Lett. 66, 203–207 (1990).
39. Hanaki, H. et al. Activated cell-wall synthesis is associated with vancomycin
resistance in methicillin-resistant Staphylococcus aureus clinical strains Mu3 and
Mu50. J. Antimicrob. Chemother. 42, 199–209 (1998).
40. Neoh, H. M. et al. Mutated response regulator graR is responsible for phenotypic
conversion of Staphylococcus aureus from heterogeneous vancomycin-
intermediate resistance to vancomycin-intermediate resistance. Antimicrob.
Agents Chemother. 52, 45–53 (2008).
41. Shinji, H. et al. Lipopolysaccharide-induced biphasic inositol 1,4,5-trisphosphate
response and tyrosine phosphorylation of 140-kilodalton protein in mouse
peritoneal macrophages. J. Immunol. 158, 1370–1376 (1997).
42. Vuong, C. et al. Polysaccharide intercellular adhesin (PIA) protects Staphylococcus
epidermidis against major components of the human innate immune system. Cell.
Microbiol. 6, 269–275 (2004).
43. Midorikawa, K. et al. Staphylococcus aureus susceptibility to innate antimicrobial
peptides, beta-defensins and CAP18, expressed by human keratinocytes. Infect.
Immun. 71, 3730–3739 (2003).
44. Machin, D. C. M., Fayers, P. & Pinol, A. In Sample Size Tables for Clinical Studies 2nd
edn, chap. 3 (Blackwell Science, 1997).
45. Machin, D. C. M., Fayers, P. & Pinol, A. In Sample Size Tables for Clinical Studies 2nd
edn, chap. 9 (Blackwell Science, 1997).
46. Uehara, Y. et al. Bacterial interference among nasal inhabitants: eradication of
Staphylococcus aureus from nasal cavities by artificial implantation of
Corynebacterium sp. J. Hosp. Infect. 44, 127–133 (2000).
47. Paule, S. M., Pasquariello, A. C., Thomson, R. B. Jr, Kaul, K. L. & Peterson, L. R. Real-
time PCR can rapidly detect methicillin-susceptible and methicillin-resistant
Staphylococcus aureus directly from positive blood culture bottles. Am. J. Clin.
Pathol. 124, 404–407 (2005).
48. Ikeda, Y., Ohara-Nemoto, Y., Kimura, S., Ishibashi, K. & Kikuchi, K. PCR-based
identification of Staphylococcus epidermidis targeting gseA encoding the glutamic-
acid-specific protease. Can. J. Microbiol. 50, 493–498 (2004).
 Vol 465 | 20 May 2010 | doi:10.1038/nature08997

LETTERS
Effects of thymic selection of the T-cell repertoire on
HLA class I-associated control of HIV infection
Andrej Košmrlj1,2*, Elizabeth L. Read1,3,4*, Ying Qi5, Todd M. Allen1, Marcus Altfeld1, Steven G. Deeks6,
Florencia Pereyra1, Mary Carrington1,5, Bruce D. Walker1,7 & Arup K. Chakraborty1,3,4,8

Without therapy, most people infected with human immunodefi-
ciency virus (HIV) ultimately progress to AIDS. Rare individuals
(‘elite controllers’) maintain very low levels of HIV RNA without
therapy, thereby making disease progression and transmission
unlikely. Certain HLA class I alleles are markedly enriched in elite
controllers, with the highest association observed for HLA-B57
(ref. 1). Because HLA molecules present viral peptides that activate
CD81 T cells, an immune-mediated mechanism is probably
responsible for superior control of HIV. Here we describe how
the peptide-binding characteristics of HLA-B57 molecules affect
thymic development such that, compared to other HLA-restricted
T cells, a larger fraction of the naive repertoire of B57-restricted
clones recognizes a viral epitope, and these T cells are more cross-
reactive to mutants of targeted epitopes. Our calculations predict
that such a T-cell repertoire imposes strong immune pressure on
immunodominant HIV epitopes and emergent mutants, thereby
promoting efficient control of the virus. Supporting these predic-
tions, in a large cohort of HLA-typed individuals, our experiments
show that the relative ability of HLA-B alleles to control HIV
correlates with their peptide-binding characteristics that affect
thymic development. Our results provide a conceptual framework
that unifies diverse empirical observations, and have implications
for vaccination strategies.
HIV infection leads to acute high level viraemia, which is subse-
quently reduced to a set-point viral load. Without therapy, most
patients experience a subsequent increase in viral load, and ultimately
the development of AIDS. Viraemia levels and time to disease vary
widely, and the differences correlate with the expression of different
HLA class I molecules (reviewed in ref. 2). Effector CD81 T cells
(CTLs) are implicated in viral control because T-cell antigen recep-
tors (TCRs) on CD81 T cells recognize complexes of viral peptides
and class I HLA molecules presented on the surface of infected cells,
and depletion of CD81 T cells leads to increased viraemia in animal
models of HIV infection3. We describe a feature of the HLA-B57-
restricted CD81 T-cell repertoire that contributes to enhanced con-
trol of viraemia.
Algorithms4 based on experimental data predict whether a particular
peptide will bind to a given HLA molecule5. We tested four predictive
algorithms against available experimental data on peptide binding to
diverse HLA molecules and found that, in most cases, they are highly
accurate (Supplementary Fig. 1 and Supplementary Table 1). For
example, predictions using the best algorithm for HLA-B*5701 were
97% accurate. Using these algorithms, we computed the fraction of
peptides derived from the human proteome6 that bind to various HLA
1
Ragon Institute of MGH, MIT and Harvard, Boston, Massachusetts 02114, USA. 2Department of Physics, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139,
USA. 3Department of Chemical Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. 4Department of Chemistry, Massachusetts Institute of
Technology, Cambridge, Massachusetts 02139, USA. 5Cancer and Inflammation Program, Laboratory of Experimental Immunology, SAIC-Frederick, Inc., NCI-Frederick, Frederick,
Maryland 21702, USA. 6University of California, San Francisco, California 94110, USA. 7Howard Hughes Medical Institute, Chevy Chase, Maryland 20815, USA. 8Department of
Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA.
*These authors contributed equally to this work.
molecules. Of the ,107 unique peptide sequences, only 70,000 are
predicted to bind to HLA-B*5701, and 180,000 bind to HLA-
B*0701 (an allele that is not protective against HIV). Essentially iden-
tical results were obtained for randomly generated peptides (data not
shown). The protective allele in macaques, Mamu-B*17, also binds
fewer self peptides than other Mamu molecules for which data are
available (Mamu-B*17 binds 4, 6 and 13 times fewer self peptides than
Mamu-A*11, Mamu-A*01 and Mamu-A*02, respectively; Sup-
plementary Table 1).
The intrinsic differences in self-peptide binding among HLA mole-
cules are important during T-cell repertoire development. Immature
T cells are exposed to diverse host-derived peptide–HLA complexes
presented in the thymus. As fewer self peptides are able to bind to
HLA-B*5701 (and Mamu-B*17) molecules, a smaller diversity of self-
peptide TCR contact sequences will be encountered by HLA-B*5701/
Mamu-B*17-restricted T cells in the thymus (Supplementary
Discussion 1).
The diversity of self peptides presented in the thymus shapes the
characteristics of the mature T-cell repertoire. Experiments7,8 and
theoretical studies9,10 show that T cells that develop in mice with only
one type of peptide in the thymus are more cross-reactive to point
mutants of peptide epitopes that they recognize than T cells from
mice that express diverse self peptides. Thus, by encountering fewer
self peptides during thymic development, HLA-B57-restricted CD81
T cells should be more cross-reactive to point mutants of targeted
viral peptides.
We carried out in silico thymic selection experiments to test this
hypothesis. We chose an HLA-dependent number of thymic self pep-
tides, each with amino acids of the TCR contact residues picked
according to the frequency with which they appear in the human
proteome6,9. A diverse set of immature CD81 T cells (thymocytes)
was generated by choosing the sequences of their peptide contact
residues in the same way, and by varying the TCR–HLA interactions.
A thymocyte emerges from the thymus as a mature CD81 T cell if its
TCR binds to at least one self-peptide–major histocompatibility com-
plex (pMHC; human MHC is called HLA) molecule with an affinity
that exceeds the positive selection threshold, and does not interact
with any pMHC more strongly than the negative selection threshold.
Using a computational model9,10 in the class of ‘string models’11, we
assessed the affinity of TCR–self-peptide–HLA complexes (Methods)
to determine which T cells survive positive and negative selection, and
become a part of the mature repertoire. Our qualitative results are
independent of the parameters used to determine these interaction
strengths (Supplementary Figs 2 and 3)9,10.

350
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
The mature T cells that emerged from these in silico thymic selec-
tion experiments were then computationally challenged by a viral
peptide (that is, not seen in the thymus) bound to the same HLA
type. T cells that recognize this peptide–HLA complex were obtained
by assessing whether the interaction strength exceeded the negative
selection threshold (shown to be equal to the recognition threshold in
mouse models12); qualitative results are invariant if the recognition
threshold is not much weaker than that corresponding to negative
selection (Supplementary Fig. 3). Cross-reactivity of these T cells was
then determined in silico by mutating each TCR contact residue of the
peptide to the other 19 possibilities. Sites on the viral peptide were
called ‘important contacts’ if half the mutations therein abrogated
recognition by T cells that target this epitope. The frequency of the
number of important contacts in viral peptides that determine T-cell
recognition was obtained by repeating this procedure 1,000 times
with different choices of thymocytes and self and foreign peptides.
Our calculations predict that a T-cell repertoire restricted by an
HLA molecule such as HLA-B*5701, which presents fewer self pep-
tides in the thymus, has a higher frequency of occurrence of T cells
that recognize viral peptides through smaller numbers of important
contacts (Fig. 1a). In contrast, the frequency of occurrence of T cells
a
small number of important contacts
10–2
Number of self peptides
Frequency of occurrence of
encountered in the thymus
10–3 5,000
2,500
1,000
10–4
10–5
10–6
10–7
0 1
Number of important contacts for T-cell recognition of viral peptide
b HLA molecules. Thus, more HLA-B*5701-restricted T cell clones are
TCR–viral-peptide interaction strengths
Probability of occurrence of different
100 Number of self peptides
encountered in the thymus
5,000
2,500
1,000
10–2
10–4
10–6
0 5 10
Magnitude of TCR−viral-peptide interaction strength
Figure 1 | Thymic selection against fewer self peptides leads to a more
cross-reactive T-cell repertoire. a, Histogram of the frequency with which T
cells recognize viral peptides (that is, not seen in the thymus) through only a
small number (0, 1, 2, 3) of important contacts is shown for three T-cell
repertoires that developed with different numbers of self-peptide–HLA
complexes in the thymus. Important contacts were determined by making
single point mutations. If the TCR–peptide–HLA interaction is sufficiently
strong, no single point mutation can abrogate recognition, resulting in zero
important contacts. A higher frequency of occurrence of a small number of
important contacts indicates a more cross-reactive T-cell repertoire because
only mutations at these contacts are likely to abrogate recognition. The
frequency with which T cells recognize viral peptides through many
significant contacts (greater than four) is larger for T-cell repertoires
restricted by HLA alleles that present more self peptides in the thymus (not
shown). b, The probability that a TCR binds to viral peptides with a certain
interaction strength is shown for three T-cell repertoires (as in a). A
particular TCR recognizes a viral peptide when the binding strength exceeds
the recognition threshold (dotted black line). Members of a T-cell repertoire
selected against fewer self peptides are more likely to recognize a viral
peptide. The model we used describes qualitative trends robustly9,10
(Methods), but is not meant to be quantitatively accurate.
LETTERS
that recognize viral peptides through many important contacts is
larger for repertoires restricted by HLA alleles that present a greater
diversity of self peptides in the thymus (data not shown for .four
contacts). Mutations at sites different from the important contacts do
not affect binding strength substantially. Therefore, when the inter-
action between peptide–HLA and TCR is mediated by fewer import-
ant contacts, a larger number of possible point mutations of the
peptide do not affect peptide recognition, thereby making the T cells
more cross-reactive to mutants that arise. Thus, the HLA-B57-
restricted T-cell repertoire is expected to be more cross-reactive to
mutants of targeted viral peptides than repertoires restricted by HLA
alleles that present a greater diversity of self peptides.
Our computational models give this qualitative mechanistic
insight, but do not provide quantitative estimates of the extent of
this enhanced cross-reactivity of T cells. However, compelling experi-
mental data13 has shown that the effect revealed by our studies is
important in humans. Peripheral blood mononuclear cells from
patients expressing HLA-B57 contained CTLs that were more
cross-reactive to various HIV epitopes and their point mutants than
those of HLA-B8-positive patients. HLA-B8 is associated with rapid
progression to disease13, and the most accurate algorithm for peptide
binding suggests that the HLA-B8 molecule binds a greater diversity
of self peptides than HLA-B57 (Supplementary Fig. 4 and
Supplementary Table 1). Other experimental studies also show that
patients expressing HLA-B57 cross-recognize point mutants of the
dominant epitope and use more public TCRs14,15.
Next, we computed interaction strengths between diverse viral
peptides and members of T-cell repertoires restricted by HLA mole-
cules that present differing numbers of self peptides in the thymus.
This allowed us to obtain the probability with which a randomly
picked T-cell clone and viral peptide will interact sufficiently strongly
for recognition to occur. The results (Fig. 1b) indicate that a typical
CD81 T cell restricted by an HLA molecule such as HLA-B*5701,
2 3 which presents fewer peptides in the thymus, has a higher probability
of recognizing a viral epitope compared to a T cell restricted by other
likely to recognize a viral epitope, making effective precursor fre-
Recognition quencies higher in an HLA-B*5701-restricted repertoire (a strong
predictor of response magnitude16). A greater precursor frequency
for viral epitopes in the naive repertoire restricted by HLA-B57 is
indicated by experimental results showing that HLA-B*5701 contri-
butes the most to acute-phase CTL responses of all HLA alleles
tested17.
The results in Fig. 1 stem from the constraint that thymocytes must
avoid being negatively selected by each self-peptide–HLA complex
encountered during development in the thymus. T cells expressing
15 20 25 TCRs with peptide contact residues composed of amino acids that
interact strongly with other amino acids (for example, charged resi-
dues, flexible side chains) have a high probability of binding to a self
peptide strongly. The greater the diversity of self peptides presented
in the thymus, the higher the chance that a TCR with such peptide
contact residues will encounter a self peptide with which strong
interactions will result in negative selection. Thus, as the diversity
of self peptides presented in the thymus increases, the peptide contact
residues of TCRs in the mature T-cell repertoire are increasingly
enriched in weakly interacting amino acids (Supplementary Fig. 5).
T cells bearing TCRs with weakly interacting peptide contact residues
recognize viral peptides by means of several moderate interactions,
making many contacts important for recognition. In contrast, TCRs
with peptide contact residues containing strongly interacting amino
acids are more likely to recognize viral peptides through a few
important contacts mediated by these residues, making recognition
cross-reactive to mutations at other peptide sites. These mechanistic
insights are supported by experimental results7,9 (Supplementary
Discussion 2).
By studying a model of host–pathogen dynamics that builds on past
models of host–HIV interactions18–20, we explored the consequences
351
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

of the HLA-B57-restricted CD81 T-cell repertoire having a higher
precursor frequency for viral peptides and being more cross-reactive
to point mutants of targeted epitopes on the control of HIV. Because
of the importance of immune control exerted by CD81 T cells17,21, we
focused on the interaction between a mutating virus quasispecies and
epitope-directed, variably cross-reactive, host CTL responses.
The essential features of the model are depicted in Fig. 2a (details in
Methods). The virus is modelled as a number of epitopes consisting of
strings of amino acids, and new viral strains (point mutations of
epitopes), which differ in replicative fitness, arise over the course of
infection. For each individual, an HLA-dependent CD81 T-cell re-
pertoire was chosen. To mimic the results obtained from our thymic
selection calculations (Fig. 1b), more or less cross-reactive repertoires
were chosen (Supplementary Fig. 6) to represent HLA-B57-restricted
T cells and those restricted by other HLAs, respectively. Infection rates
were limited by target CD41 T cells, and CTL contraction and
memory were included. Other dynamic models were studied, includ-
ing one that does not incorporate target cell limitation or CTL con-
traction. Our qualitative results about the effects of cross-reactivity
are robust to variations in parameters and model assumptions
(Supplementary Figs 7–16).
a km
mutation
Vm
pMHC
Vn presentation
ks
We find that individuals with a more cross-reactive CTL repertoire
control viral loads better during the acute phase of the infection
(Fig. 2b). This is in agreement with findings in simian immunodefi-
ciency virus (SIV)-infected rhesus macaques22, where the number of
cross-reactive TCR clones negatively correlates with viral load. Our
simulations show that a larger number of CTL clones in a more cross-
reactive T-cell repertoire recognize epitopes from the infecting viral
strain (Fig. 2c). This is because the predicted higher precursor fre-
quency for viral epitopes (Fig. 1b) leads to a greater response mag-
nitude (as in mouse models16). This conclusion is supported by data
showing that in people with a protective HLA allele, the initial T-cell
response to HIV is dominated by T cells restricted by the protective
HLA and not those restricted by other HLAs expressed17. Our simu-
lations also show that enhanced cross-reactivity of the T-cell re-
pertoire leads to greater immune pressure on the emergent viral
mutants by individuals expressing HLA-B57 compared to those with
T cells restricted by HLA molecules that bind more types of self
peptides. The stronger immune pressure on infecting and emerging
viral strains results in superior control of viral load. Thus, we predict
that HIV-infected individuals with HLA alleles that bind fewer self
peptides are more likely to control viral loads to low values.
To test this prediction, we studied two large HLA-typed cohorts:
1,110 controllers with less than 2,000 HIV particles ml21 and 628
ko
Ti0...D–1 kp progressors (or non-controllers) with viral loads exceeding 104 ml21
CD8+ expansion
pMHC off (Methods). From these data, we obtained the odds ratio (OR) for
individual HLA alleles. People with HLA alleles associated with an
Effector
C

OR value greater or less than one are more likely to be progressors

pMH
j
Pn,

kv CD8+ T cell kdt

Virus
replication Ti* cell death or controllers, respectively. We focused on HLA-B alleles because they
kk
kc Infected kill are associated with control of HIV23. Of 40 HLA-B alleles that were
clearance kt CD4+ T cell
infection kd ′ studied, significant results (P value ,0.05) were obtained for five
In cell death
activation km
ka memory
HLA-B alleles (Supplementary Table 2) and peptide-binding data
Naive
are available for four of them. In support of our predictions, those
ks′ pMHC presentation
Infected on APC P APC
CD8+ T cell HLA-B alleles associated with higher OR values also bind more self
C

n, j
Ti
peptides (Fig. 3).
pMH

CD4+ T cell kra

It reactivation
Superior control of viral load due to the greater precursor frequency
kb kd APC Memory and cross-reactivity of those T-cell repertoires restricted by HLA
cell birth cell death
kdm CD8+ T cell molecules that bind to few self peptides (for example, HLA-B57)
cell death Mi
ko′
pMHC off Figure 2 | Model of host–pathogen interactions shows superior viral
control by cross-reactive CD81 T-cell repertoires. a, Dynamic model: the
b virus mutates, infects limited-target CD41 T cells, and is cleared. Infected
High CR CD41 T cells produce more free virus and die. Infected cells present viral
Viral load (per ml plasma)

106 Low CR peptides in complex with HLA molecules (until peptides unbind from HLA).
Activated CD81 T cells produced by recognition of viral epitopes on antigen-
presenting cells (APCs) proliferate and differentiate into effector CTLs.
105 CTLs kill infected cells bearing cognate peptide–HLA complexes, and turn
into memory cells that are activated after re-exposure to antigen.
b, Simulated HIV viral loads versus time for different cross-reactivities (CR)
104 of the CD81 T-cell repertoire. Black curve, high cross-reactivity; red curve,
low cross-reactivity. Each curve is averaged over 500 simulations (each
simulation represents a person). The model shows a reduced set-point viral
103 load for people with a more cross-reactive T-cell repertoire. Other models of
0 50 100 150 200
Time (days) host–pathogen dynamics show similar effects of T-cell cross-reactivity
(Supplementary Figs 7 and 8). c, Virus diversity and immune pressure for
c High CR Low CR representative people (that is, representative simulations) with high cross-
100 100 reactivity (left) and low cross-reactivity (right) of CD81 T-cell repertoires.
(% contribution)

Top panels show the relative population sizes of two dominant viral strains:
Viral strain

Infecting
strain the infecting strain (black), and an emerging, less fit strain (green) (other less
50 50
populous viral strains are not shown). For people with a more cross-reactive
Mutant
T-cell repertoire, the emergent mutant strain only begins to dominate the
0 0 infecting strain after 175 days, whereas for low cross-reactivity the mutant
0 100 200 0 100 200
increases to nearly 100% of the viral population within 100 days after
0.2 1 infection. Bottom panels show the relative immune pressure, defined as the
Relative immune

rate of killing of an infected cell (see equation (4), Methods), imposed on

pressure

each viral strain by different CD81 T-cell clones. Each curve represents the
0.1 0.5 relative immune pressure exerted on that viral strain by a particular T-cell
clone. For people with a more cross reactive T-cell repertoire, several T-cell
clones exert immune pressure on both the infecting and emergent strains.
0 0
0 100 200 0 100 200 For people with a low-cross-reactivity T-cell repertoire, the emergent strain
Time (days) Time (days) is not recognized, and thus escapes.
352
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
OR (progressors to controllers)
HLA-B*0702 HLA-B*3501
100
HLA-B*2705
HLA-B*5701
10–1
0 0.005 0.010 0.015 0.020 0.025
Fraction of self peptides that can bind to HLA molecules
Figure 3 | HLA-B alleles associated with greater ability to control HIV
correlate with smaller self-peptide binding propensities. The odds ratio
pw =pwo
(OR) for an allele is defined as: , where pw and pwo are the numbers of
cw =cwo
individuals in the progressor cohort with and without this HLA, respectively;
and cw and cwo are the numbers of individuals in the controller cohort with
and without this HLA, respectively. This definition suggests that the OR
measures the likelihood of an allele being correlated with progressors versus
controllers, with an OR greater than one indicating association with the
progressor cohort. The fraction of peptides derived from the human
proteome that bind to a given HLA allele was determined using the most
accurate predictive algorithms (Methods and Supplementary Table 1).
Compared to experimental data, the predictive algorithms for peptide
binding by HLA-B*3501 are less accurate than algorithms for the other three
alleles (Supplementary Fig. 17 and Supplementary Table 1); the number
reported here for HLA-B*3501 using the most accurate algorithm
underestimates the binding fraction. The error bars represent the 95%
confidence intervals for OR. The dotted line corresponds to equal odds for
an allele being associated with progressors and controllers.
should also confer protection against diseases caused by other fast-
mutating viruses. Indeed, HLA-B57 is protective against hepatitis C
virus (HCV)24, another highly mutable viral disease in which CD81 T
cells are important. Also, HLA-B8, which binds a greater diversity of
self peptides, is associated with faster disease progression in HCV25
and HIV13. Thus, the correlation between the diversity of peptides
presented in the thymus during T-cell development and control or
progression of disease may be general.
Undoubtedly, many complex factors influence the relationship
between HLA type and disease outcome. The effect of the new factor
we have identified should be greatest for HLA molecules that bind
relatively few (for example, HLA-B57) or many (for example, HLA-
B7, -B35, -B8) self peptides. The strong association of HLA-B27—
which binds an intermediate number of self peptides (twice as many as
HLA-B57)—with viral control indicates that, in this case, the effects of
T cell cross-reactivity are reinforced by this molecule binding HIV
epitopes that are subject to very strong structural constraints.
Our results also point to a mechanistic explanation for as yet
unexplained associations between HLA alleles that confer protection
against HIV and autoimmune diseases. T cells restricted by HLA
alleles that bind to few self peptides are subject to less stringent nega-
tive selection in the thymus, and should therefore be more prone to
recognizing self peptides. Indeed, HLA-B57 has been associated with
autoimmune psoriasis26 and hypersensitivity reactions27. Enhanced
cross-reactivity of HLA-B27-restricted T cells and other unique prop-
erties of this molecule (misfolding, homodimers28) probably contrib-
ute to the enhanced risk of autoimmunity associated with this allele29.
Our results shed light on another intriguing observation; acutely
infected patients with low viral loads (and protective HLAs) tend to
target an immunodominant epitope that makes a larger relative con-
tribution to the total CTL response as compared to individuals pre-
senting with higher levels of viraemia30. This is counterintuitive as the
most protective responses appear most focused, rather than broadly
distributed over many epitopes. We calculated how viral load corre-
lates with the number of CTLs responding to the immunodominant
epitope divided by the total number of CTLs activated by the virus (a
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
quantity analogous to relative contribution30). Mirroring experi-
mental data, HLA alleles that restrict a more cross-reactive repertoire
and are more protective also make a larger relative contribution
(Supplementary Fig. 13). This result unifies the idea of both a broad
and a focused response. The more cross-reactive repertoire targets
more epitopes and emergent mutants, but a larger number of clones
also recognize the dominant epitope (Fig. 2c).
Cross-reactive T cells are rare in people with HLA alleles that
present more self peptides in the thymus than the B57 allele, but they
do exist. Our results suggest that a T-cell vaccine for a diverse popu-
0.030 lation must aim to activate these rare cross-reactive T cells that also
target epitopes from a conserved region of the HIV genome (like
HLA-B57 Gag epitopes). This will enable robust responses to infect-
ing and mutant strains until a strain with low replicative fitness
emerges, enhancing control of viral load.
METHODS SUMMARY
Predictive algorithm tools for peptide binding to HLA and Mamu molecules
were obtained from the Immune Epitope Database (IEDB)4 and were used to
predict the fraction of bound peptide derived from the human and macaque
proteomes6. Accuracies of these tools were tested on experimental data obtained
from the IEDB4. To assess the effects of thymic selection on TCRs restricted by
different MHC molecules (HLA or Mamu), we used a computational model of
thymic selection described in Methods (and previously9,10).
To explore host–pathogen dynamics, we constructed a small model of the HIV
virus with distinct epitopes and sequence diversity, based in part on past
work18–20. We carried out numerical simulations of ordinary differential equa-
tion models, shown schematically in Fig. 2a and Supplementary Fig. 7.
Parameters and their justification are given in Supplementary Tables 3 and 4
and in the Supplementary Methods. To explore cross-reactivity, we varied the
distribution of pairwise-interaction free energies of TCR–pMHC contacts. Our
goal was not to obtain precise numbers, but to examine the qualitative effects of
variation in repertoire cross-reactivity on virus control. Qualitative results are
robust to variations in parameters and assumptions (Supplementary Figs 8–16).
HLA-typed cohorts of people of diverse races were divided into HIV control-
lers and HIV non-controllers, and analysed for HLA association with the ability
to control HIV. The results (Fig. 3 and Supplementary Table 2) were adjusted for
the effects of HLA-B*0702, HLA-B*3501, HLA-B*2705 and HLA-B*5701.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 13 October 2009; accepted 11 March 2010.
Published online 5 May 2010.
1. Migueles, S. A. et al. HLA B*5701 is highly associated with restriction of virus
replication in a subgroup of HIV-infected long term nonprogressors. Proc. Natl
Acad. Sci. USA 97, 2709–2714 (2000).
2. Deeks, S. G. & Walker, B. D. Human immunodeficiency virus controllers:
mechanisms of durable virus control in the absence of antiretroviral therapy.
Immunity 27, 406–416 (2007).
3. Jin, X. et al. Dramatic rise in plasma viremia after CD81 T cell depletion in simian
immunodeficiency virus-infected macaques. J. Exp. Med. 189, 991–998 (1999).
4. Peters, B. et al. The immune epitope database and analysis resource: from vision
to blueprint. PLoS Biol. 3, e91 (2005).
5. Rao, X., Fontaine Costa, A. I. C. A., van Baarle, D. & Keşmir, C. A comparative study
of HLA binding affinity and ligand diversity: implications for generating
immunodominant CD81 T cell responses. J. Immunol. 182, 1526–1532 (2009).
6. Hubbard, T. J. et al. Ensembl 2009. Nucleic Acids Res. 37, D690–D697 (2009).
7. Huseby, E. S., Crawford, F., White, J., Marrack, P. & Kappler, J. W. Interface-
disrupting amino acids establish specificity between T cell receptors and
complexes of major histocompatibility complex and peptide. Nature Immunol. 7,
1191–1199 (2006).
8. Huseby, E. S. et al. How the T cell repertoire becomes peptide and MHC specific.
Cell 122, 247–260 (2005).
9. Košmrlj, A., Jha, A. K., Huseby, E. S., Kardar, M. & Chakraborty, A. K. How the
thymus designs antigen-specific and self-tolerant T cell receptor sequences. Proc.
Natl Acad. Sci. USA 105, 16671–16676 (2008).
10. Košmrlj, A., Chakraborty, A. K., Kardar, M. & Shakhnovich, E. I. Thymic selection of
T-cell receptors as an extreme value problem. Phys. Rev. Lett. 103, 068103
(2009).
11. Chao, D. L., Davenport, M. P., Forrest, S. & Perelson, A. S. The effects of thymic
selection on the range of T cell cross-reactivity. Eur. J. Immunol. 35, 3452–3459
(2005).
12. Naeher, D. et al. A constant affinity threshold for T cell tolerance. J. Exp. Med. 204,
2553–2559 (2007).
353
LETTERS
13. Turnbull, E. L. et al. HIV-1 epitope-specific CD81 T cell responses strongly
associated with delayed disease progression cross-recognize epitope variants
efficiently. J. Immunol. 176, 6130–6146 (2006).
14. Gillespie, G. M. et al. Cross-reactive cytotoxic T lymphocytes against a HIV-1 p24
epitope in slow progressors with B*57. AIDS 16, 961–972 (2002).
15. Yu, X. G. et al. Mutually exclusive T-cell receptor induction and differential
susceptibility to human immunodeficiency virus type 1 mutational escape
associated with a two-amino-acid difference between HLA class I subtypes. J.
Virol. 81, 1619–1631 (2007).
16. Moon, J. J. et al. Naive CD41 T cell frequency varies for different epitopes and
predicts repertoire diversity and response magnitude. Immunity 27, 203–213
(2007).
17. Altfeld, M. et al. HLA alleles associated with delayed progression to AIDS
contribute strongly to the initial CD81 T cell response against HIV-1. PLoS Med. 3,
e403 (2006).
18. Althaus, C. L. & De Boer, R. J. Dynamics of immune escape during HIV/SIV
infection. PLoS Comput. Biol. 4, e1000103 (2008).
19. Nowak, M. A. et al. Antigenic oscillations and shifting immunodominance in HIV-1
infections. Nature 375, 606–611 (1995).
20. Wodarz, D. & Thomsen, A. R. Effect of the CTL proliferation program on virus
dynamics. Int. Immunol. 17, 1269–1276 (2005).
21. Cao, J. H., McNevin, J., Malhotra, U. & McElrath, M. J. Evolution of CD81 T cell
immunity and viral escape following acute HIV-1 infection. J. Immunol. 171,
3837–3846 (2003).
22. Price, D. A. et al. Public clonotype usage identifies protective Gag-specific CD81 T
cell responses in SIV infection. J. Exp. Med. 206, 923–936 (2009).
23. Kiepiela, P. et al. Dominant influence of HLA-B in mediating the potential co-
evolution of HIV and HLA. Nature 432, 769–775 (2004).
24. Thio, C. L. et al. HLA-Cw*04 and hepatitis C virus persistence. J. Virol. 76,
4792–4797 (2002).
25. McKiernan, S. M. et al. Distinct MHC class I and II alleles are associated with
hepatitis C viral clearance, originating from a single source. Hepatology 40,
108–114 (2004).
26. Bhalerao, J. & Bowcock, A. M. The genetics of psoriasis: a complex disorder of the
skin and immune system. Hum. Mol. Genet. 7, 1537–1545 (1998).
354
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
27. Chessman, D. et al. Human leukocyte antigen class I-restricted activation of
CD81 T cells provides the immunogenetic basis of a systemic drug
hypersensitivity. Immunity 28, 822–832 (2008).
28. López de Castro, J. A. HLA-B27 and the pathogenesis of spondyloarthropathies.
Immunol. Lett. 108, 27–33 (2007).
29. Bowness, P. HLA B27 in health and disease: a double-edged sword? Rheumatology
41, 857–868 (2002).
30. Streeck, H. et al. Human immunodeficiency virus type 1-specific CD81 T-cell
responses during primary infection are major determinants of the viral set point
and loss of CD41 T cells. J. Virol. 83, 7641–7648 (2009).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements Financial support was provided by the Mark and Lisa Schwartz
Foundation, the National Institutes of Health (NIH) Director’s Pioneer award
(A.K.C.), Philip T and Susan M Ragon Foundation, Jane Coffin Childs Foundation
(E.L.R.), the Bill and Melinda Gates Foundation, and the NIAID (B.D.W., T.M.A. and
M.A.). This project has been funded in whole or in part with federal funds from the
National Cancer Institute, NIH, under contract no. HHSN261200800001E. The
content of this publication does not necessarily reflect the views or policies of the
Department of Health and Human Services, nor does mention of trade names,
commercial products, or organizations imply endorsement by the US Government.
This research was supported in part by the Intramural Research Program of the NIH,
National Cancer Institute, Center for Cancer Research.
Author Contributions A.K. and E.L.R. contributed equally to this work. A.K.C. and
B.D.W. initiated the project. A.K., E.L.R. and A.K.C. developed the computational
models. A.K., E.L.R., A.K.C. and B.D.W. analysed computational results. Y.Q., F.P.,
M.C., S.G.D. and B.D.W. collected and analysed the data from cohorts of
HIV-infected people. A.K., E.L.R., T.M.A., M.A., M.C., B.D.W. and A.K.C.
contributed to the writing of the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to A.K.C.
(arupc@mit.edu) or B.D.W. (bwalker@partners.org).
doi:10.1038/nature08997

METHODS dVn X
~kvn In {kc Vn zkm ðVm {Vn Þ ð2Þ
HLA–peptide binding predictions. There are at present several HLA–peptide dt n:m
binding prediction methods. The performance of these algorithms to identify
new epitopes has recently been benchmarked against experimental data31. In dI t X
general, artificial neural networks (ANN)32 and the stabilized matrix method ~kb {kd I t {kt I t Vn ð3Þ
dt n
(SMM)33 were found to be superior to other methods31. We used ANN and the
SMM (versions 2009-09-01 and 2007-12-27) prediction tools provided by the dIn XX
IEDB4. Accuracy of prediction tools was tested against experimental data down- ~kt Vn I t {kd 0 In { si,j kk Pn,j Ti
ð4Þ
dt i j
loaded from the IEDB in September 2009 (Supplementary Fig. 1, Supplementary
Table 1 and Supplementary Notes 1). These experimental data were obtained by
two methods: competition assays, in which purified MHC and radioactive label- dPn,j dI ðkillÞ Pn,j
~ks In {ko Pn,j { n ð5Þ
ling are used; and association studies, in which purified MHC and fluorescence dt dt In
labelling are used. Data obtained from the two methods show significant corre-
APC
lations of measured binding affinities (as measured by half-maximum inhibitory dPn,j
~ks 0 In {ko 0 Pn,j
APC
ð6Þ
concentration (IC50) and half-maximum effective concentration (EC50))5. dt
Prediction tools were tested against experimental data for accuracy of classifying X
peptides into binders (IC50 , 500 nM) and non-binders (IC50 $ 500 nM); the dTi APC
~{k a Ti si,j Pn,j ð7Þ
chosen thresholds are commonly accepted values5. We also tested how well these dt n,j
tools predict absolute measured affinity values, not just classification of binders
and non-binders, which is dependent on the chosen thresholds. The accuracy of dTi0 X X
~{kp Ti0 zka Ti APC
si,j Pn,j zkra Mi APC
si,j Pn,j ð8Þ
the prediction tools thus determined are summarized in Supplementary Table 1 dt n,j n,j
and Supplementary Fig. 1. We excluded all HLA and Mamu alleles for which
there was not enough experimental data (at least 50 binders and 50 non-binders) dTim
or prediction tools were not sufficiently accurate (Supplementary Notes 1). For ~2kp Ti(m{1) {kp Tim ð9Þ
dt
each HLA and Mamu allele, the most accurate prediction tool was used to predict
the fraction of unique peptides derived from the human and macaque proteome dTi

(Homo_sapiens.GRCh37.55.pep.all.fa and Macaca_mulatta.MMUL_1.56.pep. ~2kp Ti(D{1) {kdt Ti
{km Ti
ð10Þ
dt
all.fa obtained from Ensembl6) that can bind to that allele. We focused only
on the binding abilities of peptides of 9 amino acids to HLA molecules, because dMi X
~km Ti
{kdm Mi {kra Mi APC
si,j Pn,j ð11Þ
there is not enough experimental data available for the binding affinities of dt n,j
peptides of 8, 10 and 11 amino acids to HLA-B*5701 and the other relevant
HLA-B alleles that emerged from our analyses (HLA-B*2705, HLA-B*0702 and Target CD41 T cells, I t , are infected by free virus particles, where Vn denotes
HLA-B*3501). virions of strain n. In denotes CD41 T cells infected by virus of strain n, Pn,j is a
Thymic selection model and antigen recognition. The TCR contact residues of pMHC complex of peptide j derived from viral strain n, displayed on the surface
peptides and the peptide contact residues of TCRs are represented as strings of
APC
of the infected cell, Pn,j is a pMHC displayed by APCs and Ti is a naive CD81 T
sites of length N. One-million sequences of TCR peptide contact residues were cell of clonotype i. Activated T cells undergo D rounds of cell division before
subject to development in a thymus containing M self peptides with TCR contact becoming effector CTLs; Ti0 is an activated CD81 T cell of type i that has not yet
residues generated according to their frequency of occurrence in the human begun dividing and Tim are the dividing cells, where m runs from 1 to D 2 1.
Effector CTLs, Ti
, differentiate into memory CD81 T cells, Mi , which are acti-
proteome. A particular TCR with the sequence of peptide contact residues ~ t
vated upon re-exposure to pMHC.
successfully matures in the thymus if it avoids negative selection with all self
T-cell clone i recognizes pMHC j, si,j is 1, and 0 otherwise. In equation
If X
peptides (Eint . En) and is positively selected by at least one self peptide
(2), denotes the sum over viral strains m that are Hamming distance 1 away
(Eint , Ep). Interaction free energy between sequences of TCR and peptide con-
n:m
tacts, ~
t and~s, is, respectively: from strain n. That is, only point mutations are allowed. The third term of
equation (5) ensures that if an infected cell is killed, the pMHC bound on its
X
N
t ,~s Þ~Ec z
Eint ð~ J ðti ,si Þ ð1Þ dI ðkillÞ
surface must also disappear; n denotes the third term of equation (4),
i~1 dt
which describes killing of an infected cell by CTLs that recognize pMHC on its
where Ec represents an interaction between a TCR and an HLA molecule, and J is surface. Simulations were performed using ode45 and ode15s solvers in
an empirically determined statistical potential between interacting amino acids MATLAB. A further dynamic model, which does not incorporate target cell
on a TCR and a peptide. Antigenic peptides are recognized by a mature TCR if limitation and allows unlimited expansion of activated CTLs, was also developed
binding is stronger than the threshold for recognition (Eint , Er). The statistical to show robustness of our results to model assumptions. It is discussed in the
potentials do not necessarily provide quantitatively accurate values of the inter- Supplementary Information (Supplementary Figs 7–12).
action free energies. However, theoretical analyses and computational results9,10 Rate constants used in the models are given in Supplementary Tables 3–4, and
show that the following qualitative result is true regardless of the choice of the are in keeping with values reported in the literature. We assume a concentration
statistical potentials: the smaller the diversity of self peptides presented in the of 106 CD41 T cells per ml blood before infection, with 1% of these cells activated
thymus, the greater the cross-reactivity of the mature T-cell repertoire that and thus initial targets for HIV infection35,36. The initial conditions of infection
develops therein. More details of the model and the insensitivity of our results in the simulations were one infected CD41 T cell per ml of plasma and a naive-
to parameter variations (for example, qualitative results do not depend on the CD81 repertoire size of one cell per ml of each clonotype. We assume that the
choice of J or Ec (as long as Ec is not too small or large)) are described in number of epitopes, length of each epitope, and number of amino acids (L, M, N)
Supplementary Information (Supplementary Figs 2 and 3) and elsewhere9,10. are all 2, giving 8 pMHC types and 16 possible viral strains. The number of CD81
The parameters used for the results in the main text are: N 5 5; En 2 Ec 5 221 clonotypes was chosen to be 20.
kBT; Ep 2 En 5 2.5 kBT; Er 5 En and Miyazawa–Jernigan statistical potential J34. The interplay between antigen and immune receptor diversity is captured in this
Numbers of self peptides presented in the thymus, M, were varied to represent model through variability in si,j and viral fitness. Different fitness levels for dif-
different HLA alleles. ferent strains of the virus are modelled by randomly selecting kvn , the virus proli-
Host–pathogen interaction dynamics. We constructed a small model of HIV feration rate, for each strain from a uniform distribution between 0 and 2,000 per
with distinct epitopes and sequence diversity, based in part on models developed day18,37, with the assumption that the infecting strain has the maximum fitness. The
previously18,19. The virus is modelled as displaying L epitopes, each consisting of matrix si,j encodes the ability of T cells to recognize pMHCs. We generate si,j in
M amino acid residues that may be of N types. Different viral strains arise such a way as to mimic the results of the thymic selection model (Fig. 1b), to
through point mutations at the amino-acid sites, giving (NM)L distinct strains. investigate the effects of those predictions on host–pathogen dynamics. That is, we
The number of different pMHC types is L 3 NM, because peptide sequences at assume that T-cell repertoires restricted by different HLA types differ in the inter-
epitope positions 1…L are considered to be distinct. The system of ordinary action free energies of their TCR–pMHC contacts, and generate si,j accordingly
differential equations corresponding to the model in Fig. 2 and based on pre- using a type of random-energy-like model (Supplementary X Fig. 6). The interaction
vious work20 is as follows: free energy between a T cell and an epitope is given by J ði,ja Þ, where J ði,ja Þ is
a

©2010 Macmillan Publishers Limited. All rights reserved

doi:10.1038/nature08997
the interaction free energy between T cell of clonotype i and residue a on epitope j.
Similar to the models used for thymic selection, the total interaction free energy is
taken to be the sum of the individual residue interactions and recognition is said to
occur when it exceeds a recognition threshold (in the dynamic model, T-cell
sequences are not specified explicitly). J ði,ja Þ is a random variable chosen from
a uniform distribution, and the width of the distribution determines the probabil-
ity that the summed interaction energy falls above the threshold, and thus the
probability that a peptide is recognized by a given T cell. Repertoires generated in
this way approach a Gaussian distribution of interaction energies, and the distri-
bution shifts and thus cross-reactivity increases when the uniform distribution
from which J (i,ja ) is selected is wider. Generating si,j in this way allows us to
describe variable cross-reactivities of the T-cell repertoire (both intra- and inter-
epitope), and also accounts for correlated interaction energies and thus recog-
nition probabilities of similar peptide sequences.
HLA-allele association with ability to control HIV. SAS 9.1 (SAS Institute) was
used for data management and statistical analyses. Odds ratios and 95% confid-
ence intervals were determined using PROC LOGISTIC in a comparison of HIV
controllers (those individuals who maintained viral loads of less than 2,000
copies of the virus per ml plasma on three determinations over at least a year
of follow-up and, on average, for approximately 15 years38) to HIV non-con-
trollers (those individuals whose viral loads exceeded 10,000 copies of the virus
per ml plasma). To eliminate the confounding effects of B*0702, B*3501, B*2705
and B*5701, alleles strongly associated with progression or control, these factors
were used as covariates in the logistic regression model for the analysis of all other
HLA class I types39. All ethnic groups were included in the analyses shown
(European, African-American and others) and we adjusted for ethnicity in the
©2010 Macmillan Publishers Limited. All rights reserved
logistical regression model. All P values were corrected for multiple tests using
the Bonferroni correction, a stringent and commonly used approach for mul-
tiple comparisons40.
31. Peters, B. et al. A community resource benchmarking predictions of peptide
binding to MHC-I molecules. PLoS Comput. Biol. 2, e65 (2006).
32. Gulukota, K., Sidney, J., Sette, A. & DeLisi, C. Two complementary methods for
predicting peptides binding major histocompatibility complex molecules. J. Mol.
Biol. 267, 1258–1267 (1997).
33. Peters, B., Tong, W., Sidney, J., Sette, A. & Weng, Z. Examining the independent
binding assumption for binding of peptide epitopes to MHC-I molecules.
Bioinformatics 19, 1765–1772 (2003).
34. Miyazawa, S. & Jernigan, R. L. Residue-residue potentials with a favorable contact
pair term and an unfavorable high packing density term, for simulation and
threading. J. Mol. Biol. 256, 623–644 (1996).
35. Sachsenberg, N. et al. Turnover of CD41 and CD81 T lymphocytes in HIV-1
infection as measured by Ki-67 antigen. J. Exp. Med. 187, 1295–1303 (1998).
36. Stafford, M. A. et al. Modeling plasma virus concentration during primary HIV
infection. J. Theor. Biol. 203, 285–301 (2000).
37. Parera, M., Fernandez, G., Clotet, B. & Martinez, M. A. HIV-1 protease catalytic
efficiency effects caused by random single amino acid substitutions. Mol. Biol.
Evol. 24, 382–387 (2007).
38. Pereyra, F. et al. Genetic and immunologic heterogeneity among persons who
control HIV infection in the absence of therapy. J. Infect. Dis. 197, 563–571 (2008).
39. Hosmer, D. W., Jovanovic, B. & Lemeshow, S. Best subsets logistic-regression.
Biometrics 45, 1265–1270 (1989).
40. Cheverud, J. M. A simple correction for multiple comparisons in interval mapping
genome scans. Heredity 87, 52–58 (2001).
Vol 465 | 20 May 2010 | doi:10.1038/nature08970

LETTERS
Modulation of Shigella virulence in response to
available oxygen in vivo
Benoit Marteyn1, Nicholas P. West1{, Douglas F. Browning2, Jeffery A. Cole2, Jonathan G. Shaw3, Fredrik Palm4,
Joelle Mounier5, Marie-Christine Prévost6, Philippe Sansonetti5,7 & Christoph M. Tang1

Bacteria coordinate expression of virulence determinants in res- higher than in an aerobic cabinet (P , 0.05, Fig. 1e). This was not
ponse to localized microenvironments in their hosts. Here we caused by cell death or mediated by the host transcription factor
show that Shigella flexneri, which causes dysentery, encounters
varying oxygen concentrations in the gastrointestinal tract, which
a b
govern activity of its type three secretion system (T3SS). The T3SS
is essential for cell invasion and virulence1. In anaerobic environ-
ments (for example, the gastrointestinal tract lumen), Shigella is
primed for invasion and expresses extended T3SS needles while
reducing Ipa (invasion plasmid antigen) effector secretion. This is
mediated by FNR (fumarate and nitrate reduction), a regulator of
anaerobic metabolism that represses transcription of spa32 and
spa33, virulence genes that regulate secretion through the T3SS.
M90T M90TΔfnr
We demonstrate there is a zone of relative oxygenation adjacent
to the gastrointestinal tract mucosa, caused by diffusion from c d
the capillary network at the tips of villi. This would reverse the
anaerobic block of Ipa secretion, allowing T3SS activation at its
precise site of action, enhancing invasion and virulence.
Shigella virulence depends on its ability to enter epithelial cells by
delivering Ipa effectors via its T3SS (which acts as a molecular syringe)
into the host cell cytoplasm1. Secretion through T3SSs is highly regu-
lated. Initially, T3SS needle components are secreted until it reaches a
pre-defined length2,3. In inducing conditions, a switch then occurs M90TΔmxiD M90TΔfnr pBM2
allowing Ipa secretion through needles, mediating bacterial entry4.
Using the rabbit ligated intestinal loop model5,6, we identified a e * *
colonisation-defective S. flexneri M90T mutant with a transposon in 105
fnr, which encodes a regulator of anaerobic metabolism (Supplemen-
tary Fig. 1). FNR is only active as a dimer containing a [4Fe–4S] cluster.
c.f.u. ml–1

The cluster dissociates in the presence of oxygen (O2), destabilizing 104

the dimer, with loss of FNR activity7. M90TDfnr was substantially
attenuated for colonisation compared with the wild-type strain,
M90T (competitive index (C.I.) 5 0.05), and this was restored by
103
complementation (C.I. of M90TDfnr pBM2 5 0.6). Furthermore, we + – + – + – O2 in cabinet
challenged intestinal loops with strains individually. Infection with M90T M90TΔfnr M90TΔfnr
M90T led to abscesses, and shortening and destruction of villi pBM2
(Fig. 1a). These alterations were significantly reduced following infec- Figure 1 | Influence of anaerobiosis and FNR on Shigella invasion.
tion with M90TDfnr (P , 0.01, Fig. 1b and Supplementary Table 1) Histological analysis of the gastrointestinal tract infected with S. flexneri
and absent after challenge with a T3SS- mutant, M90TDmxiD (Fig. 1c). strains as indicated. a, There was marked destruction of the mucosal surface
Restoration of the invasive phenotype following complementation with loss of villi and infiltration of the submucosa with inflammatory cells
(M90TDfnr pBM2) confirmed the requirement of FNR in vivo following challenge with the wild-type strain, M90T. b, c, These changes are
(Fig. 1d). less pronounced after challenge with M90TDfnr (b) or a strain lacking the
T3SS (M90TDmxiD, c). d, Complementation of fnr (M90TDfnr pBM2)
To define the contribution of O2 to Shigella virulence, epithelial cells restores the pathological changes. Sections were stained with Evans blue;
were propagated in an aerobic environment, and either left there or bars, 50 mm. e, Epithelial cell invasion by M90T, M90TDfnr and M90TDfnr
transferred to an anaerobic cabinet for 30 min and then challenged pBM2 in an aerobic or anaerobic cabinet (n 5 5 independent experiments).
with bacteria grown in aerobic or anaerobic conditions, respectively. c.f.u., colony-forming units. Error bars show the s.d. and asterisks indicate
Bacterial entry into cells in an anaerobic cabinet was significantly P , 0.05 (Student’s t-test).
1
Centre for Molecular Microbiology and Infection, Department of Microbiology, Flowers Building, Imperial College London, London SW7 2AZ, UK. 2School of Biosciences, The
University of Birmingham, Edgbaston, Birmingham B15 2TT, UK. 3Division of Genomic Medicine, University of Sheffield Medical School, Beech Hill Road, Sheffield S10 2RX, UK.
4
Department of Medical Cell Biology, Biomedical Center, Uppsala University, 751 23 Uppsala, Sweden. 5Unité de Pathogénie Microbienne Moléculaire, and Unité INSERM786, Institut
Pasteur, 28 rue du Dr Roux, F-75724 Paris Cédex 15, France. 6Plateforme de Microscopie électronique, Institut Pasteur, 25 Rue du Docteur Roux, F-75724 Paris cedex 15, France.
7
Collège de France, 11 Place Marcelin Berthelot, F-75231, Paris Cédex 05, France. {Present address: Centenary Institute, Newtown, New South Wales 2042, Australia.
355
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

HIF-1a (refs 8, 9; Supplementary Fig. 2). Instead, the enhanced cell a M90T Δfnr Δfnr pBM2
entry of Shigella in the absence of O2 was dependent on FNR (Fig. 1e). O2 + + – + – + –
To understand the basis of the increased cell entry, we investigated CR – + + + + + +
T3SS structure and function of Shigella grown with or without O2.
IpaB
With O2, M90T and M90TDfnr secreted the effectors IpaB, IpaC and
IpaD following exposure to the inducer of secretion Congo red4. Congo Sec. IpaC
red-induced secretion of Ipas by M90T was reduced in anaerobic
conditions and was accompanied by increased intra-bacterial Ipa levels IpaD
(P , 0.01, Fig. 2a and Supplementary Fig. 3). The reduced effector
secretion in the absence of O2 was not detected with M90TDfnr IpaB
(Fig. 2a), whereas complementation of the fnr mutant restored low- IpaC
level Ipa secretion in anaerobiosis as observed with M90T. Addi- Cell
tionally, scanning electron microscopy demonstrated that the number IpaD
of visible needle tips was significantly higher on bacteria after growth in RecA
anaerobic compared with aerobic conditions (Supplementary Fig. 4).
This was not associated with any detectable change in the lipopolysac- b
charide profile, which affects the presentation of T3SS needles5 10

Percentage of needles
M90T +O2 M90T +O2
(Supplementary Fig. 5). Instead, transmission electron microscopy 4 Loess model
revealed a 25% increase in the average needle length during growth 48 ±11 nm
2
in anaerobic compared with aerobic conditions (62 versus 48 nm, 5
respectively; P , 0.001, Fig. 2b), with less rigorous control of needle 0
0 100 200
length in the absence of O2 (s.d. of needle length 28 and 11 nm in
anaerobic and aerobic environments, respectively). In contrast, needles 0
0 50 100 150 200 250
produced by M90TDfnr were similar regardless of the presence of O2 10

Percentage of needles
M90T –O2 M90T –O2
(average length 53 and 58 nm with and without O2, respectively, 4
Fig. 2c). Therefore in the absence of O2, FNR mediates reduced Ipa
secretion and elongation of T3SS needles. 2 62 ±28 nm
5
As FNR is a transcription regulator, we next examined mRNA 0
levels of mxi/spa pathogenicity island genes and their regulators by 0 100 200

quantitative PCR with reverse transcription (qRT–PCR, ref. 10)

0
(Fig. 3a). There was no significant alteration in mRNA levels of genes 0 50 100 150 200 250
encoding effectors, T3SS components, or regulators in the presence Needle length (nm)
c
M90TΔfnr +O2
or absence of O2. However, there was a reduction in spa32 and spa33 10
Percentage of needles
M90TΔfnr +O2
mRNA levels (4.7- and 5.5-fold reduction, respectively; P , 0.01)
during anaerobic growth that was FNR-dependent (Fig. 3a). Spa32 4
is required for control of needle length and the selection of substrates 5
53 ±18 nm
for secretion2,3, whereas dysregulation of Spa33 expression blocks Ipa 2
secretion11. The activity of lacZ reporter fusions confirmed that the 0
reduction of spa32 and spa33 mRNA levels is associated with 0
0 100 200

decreased transcription in an FNR-dependent fashion (Supplemen- M90TΔfnr –O2 0 50 100 150 200 250
10
Percentage of needles

M90TΔfnr –O2
tary Fig. 6). The O2 regulation of spa32 and spa33 fusions was 4

abolished by modifying the two predicted FNR binding sites12 in 2 58 ±15 nm

the promoters of spa32 and spa33 (2156 and 267, and 2205 and
2116 upstream of the initiation codons, respectively; Supplementary
Fig. 6), while FNR binds sequences upstream of spa32 and spa33
(Fig. 3b). Binding upstream of spa33 was abolished by modification
of the FNR binding sites (Supplementary Fig. 7). These data demon-
strate that FNR represses the Shigella virulence genes spa32 and spa33
in anaerobic conditions.
To establish whether FNR-mediated repression of spa32 and spa33
contributes to increased Shigella entry into cells in an anaerobic
cabinet, we constructed M90T containing spa32 and spa33 under
the control of either their native promoters (M90Tp32/33) or pro-
moters with disrupted FNR boxes (M90Tp#32/#33). The presence in
M90T of spa32 and spa33 alleles that are not repressed by FNR led to
IpaB secretion during anaerobic growth (Fig. 3c), and loss of the
increased entry of Shigella into cells in an anaerobic cabinet (invasion
of strain with p#32/#33 versus p32/33, P , 0.01, Fig. 3d); similar
findings were obtained by inappropriate expression of spa32 and
spa33 in anaerobic conditions with isopropyl-b-D-thiogalactoside
(IPTG; Supplementary Fig. 7).
To explain the enhanced entry of Shigella into epithelial cells fol-
lowing transfer into anaerobic cabinets, we proposed that bacteria
encounter O2 in the vicinity of host cells that promotes T3SS activa-
tion. We used microelectrodes to measure concentrations of O2
directly above cells after they had been moved into an anaerobic
cabinet13. There was sufficient O2 (approximately 0.4%) 10 mm above
356
©2010 Macmillan Publishers Limited. All rights reserved
5
0
0 100 200
0
0 50 100 150 200 250
Needle length (nm)
Figure 2 | T3SS structure and function are modified by ambient O2.
a, Secreted (Sec.) and cytoplasmic (Cell) Ipas detected by western analysis.
Secretion induced by Congo red (CR) was diminished in anaerobic
conditions, with effectors retained in the bacteria; this is FNR-dependent.
Strains and growth conditions are indicated. b, c, Detection of T3SS needles
(white triangles) on M90T (b) and M90TDfnr (c) by transmission electron
microscopy. The average needle length (.200 needles counted in n 5 3
experiments, 6 s.d.) on strains is shown and results displayed applying a
Loess model. Needles on M90T were significantly longer following growth in
anaerobic than aerobic conditions (P , 0.001, Student’s t-test), and the
needle length was more varied (s.d. 6 28 versus 11 nm).
cells to inactivate FNR14 and allow IpaB secretion by S. flexneri (Fig. 4a
and b). Whereas no O2 was detected above cells following addition of
the O2 scavenger deoxy-haemoglobin, O2 levels were unaffected by
the addition of cytochrome c, which contains haem but does not bind
O2 (ref. 15). These data provide an explanation for the hyper-
invasion of Shigella following transfer of cells to an anaerobic cham-
ber, and indicate that remote from the cells, bacteria would assume
a ‘primed state’ (with extended T3SS needles and impaired Ipa
NATURE | Vol 465 | 20 May 2010 LETTERS

a rs nt b spa32 spa33 a b

Regulators
to ne
100
ec po 4
eff m Control Percentage O2 0 0.4 1 20 20
Fold expression compared

co
SS

Oxygen tension
+Haemoglobin
SS as FNR–DNA
T3 T3 Sp complexes
3 +Cytochrome c Sec.

(mm Hg)
to M90T in O2

10
M90T –O2 2
M90TΔfnr –O2
Free Cell
DNA 1
1
FNR 0 0.5 1 2 3 4 0 0.5 1 2 Grown – O2 Grown + O2
(μM)
** ** M90T M90T 0
d After 30 min After 60 min
M90T p32/33 p#32/#33 c
0.1 106
** 15 min unclamped 30 min unclamped 15 min clamped 30 min clamped
ni
rB aB paC paD xiH a47 a13 a32 a33 a24 virB virF
ip i i m sp sp sp sp sp
c

c.f.u. ml–1
Percentage O2 0 5 10 20
105
M90T

IpaB
Figure 3 | Regulation of Shigella virulence genes by O2 and FNR.
a, qRT–PCR analysis of relative mRNA levels in wild-type M90T in the
presence (dashed line) and absence (black boxes) of O2; the reduction of
spa32 and spa33 mRNA levels under anaerobic conditions are FNR
dependent (n 5 3 independent experiments). Grey boxes, mRNA levels in
M90TDfnr. b, Binding of FNR to regions upstream of spa32 and spa33 by
electrophoretic mobility shift assay. c, IpaB secretion detected by western
blot analysis 30 min after transferring anaerobically grown bacteria to
different O2 concentrations (indicated). Strains including M90T and M90T
containing spa32 and spa33 under the control of their native promoters
(M90Tp32/33) or promoters with disrupted FNR boxes (M90Tp#32/#33).
d, Epithelial cells entry of bacteria in cabinets with (white bars) or without
(black bars) O2. Error bars, s.d.; ** indicates P , 0.01; n 5 3 independent
experiments.
secretion), while residual O2 in proximity to cells would allow T3SS
activation and enhanced entry.
To establish the relevance of these findings to Shigella invasion in
vivo, we constructed two bacterial reporters for the presence of O2
during host–pathogen interactions as introduction of a microelec-
trode into the gastrointestinal lumen might disrupt O2 homeostasis.
Formation of the green fluorescent protein (GFP) fluorophore
requires covalent bonding with O2 (refs 16, 17), and luciferase activity
is dependent on O2 and ATP18. M90TDmxiD pFPV25.1 expresses GFP
with O2-dependent fluorescence16,17, whereas Lux-based lumin-
escence of Escherichia coli pXen5 provides a distinct reporter for the
presence of O2 (Supplementary Fig. 8 and 9); E. coli was used as the
host because this construct is not functional in S. flexneri. The lumin-
escence of anaerobically grown E. coli pXen5 was measured in intest-
inal loops with an intact vascular supply or in loops in which the
vessels had been occluded by clamping before challenge. Bacteria
emitted significant luminescence in intestinal loops with an intact
vascular supply within 15 min, whereas barely any signal was detected
from loops that had had their afferent vessels occluded (P , 0.001,
Fig. 4c and Supplementary Fig. 10). To determine the location of
intra-luminal O2 within the gastrointestinal tract, intestinal loops
were infected with M90TDmxiD pFVP25.1, and tissue recovered
18 h later. Even though bacteria were present throughout the gut
lumen (detected with an anti-lipopolysaccharide polyclonal anti-
body), only Shigella located within approximately 70 mm of the epi-
thelial surface (particularly at the villous tips) emitted significant
fluorescence (Fig. 4d). This zone of relative oxygenation is provided
by the vascular supply to the gastrointestinal tract as the GFP signal
is abolished on clamping the afferent vasculature (Supplementary
Fig. 11).
The status of the Shigella T3SS is precisely modulated by ambient
O2 tensions that vary at specific sites in the gastrointestinal tract. In
the anaerobic lumen of the intestine, T3SS needles will be extended,
but not competent for secretion. In contrast, the relatively aerobic
zone adjacent to the mucosal surface would allow activation of the
T3SS on bacteria at this site (Fig. 4e). This zone of oxygenation
depends on the vascular supply, probably by diffusion of O2 from
the capillary network at villous tips19. Expression of elongated needles
would optimise the presentation of the T3SS translocon for engaging
M90Tp32/33 d e
Gastrointestinal lumen
M90Tp#32/#33 104
O2 in cabinet + – + – + –
O2–/GFP–
Primed Shigella No FNR
LPS/actin - long needles - short needles
- no secretion - secretion
Lumen
70 μm
O2+/GFP+
Villi Secretion
GFP Invasive Shigella
Host
Merge/DAPI epithelium
Figure 4 | The presence of O2 encountered by Shigella in vitro and in vivo
a, HeLa cells were transferred to an anaerobic cabinet with or without deoxy-
haemoglobin or cytochrome c (6 mM) in the medium, and oxygen tensions
measured 10 mm above cells at times afterwards (n 5 3 independent
experiments, error bars, s.d.). b, M90T grown anaerobically or aerobically
was transferred to different O2 concentrations with Congo red. Secreted
(Sec.) and bacterial (Cell) IpaB was detected 30 min later. c, Luminescence of
E. coli pXen5 after introduction into loops (arrowed) with unclamped or
clamped vascular supply. d, Loops were challenged with M90TDmxiD
pFPV25.1, and tissue obtained 18 h later; host cell nuclei stained blue, and
bacteria and actin stained red. GFP-positive bacteria were detected in a zone
within 70 mm of villi (arrowed). e, Shigella is ‘primed’ within the anaerobic
lumen of the gastrointestinal tract; O2 from capillaries creates a permissive
environment for T3SS activation.
host cells20. Furthermore precise regulation of secretion of effectors
would allow bacteria to deliver maximal amounts of Ipas at the exact
time and place in vivo needed to trigger efficient manipulation of the
cytoskeleton and cell entry21.
The presence of available O2 in localized regions in the intestine
may be under-appreciated and could have important biological con-
sequences. This ‘aerobic zone’ might dictate the outcome of interac-
tions with the extensive intestinal microbiota, acting as an innate
immune barrier to protect the mucosal surface from anaerobic bac-
teria, while being recognized as a signal to promote invasion by
pathogens such as Shigella, Salmonella and Listeria monocytogenes22.
FNR and possibly other regulators (such as the ArcBA system) may
mediate other effects aside from bacterial invasion, especially in the
context of the inflammatory response to infection. Similar mechan-
isms of modulating virulence in response to O2 may also have evolved
in other intestinal pathogens, as FNR boxes are present upstream of
genes required for T3SS function in Yersinia spp., and serovars of
Salmonella enteritidis (Supplementary Table 2).
METHODS SUMMARY
The bacterial strains and plasmids used in this study are shown in Supplementary
Table 3. S. flexneri was grown in trypticase soy (TCS) broth or on TCS agar plates
supplemented with 0.01% Congo red (Sigma), when necessary. The mutant with
a transposon inserted in fnr was isolated by signature tagged mutagenesis as
357
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
described previously5. Complementation was performed by amplifying fnr and
its own promoter with primers NG825 and NG826, then ligating the product
into pBBR1MCS-4 to generate pBM2 (Supplementary Table 3). The sequence of
all oligonucleotide primers is given in Supplementary Table 4.
Invasion assays were performed with HeLa cells and human Rcc10 1/2 von
Hippel-Lindau protein (VHL; provided by P. Maxwell) seeded into 12-well
tissue culture plates (Life Technologies), and grown until the cells were semi-
confluent. S. flexneri strains were diluted from overnight liquid cultures and
grown in fresh Luria–Bertani medium at 37 uC with agitation until the attenu-
ance reached a D600 of 0.2; cells were infected at a multiplicity of infection of 100.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 16 September 2009; accepted 2 March 2010.
Published online 2 May 2010.
1. Phalipon, A. & Sansonetti, P. J. Shigella’s ways of manipulating the host intestinal
innate and adaptive immune system: a tool box for survival? Immunol. Cell Biol. 85,
119–129 (2007).
2. Magdalena, J. et al. Spa32 regulates a switch in substrate specificity of the type III
secreton of Shigella flexneri from needle components to Ipa proteins. J. Bacteriol.
184, 3433–3441 (2002).
3. Tamano, K., Katayama, E., Toyotome, T. & Sasakawa, C. Shigella Spa32 is an
essential secretory protein for functional type III secretion machinery and
uniformity of its needle length. J. Bacteriol. 184, 1244–1252 (2002).
4. Bahrani, F. K., Sansonetti, P. J. & Parsot, C. Secretion of Ipa proteins by Shigella
flexneri: inducer molecules and kinetics of activation. Infect. Immun. 65,
4005–4010 (1997).
5. West, N. P. et al. Optimization of virulence functions through glucosylation of
Shigella LPS. Science 307, 1313–1317 (2005).
6. Sansonetti, P. J. et al. Alterations in the pathogenicity of Escherichia coli K-12 after
transfer of plasmid and chromosomal genes from Shigella flexneri. Infect. Immun.
39, 1392–1402 (1983).
7. Kiley, P. J. & Beinert, H. The role of Fe–S proteins in sensing and regulation in
bacteria. Curr. Opin. Microbiol. 6, 181–185 (2003).
8. Maxwell, P. H. et al. The tumour suppressor protein VHL targets hypoxia-
inducible factors for oxygen-dependent proteolysis. Nature 399, 271–275 (1999).
9. Esteban, M. A. et al. Regulation of E-cadherin expression by VHL and hypoxia-
inducible factor. Cancer Res. 66, 3567–3575 (2006).
10. Tobe, T. et al. Temperature-regulated expression of invasion genes in Shigella
flexneri is controlled through the transcriptional activation of the virB gene on the
large plasmid. Mol. Microbiol. 5, 887–893 (1991).
11. Schuch, R. & Maurelli, A. T. Spa33, a cell surface-associated subunit of the Mxi-
Spa type III secretory pathway of Shigella flexneri, regulates Ipa protein traffic.
Infect. Immun. 69, 2180–2189 (2001).
12. Hoitink, C. W., Woudt, L. P., Turenhout, J. C., van de Kamp, M. & Canters, G. W.
Isolation and sequencing of the Alcaligenes denitrificans azurin-encoding gene:
comparison with the genes encoding blue copper proteins from Pseudomonas
aeruginosa and Alcaligenes faecalis. Gene 90, 15–20 (1990).
358
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
13. Melican, K. et al. Bacterial infection-mediated mucosal signalling induces local
renal ischaemia as a defence against sepsis. Cell. Microbiol. 10, 1987–1998
(2008).
14. Becker, S., Holighaus, G., Gabrielczyk, T. & Unden, G. O2 as the regulatory signal
for FNR-dependent gene regulation in Escherichia coli. J. Bacteriol. 178, 4515–4521
(1996).
15. Marti, M. A. et al. Dioxygen affinity in heme proteins investigated by computer
simulation. J. Inorg. Biochem. 100, 761–770 (2006).
16. Hansen, M. C., Palmer, R. J. Jr, Udsen, C., White, D. C. & Molin, S. Assessment of
GFP fluorescence in cells of Streptococcus gordonii under conditions of low pH and
low oxygen concentration. Microbiology 147, 1383–1391 (2001).
17. Takahashi, E. et al. In vivo oxygen imaging using green fluorescent protein. Am. J.
Physiol. Cell Physiol. 291, C781–C787 (2006).
18. Ghisla, S., Hastings, J. W., Favaudon, V. & Lhoste, J.-M. Structure of the oxygen
adduct intermediate in the bacterial luciferase reaction: 13C nuclear magnetic
resonance determination. Proc. Natl Acad. Sci. USA 75, 5860–5863 (1978).
19. Torres Filho, I. P., Leunig, M., Yuan, F., Intaglietta, M. & Jain, R. K. Non-invasive
measurement of microvascular and interstitial oxygen profiles in a human tumor
in SCID mice. Proc. Natl Acad. Sci. USA 91, 2081–2085 (1994).
20. Tamano, K. et al. Supramolecular structure of the Shigella type III secretion
machinery: the needle part is changeable in length and essential for delivery of
effectors. EMBO J. 19, 3876–3887 (2000).
21. Jaumouillé, V., Francetic, O., Sansonetti, P. J. & Tran Van Nhieu, G. Cytoplasmic
targeting of IpaC to the bacterial pole directs polar type III secretion in Shigella.
EMBO J. 27, 447–457 (2008).
22. Pentecost, M., Otto, G., Theriot, J. A. & Amieva, M. R. Listeria monocytogenes
invades the epithelial junctions at sites of cell extrusion. PLoS Pathog. 2, e3
(2006).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements This work was supported by the Fondation pour la Recherche
Médicale, the Royal Society, an ERC Advanced Grant (HOMEOEPITH), and the
European Union (QLRT-1999-00938). Work in CMT’s laboratory is supported by
the Wellcome Trust and The Medical Research Council, and PJS is a Howard
Hughes Medical Institute scholar. Imaging was performed at the PFID station
(Institut Pasteur). We are grateful to J. Green for the anti-FNR pAbs and advice.
R. Exley, D. Holden and K. Ray provided suggestions and reviewed the manuscript;
we thank M.-A. Nicola for technical help.
Author Contributions B.M., N.P.W. and J.M. performed experiments with Shigella.
D.F.B. and J.A.C. provided technical support and advice for electrophoretic mobility
shift assays and analysis of the fusions. J.G.S. analysed the lipopolysaccharide,
M.-C.P. carried out EM, and F.P. performed the oxygen measurements. C.M.T. and
P.S. provided advice and overall direction, and wrote the paper.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to C.M.T.
(c.tang@imperial.ac.uk) and P.S. (psanson@pasteur.fr).
doi:10.1038/nature08970
METHODS
Bacterial strains, growth conditions and lipopolysaccharide analysis. All
strains and plasmids in this study are described in Supplementary Table 2.
E. coli was grown in Luria–Bertani broth whereas Shigella was propagated in
trypticase soy medium (TCS; Oxoid) or on TCS plates with Congo red (0.01%,
Sigma). Experiments under different O2 concentrations were performed in an
anaerobic cabinet (Whitley DG250), or in a cabinet in which O2 tensions can be
controlled (Plaslabs). Antibiotics were added at the following concentrations:
chloramphenicol, 50 mg ml21; ampicillin, 100 mg ml21; kanamycin 50 mg ml21.
For lipopolysaccharide analysis, whole-cell lysates of bacteria grown in liquid
culture to a D600 of ,0.2 were prepared in lipopolysaccharide loading buffer,
treated with proteinase K, then subjected to tricine SDS–PAGE analysis23, and
visualized by silver staining (Pierce).
DNA manipulations. The regulatory regions of spa32 and spa33 (from 300 bp
upstream of the start codon to 15 bp downstream) were amplified by PCR with
NG900/901 and NG902/903, respectively, and ligated into the SmaI and BamHI
sites of pRS415 (ref. 24, Supplementary Fig. 6). All oligonucleotides are shown in
Supplementary Table 4. For transcriptional fusions, the regulatory regions of
spa32, spa33 and P6 (located upstream of spa47, ref. 25 and Supplementary Fig.
6) were amplified with primers SG87/88, SG81/89 and SG94/26, and ligated in
pQF50.1 which contains a promoterless lacZ as a reporter (gift from C. Parsot26).
Putative FNR binding sites upstream of spa32 and spa33 were modified by a two-
step PCR procedure27. spa32 FNRbox1 (TTGAAGAAATTCAA, inverted repeats
in bold) and spa32 FNRbox2 (TTGATTAAAAGAAA) were altered to
TCGAGGAAATACAG and CTCATTAAACGGAA using primers SG25/26 and
SG27/28, respectively (for details see Supplementary Fig. 6). spa33 FNRbox1
(TTGATAGCAGTCAA) and spa33 FNRbox2 (TTGACGCTAACGAA) were
altered to TCGATAGCACAGCA and TCCACGCTAAGGAA with primers
SG29/30 and SG31/32, respectively. To generate p32/33, the open reading frames
and promoters of spa32 and spa33 were amplified with SG13 and SG14, and
ligated into pBBR1MCS-4. The FNR boxes in this plasmid were modified as
above to generate p#32/#33. For IPTG-inducible overexpression, spa32 and
spa33 were amplified with NG1390/1391 and NG1369/1370, and ligated into
the NdeI/BamHI sites of pET21b and the NdeI/XhoI sites of pET28b
(Invitrogen), respectively. Constructs were verified by DNA sequencing.
Antibodies and fluorescent dyes. Murine anti-Shigella lipopolysaccharide poly-
clonal antibodies28 were diluted 1:500 for immunofluorescence microscopy.
FITC (fluorescein isothiocyanate)-conjugated goat anti-rabbit and phalloidin-
rhodamine red X (RRX)-conjugated donkey anti-mouse antibodies (Jackson
Immunoresearch Antibodies) were used at a final dilution of 1:500. DAPI
(49,6-diamidino-2-phenylindole; Invitrogen) was used at 1 mg ml21. Rabbit
anti-FNR polyclonal antibodies (gift from J. Green), anti-RecA (MBL), anti-
IpaB, anti-IpaC and anti-IpaD29 were used at a 1:10,000 final dilution.
Horseradish peroxidase-conjugated anti-His monoclonal antibody (Qiagen)
was used at a final dilution of 1:2,000.
Cell culture. HeLa epithelial cells were grown to semi-confluency in aerobic
conditions in 24-well tissue culture plates. Before challenge, cells were either
maintained under aerobic conditions or transferred to an anaerobic cabinet
(Whitley DG250) for 30 min and the media replaced with anaerobically condi-
tioned media. Cells were challenged at a multiplicity of infection of 100 with
bacteria grown in liquid culture under aerobic or anaerobic conditions. For
aerobic growth, bacteria were grown in air overnight in 5 ml Luria–Bertani with
shaking in a 37 uC incubator. Cultures were then diluted 1 in 100 into 5 ml fresh
medium in a 20 ml universal and then grown with vigorous shaking until the D600
reached 0.2. For anaerobic growth, bacteria were propagated in exactly the same
way except cultures were performed in an anaerobic cabinet in media which had
been pre-incubated in the cabinet for at least 16 h. Bacteria were spun onto cells by
centrifugation at 2,000g for 10 min. For adhesion assays, cells were immediately
washed three times in PBS, then fixed in 3% paraformaldehyde (PFA) for 15 min.
To measure bacterial entry, cells were challenged as above, incubated for
30 min at 37 uC, after which gentamicin (50 mg ml21) was added for 1 h to kill
extracellular bacteria. Cells were washed with PBS, lysed with 500 ml 1% saponin
in PBS, and bacteria recovered by plating. Results are expressed as the number of
c.f.u. ml21 of cell lysate, and are the average of at least three independent experi-
ments performed in triplicate. Statistical significance was calculated using the
Student’s t-test. When required, IPTG (1 mM final concentration) was added to
the cells in tissue culture wells at the time of bacterial challenge. Cells were
exposed to mimosine (Sigma, final concentration 800 mM), which inhibits
degradation of HIFa, for 1 h before challenge with bacteria30. The constitutively
HIFa-expressing epithelial cell lines Rcc10 VHL2/2 and corresponding control
VHL1/1 (provided by P. Maxwell8,31) were cultured and infected as above.
For immunofluorescence microscopy, cells were grown on 12 mm glass cover-
slips, challenged as above, washed, then fixed in 3% PFA for 15 min, and washed
©2010 Macmillan Publishers Limited. All rights reserved
again. Coverslips were incubated for 30 min with primary antibodies diluted in
10% horse serum/0.1% saponin in PBS, washed with PBS containing 0.1%
saponin, then incubated with the secondary conjugated antibodies for 30 min.
Samples were washed three times in 0.1% saponin in PBS, then in PBS, and
finally in water, mounted in Prolong (Molecular Probes), and examined by laser-
scanning confocal microscopy (Zeiss Axiovert 100M or LSM510).
Cell viability was determined by staining with 0.01% propidium iodide (PI,
Sigma) in PBS for 10 min (ref. 32). PI only penetrates into and stains the DNA of
non-viable cells. As a positive control, cells were killed by incubation in 3% PFA
for 15 min. Fluorescence was measured using a FACS flow cytometer (BD sys-
tems) recording at least 10,000 events. Data were analysed with CellQuest Pro
software (BD Biosciences), and expressed as percentage survival.
To evaluate the response of bacterial O2 reporters, M90TDmxiD pFPV25.1 or
E. coli pXen were grown anaerobically in a controlled atmosphere chamber
(Plaslabs) in TCS medium at 37 uC until the D600 reached 0.2. The cabinet was
then adjusted to 5%, 10%, 15% or 20% O2, with the ambient oxygen tension
monitored with an oxygen meter. At each step, bacteria and PFA were incubated
for 15 min. To measure fluorescence an aliquot of the culture was mixed with an
equal volume of equilibrated PFA. For luminescence, bacteria were transferred
into an air-tight tube within the chamber. Fluorescence of bacteria was assessed
by confocal microscopy, and the results are the average of at least 150 bacteria
from three different fields. Luminescence was measured with an IVIS camera
(Xenogen). Experiments were performed on three different occasions.
The luxABCDE operon present on pXen5 (Xenogen) was introduced into
E. coli by transformation. Luminescence was measured using an IVIS camera
(Xenogen), and images quantified using the Living Image 3.1 software
(Xenogen) and expressed as a total flux (photons s21), as recommended by
the manufacturer.
Electron microscopy. For transmission electron microscopy, bacteria were
grown on solid medium overnight and a single colony was resuspended in
distilled water. Bacteria were negatively stained with 2% uranyl acetate on
glow-discharged copper grids. The samples were observed in a Jeol 1200EXII
or a JEM 1010 microscope (Jeol Company) at 80 kV with an Eloise Keenview
camera. Images were recorded with Analysis Pro Software version 3.1 (Eloise
SARL). The length of at least 200 needles was measured on three separate occa-
sions, and the results presented using a nonlinear regression model (Loess,
locally weighted scatterplot smoothing, GraphPad, Prism software).
For scanning electron microscopy, bacteria were fixed in 2.5% glutaraldehyde
in 0.1 M cacodylate buffer (pH 7.2) for 1 h on coverslips coated with poly
L-lysine, washed three times in 0.2 M cacodylate buffer (pH 7.2), then post-fixed
for 1 h in 1% (w/v) osmium tetroxide in 0.2 M cacodylate buffer (pH 7.2).
Samples were rinsed with distilled water, dehydrated through a graded series
of ethanol then critical-point-dried with CO2. Dried specimens were sputtered
with 10 nm gold palladium with a GATAN Ion Beam Coater and examined with
a JEOL JSM 6700F field emission scanning electron microscope operating at
5 kV. Images were acquired with the upper scanning electron detector and the
lower secondary detector.
Western blot analysis. Bacteria were grown in 500 ml TCS liquid medium in the
presence or absence of O2 until the D600 reached 0.2; IPTG was added to a final
concentration of 1 mM as required. Bacteria were collected by centrifugation for
15 min at 3,000g, washed, then re-suspended in 10 ml PBS. Cells were spun again
for 5 min at 3000g, and resuspended in 200ml of PBS. Secretion was induced by
adding Congo red (final concentration, 0.05%) under aerobic or anaerobic
conditions, and incubated for 20 min at 37 uC. Bacteria (representing the cellular
fraction) were collected by centrifugation at 3000g, and the supernatants (repre-
senting the secreted fraction) retained. Total protein concentrations were mea-
sured by the method of Bradford (Bio-Rad). Proteins were separated by 10%
SDS–PAGE and either visualized by Coomassie blue staining or transferred to
nitrocellulose membranes using semi-dry transfer, and incubated with the prim-
ary antibodies diluted in PBS/5% milk/0.01% Tween20 (Sigma) for 1 h.
Membranes were washed in PBS three times, then incubated with secondary
antibodies for 1 h before washing. Antibody binding was detected with chemi-
luminescence (ECL kit, GE Healthcare). Quantification of expression levels was
performed by densitometry of scanned blots using the ImageJ software and
performed at least on two independent occasions. Statistical significance was
calculated by the Student’s t-test.
Electrophoretic mobility shift assays. Gel retardation assays were carried out as
described previously using purified FNRD154A, an allele that forms dimers under
aerobic conditions33. In brief, purified 300 bp spa32 and spa33 promoter fragments
(Supplementary Fig. 6), and the modified alleles were end-labelled with [c-32P]-
ATP with 10 U ml21 T4 polynucleotide kinase (New England BioLabs). Approxi-
mately 0.5 ng of each labelled fragment was incubated with varying amounts of
FNRD154A in 10 mM potassium phosphate (pH 7.5), 100 mM potassium gluta-
mate, 1 mM EDTA, 50 mM dithiothreitol, 5% glycerol and 25 mg ml21 herring
doi:10.1038/nature08970
sperm DNA. After incubation at 37 uC for 20 min, samples were separated on 6%
polyacrylamide gels containing 2% glycerol. Gels were analysed using a Bio-Rad
Molecular Imager FX and Quantity One software (Bio-Rad).
Quantitative real-time PCR with reverse transcription. Bacteria were grown in
liquid Luria–Bertani medium until D600 5 0.2. The cultures were immediately
transferred to a new tube containing a 1 in 5 volume of 5% phenol pH 4.3
(Sigma) in ethanol. Samples were kept on ice for 30 min, then bacteria were
collected by centrifugation at 10,000g for 10 min. The pellets were re-suspended
in TE buffer containing 50 mg ml21 lysozyme. RNA was isolated with the RNeasy
method (Qiagen), and quantified by measuring ratio of the absorbance A260/A280
of samples. DNA was digested by treatment with DNase I (Sigma), then cDNA
was synthesized with reverse transcriptase (Quantitect, Qiagen). polA, ipaB,
ipaC, ipaD, mxiH, spa47, spa13, spa32, spa33, spa24, virF and virB were amplified
using the SensiMix DNA kit (Quantace) with primers described in
Supplementary Table 4 with a Rotor-gene RT–PCR machine (Corbett
Research). Results are the average of duplicate experiments each performed on
three independent occasions. Data were expressed relative to polA mRNA levels,
and analysed with Rotor Gene 3000 (Corbett Research) software.
Analysis of promoter fusion activity. Bacteria were grown to a D600 5 0.2 in
10 ml of TCS medium and collected by centrifugation at 5,000g for 10 min.
Pellets were resuspended in 500 ml of PBS, sonicated and further centrifuged at
13,000g for 20 min and the supernatants collected. b-galactosidase activity was
measured at 37 uC by assaying degradation of ortho-nitrophenyl-b-galactoside
(ONPG)34. The protein concentration of lysates was determined by the method
of Bradford, and the results are the mean values of three measurements per-
formed on two independent samples, and are expressed in Miller units mg21 of
total cellular protein.
Rabbit ligated intestinal loop model. For evaluation of virulence and measure-
ment of competitive indices (C.I.), experiments were performed on four or more
independent occasions in at least two loops in four animals. Bacteria were grown
in liquid media for 3 h at 37 uC, and a total 107 c.f.u. of bacteria consisting of
equal numbers of two strains in 500 ml of physiological saline were injected into
loops. Bacteria were recovered from the loops by plating homogenates to solid
media with (to recover mutant strains) or without antibiotics (to recover the
mutant and wild-type strains). The competitive index was calculated as the ratio
of mutant to wild-type bacteria recovered from animals divided by the ratio in
the inoculum; the results are the average of at least four samples from four
different rabbits.
For histopathological analysis, ileal loops were fixed in 4% buffered formalin,
embedded in paraffin, 5 mm sections obtained, and stained with haematoxylin-
eosin-safranin or Giemsa. The extent of damage to villi was quantified by mea-
suring their volume-to-length ratio and the presence of indentations, ruptures of
the mucosal surface, and abscesses. Results are the average of measurements
from at least 25 villi from two independent samples for each strain35. For
immunohistochemical staining, samples were washed in PBS, incubated at
4 uC PBS containing 12% sucrose for 90 min, then in PBS with 18% sucrose
overnight, and frozen in OCT (Sakura) on dry ice. 7 mm sections were obtained
using a cryostat CM-3050 (Leica). The immunostaining was performed as for
immunofluorescence of infected cells, and the antibodies were used at the same
concentrations. Stained samples were examined by laser-scanning confocal
microscopy (Zeiss Axiovert 100M or LSM510), as described above.
The vascular supply to intestinal loops was interrupted by exposing the afferent
mesenteric vessels, which were then occluded with a clamp. To detect luciferase
and GFP activity in the loops, bacteria were grown under anaerobic conditions for
4 h in LB medium then 107 c.f.u. were injected into the lumen of individual ligated
©2010 Macmillan Publishers Limited. All rights reserved
loops. The total luminescence of each injected loop was measured with an IVIS
camera system (Xenogen) and expressed as a total flux (photons s21), as recom-
mended by the manufacturer. To detect GFP fluorescence, the corresponding
loops were sectioned and fixed in 3% PFA and processed as above. Stained
samples were observed by laser-scanning confocal microscopy (Zeiss Axiovert
100M). The results are the average of duplicate experiments each performed on
three independent occasions.
Measurement of oxygen. HeLa cells were grown at 37 uC in room air supple-
mented with 5% CO2. Plates with cells were transferred into an anaerobic cab-
inet, and medium discarded and replaced with medium that had been
equilibrated in the cabinet for over 18 h. An oxygen microelectrode sensor
(Oxy-4, Unisense) connected to a PA-2000 (Unisense) was advanced into the
medium to 10 mm from the top of the cells. The oxygen sensor was calibrated at
37 uC as described previously36. Measurements were performed in triplicate on
three occasions at 30 and 60 min after replacement of the medium. For experi-
ments in the presence of deoxy-haemoglobin (Sigma) and cytochrome c
(Sigma), saturated solutions were prepared in deionised water, and added to
cells at a final concentration of 6 mM to medium kept under aerobic or anaerobic
conditions.
23. Exley, R. M. et al. Available carbon source influences the resistance of Neisseria
meningitidis against complement. J. Exp. Med. 201, 1637–1645 (2005).
24. Podkovyrov, S. M. & Larson, T. J. A new vector-host system for construction of
lacZ transcriptional fusions where only low-level gene expression is desirable.
Gene 156, 151–152 (1995).
25. Sasakawa, C., Komatsu, K., Tobe, T., Suzuki, T. & Yoshikawa, M. Eight genes in
region 5 that form an operon are essential for invasion of epithelial cells by Shigella
flexneri 2a. J. Bacteriol. 175, 2334–2346 (1993).
26. Farinha, M. A. & Kropinski, A. M. Construction of broad-host-range plasmid
vectors for easy visible selection and analysis of promoters. J. Bacteriol. 172,
3496–3499 (1990).
27. Datsenko, K. A. & Wanner, B. L. One-step inactivation of chromosomal genes in
Escherichia coli K-12 using PCR products. Proc. Natl Acad. Sci. USA 97, 6640–6645
(2000).
28. Phalipon, A. et al. Identification and characterization of B-cell epitopes of IpaC, an
invasion-associated protein of Shigella flexneri. Infect. Immun. 60, 1919–1926
(1992).
29. Ménard, R., Sansonetti, P., Parsot, C. & Vasselon, T. Extracellular association and
cytoplasmic partitioning of the IpaB and IpaC invasins of S. flexneri. Cell 79,
515–525 (1994).
30. Warnecke, C. et al. Activation of the hypoxia-inducible factor-pathway and
stimulation of angiogenesis by application of prolyl hydroxylase inhibitors. FASEB
J. 17, 1186–1188 (2003).
31. Esteban, M. A. et al. Regulation of E-cadherin expression by VHL and hypoxia-
inducible factor. Cancer Res. 66, 3567–3575 (2006).
32. Darzynkiewicz, Z. et al. Features of apoptotic cells measured by flow cytometry.
Cytometry 13, 795–808 (1992).
33. Browning, D. F., Beatty, C. M., Wolfe, A. J., Cole, J. A. & Busby, S. J. Independent
regulation of the divergent Escherichia coli nrfA and acsP1 promoters by a
nucleoprotein assembly at a shared regulatory region. Mol. Microbiol. 43,
687–701 (2002).
34. Miller, J. Experiments in Molecular Genetics. (1972).
35. D’Hauteville, H. et al. Two msbB genes encoding maximal acylation of lipid A are
required for invasive Shigella flexneri to mediate inflammatory rupture and
destruction of the intestinal epithelium. J. Immunol. 168, 5240–5251 (2002).
36. Liss, P., Nygren, A., Revsbech, N. P. & Ulfendahl, H. R. Intrarenal oxygen tension
measured by a modified Clark electrode at normal and low blood pressure and
after injection of x-ray contrast media. Pflugers Arch. 434, 705–711 (1997).
Vol 465 | 20 May 2010 | doi:10.1038/nature09022
Calcium-dependent protein kinase 1 is an essential
regulator of exocytosis in Toxoplasma
Sebastian Lourido1, Joel Shuman1, Chao Zhang2, Kevan M. Shokat2, Raymond Hui3 & L. David Sibley1
Calcium-regulated exocytosis is a ubiquitous process in eukaryotes,
whereby secretory vesicles fuse with the plasma membrane and
release their contents in response to an intracellular calcium surge1.
This process regulates various cellular functions such as plasma
membrane repair in plants and animals2,3, the discharge of defensive
spikes in Paramecium4, and the secretion of insulin from pancreatic
cells, immune modulators from lymphocytes, and chemical trans-
mitters from neurons5. In animal cells, serine/threonine kinases
including cAMP-dependent protein kinase, protein kinase C and
calmodulin kinases have been implicated in calcium-signal trans-
duction leading to regulated secretion1,6,7. Although plants and pro-
tozoa also regulate secretion by means of intracellular calcium, the
method by which these signals are relayed has not been explained.
Here we show that the Toxoplasma gondii calcium-dependent
protein kinase 1 (TgCDPK1) is an essential regulator of calcium-
dependent exocytosis in this opportunistic human pathogen.
Conditional suppression of TgCDPK1 revealed that it controls
calcium-dependent secretion of specialized organelles called micro-
nemes, resulting in a block of essential phenotypes including para-
site motility, host-cell invasion, and egress. These phenotypes were
recapitulated by using a chemical biology approach in which
pyrazolopyrimidine-derived compounds specifically inhibited
TgCDPK1 and disrupted the parasite’s life cycle at stages dependent
on microneme secretion. Inhibition was specific to TgCDPK1,
because expression of a resistant mutant kinase reversed sensitivity
to the inhibitor. TgCDPK1 is conserved among apicomplexans
and belongs to a family of kinases shared with plants and ciliates8,
suggesting that related CDPKs may have a function in calcium-
regulated secretion in other organisms. Because this kinase family
is absent from mammalian hosts, it represents a validated target that
may be exploitable for chemotherapy against T. gondii and related
apicomplexans.
The apicomplexan parasite T. gondii has been used as a model for
the secretion of numerous proteins from specialized organelles, called
micronemes, in response to an increase in intracellular calcium con-
centration9. Microneme secretion can be blocked by broad-spectrum
serine/threonine kinase inhibitors, and this is not circumvented by the
addition of calcium ionophores, suggesting that kinases mediate the
transduction of the calcium signal10. CDPKs have been identified in
plants, ciliates and apicomplexans but are absent from fungi and
animals11. CDPKs can respond to calcium when their calmodulin-like
domain binds calcium and releases the kinase domain from an inactive
conformation11. Recent structural studies illustrate a novel mechanism
of CDPK activation that results from a large-scale intramolecular
rearrangement12. Apicomplexans contain a diverse family of CDPKs;
some of these have canonical domain structures, whereas others are
more diverse8. In Plasmodium, the causative agent of malaria, gene
knockouts of several individual CDPKs have revealed important roles
1
Department of Molecular Microbiology, Washington University School of Medicine, 660 S. Euclid Avenue, St Louis, Missouri 63110, USA. 2Howard Hughes Medical Institute,
Department of Cellular and Molecular Pharmacology, University of California at San Francisco, San Francisco, California 94158, USA. 3Structural Genomics Consortium, University
Toronto, MaRS South Tower, Suite 732, 101 College Street, Toronto, Canada M5G 1L7.
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
at specific developmental stages8. For example, disruption of CDPK4
in asexual stages of Plasmodium berghei leads to differentiation defects
in male gametocytes13, a block that currently precludes analysis of its
role in other motile stages such as sporozoites. The orthologue of this
kinase is called TgCDPK1 in T. gondii, and a previous study suggested
that the ability of KT5926, a pan-specific S/T kinase inhibitor related to
staurosporine, to block cell attachment may result from inhibition of
this target14. Although KT5926 inhibits TgCDPK1 activity in vitro14,
it is unlikely to provide specific inhibition in the parasite, which
harbours 11 distinct CDPKs8 that control largely unexplored cellular
pathways.
To precisely define the role of TgCDPK1 in the parasite’s life cycle,
we generated a conditional knockout (cKO) using the tetracycline
transactivator system, previously developed for the study of essential
genes in Toxoplasma15. We first engineered a strain expressing an allele
encoding TgCDPK1 tagged with the HA9 epitope (YPYDVPDYA)
driven by a modified TetOSAG1 promoter, permitting suppression
of the transgene during growth in anhydrotetracycline (ATc; Fig. 1a).
The endogenous TgCDPK1 gene was then replaced by double
homologous recombination, generating the cKO as confirmed by
PCR analysis (Fig. 1a, b). Different alleles, expressed under the
SAG1 constitutive promoter, were subsequently introduced into the
cKO to test for complementation. Growth of the cKO in ATc resulted
in nearly undetectable levels of the HA9-tagged regulatable protein,
whereas the c-Myc-tagged constitutive proteins were stably expressed
in the complemented strains (Fig. 1c, d). As a first assessment of the
essential nature of TgCDPK1, we tested the ability of parasites to form
plaques on host cell monolayers. Both the wild-type and cKO lines
grew normally in the absence of ATc, whereas the presence of ATc led
to a complete block in plaque formation in the cKO only (Fig. 1e). The
phenotype was fully rescued when complemented with the wild-type
allele (cKO/WT; Fig. 1e) but not by a mutant allele in which the
catalytic aspartic residue was mutated to alanine (cKO/D174A;
Fig. 1e), indicating that TgCDPK1 function requires an active kinase.
Motility in apicomplexan parasites depends on a unique system
whereby adhesins contained in the micronemes are released onto the
apical end of the parasite and translocated to the posterior of the cell,
thus propelling the parasite forward16. Downregulation of TgCDPK1
by the addition of ATc during intracellular growth did not affect
parasite replication, yet parasites harvested from such cultures were
significantly impaired in all forms of gliding motility (Fig. 2a). Those
few cKO parasites that where able to glide, presumably as a result of
leaky suppression, exhibited wild-type speeds of motility, indicating
that the motor complex itself was unaffected (Supplementary Fig. 1).
Gliding motility is normally a prerequisite for cell invasion16, and
consistent with this was our observation that the cKO experienced
a decrease of more than 90% in invasion when grown in the presence
of ATc (Fig. 2b). Just as with plaque formation, invasion could be
359
LETTERS NATURE | Vol 465 | 20 May 2010

1
a 5′ UTR bleR 3′ UTR YFP b
Regulatable E1 E2
2

SAG1 CDPK1–c-myc 3′ UTR

5′ UTR CDPK1 3′ UTR 5′ UTR CDPK1 3′ UTR 5′ UTR bleR 3′ UTR 5′ UTR bleR 3′ UTR 335 bp
1
Endogenous E1 E2
3
Conditional Wild type
Wild type Merodiploid knockout complemented 263 bp

WT mDip cKO
bp
TetOSAG1 CDPK1–HA9 3′ UTR TetOSAG1 CDPK1–HA9 3′ UTR TetOSAG1 CDPK1–HA9 3′ UTR 335
263

c d e WT cKO cKO/WT cKO/D174A

cKO
cKO cKO/WT cKO/D174A
–ATc +ATc
ATc – + – + – +
–ATc
kDa
HA9
50

c-Myc 50
+ATc
Aldolase
37
2 cm

Figure 1 | TgCDPK1 is essential for the lytic cycle. a, Regulatable, HA9- pairs. c, Immunofluorescence analysis of the cKO in the presence or absence
tagged TgCDPK1 was added to the wild type (WT) to create a merodiploid of ATc; green, endogenous MIC2; red, HA9 tag; blue, DNA. Scale bar, 5 mm.
(mDip). Endogenous TgCDPK1 was replaced with phleomycin resistance d, Immunoblot of HA9-tagged regulatable and c-Myc-tagged constitutive
(bleR) to generate the cKO. Complementation with c-Myc-tagged mutant TgCDPK1 in cKO and complemented strains in the presence or absence of
alleles (denoted by cKO/‘allele’). UTR, untranslated region; YFP, yellow ATc. Aldolase, loading control. e, Plaque formation on fibroblast
fluorescent protein. b, Multiplexed PCR analysis of TgCDPK1. bp, base monolayers, in the presence or absence of ATc for 7 days.

rescued by expression of the constitutive WT allele but not the kinase- Egress from host cells depends on many of the same cellular pathways
dead allele (Fig. 2b). Suppression of TgCDPK1 also resulted in a required for invasion, and this response may naturally be triggered by
strong decrease in attachment to host cells (Fig. 2b), suggesting that accumulation of the plant-like hormone abscisic acid17. The cKO para-
TgCDPK1 affects an early step in invasion. sites grown in the absence of ATc behaved like the wild type and rapidly
egressed from host cells in response to treatment with calcium iono-
a c cKO –ATc cKO +ATc phore, an artificial but potent trigger of egress18 (Fig. 2c, Supplementary
60 Movie 1 and Supplementary Table 1). In contrast, almost all (namely
* 98%) cKO parasites grown in the presence of ATc did not respond to
Parasites (% of total)

Twirling
ionophore, remaining immotile within the vacuole (Fig. 2c, Sup-
40
Helical gliding 15 μm 15 μm plementary Movie 2 and Supplementary Table 1). Taken together, these
Circular gliding 0:30 0:30
experiments indicate that TgCDPK1 is essential for the transduction of
20 the calcium signals regulating gliding motility, invasion and egress.
All of the above TgCDPK1-dependent phenotypes share a require-
ment for adhesins stored in micronemes, which undergo calcium-
0
ATc – + – + 1:45 1:45 regulated exocytosis, in contrast to other secretory compartments
WT cKO such as dense granules, which are constitutively released9. TgCDPK1
shares a similar expression pattern to that of known microneme
proteins, as detected by microarray analysis of synchronized parasites
b 2 Intracellular
Extracellular (95% confidence interval; M. Behnke and M. White, personal com-
** 3:00 3:00
munication). Taken together, these data suggested that TgCDPK1
Parasites per host cell

*** might regulate microneme secretion, releasing, among other proteins,

the well-studied adhesin microneme protein 2 (MIC2) (ref. 9). After
1 secretion onto the cell surface, MIC2 is translocated to the cell
4:15 4:15 posterior and shed from the parasite surface by proteolysis, permitting
the detection of secreted MIC2 in the supernatant19. As expected,
MIC2 was detected in the supernatant of wild-type parasites stimu-
0 lated with ethanol, which is another potent secretagogue that is
ATc – + – + – + – +
thought to act through phospholipase C (ref. 20) (Fig. 3a). In contrast,
WT cKO cKO/WT cKO/D174A 5:30 5:30
the amount of MIC2 secreted by the cKO parasites grown in the
Figure 2 | TgCDPK1 is required for phenotypes associated with microneme presence of ATc was nearly undetectable, demonstrating a severe
secretion. a, Types of gliding motility as quantified by videomicroscopy. defect in calcium-regulated exocytosis (Fig. 3a). Secretion of MIC2
Student’s t-test; asterisk, P , 0.05; means 6 s.e.m. for n 5 4 experiments. was restored to wild-type levels by the constitutive WT allele but not
b, Invasion of fibroblasts by wild-type, cKO and complemented strains. by the kinase-dead allele (Fig. 3a). Growth of the cKO in ATc did not
Extracellular and intracellular parasites were stained differentially and
affect the release of dense granules (Fig. 3a), demonstrating that
enumerated as described in Supplementary Information. Student’s t-test;
three asterisks, P , 0.0005; two asterisks, P , 0.005; means 6 s.e.m. for TgCDPK1 specifically regulates calcium-dependent exocytosis from
n 5 3 experiments. c, Ionophore-induced egress of the cKO in the presence micronemes, not other secretory pathways.
or absence of ATc. The time stamps are coded as minutes:seconds after the Microneme secretion is also important in parasite egress, releasing
addition of calcium ionophore. See Supplementary Movies 1 and 2. the perforin-like protein TgPLP1 that aids in permeabilization of the
360
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 LETTERS

a 200 ** b PVM integrity by live videomicroscopy. To avoid premature rupture

WT +ATc cKO +ATc
of the vacuole, parasites were treated with cytochalasin D to immobi-
MIC2 secretion
(% of wild type)

150 lize them and allow selective monitoring of the kinetics of PVM rup-
** ture by leakage of DsRed. As reported previously21, wild-type parasites
100 rapidly permeabilized the PVM on treatment with calcium ionophore,
20 μm
1:00 20 μm 1:00 releasing DsRed into the host cell’s cytoplasm (Fig. 3b and Sup-
50
plementary Movie 3). All wild-type vacuoles showed rapid PVM per-
0 kDa
meabilization (1.7 6 0.5 min (mean 6 s.d.)), whereas about 30% of
MIC2 100 cKO parasites grown in the presence of ATc failed to rupture the PVM
3:00 3:00 (Fig. 3c and Supplementary Movie 4; data not shown). Analysis of
GRA1 25 those cKO parasite vacuoles that did rupture showed a significant
ATc – + – + – + – +
delay in the timing and rate of DsRed release when compared with
WT cKO cKO/WT cKO/D174A
wild-type vacuoles (Fig. 3c, d). Taken together, these results demon-
c ** d 400 *** strate a requirement for TgCDPK1 in controlling the release of micro-
5:00 5:00
neme contents, including TgPLP1, during egress. Moreover, the
Maximum rate of dimming

Time to maximum rate of

3
(% fluorescence s–1)

300 inability of calcium ionophore to circumvent the requirement for

TgCDPK1 places this kinase as the critical transducer downstream
dimming (s)

2 of the calcium signal regulating microneme exocytosis.

200
1 100
0
0
WT cKO
ATc
Figure 3 | Calcium-dependent microneme secretion requires TgCDPK1.
a, Western blot analysis of microneme protein MIC2 secretion after
induction with ethanol for 15 min. Dense granule protein-1 (GRA1) shows
constitutive secretion of dense granules. Student’s t-test; two asterisks,
P , 0.005; means 6 s.e.m. for n 5 3 experiments. b, Ionophore-induced
permeabilization detected by vacuolar DsRed leakage monitored by
fluorescence videomicroscopy of strains after treatment with ATc for 90 h.
The time stamps are coded as minutes:seconds after the addition of calcium
ionophore. Cytochalasin D was added to prevent egress. c, d, Quantification
of maximal rate (c) and timing (d) of fluorescence loss from rupturing
vacuoles. Mann–Whitney test; three asterisks, P , 0.0005; two asterisks,
P , 0.005; n 5 3 experiments.
parasitophorous vacuole membrane (PVM)21. To assess the role of
TgCDPK1 in controlling microneme secretion during egress, we
generated wild-type and cKO lines expressing a constitutively secreted
form of the fluorescent protein DsRed, permitting the detection of
a b
7:00 7:00
Having established the crucial role of TgCDPK1, we took advantage
of the atypical nucleotide-binding pocket of TgCDPK1 (ref. 12) to
develop a chemical biology approach to further evaluate the essential
nature of this kinase. It has been previously reported that the amino-
WT cKO 9:00 9:00 acid residue at the ‘gatekeeper’ position within the nucleotide-binding
ATc pocket radically affects inhibition by pyrazolopyrimidine (PP1) deri-
vatives, which have limited activity against most S/T protein kinases22.
Insensitivity is conferred by bulky gatekeeper residues in nearly all
kinases of both animal and parasite cells; however, kinases can be
rendered fully sensitive by mutation to a small gatekeeper22 (Sup-
plementary Table 2). Fortuitously, TgCDPK1 has a glycine residue
at this position, which is unique among canonical CDPKs (Fig. 4a)
and all other protein kinases in T. gondii (L. Peixoto and D. Roos,
personal communication). This finding predicted that wild-type
TgCDPK1 would be naturally sensitive to PP1-based inhibitors;
consistent with this, a pilot screen of selected derivatives inhibited
parasite lytic growth in vitro (Supplementary Fig. 2). Furthermore,
purified TgCDPK1 enzyme was extremely sensitive to the compound
3-methylbenzyl-PP1 (3-MB-PP1) (Fig. 4b), whereas mutation of
the glycine gatekeeper to methionine shifted the sensitivity by more
than 4 log units (Supplementary Table 2). Complementation of the
TgCDPK1 cKO with either wild-type or gatekeeper-mutant kinase
c 3 Intracellular
Extracellular
Parasites per host cell

Subdomain V 3-MB-PP1 3-BrB-PP1

TgCDPK1 FYLVGEVYTGGELFDEII Br
2
* *
TgCDPK3 YYLVMEVYRGGELFDEII
TgCDPK2A IYLVMELCTGGELFDRII NH2 NH2
TgCDPK2B IYLVLELCKGGELFDRII N N 1
TgCDPK5 LYLVMELCTGGELFERII N N
N N N
TgCDPK2 FYLVMEYCTGGELFDRLV N
TgPKG LYFLTELVTGGELYDAIR 0
DMSO + – + – + –
3-MB-PP1 – + – + – +
WT cKO/WT cKO/G128M
d e 120 f 120
200 DMSO
100 WT 100 WT
3-MB-PP1
cKO/WT cKO/WT
Inhibition (%)

Inhibition (%)
MIC2 secretion
(% of wild type)

* 80 80
150 * cKO/G128M cKO/G128M
60 60
100 40 40
20 20
50
0 0
0 –20 –20
WT cKO/WT cKO/G128M 0.01 0.1 1 10 100 0.01 0.1 1 10 100
3-MB-PP1 (μM) 3-BrB-PP1 (μM)

Figure 4 | PP1 derivatives specifically inhibit TgCDPK1 and block 3-MB-PP1 on MIC2 secretion. Student’s t-test; asterisk, P , 0.05;
microneme-mediated functions. a, Alignment of the kinase subdomain V, means 6 s.e.m. for n 5 3 experiments. e, f, Effect of PP1 derivatives on host
highlighting the gatekeeper residue. b, Structures of 3-MB-PP1 and 3-BrB- lysis by T. gondii in the presence or absence of 3-MB-PP1 (e) and 3-Br-PP1
PP1. c, Effect of 5 mM 3-MB-PP1 on host cell invasion. Student’s t-test; (f). Means 6 s.e.m. for n 5 3 experiments.
asterisk, P , 0.05; means 6 s.e.m. for n 5 3 experiments. d, Effect of 5 mM
361
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
alleles (cKO/WT or cKO/G128M, respectively) restored plaque forma-
tion, demonstrating that the mutation had no major deleterious effect
(Supplementary Fig. 3). When treated with PP1 analogues, both wild-
type and cKO/WT parasites were strongly inhibited in attachment to
and invasion of host cells (Fig. 4c), which is consistent with a recent
report that appeared online during the revision of the present work23.
In contrast with this recent report, which did not provide a quantita-
tive analysis of secretion23, we observed that microneme secretion by
extracellular parasites and ionophore-induced egress (Supplementary
Fig. 4) were also strongly inhibited by PP1 analogues (Fig. 4d). The
reversal of these phenotypes in the G128M mutant confirms that the
primary in vivo target of PP1 derivatives is TgCDPK1. Consistent with
these effects was our observation that 3-MB-PP1 and the related com-
pound 3-bromobenzyl-PP1 (3-BrB-PP1) blocked the ability of the
parasite to lyse host-cell monolayers (Fig. 4e, f), demonstrating the
essential role of TgCDPK1 during in vitro infection. Chemical genetic
studies indicate that TgCDPK1 acts independently of the previously
characterized cGMP-dependent protein kinase (PKG), the primary
target of tri-substituted pyrrole and imidazopyridine kinase inhibitors
that also block microneme secretion in T. gondii24,25. Correspondingly,
PKG is predicted to be insensitive to PP1 derivatives (Fig. 4a)22. Taken
together, these findings indicate that both kinases are essential for
efficient microneme secretion, possibly reflecting a hierarchical con-
trol of this important cellular pathway.
Our findings demonstrate that TgCDPK1 acts downstream of the
second messenger calcium to regulate exocytosis in T. gondii, thus
controlling several essential biological steps in the life cycle. Nearly all
mammalian protein kinases normally show very low sensitivity to
PP1 derivatives22, making them potential lead compounds for the
development of specific anti-parasitic drugs. In addition, CDPKs
may regulate calcium-dependent exocytosis in related parasites or
other organisms such as ciliates and plants, representing an evolu-
tionary precedent for calmodulin-dependent kinases that regulate
exocytosis in animals.
METHODS SUMMARY
Growth of host cells and parasite strains. T. gondii tachyzoites were maintained
by growth in human foreskin fibroblasts (HFFs), as described previously26.
Complemented strains were grown in 3 mM pyrimethamine (Sigma), and ATc
(Clontech) was added at 1.5 mg ml21 for 72 h unless indicated otherwise. For
inhibitor studies, parasites were incubated in the indicated concentration of
3-MB-PP1, 3-BrB-PP1, or dimethylsulphoxide (DMSO) control, for 20 min at
room temperature (20–25 uC), before use in assays.
Cellular assays. Plaque formation and invasion assays were performed as
described previously27,28. Microneme secretion was assayed by monitoring the
release of MIC2 into the culture medium after stimulation for 15 min with 3%
FBS and 2% ethanol at 37 uC, as described previously20. Samples were resolved by
SDS–PAGE, blotted and probed with mouse anti-MIC2 (monoclonal antibody
6D10), and mouse anti-GRA1 (monoclonal antibody Tg17-43) and quantified
by PhosphorImager analysis. Egress and PVM permeabilization were monitored
by videomicroscopy after stimulation with the calcium ionophore A23187
(Calbiochem) at 8 mM. When noted, parasites were pretreated for 10 min with
2 mM cytochalasin D (Calbiochem) to block motility. The extent and rate of
egress, and the degree of vacuole permeability, were quantified with Openlab
v. 4.1 (Improvision) as described in Supplementary Information. Host cell lysis
was assayed by staining monolayers with crystal violet, 3 days after infection at a
multiplicity of infection of 1.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 18 December 2009; accepted 17 March 2010.
1. Barclay, J. W., Morgan, A. & Burgoyne, R. D. Calcium-dependent regulation of
exocytosis. Cell Calcium 38, 343–353 (2005).
2. Schapire, A. L., Valpuesta, V. & Botella, M. A. Plasma membrane repair in plants.
Trends Plant Sci. 14, 645–652 (2009).
3. Bansal, D. & Campbell, K. P. Dysferlin and the plasma membrane repair in
muscular dystrophy. Trends Cell Biol. 14, 206–213 (2004).
4. Vayssie, L., Skouri, F., Sperling, L. & Cohen, J. Molecular genetics of regulated
secretion in paramecium. Biochimie 82, 269–288 (2000).
362
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
5. Chieregatti, E. & Meldolesi, J. Regulated exocytosis: new organelles for non-
secretory purposes. Nature Rev. Mol. Cell Biol. 6, 181–187 (2005).
6. Choi, W. S., Chahdi, A., Kim, Y. M., Fraundorfer, P. F. & Beaven, M. A. Regulation of
phospholipase D and secretion in mast cells by protein kinase A and other protein
kinases. Ann. NY Acad. Sci. 968, 198–212 (2002).
7. Easom, R. A. CaM kinase II: a protein kinase with extraordinary talents germane to
insulin exocytosis. Diabetes 48, 675–684 (1999).
8. Billker, O., Lourido, S. & Sibley, L. D. Calcium-dependent signaling and kinases in
apicomplexan parasites. Cell Host Microbe 5, 612–622 (2009).
9. Carruthers, V. B. & Sibley, L. D. Mobilization of intracellular calcium stimulates
microneme discharge in Toxoplasma gondii. Mol. Microbiol. 31, 421–428 (1999).
10. Carruthers, V. B., Giddings, O. K. & Sibley, L. D. Secretion of micronemal proteins
is associated with Toxoplasma invasion of host cells. Cell. Microbiol. 1, 225–236
(1999).
11. Harper, J. F. & Harmon, A. C. Plants, symbiosis and parasites: a calcium signalling
connection. Nature Rev. Mol. Cell Biol. 6, 555–566 (2005).
12. Wernimont, A. et al. Structural analysis of calcium-dependent protein kinases
reveal mechanism of activation by calcium. Nature Struct. Mol. Biol. doi:10.1038/
nsmb.1795 (2 May 2010).
13. Billker, O. et al. Calcium and a calcium-dependent protein kinase regulate gamete
formation and mosquito transmission in a malaria parasite. Cell 117, 503–514 (2004).
14. Kieschnick, H., Wakefield, T., Narducci, C. A. & Beckers, C. Toxoplasma gondii
attachment to host cells is regulated by a calmodulin- like domain protein kinase.
J. Biol. Chem. 276, 12369–12377 (2001).
15. Meissner, M., Schluter, D. & Soldati, D. Role of Toxoplasma gondii myosin A in
powering parasite gliding and host cell invasion. Science 298, 837–840 (2002).
16. Sibley, L. D. Invasion strategies of intracellular parasites. Science 304, 248–253
(2004).
17. Nagamune, K. et al. Abscisic acid controls calcium-dependent egress and
development in Toxoplasma gondii. Nature 451, 207–211 (2008).
18. Endo, T., Sethi, K. K. & Piekarski, G. Toxoplasma gondii: calcium ionophore A23187-
mediated exit of trophozoites from infected murine macrophages. Exp. Parasitol.
53, 179–188 (1982).
19. Carruthers, V. B., Sherman, G. D. & Sibley, L. D. The Toxoplasma adhesive protein
MIC2 is proteolytically processed at multiple sites by two parasite-derived
proteases. J. Biol. Chem. 275, 14346–14353 (2000).
20. Carruthers, V. B., Moreno, S. N. J. & Sibley, L. D. Ethanol and acetaldehyde elevate
intracellular [Ca21] and stimulate microneme discharge in Toxoplasma gondii.
Biochem. J. 342, 379–386 (1999).
21. Kafsack, B. F. et al. Rapid membrane disruption by a perforin-like protein facilitates
parasite exit from host cells. Science 323, 530–533 (2009).
22. Bishop, A. C. et al. A chemical switch for inhibitor-sensitive alleles of any protein
kinase. Nature 407, 395–401 (2000).
23. Sugi, T. et al. Use of the kinase inhibitor analog 1NM-PP1 reveals a role for
Toxoplasma gondii CDPK1 in the invasion step. Eukaryot. Cell 9, 667–670 (2010).
24. Gurnett, A. M. et al. Purification and molecular characterization of cGMP-
dependent protein kinase from Apicomplexan parasites. A novel
chemotherapeutic target. J. Biol. Chem. 277, 15913–15922 (2002).
25. Wiersma, H. I. et al. A role for coccidian cGMP-dependent protein kinase in
motility and invasion. Int. J. Parasitol. 34, 369–380 (2004).
26. Starnes, G. L., Coincon, M., Sygusch, J. & Sibley, L. D. Aldolase is essential for
energy production and bridging adhesin–actin cytoskeletal interactions during
parasite invasion of host cells. Cell Host Microbe 5, 353–364 (2009).
27. Roos, D. S., Donald, R. G. K., Morrissette, N. S. & Moulton, A. L. Molecular tools for
genetic dissection of the protozoan parasite Toxoplasma gondii. Methods Cell Biol.
45, 28–61 (1994).
28. Huynh, M. H. et al. Rapid invasion of host cells by Toxoplasma requires secretion of
the MIC2–M2AP adhesive protein complex. EMBO J. 22, 2082–2090 (2003).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank O. Billker, G. Ward, S. Moreno and V. Carruthers for
discussions; F. Dzierszinski for the DsRed plasmid; D. Soldati for the
Tet-transactivator system; and K. Tang for technical assistance. This work was
supported by a predoctoral fellowship from the American Heart Association (S.L.)
and a grant from the National Institutes of Health (L.D.S.).
Author Contributions S.L. designed and performed the majority of experiments,
analysed the data, generated the figures and wrote the manuscript. J.S. performed
the video miscopy measurements of motility and analysed the data. R.H. provided
key insight into the regulation of CDPKs by calcium. C.Z. and K.M.S. provided
inhibitors and insight into the strategy for chemical biology experiments. L.D.S.
supervised the project, assisted with experimental design and analyses, and
contributed to writing the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Readers are welcome to comment on the online version of this article at
www.nature.com/nature. Correspondence and requests for materials should be
addressed to L.D.S. (sibley@borcim.wustl.edu).
doi:10.1038/nature09022
METHODS
Growth of host cells and parasite strains. T. gondii tachyzoites were maintained
by growth in monolayers of HFFs cultured in DMEM medium containing 10%
tetracycline-free fetal bovine serum (HyClone), 2 mM glutamine, 10 mM HEPES
pH 7.5 and 20 mg ml21 gentamicin, as described26. Chloramphenicol
(20 mg ml21; Sigma), phleomycin (5 mg ml21; InvivoGen), ATc (1.5 mg ml21;
Clontech) and pyrimethamine (3 mM; Sigma) were added to the medium as
indicated, and for the maintenance of merodiploid or complemented strains.
When noted, parasites were treated with ATc for 72 h.
Plaque assay. Plaque assays were performed as described previously27. Confluent
monolayers of HFFs in six-well plates were infected with 200 parasites per well in
medium with or without 1.5 mg ml21 ATc (Clontech). At 24 h after infection,
additional medium was added to decrease the concentration of ATc to 1 mg ml21.
Monolayers were fixed 7 days after infection and stained with crystal violet.
Experiments were repeated three times with triplicate wells per experiment.
Invasion assay. Parasites were harvested in invasion medium (DMEM contain-
ing 20 mM HEPES pH 7.5, supplemented with 3% FBS). Parasites (5 3 106) in a
250 ml volume were added to subconfluent HFF monolayers in 24-well plates and
allowed to invade for 20 min. Monolayers were then fixed and stained as
described previously28 to distinguish extracellular from total parasites. Three
experimental replicates were performed for each strain in each of three separate
experiments, and parasite numbers per field were normalized to host-cell nuclei.
For the inhibitor studies, parasites were incubated in 5 mM 3-MB-PP1 or vehicle-
only control (DMSO), for 20 min at room temperature, before invasion.
MIC2 secretion assay. Microneme secretion was assayed as described previ-
ously20 by monitoring the release of MIC2 into the culture medium. Secretion
was stimulated by treatment for 15 min with 3% FBS and 2% ethanol at 37 uC.
Parasite lysis was monitored by the release of actin into the medium and
remained undetectable in all experiments presented. GRA1 secretion was used
as a control for constitutive secretion. Samples were resolved by SDS–PAGE,
blotted and probed with mouse anti-MIC2 (mAb 6D10), rabbit anti-TgACT1
and mouse anti-GRA1 (mAb Tg17-43, provided by M. F. Cesbron).
Quantification was performed by densitometry with a FLA-5000
PhosphorImager (Fuji Medical Systems). For the inhibitor studies, parasites
were pretreated for 20 min with 5 mM 3-MB-PP1 or vehicle-only control
(DMSO) at 37 uC before stimulation.
Strain construction. The TgCDPK1 cKO was constructed with a tetracycline
transactivator system15, as described previously for TgALD1 (ref. 26). In brief,
TgCDPK1 (GenBank accession number AF333958) was cloned with a carboxy-
terminal HA9 tag into the pTetO7SAG1 vector (obtained from D. Soldati),
downstream of the inducible promoter, and the CAT selectable marker driven
by the SAG1 promoter was introduced at a different site. The TATi-1 strain
(obtained from D. Soldati), used as the wild-type background in this study,
was transfected with the regulatable construct, and stable merodiploids were
selected with chloramphenicol29 and cloned by limiting dilution. To generate
the knockout construct the Ble selectable marker30 was flanked with 1.5 kilobases
(kb) upstream of the TgCDPK1 start codon and 1.5 kb downstream of the stop
codon, followed by a YFP expression cassette26. The knockout construct was
linearized and transfected into the merodiploid strain and stable pools were
selected through two rounds of phleomycin selection30. Sorting for YFP-negative
parasites was used to enrich for successful knockouts, and individual clones were
isolated by limiting dilution. Knockout of the endogenous TgCDPK1 gene was
confirmed in clones by PCR, using primers against consecutive exons and the
intervening intron, to distinguish between the endogenous and regulatable
genes. Complementing plasmids were constructed by cloning TgCDPK1 with a
carboxy-terminal c-Myc tag, under the regulation of the SAG1 promoter. The
DHFR selectable marker conferring pyramethamine resistance31 was cloned into
the complementing vectors. For the inhibitor studies, SAG1 was replaced with the
1.5-kb region preceding the TgCDPK1 start codon. Co-transfection with pDHFR-
TS (ref. 31) was used to generate stable clones. Mutations were generated by site-
directed mutagenesis. Complementing plasmids were transfected into the cKO,
stable lines were selected with pyramethamine, and clones were isolated by lim-
iting dilution. To monitor PVM permeabilization, wild-type and cKO strains
were transfected with p30-DsRed (ref. 21) (obtained from F. Dzierszinski) and
pDHFR-TS for isolation of stable transgenic lines as described above.
©2010 Macmillan Publishers Limited. All rights reserved
Host lysis assay. Parasites were harvested and incubated in the indicated inhibitor
or DMSO concentrations for 20 min at room temperature, before incubation with
confluent monolayers in 96-well plates at a multiplicity of infection of 1. For
experiments comparing complemented strains, parasites were grown in
1.5 mg ml21 ATc for 72 h before harvesting. Parasites were allowed to invade for
1 h, monolayers were washed three to five times, and fresh medium containing
1 mg ml21 ATc was added. The infection was allowed to progress for 3 days before
fixing with 70% ethanol and staining with crystal violet. Host cell lysis was deter-
mined by measuring absorbance at 570 nm in an EL800 microplate reader
(BioTek Instruments, Inc.).
Immunofluorescence microscopy. Intracellular parasites were stained as
described previously26. MIC2 staining within the micronemes required permea-
bilization for 2 min with 100% ethanol on ice. Staining was performed with
rabbit anti-HA9 (Invitrogen) and mouse anti-MIC2 (mAb 6D10), followed by
Alexa564-goat anti-rabbit IgG (Invitrogen), Cy5-goat anti-mouse IgG (Jackson)
and Sytox green (Invitrogen) for the nuclear stain. Images were collected on a
Zeiss LSM 510 confocal microscope.
Videomicroscopy and quantification of gliding motility. Gliding and egress
were analysed by videomicroscopy as described previously32. For gliding, 75
images were taken with exposure times ranging from 50 to 100 ms with 1 s
between exposures. Images were collected and combined into composites with
Openlab v. 4.1 (Improvision). ImageJ was used to analyse the images. The
ParticleTracker plug-in was used to track cell motility and Cell Counter was used
to quantify percentage motility.
Ionophore-induced egress and PVM permeabilization. Egress and PVM per-
meabilization were monitored by videomicroscopy as described above. Where
noted, parasites were preincubated for 10 min in medium containing 2 mM
cytochalasin D (Calbiochem) at 37 uC. All dishes were allowed to equilibrate
for 5 min on the heated stage before the addition of 8 mM calcium ionophore
A23187 (Calbiochem). Vacuoles were imaged for up to 10 min after the addition
of ionophore. To quantify vacuole permeabilization the fluorescence intensity
within an 80-mm2 region of each vacuole was measured with Openlab. The values
for each vacuole were normalized against the starting (100%) and ending (0%)
values for that particular vacuole, and the derivative of the curve was used to find
the maximal rate of fluorescence loss and the time when that rate occurred. For
the inhibitor studies, parasites were pretreated for 20 min with 5 mM 3-MB-PP1
or vehicle-only control (DMSO) at 37 uC before the addition of ionophore.
In vitro determination of IC50. Full-length His-tagged TgCDPK1 was cloned
into the pET-22b(1) vector (Novagen) and expressed in Escherichia coli BL21.
Point mutations were generated by QuickChange site-directed mutagenesis
(Stratagene). Proteins were induced with isopropyl b-D-thiogalactoside and
purified by nickel-affinity chromatography. Activities of WT and G128M
TgCDPK1 were determined with the CycLex CaM Kinase II Assay Kit (MBL
International Corporation) in accordance with the manufacturer’s instructions.
c-Src proteins were expressed and assayed in the presence of various concentra-
tions of the inhibitors as described previously33. IC50 was determined by fitting
the dose–response curve with GraphPad Prism software.
Statistics. Experiments were repeated three or more times and statistical analyses
were conducted in Excel with Students’s t-test (unpaired, equal variance, two-
tailed test) for comparisons with data that fit a normal distribution or the Mann–
Whitney test for non-parametric comparisons.
29. Kim, K., Soldati, D. & Boothroyd, J. C. Gene replacement in Toxoplasma gondii with
chloramphenicol acetyltransferase as selectable marker. Science 262, 911–914
(1993).
30. Messina, M., Niesman, I. R., Mercier, C. & Sibley, L. D. Stable DNA transformation
of Toxoplasma gondii using phleomycin selection. Gene 165, 213–217 (1995).
31. Donald, R. G. K. & Roos, D. S. Stable molecular transformation of Toxoplasma
gondii: a selectable dihydrofolate reductase–thymidylate synthase marker based
on drug resistance mutations in malaria. Proc. Natl Acad. Sci. USA 90, 11703–11707
(1993).
32. Håkansson, S., Morisaki, H., Heuser, J. E. & Sibley, L. D. Time-lapse video
microscopy of gliding motility in Toxoplasma gondii reveals a novel, biphasic
mechanism of cell locomotion. Mol. Biol. Cell 10, 3539–3547 (1999).
33. Blair, J. A. et al. Structure-guided development of affinity probes for tyrosine
kinases using chemical genetics. Nature Chem. Biol. 3, 229–238 (2007).
Vol 465 | 20 May 2010 | doi:10.1038/nature08973

LETTERS
A three-dimensional model of the yeast genome
Zhijun Duan1,2*, Mirela Andronescu3*, Kevin Schutz4, Sean McIlwain3, Yoo Jung Kim1,2, Choli Lee3, Jay Shendure3,
Stanley Fields2,3,5, C. Anthony Blau1,2,3 & William S. Noble3

Layered on top of information conveyed by DNA sequence and
chromatin are higher order structures that encompass portions of
chromosomes, entire chromosomes, and even whole genomes1–3.
Interphase chromosomes are not positioned randomly within the
nucleus, but instead adopt preferred conformations4–7. Disparate
DNA elements co-localize into functionally defined aggregates or
‘factories’ for transcription8 and DNA replication9. In budding
yeast, Drosophila and many other eukaryotes, chromosomes adopt
a Rabl configuration, with arms extending from centromeres adja-
cent to the spindle pole body to telomeres that abut the nuclear
envelope10–12. Nonetheless, the topologies and spatial relationships
of chromosomes remain poorly understood. Here we developed a
method to globally capture intra- and inter-chromosomal inter-
actions, and applied it to generate a map at kilobase resolution of
the haploid genome of Saccharomyces cerevisiae. The map recapi-
tulates known features of genome organization, thereby validating
the method, and identifies new features. Extensive regional and
higher order folding of individual chromosomes is observed.
Chromosome XII exhibits a striking conformation that implicates
the nucleolus as a formidable barrier to interaction between DNA
sequences at either end. Inter-chromosomal contacts are anchored
confirmed them with interactions from the EcoRI libraries at the
same threshold.
From our HindIII libraries, we identified 2,179,977 total interac-
tions at an FDR of 1%, corresponding to 65,683 interactions between
distinct pairs of HindIII fragments. We used these data to generate
conformational maps of all 16 yeast chromosomes. The overall pro-
pensity of HindIII fragments to engage in intra-chromosomal inter-
actions varied little between chromosomes, ranging from 436
interactions per HindIII fragment on chromosome XI to 620 inter-
actions per HindIII fragment on chromosome IV (Supplementary
Table 9). These results indicate broadly similar densities of self-
interaction (intra-chromosomal interaction) between chromosomes
and indicate that the density of self-interaction does not vary with
chromosome size (Supplementary Fig. 7).
Some large segments of chromosomes showed a striking propen-
sity to interact with similarly sized regions of the same chromosome.
For example, two regions on chromosome III (positions 30–90 kilo-
bases (kb), and 105–185 kb) showed an excess of interactions (Fig. 3
1. Crosslinking 5. RE2 cutting 6. Circularization
7. RE1 cutting

by centromeres and include interactions among transfer RNA Nucleus

RE2
RE2
RE1
genes, among origins of early DNA replication and among sites RE2
RE1

where chromosomal breakpoints occur. Finally, we constructed a

three-dimensional model of the yeast genome. Our findings pro- 2. RE1 cutting 8. EcoP15l adaptor
EcoP15l EcoP15l
vide a glimpse of the interface between the form and function of a 2. Deproteinization
eukaryotic genome. RE1 RE1
RE1
Chromosome conformation capture (3C) and its derivatives have RE1 9. Biotinylated adaptor & circularization
RE1 RE1
been used to detect long-range interactions within and between chro-
mosomes13–20. We developed a method for identifying chromosomal Biotinylated adaptor
interactions genome-wide by coupling chromosome conformation
capture-on-chip (4C)14 and massively parallel sequencing (Fig. 1 and RE1
RE1 10. EcoP15l cutting
RE1
Supplementary Methods). Because all 3C-based technologies are
encumbered by low signal-to-noise ratios18,21, we established the 3. Intra-molecular ligation
EcoP15l EcoP15l
method’s reliability by assessing: (1) random intermolecular ligations 11. Biotin isolation N25–27 N25–27
RE1
from each of five control libraries (Fig. 2a, Supplementary Tables 1 RE1 Biotinylated RE1
and 2 and Supplementary Methods); (2) restriction site-based biases adaptor

(Fig. 2b, Supplementary Figs 1 and 2 and Supplementary Table 3); (3) Figure 1 | Schematic depiction of the method. Our method relies on the 4C
reproducibility between independent sets of experimental libraries procedure by using cross-linking, two rounds of alternating restriction
that differed in DNA concentration at the 3C step, which critically enzyme (RE) digestion (6-bp-cutter RE1 for the 3C-step digestion and 4-bp-
influences signal-to-noise ratios (Supplementary Table 1, Fig. 2b and cutter RE2 for the 4C-step digestion) and intra-molecular ligation. At step 7,
c and Supplementary Fig. 2); (4) consistency between the HindIII and each circle contains the 6-bp restriction enzyme recognition site originally
EcoRI libraries (Supplementary Figs 3–5 and Supplementary Tables used to link the two interacting partner sequences (RE1). Diverging from 4C,
we relinearize the circles using RE1, then sequentially insert two sets of
4–8), and (5) a set of 24 chromosomal interactions using conven- adaptors, one of which permits digestion with a type IIS or type III
tional 3C (Fig. 2d, Supplementary Fig. 6). These results show that our restriction enzyme (such as EcoP15I). Following EcoP15I digestion,
method is reliable and robust (detailed in Supplementary Methods). fragments are produced that incorporate interacting partner sequence at
We established yeast genome architecture features using interactions either end, which can be rendered suitable for deep sequencing (see
from the HindIII libraries at a false discovery rate (FDR) of 1%, and Supplementary Methods).
1
Institute for Stem Cell and Regenerative Medicine, University of Washington, Seattle, Washington 98195-8056, USA. 2Department of Medicine, University of Washington Seattle,
Washington 98195-8056, USA. 3Department of Genome Sciences, University of Washington, Seattle, Washington 98195-5065, USA. 4Graduate Program in Molecular and Cellular
Biology, University of Washington, Seattle, Washington 98195-5065, USA. 5Howard Hughes Medical Institute.
*These authors contributed equally to this work.
363
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

a 10
H-Mp
b

Mean interaction frequency

H-Me
E-Mp r = 0.85, P = 10–16 r = 0.18, P = 0.2
8 0.07 0.07

Percentage of interactions
Percentage of interactions
E-Me
Controls 0.06 0.06
6

(H-Mp-A), chr I
(H-Mp-A), chr I
0.05 0.05
0.04 0.04
4 0.03 0.03
0.02 0.02
2 0.01 0.01
0 0
0 0.02 0.04 0.06 0.08 0 0.02 0.04 0.06 0.08
0 Percentage of interactions Percentage of interactions
20 50 100 150 200 250 300 350 400 450 500 (H-Mp-B), chr I (H-Mp control), chr I
Interaction distance (kb)
c d 50
H−Mp−A × 10–8 H−Mp−B1 × 10–8 5

in HindIII libraries (symbols)

Genomic position (kb) of chr I

Genomic position (kb) of chr I

frequencies by 3C (bars)
7 9

Interaction frequencies
Relative interaction
200 200 4
6 8
150 150 7
5 3
6
25
100 4 100 5
3 4 2
50 50 3
2
2 1
0 1 0
1
0 0 0 0
0 50 100 150 200 0 50 100 150 200

-B

-D

-F

-G

-I
-J

-L
-M

-O

-Q
Genomic position (kb) of chr I Genomic position (kb) of chr I

H
H

K
E
A

P
K
Figure 2 | Validation of the assay. a, Graph showing an inverse relationship mappable (black hatches) HindIII fragments are indicated. The binary
between interaction frequency and genomic distance (20 kb or larger, interaction matrix of all interactions with an FDR threshold of 1% has been
excluding self-ligations and adjacent ligations) separating interacting smoothed with a Gaussian of width 3 kb. d, High degree of correlation
restriction fragments (either HindIII or EcoRI) in each of four experimental between absolute interaction frequencies as determined by our method
but none of five control libraries. Note, the five lines representing the five (symbols) versus relative interaction frequencies as determined by
control libraries are very close to each other. H-Mp, HindIII-MspI; H-Me, conventional 3C using cross-linked (dark grey bars) and uncross-linked
HindIII-MseI; E-Mp, EcoRI-MspI and E-Me, EcoRI-MseI library. b, The (light grey bars) libraries. Results for 10 potential long-range intra-
fraction of instances that each HindIII site along chromosome I (chr I) was chromosomal interactions are depicted, of which 6 passed (circles) and 4 did
engaged in an intra-chromosomal interaction was highly correlated between not pass (triangles) an FDR threshold of 1%. Error bars denote standard
two independently derived experimental H-Mp (HindIII-MspI) libraries deviations over three experiments. Interaction sites are as follows. A, Chr III
(designated A and B, left panel) but was not correlated between experimental position 11811; B, chr III position 290056; C, chr III position 15939; D, chr
and non-cross linked control H-Mp libraries (right panel). c, Two- III position 314440; E, chr I position 26147; F, chr I position 191604; G, chr I
dimensional heat maps demonstrating broad reproducibility of interaction position 204567; H, chr VI position 12007; I, chr VI position 243206; J, chr
patterns within chromosome I for two independently derived H-Mp VI position 249743; K, chr II position 238203; L, chr II position 502988; M,
libraries (H-Mp-A and the equivalent sequence depth of H-Mp-B, H-Mp- chr II position 512024; N, chr IV position 236977; O, chr IV position 447899;
B1). The chromosomal positions of mappable (green hatches) and un- P, chr IV position 239805; Q, chr IV position 461284.

× 10–7 × 10–7 Figure 3 | Folding patterns of chromosomes.

a c
1.5 Chromosomes III (a, b) and XII (c, d) are shown.
Genomic position (kb) of chr XII
Genomic position (kb) of chr III

300 1000 2.5

250 The heat maps (a, c) and Circosdiagrams (b, d) were
800
2 generated using the intra-chromosomal
200 1 600 interactions identified from the HindIII libraries at
150 1.5
400 an FDR threshold of 1%. In the heat maps (a, c), the
100 chromosomal positions of centromeres (dashed
1
0.5 200
50 pink lines), telomeres (pink hatches), mappable
0 0 0.5 (green hatches) and un-mappable (black hatches)
−50 −200 HindIII fragments are indicated. Circos diagrams
0
(b, d) depict each chromosome as a circle. Each arc
00

0
00
0
0
50

0
0
0
0
0

20

40

60

80
−5

10
15
20
25
30

−2

10

Genomic position (kb) of chr III Genomic position (kbp) of chr XII connects two HindIII fragments and represents a
distinct interaction. The shade of each arc, from
1070
1060
1050

very light grey to black, is proportional to the

10400
310

10
10320

20
30
10 10

b d
40
0
10

50
300

10 0

60
0

10 990
20

70

negative log of the P-value of the interaction. The

80
0

1 0
29

1 00
9
98 70

1 10
30

9 60
0

13 20
0

9 chromosomal positions of centromeres (red

9

0
28

9 50 0
1450 50
40

9 40
0

27 9 30 11 0
16 70 rectangles), telomeres (red coloured areas), tRNA
0 50 9120 1 80
90 0
260 89 0
1 0
19 0 genes (blue outer hatches), mappable (green inner
60 8800 20 0
870
860
21 0
22
hatches) and un-mappable (black inner hatches)
230
250 70 850
840
240 HindIII fragments are indicated. Black outer
250
830
820
260
270
hatches and numbers mark genomic positions.
240 80 810 280
800 290 Note that the two ends of chromosome XII
790 300
230 90 780 310
320
(c, d) exhibit extensive local interactions, but very
770 33
7600 3400 little interaction with each other. Separating the
100 75 0
220 3
74 0 3 50 ends of chromosome XII are 100–200 rDNA
73 0 3 60
0
11 72 10 3 70
21
0 7 00 39 80 repeats, of which only two copies are depicted here
7 90 40 0
0 6 0
6 80
12

41 0

20 (from coordinates 450 to 470 kb). Additional heat

6 70

42 30
6

0
0

65 60

4 40
64 0

4 0
13
0

63 0

4560

maps and Circos diagrams for all chromosomes are

62 0
19

4
6100

4700
0

48 0
140
180

600

49
590

500
580

510
570

520
150

560
170

530
550 540 XII
160

shown in Supplementary Fig. 8.

III

364
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
a, b). Such regions may represent a ‘zippering’ of chromosomal seg-
ments, in which a large segment of DNA lies juxtaposed to a similar
length segment (see also chromosome II, 20–200 kb and 250–430 kb;
Supplementary Fig. 8). In other cases, large segments of chromo-
somes were enriched for local interactions, such that a series of con-
secutive HindIII fragments spanning tens of kilobases interacted
frequently with other HindIII fragments within the same segment
(for example, regions in chromosomes IV, XIII and XVI, Supplemen-
tary Fig. 8). Conversely, many combinations of fragments showed
few or no interactions, indicating highly improbable chromosome
conformations. For example, centromeric regions tended to engage
in relatively few long-range intra-chromosomal interactions (Sup-
plementary Fig. 8). Overall, the number of interactions involving any
given HindIII fragment was strongly influenced by the interaction fre-
quencies of neighbouring HindIII fragments, demonstrating regional
differences in the tendency for interaction within chromosomes.
Intra-chromosomal interactions between telomeric ends varied
markedly from one chromosome to another. Consistent with previous
observations10,11, chromosomes III and VI exhibited high levels of
enrichment of intra-chromosomal interactions between their telo-
meres (Supplementary Table 10). In contrast, the ends of chromosomes
IV and XII showed no intra-chromosomal telomeric interactions (Sup-
plementary Table 10). Also as previously observed10, the ratios of
observed/possible intra-chromosomal telomeric interactions between
the two ends of chromosomes V and XIV were less than 1/25 that of
chromosome III (0.4 and 0.5 versus 13, respectively; Supplementary
Table 10).
The conformation of chromosome XII differed strikingly from its
counterparts. In contrast to the typical pattern of intra-chromosomal
interactions enveloping the lengths of entire chromosomes (Fig. 3 a, b
and Supplementary Fig. 8), chromosome XII segregated into three
distinct segments (Fig. 3 c, d). Regions of 430 kb at one end and
550 kb at the other end engaged in extensive local interactions;
however, these two regions did not interact with each other.
Extensive local interactions at either end of chromosome XII termi-
nated abruptly at the boundaries of nucleolus-associated ribosomal
DNA (rDNA), where 100–200 rDNA repeats comprise 1–2 mega-
bases of DNA22. This finding indicates that rDNA, and by inference
the nucleolus, acts as a near absolute barrier, blocking interactions
between the chromosome ends.
For HindIII, a total of 639,607 intra-chromosomal and 8,119,614
inter-chromosomal interactions are possible. Thus, any given HindIII
fragment end has a much larger universe of candidate fragments on
other chromosomes with which to partner than fragments within
the same chromosome. Nonetheless, a strong tendency for intra-
chromosomal ligation resulted in 53.2% of observed interactions
occurring between HindIII fragments within the same chromosome.
The frequency of inter-chromosomal interactions was significantly
enriched in the experimental versus control libraries, especially
among inter-centromeric and inter-telomeric interactions (Sup-
plementary Fig. 10). In budding yeast, clustering of centromeres adja-
cent to the spindle pole body persists throughout the cell cycle23. A
clustering of centromeres marked the primary point of engagement
between different chromosomes and was the most striking feature of
the inter-chromosomal contacts (Fig. 4, Supplementary Figs 9, 10).
Of interactions of chromosome I with other chromosomes (at FDR of
1%) in both the HindIII and EcoRI libraries, the overwhelming
majority lay within narrow 20 kb windows centred around their cen-
tromeres (Fig. 4a, b). The centromeres of the other fifteen chromo-
somes demonstrated similar clustering (Supplementary Fig. 9).
Another chromosomal landmark that mediates inter-chromosomal
contacts are telomeres11,24, which congregate and form five to eight foci
within the interphase nucleus. Our data show widespread associations
between pairs of telomeres on different chromosomes, with 88 of 450
possible telomere pairs associating (P # 0.02; another 30 pairs were not
analysed owing to lack of mappable HindIII sites; Fig. 4d, Supplemen-
tary Fig. 11 and Supplementary Table 11). The average size differences
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
between the two corresponding chromosome arms of each of the 88
associated telomere pairs was much smaller than that of the remaining
450 pairs (199.0 kb versus 373.9 kb, P , 1026, unpaired two-tailed
t-test). These results indicate that two telomeres positioned at similar
distances from their corresponding centromeres are more likely to
interact, as described previously11. Nevertheless, there were excep-
tions. For example, the left arm of chromosome V and the right arm
of chromosome XIV are of similar size (152 and 155 kb, respectively),
but their telomeres did not associate (P 5 0.142, Supplementary
Table 11), consistent with previous observations11.
We assessed whether specific categories of genes or other chromo-
somal features were enriched in interactions among their members.
The 274 tRNA genes are dispersed throughout the yeast genome, yet
clustering of tRNA genes has been observed in the nucleolus25,26.
Consistent with this finding, HindIII sites adjacent to tRNA genes
were significantly enriched for interaction with sites neighbouring
other tRNA genes (Supplementary Fig. 11). Using a hierarchical clus-
tering algorithm, we identified two clusters of co-localized tRNA
genes (Supplementary Fig. 12), one that seems to be co-localized with
the rDNA region on chromosome XII, consistent with nucleolar
localization, and another that seems to be clustered with centromeres.
There was an enrichment of interactions among early (but not late)
origins of DNA replication (Fig. 4d and Supplementary Fig. 11).
These early replication origins clustered into at least two discrete
regions, consistent with their co-localization in replication factories
(Supplementary Fig. 13). Both tRNA genes and origins of early DNA
replication associate with chromosomal breakpoints27, and we
detected a significant co-localization of breakpoint sites (Fig. 4d
and Supplementary Fig. 11). Finally, we investigated whether other
groups of genes were significantly enriched in interactions, and found
that they were not (Supplementary Fig. 11).
The ratio of non-self to self interactions correlated inversely with
chromosome size (Supplementary Fig. 14). Smaller chromosomes
(I, III and VI) had the strongest propensity to interact with other
chromosomes, whereas the large chromosomes XII and IV were the
most isolated. Considering each chromosome pair for the ratio of
observed versus expected interactions, we found that interactions
were much more prevalent between smaller chromosomes (I, III,
VI, IX and VIII) (Supplementary Fig. 15). Only three pairs of larger
chromosomes (IV and VII, IV and XII, and IV and XV) displayed
relatively high enrichment ratios. Analysing inter-chromosomal
interactions among the 32 chromosome arms, we found that chro-
mosome arms ,250 kb in size were much more likely to interact with
one another (Fig. 4c). Notably, among the larger chromosome arms,
the right arms of chromosomes IV and XII showed the highest inter-
action enrichments (Fig. 4c). Similarly, the interaction pattern
between any given chromosome pair was strongly influenced by
the relative sizes of the partners. Chromosomes of similar size inter-
acted along their entire lengths (Supplementary Fig. 9); however, a
smaller chromosome tended to interact along its length with a region
of corresponding size within its larger partner. For example, chro-
mosome I (230 kb in length) interacted preferentially within a region
of chromosome XIV approximately 270 kb in length (between 510
and 780 kb) (Fig. 4b and Supplementary Fig. 9).
These observations can be explained by the Rabl configuration of
yeast chromosomes. Tethered by their centromeres to one pole of the
nucleus, the chromosomes extend outward towards the nuclear
membrane. Small chromosome arms are crowded within the thicket
of the entire set of 32 arms, thereby making frequent contacts with
other chromosomes. In contrast, the distal regions of the larger chro-
mosome arms occupy relatively uncrowded terrain, making fewer
contacts with other chromosomes.
To address the question of chromosome territories28,29, we com-
pared the observed/expected ratios for intra-chromosomal versus
inter-chromosomal interactions for all 32 chromosome arms
(Supplementary Fig. 16). Examining the entirety of each arm, we
found a higher enrichment for the 16 intra-chromosomal pairings
365
LETTERS NATURE | Vol 465 | 20 May 2010

XVI

0
I
0

II
Chromosome
a c

XVI
XIV
XIII
III

XV
I II III IV V VI VII VIII IX X XI XII

0
XV
I 1
0 II
IV III 0.5
XIV IV
V
0 0

Chromosome
0 VI
VII
XIII V VIII –0.5
0
IX
0 0
VI X −1
XI
I VI
XI I XII
–1.5
0 XIII

0
XIV

VII
XI

0
0
−2

IX

I
0
XV

X
XVI
780
770
760

10
20
30
7500

40
0
74 0

50
60
73

70
7210

80
7 00

1 0
1 00
0
7 90

b d

1 10
0 I
1
6 80

13 2 0
6
6

6 70 0
6 60 14

Percentage of positive interactions

6 50 15
0
0
6340 1670
62 0 1 80 8
61 0 1 0
19 0
60 0 20 0
5900 21
580 220 6
570 230
560
550 0
540 10
20 4
530 30 Centromeres
520 40
510 50 Telomeres
500 60 Early origins
4900 70 2 Late origins
48 0 80
47 0 90 Breakpoints (Scer)
46 0 1
1 00 Breakpoints (Scer and Kwal)
4540 1 10
4 30 1 20 0
4 20 14 30 0 0.2 0.4 0.6 0.8 1
4 10 0
15 60

4 Percentage of negative interactions

3 0

1 70
V 0

0
38 90

1 0
XI 4

37 0

1890
36 0

1 00
0

2 0
34 0

21 0
3300

22
35

230
320

240
310

250
300

270
260
290
280

Figure 4 | Inter-chromosomal interactions. a, Circos diagram showing chromosome I, and a distinct region of corresponding size on chromosome
interactions between chromosome I and the remaining chromosomes. All 16 XIV. c, Inter-chromosomal interactions between all pairs of the 32 yeast
yeast chromosomes are aligned circumferentially, and arcs depict distinct chromosomal arms (the 10 kb region starting from the midpoint of the
inter-chromosomal interactions. Bold red hatch marks correspond to centromere in each arm is excluded). For each chromosome, the shorter arm
centromeres. To aid visualization of centromere clustering, these is always placed before the longer arm. Note that the arms of small
representations were created using the overlap set of inter-chromosomal chromosomes tend to interact with one another. The colour scale
interactions identified from both HindIII and EcoRI libraries at an FDR corresponds to the natural log of the ratio of the observed versus expected
threshold of 1%. Additional heat maps and Circos diagrams are provided in number of interactions (see Supplementary Materials). d, Enrichment of
Supplementary Fig. 9. b, Circos diagram, generated using the inter- interactions between centromeres, telomeres, early origins of replication,
chromosomal interactions identified from the HindIII libraries at an FDR and chromosomal breakpoints. To measure enrichment of strong
threshold of 1%, depicting the distinct interactions between a small and a interactions with respect to a given class of genomic loci, we use receiver
large chromosome (I and XIV, respectively). Most of the interactions operating curve (ROC) analysis.
between these two chromosomes primarily involve the entirety of
than all inter-chromosomal pairings, except for pairing between the depict intra-chromosomal folding, we incorporated a metric that
two smallest arms (1R and 9R) (Supplementary Fig. 16a). However, converts interaction probabilities into nuclear distances (assigning
the preference for intra-chromosomal arm pairing versus inter- 130 bp of packed chromatin a length of 1 nm, ref. 30) (Supplemen-
chromosomal arm pairing decreased with increasing distance from tary Figs 17 and 18 and Supplementary Methods). Using this ruler,
centromeres (Supplementary Fig. 16 b–d). These observations indi- we calculated the spatial distances between all possible pairings of the
cate that yeast chromosome arms are highly flexible. 16 centromeres (Supplementary Tables 14 and 15) The results are
Combining our set of 4,097,539 total and 306,312 distinct inter- consistent with previous observations12.
actions with known spatial distances that separate sub-nuclear land- The resulting map resembles a water lily, with 32 chromosome
marks12, we derived a three-dimensional map of the yeast genome. To arms jutting out from a base of clustered centromeres (Fig. 5).

I I
Figure 5 | Three-dimensional model of the yeast
II
III
II
III
genome. Two views representing two different
IV IV angles are provided. Chromosomes are coloured
V V
VI VI as in Fig. 4a (also indicated in the upper right). All
VII VII
VIII VIII chromosomes cluster via centromeres at one pole
IX IX
X X of the nucleus (the area within the dashed oval),
XI XI
XII XII while chromosome XII extends outward towards
XIII XIII the nucleolus, which is occupied by rDNA repeats
XIV XIV
XV
XVI
XV
XVI
(indicated by the white arrow). After exiting the
nucleolus, the remainder of chromosome XII
interacts with the long arm of chromosome IV.

366
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
Chromosome XII stretches its long arm across to the opposite nuc-
lear pole, incorporating its rDNA repeats into the nucleolus, with the
remainder of its long arm interacting with the long arm of chro-
mosome IV. The map represents a coarse-grained image, a snapshot
that ignores the dynamic nature of chromosomes. An additional
feature constraining the resolution of the map is the population-
based nature of the 3C technology, which cannot distinguish between
interactions that occur at high probability in a small fraction of cells
versus those that occur at low probability in a majority of cells. Our
results provide the first glimpse into the architecture of a eukaryotic
genome at high resolution, highlighting the three-dimensional com-
plexity of the genome of even this simple organism. Although we do
not understand how DNA sequence specifies this structure, further
work should unveil its general organizing principles. With continu-
ing developments in high throughput DNA sequence analysis, both
the definition and comparative analysis of the high-resolution archi-
tectures of additional organisms will be increasingly feasible.
METHODS SUMMARY
A culture of a bar1 derivative of Saccharomyces cerevisiae BY4741 (MATa his3D1
leu2D0 met15D0 ura3D0 bar1::KanMX) was crosslinked with 1% formaldehyde
for 10 min. Two sequential rounds of alternating restriction enzyme digestion,
intra-molecular ligation, biotin-streptavidin-mediated purification, linear PCR
amplification, and gel purification were carried out before construction of
paired-end libraries. Libraries were paired-end sequenced using the Illumina
Genome Analyzer 2, and sequence reads were mapped to the S. cerevisiae ref-
erence genome. To identify signal from background noise, we performed stat-
istical confidence estimation. To estimate a false discovery rate (FDR), we
eliminated self-ligations, ligations between adjacent restriction fragments, and
ligations between restriction fragments separated by less than 20 kb at their
midpoints. To account for the strong influence of genomic proximity on ligation
frequency, we subdivided the remaining intra-chromosomal interactions into
5 kb bins as measured by the genomic distance between the midpoints of the two
ligated fragments. Inter-chromosomal interactions were placed into a separate
bin. In each bin, the observed interactions were ranked according to their
sequence frequency and assigned a P-value relative to all other possible interac-
tions in the same bin. Lastly, the P-value of each interaction was converted into a
q value (defined as the minimal FDR threshold at which the interaction is
deemed significant), and we used these values to rank interactions library-wide.
After the true interaction sets were derived, further computational analyses were
performed as described in detail in the online supplementary information.
Received 17 November 2009; accepted 1 March 2010.
Published online 2 May 2010.
1. Misteli, T. Beyond the sequence: cellular organization of genome function. Cell
128, 787–800 (2007).
2. Lanctôt, C., Cheutin, T., Cremer, M., Cavalli, G. & Cremer, T. Dynamic genome
architecture in the nuclear space: regulation of gene expression in three
dimensions. Nature Rev. Genet. 8, 104–115 (2007).
3. Zhao, R., Bodnar, M. S. & Spector, D. L. Nuclear neighborhoods and gene
expression. Curr. Opin. Genet. Dev. 19, 172–179 (2009).
4. Heun, P., Laroche, T., Shimada, K., Furrer, P. & Gasser, S. M. Chromosome
dynamics in the yeast interphase nucleus. Science 294, 2181–2186 (2001).
5. Gasser, S. M. Visualizing chromatin dynamics in interphase nuclei. Science 296,
1412–1416 (2002).
6. Stone, E. M., Heun, P., Laroche, T., Pillus, L. & Gasser, S. M. MAP kinase signaling
induces nuclear reorganization in budding yeast. Curr. Biol. 10, 373–382 (2000).
7. Casolari, J. M., Brown, C. R., Drubin, D. A., Rando, O. J. & Silver, P. A.
Developmentally induced changes in transcriptional program alter spatial
organization across chromosomes. Genes Dev. 19, 1188–1198 (2005).
8. Osborne, C. S. et al. Active genes dynamically colocalize to shared sites of ongoing
transcription. Nature Genet. 36, 1065–1071 (2004).
9. Kitamura, E., Blow, J. J. & Tanaka, T. U. Live-cell imaging reveals replication of
individual replicons in eukaryotic replication factories. Cell 125, 1297–1308
(2006).
10. Bystricky, K., Laroche, T., van Houwe, G., Blaszczyk, M. & Gasser, S. M.
Chromosome looping in yeast: telomere pairing and coordinated movement
reflect anchoring efficiency and territorial organization. J. Cell Biol. 168, 375–387
(2005).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
11. Schober, H. et al. Controlled exchange of chromosomal arms reveals principles
driving telomere interactions in yeast. Genome Res. 18, 261–271 (2008).
12. Berger, A. B. et al. High-resolution statistical mapping reveals gene territories in
live yeast. Nature Methods 5, 1031–1037 (2008).
13. Dekker, J., Rippe, K., Dekker, M. & Kleckner, N. Capturing chromosome
conformation. Science 295, 1306–1311 (2002).
14. Simonis, M. et al. Nuclear organization of active and inactive chromatin domains
uncovered by chromosome conformation capture-on-chip (4C). Nature Genet.
38, 1348–1354 (2006).
15. Murrell, A., Heeson, S. & Reik, W. Interaction between differentially methylated
regions partitions the imprinted genes Igf2 and H19 into parent-specific chromatin
loops. Nature Genet. 36, 889–893 (2004).
16. Spilianakis, C. G., Lalioti, M. D., Town, T., Lee, G. R. & Flavell, R. A.
Interchromosomal associations between alternatively expressed loci. Nature 435,
637–645 (2005).
17. Zhao, Z. et al. Circular chromosome conformation capture (4C) uncovers
extensive networks of epigenetically regulated intra- and interchromosomal
interactions. Nature Genet. 38, 1341–1347 (2006).
18. Fullwood, M. J. & Ruan, Y. ChIP-based methods for the identification of long-range
chromatin interactions. J. Cell. Biochem. 107, 30–39 (2009).
19. Fullwood, M. J. et al. An oestrogen-receptor-a-bound human chromatin
interactome. Nature 462, 58–64 (2009).
20. Lieberman-Aiden, E. et al. Comprehensive mapping of long-range interactions
reveals folding principles of the human genome. Science 326, 289–293 (2009).
21. Simonis, M., Kooren, J. & de Laat, W. An evaluation of 3C-based methods to
capture DNA interactions. Nature Methods 4, 895–901 (2007).
22. Venema, J. & Tollervey, D. Ribosome synthesis in Saccharomyces cerevisiae. Annu.
Rev. Genet. 33, 261–311 (1999).
23. Jin, Q., Trelles-Sticken, E., Scherthan, H. & Loidl, J. Yeast nuclei display prominent
centromere clustering that is reduced in nondividing cells and in meiotic
prophase. J. Cell Biol. 141, 21–29 (1998).
24. Gotta, M. et al. The clustering of telomeres and colocalization with Rap1, Sir3, and
Sir4 proteins in wild-type Saccharomyces cerevisiae. J. Cell Biol. 134, 1349–1363
(1996).
25. Haeusler, R. A., Pratt-Hyatt, M., Good, P. D., Gipson, T. A. & Engelke, D. R.
Clustering of yeast tRNA genes is mediated by specific association of condensin
with tRNA gene transcription complexes. Genes Dev. 22, 2204–2214 (2008).
26. Thompson, M., Haeusler, R. A., Good, P. D. & Engelke, D. R. Nucleolar clustering of
dispersed tRNA genes. Science 302, 1399–1401 (2003).
27. Di Rienzi, S. C., Collingwood, D., Raghuraman, M. K. & Brewer, B. J. Fragile genomic
sites are associated with origins of replication. Genome. Biol. Evol. 2009, 350–363
(2009).
28. Haber, J. E. & Leung, W. Y. Lack of chromosome territoriality in yeast:
promiscuous rejoining of broken chromosome ends. Proc. Natl Acad. Sci. USA 93,
13949–13954 (1996).
29. Lorenz, A., Fuchs, J., Trelles-Sticken, E., Scherthan, H. & Loidl, J. Spatial organisation
and behaviour of the parental chromosome sets in the nuclei of Saccharomyces
cerevisiae 3 S. paradoxus hybrids. J. Cell Sci. 115, 3829–3835 (2002).
30. Bystricky, K., Heun, P., Gehlen, L., Langowski, J. & Gasser, S. M. Long-range
compaction and flexibility of interphase chromatin in budding yeast analyzed by
high-resolution imaging techniques. Proc. Natl Acad. Sci. USA 101, 16495–16500
(2004).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We appreciate the advice and assistance of M. Dorschner, the
comments of S. Di Rienzi, B. Brewer and B. Byers, and the assistance of L. Zhang and
G. Schroth (Illumina Inc.) in performing sequencing. We thank A. Brown for help
with the 3D model. Supported by NIH grants P01GM081619, P41RR0011823, a
post-doctoral fellowship (to M.A.) from the Natural Sciences and Engineering
Research Council of Canada, and the Howard Hughes Medical Institute.
Author Contributions Z.D. devised the strategy for characterizing genome
architecture, Z.D., J.S, S.F, C.A.B. and W.S.N. designed experiments, Z.D., K.S.,
Y.J.K., and C.L. performed experiments, Z.D., M.A., S.M., J.S., S.F., C.A.B. and W.S.N.
analysed experimental data, M.A., K.S., J.S. and W.S.N. commented on the
manuscript drafts, Z.D., S.F., and C.A.B. wrote the paper.
Author Information Sequencing data have been deposited in the Sequence Read
Archive under accession number SRP002120. An interactive website for yeast
chromosomal interactions can be found at http://noble.gs.washington.edu/proj/
yeast-architecture. Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to C.A.B.
(tblau@u.washington.edu) or W.S.N. (william-noble@u.washington.edu).
367

LETTERS
Opposing roles for calcineurin and ATF3 in squamous
skin cancer
Xunwei Wu1, Bach-Cuc Nguyen1, Piotr Dziunycz2, Sungeun Chang1{, Yang Brooks1, Karine Lefort3,
Günther F. L. Hofbauer2 & G. Paolo Dotto1,3
Calcineurin inhibitors such as cyclosporin A (CsA) are the mainstay
of immunosuppressive treatment for organ transplant recipients.
Squamous cell carcinoma (SCC) of the skin is a major complication
of treatment with these drugs, with a 65 to 100-fold higher risk than in
the normal population1. By contrast, the incidence of basal cell car-
cinoma (BCC), the other major keratinocyte-derived tumour of the
skin, of melanoma and of internal malignancies increases to a signifi-
cantly lesser extent1. Here we report that genetic and pharmacological
suppression of calcineurin/nuclear factor of activated T cells (NFAT)
function promotes tumour formation in mouse skin and in xeno-
grafts, in immune compromised mice, of H-rasV12 (also known as
Hras1)-expressing primary human keratinocytes or keratinocyte-
derived SCC cells. Calcineurin/NFAT inhibition counteracts p53
(also known as TRP53)-dependent cancer cell senescence, thereby
increasing tumorigenic potential. ATF3, a member of the ‘enlarged’
AP-1 family, is selectively induced by calcineurin/NFAT inhibition,
both under experimental conditions and in clinically occurring
tumours, and increased ATF3 expression accounts for suppression
of p53-dependent senescence and enhanced tumorigenic potential.
Thus, intact calcineurin/NFAT signalling is critically required for p53
and senescence-associated mechanisms that protect against skin
squamous cancer development.
The selectively increased risk of SCC formation in CsA-treated
patients1 indicated that calcineurin signalling, central to keratinocyte
growth/differentiation control2–4, has an intrinsic role in keratinocyte
tumour suppression. Mice with keratinocyte-specific deletion of the
calcineurin B1 gene (CnB1, also known as Ppp3r1), essential for
calcineurin activity3, have increased susceptibility to chemically
induced carcinogenesis, with decreased latency, higher incidence
and size of tumours, and earlier malignant conversion (Fig. 1a and
Supplementary Fig. 1).
To assess the relevance of these findings to human skin, primary
human keratinocytes (HKCs) were infected with an oncogenic
H-rasV12-transducing retrovirus, followed by skin-reconstitution graft-
ing assays onto immune compromised (Scid) mice. In control mice,
H-rasV12-expressing HKCs produced an acanthotic epithelium with
normal pattern of differentiation. In CsA-treated mice, grafted cells
gave rise to tumours with disordered proliferation and histopathologi-
cal features of moderately to poorly differentiated SCC (Fig. 1b and
Supplementary Fig. 2a). Similar results were obtained by cell injection at
the dermal-epidermal junction, that is, a location approximating that of
malignant skin tumour formation. H-rasV12-expressing HKCs formed
only differentiated epidermal cysts in control animals, whereas in
animals treated with CsA or the unrelated calcineurin inhibitor
FK506, they produced highly cellular lesions with decreased differenti-
ation (Fig. 1c and Supplementary Fig. 2b). Similar lesions were
1
Cutaneous Biology Research Center, Massachusetts General Hospital, Charlestown, MA 02129, USA; 2Department of Dermatology, University Hospital Zurich, Zürich CH-8091;
3
Department of Biochemistry, University of Lausanne, Epalinges CH-1066, Switzerland. {Present address: Department of Dermatology, University of Ulsan College of Medicine, Asan
Medical Center, Seoul, Korea.
368
©2010 Macmillan Publishers Limited. All rights reserved
Vol 465 | 20 May 2010 | doi:10.1038/nature08996
produced by H-rasV12-expressing HKCs with siRNA-mediated knock-
down of the CnB1 gene (Fig. 1c and Supplementary Fig. 2c), and in mice
treated with a calcineurin/NFAT inhibitory peptide (VIVIT)5 versus
control scrambled peptide (VEET) (Supplementary Fig. 2d–f).
Cutaneous SCCs are characterized by mutation and/or decreased
expression of p53, but only 10–20% have ras mutations6, raising the
question of the relevance of the present findings for cancer cells
without ras activation. Nucleotide sequencing showed that SCC12
and SCC13 cells, two independent lines from cutaneous SCC with
poorly aggressive properties7, have wild-type ras genes and, as
reported8,9, single p53 missense mutations. Upregulation of endo-
genous p53 in these cells still induces ‘canonical’ effectors like
p21WAF1/Cip1 (also known as CDKN1A; ref. 10), possibly through
an indirect mechanism like the reported ability of mutant p53 to
bind and titrate p63 (also known as TRP63; refs 11, 12), a negative
regulator of p21 expression and senescence in keratinocytes13,14.
Injection of SCC12 and SCC13 cells at the dermal-epidermal junction
resulted in differentiated cysts. By contrast, in mice treated with CsA
or VIVIT, or as a consequence of p53 knockdown, SCC cells formed
highly cellular and moderately differentiated infiltrating tumours
(Supplementary Figs 3 and 4a).
Cancer cell senescence is a failsafe mechanism against tumour
development15 that can be associated with increased expression of
terminal differentiation markers16. Staining for senescence-associated
b-galactosidase activity (SA-b-Gal)17 was positive in lesions formed by
H-rasV12-expressing HKCs or SCC13 cells in control, but not in CsA-
treated mice (Fig. 1d and Supplementary Fig. 5a). This was paralleled
by differences in Ki67 and K1 (also known as MKI67 and CAMK1D,
respectively) differentiation marker expression (Supplementary Fig. 6).
Similarly decreased SA-b-Gal staining was found in tumours formed
by ras-expressing HKCs with knockdown of CnB1 or, as predicted
from the literature, of p53, a key mediator of oncogene-induced sen-
escence15 (Fig. 1e and Supplementary Fig. 5b). Senescence was also
suppressed in tumours formed by SCC13 cells with p53 knockdown
(Supplementary Fig. 4b, c).
H-rasV12-induced senescence of cultured cells was also counteracted
by CsA or VIVIT treatment and p53 knockdown (Supplementary
Fig. 7a–c). Induction of terminal differentiation markers was similarly
suppressed (Supplementary Fig. 7d). Paralleling these changes and
consistent with previous reports15, oncogenic ras expression caused
increased p53 protein levels, without effects on transcription
(Supplementary Fig. 7e). Importantly, calcineurin/NFAT inhibition
counteracted the ras effects suppressing p53 expression not only at
the protein, but also at the mRNA level (Supplementary Fig. 7e).
In HKCs and SCC cells, p53 gene transcription is under negative
control of the AP-1 complex, specifically c-Jun and c-Fos10 (also
NATURE | Vol 465 | 20 May 2010 LETTERS

a Number of large b Figure 1 | Calcineurin/NFAT inhibition promotes

CnB1+/+ CnB1–/– keratinocyte tumour formation. a, Multistep skin
Mice with papillomas (%)

tumours per mouse Tumours from HKCs Tumour weight (g)

Number of papillomas
100 12 2
1.5
carcinogenesis of mice with keratinocyte-specific
P < 0.001
9 CnB1 deletion (CnB12/2)3 together with littermate

per mouse
75 1.5
1 controls (CnB11/1). Numbers of total and large
50 6 1
0.5 (.4.5 mm diameter) tumours were determined
25 3 0.5 weekly. For histology see Supplementary Fig. 1.
P < 0.01 P < 0.005 200 μm
0 0 0
0
Ctrl CsA
b, Grafting of H-rasV12-expressing HKCs onto Scid
0 5 10 15 20 0 5 10 15 20 CnB1+/+ CnB1–/–
Weeks Weeks
Ctrl CsA mice plus/minus subsequent CsA treatment.
c Tumours from HKCs Quantification of histological lesions
Epidermal–dermal junction, black arrows. For
higher magnification images and experimental
Well-differentiated Highly cellular
Conditions
cysts lesions
conditions see Supplementary Fig. 2a. c, H-rasV12-
expressing HKCs were injected at
HKC + ras 5/6 0/6
HKC + ras + CsA 1/6 4/6
dermal–epidermal junction of Scid mice plus/
HKC + ras + FK506 2/6 4/6
minus subsequent CsA or FK506 treatment.
HKC + ras + siCtrl 4/6 1/6
Similar assays were performed with H-rasV12-
200 μm expressing HKCs with siRNA-mediated CnB1
HKC + ras + siCnB1 0/6 5/6
Ctrl FK506 siCtrl siCnB1 knockdown (siCnB1) versus control (siCtrl).
d Tumours from HKCs Tumours from SCC13 cells e Tumours from HKCs Histological analysis was 10 days later, summarized
on the right. For higher magnification images and
experimental conditions, see Supplementary Fig.
2b, c. d, e, Lesions formed by H-rasV12-expressing
HKCs or SCC13 cells in mice plus/minus CsA-
treatment (d) or plus/minus CnB1 and p53
knockdown (e) were analysed by in situ
100 μm chromogenic assay for senescence-associated
b-galactosidase activity17. For additional images,
Ctrl CsA Ctrl CsA siCtrl siCnB1 sip53
data quantification and experimental conditions
see Supplementary Figs 3–5.

a b siRNA c NFATC1 – + – + f
Clinical SCCs HKCs
CnB1 Nfatc1 CsA – – + + Untr CsA Empty
Ctrl CsA VEET VIVIT – + – + 50 1 2 3 1 2 3 vector ATF3
ATF3 50
CnB1 Nfatc1 37
50 37
c-Jun 25 25
37
20 20
c-Fos 25 (kDa) (kDa)
20 ATF3 ATF3
(kDa)
Tubulin ATF3 ATF3
25
20
Tubulin Tubulin (kDa) Tubulin Tubulin
Atf3
d 9 e TATA g
6 ATF3-positive nuclei
Atf3
P < 0.001
Number of ATF3 positive nuclei

3
Relative mRNA level

–3.4 kb –2.7 kb –0.5 kb

Untr

P < 0.05
0 50
0 24 4 8 24 h Binding site Non-binding site
CsA – + + + + In vitro In vivo 40
CHX – – + + + 4 –3.4 kb –2.7 kb 0.5 kb 4 –3.4 kb –2.7 kb –0.5 kb Rcna1 Actin
30
Relative unit

P < 0.01
9 3 3
20
6 2 2
10
CsA

3 1 1
0 0
0 0 Untr CsA Untr CsA
IgG
IgG

IgG

IgG

IgG

IgG

IgG

IgG
NFAT

NFAT

NFAT

NFAT

NFAT

0 24 4 8 24 h NFATC1 NFATC1 NFATC1 In situ SCCs Invasive SCCs

VIVIT – + + + +
Ctrl siNFAT CsA Clinical SCCs
C1

C1

C1

C1

C1

CHX – – + + +

Figure 2 | Calcineurin/NFAT signalling negatively controls ATF3 binding sites (black and white boxes in the map above). Chromatin
expression. a, b, HKCs plus/minus CsA or VIVIT treatment (a) or CnB1 or immunoprecipitation assays of the NFAT-binding region of the calcipressin
Nfatc1 knockdown (b) were analysed in parallel with controls by (Rcna1) gene20, and a b-actin genomic region without NFAT binding sites
immunoblotting. Similar results were obtained at mRNA level were included. f, SCCs from three patients under CsA treatment and from
(Supplementary Fig. 8a–c). c, HKCs infected with retroviruses expressing untreated patients (Untr) from the general population were analysed by
constitutively active NFATC1 (1)19 or GFP control (2) plus/minus immunoblotting for ATF3 expression, in parallel with HKCs infected with
subsequent CsA treatment were analysed for ATF3 expression. Similar an ATF3-expressing retrovirus (ATF3) versus empty vector control. Similar
results were obtained by real time RT–PCR (Supplementary Fig. 8f). d, HKCs differences in Atf3 mRNA expression were observed with an independent set
plus/minus CsA/VIVIT treatment and cycloheximide (CHX) exposure were of patients (Supplementary Fig. 9a). g, Tissue arrays of cutaneous SCCs from
analysed at various times (hours) for Atf3 expression by real-time RT–PCR. patients under CsA treatment versus general untreated population (Untr)
Error bars represent mean 6 s.d. (n 5 3 replicates). e, Extracts of HKCs plus/ were analysed for ATF3 expression by immunohistochemistry.
minus CsA treatment or Nfatc1 knockdown (left panel) or intact human Representative staining is shown along with quantification of ATF3-positive
epidermis (right panel) were processed for chromatin immunoprecipitation nuclei in each visual field of tumour tissues. Error bars represent mean 6 s.d.
with anti-NFATC1 antibodies or non-immune IgGs, followed by real-time (n 5 18, 16, 126 and 127 tumours per group, from left to right).
PCR of Atf3 promoter regions containing and lacking high-affinity NFATC1
369
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

known as JUN and FOS, respectively). Real time PCR with reverse
transcription (RT–PCR) and immunoblotting showed that c-Jun
and c-Fos levels were unaffected by CsA or VIVIT treatment of
HKCs. By contrast, expression of ATF3, a member of the ‘enlarged’
AP-1 family previously connected with SCC progression18, was sharply
upregulated (Fig. 2a and Supplementary Fig. 8a). ATF3 expression also
increased after CnB1 or Nfatc1 knockdown (Fig. 2b and Sup-
plementary Fig. 8b,c), and in SCC13 cells treated with CsA or VIVIT
(Supplementary Fig. 8d). Enhanced ATF3 expression in CsA- or
VIVIT-treated keratinocytes was paralleled by increased binding of
the ATF3 protein to specific oligonucleotide sequences of the p53
promoter containing intact, but not mutated, ATF3 binding sites
(Supplementary Fig. 8e). Increased ATF3 expression in CsA-treated
keratinocytes was suppressed by retrovirally expressed constitutively
active NFATC1 (ref. 19; Fig. 2c and Supplementary Fig. 8f). The
kinetics of Nfatc1 knockdown and ATF3 upregulation were highly
correlated (Supplementary Fig. 8g), and induction of ATF3 by CsA
or VIVIT occurred to similar or greater extent when protein synthesis
was inhibited (Fig. 2d). Consistent with ATF3 being a direct target,
chromatin immunoprecipitation assays showed binding of endogen-
ous NFATC1 to two distinct regions of the Atf3 promoter harbouring
NFAT binding sites, such binding being abolished by Nfatc1 knock-
down or CsA treatment (Fig. 2e, left panel). In intact human epi-
dermis, we also detected NFATC1 binding to the Atf3 promoter
comparable to a well established NFATC1 target, the calcipressin gene
(Rcna1)20 (Fig. 2e, right panel).
CsA treatment caused ATF3 upregulation and p53 down-modulation
also in human skin explants and tumour xenografts (Supplementary
Fig. 8h, i). Increased ATF3 and decreased p53 expression was also
observed in SCCs from CsA-treated patients versus untreated patients,
together with decreased senescence (Fig. 2f and Supplementary Fig. 9).
Increased ATF3 expression in SCCs from CsA-treated patients was
further confirmed by tissue array/immunohistochemical analysis of a
cohort of tumour biopsies (Fig. 2g). Nucleotide sequence analysis of p53
in five SCCs from CsA-treated patients revealed missense mutations
affecting the DNA-binding domain and/or polymorphisms associated
with cancer development (72 Pro/Arg substitution) similar to those in
the general patient population (http://www-p53.iarc.fr).
Functionally, ectopic ATF3 expression, at levels comparable to
those occurring in SCCs from CsA-treated patients (Fig. 2f), blocked
expression of p53 and senescence-associated genes (Fig. 3a, b).
Conversely, ATF3 knockdown induced these genes, counteracting
the CsA and VIVIT effects (Fig. 3c,d and Supplementary Fig. 10). In
vivo, ectopic ATF3 promoted tumorigenicity of H-rasV12-expressing
HKCs and SCC cells, producing aggressive tumours with reduced
senescence as those caused by calcineurin/NFAT inhibitors (Fig. 3e, f
and Supplementary Fig. 11a, b, e). Conversely, Atf3 knockdown over-
came the tumour-promoting effects of these compounds and restored
senescence (Fig. 3g, h and Supplementary Fig. 11c, d).
Senescence serves to restrict tumorigenic potential of cells15. To
test whether calcineurin inhibition exerts opposite effects, H-rasV12-
expressing HKCs were sorted for elevated integrin a6 and low CD71
levels (a6bri CD71dim cells), enriching for cells with high self-renewal
potential21. In culture, response of a6bri CD71dim cells to VIVIT and
CsA treatment, in term of ATF3 and p53 levels, was similar to that of
total unsorted populations (data not shown). To assess their in vivo
behaviour, H-rasV12-expressing a6bri CD71dim cells were injected in
serial dilutions, together with constant amount of normal HKCs, into

a b c d Ras – + + +
p53 DcR2
Relative mRNA levels

Ras – + + CsA – – + +
p53 p21 DcR2 – – + 6.0 6.0
ATF3 siATF3 – + – +
Relative mRNA levels

1.2 100
75 4.0 4.0
0.8 ATF3
50 2.0 2.0
0.4 37
(kDa) p53 p53
0 0 0
CsA – – + + – – – – + + – –
Ctrl ATF3
VIVIT – – – – + + – – – – + +
Tubulin
Tubulin
siATF3 – + – + – + – + – + – +
e Tumours from HKCs f Tumours from SCC13 cells

200 μm 200 μm

Ctrl ATF3 Ctrl ATF3

g Quantification of histological lesions h Tumours from HKCs

Conditions Well-differentiated Highly cellular

cysts cysts
HKCs + Ras + CsA 1/6 4/6
HKCs + Ras + siCtrl + CsA 1/6 3/6
HKCs + Ras + siATF3 + CsA 4/6 0/6 50 μm

CsA CsA + siCtrl CsA + siATF3

Figure 3 | ATF3 upregulation enhances keratinocyte tumour formation and
suppresses cancer cell senescence. a, HKCs infected with ATF3-expressing
(ATF3) and control retroviruses were analysed by real time RT–PCR for
indicated genes. Error bars represent mean 6 s.d. (n 5 3 replicates). b, HKCs
plus/minus infection with ATF3 and H-rasV12-expressing retroviruses were
analysed for p53 expression by immunoblotting. c, HKCs plus/minus Atf3
knockdown and CsA/VIVIT treatment were analysed by real time RT–PCR
for p53 and DcR2 (also known as Ccr6) expression. Additional results are
shown in Supplementary Fig. 10a. d, HKCs plus/minus H-rasV12 expression,
Atf3 knockdown and CsA treatment as indicated were analysed for ATF3 and
p53 expression by immunoblotting. For densitometric quantification and
370
©2010 Macmillan Publishers Limited. All rights reserved
experimental conditions see Supplementary Fig. 10b. e, f, H-rasV12-
expressing HKCs (e) or SCC13 cells (f) infected with ATF3-expressing and
control retroviruses were injected at the dermal-epidermal junction of Scid
mice. Shown are histological images of lesions recovered 6 weeks later. For
quantification see Supplementary Fig. 11a, b. g, H-rasV12-expressing HKCs
plus/minus Atf3 knockdown were injected at the dermal-epidermal junction
of Scid mice, followed by CsA treatment. Mice were killed 10 days later. For
histological images see Supplementary Fig. 11c. h, Lesions from the previous
experiment were analysed for SA-b-Gal activity. Low magnification images
and quantification are shown in Supplementary Fig. 11d.
NATURE | Vol 465 | 20 May 2010 LETTERS

NOD/SCID interleukin-2 receptor gamma chain null (Il2rg2/2)
mice22. Addition of Matrigel allowed retention of injected cells in
well identifiable ‘nodules’. Histological analysis of nodules from con-
trol mice at 1 month after injection showed that H-rasV12-expressing
HKCs injected in low number formed only a few keratinized cysts
with limited cellularity. By contrast, in nodules recovered from CsA-
treated mice there was a striking number of ‘proliferative centers’,
composed of keratinocytes with elevated Ki67 positivity (Fig. 4a, b
and Supplementary Fig. 12). When injected in higher number,
H-rasV12-expressing HKCs formed highly keratinized and poorly
proliferative cysts in control mice, whereas in CsA-treated animals
they gave rise to overt tumours (Fig. 4c). Similar results were
obtained with sorted H-rasV12- and ATF3- expressing HKCs versus
controls (Fig. 4b,c).
Established cancer cell lines, including SCC cells, also contain
distinct sub-populations with different growth/tumorigenic poten-
tial23,24. SCC13 cells plus/minus increased ATF3 expression were
sorted for elevated integrin a6 and low CD71 levels or for expression
of CD133, a cell surface marker of putative cancer stem cell popula-
tions, including keratinocyte-derived24. Sorting resulted in .20-fold
enrichment of tumorigenic cells. As few as 500 sorted SCC13 cells
a b
Number of proliferative
overexpressing ATF3 produced detectable tumours, whereas larger
numbers of control cells were required (Fig. 4d and Supplementary
Fig. 13a). Histologically, even the lowest number of ATF3-expressing
cells produced highly cellular tumours with poor differentiation,
whereas control cells formed differentiated cystic lesions (Fig. 4d, e
and Supplementary Fig. 13a, b).
In further testing, as little as 1,000 freshly dissociated cells from
tumours formed by H-rasV12-expressing HKCs in CsA-treated mice
gave rise to secondary tumours, whereas substantially higher cell
numbers were required under control conditions (Fig. 4f). Similar
results were obtained with H-rasV12-expressing HKCs or SCC13 cells
plus/minus increased ATF3 expression (Fig. 4f, g). Histologically,
secondary tumours formed in CsA-treated mice or by ATF3-expres-
sing cells were highly cellular and poorly differentiated, whereas those
under control conditions consisted, as the primary tumours, of dif-
ferentiated cysts (Fig. 4f, g and Supplementary Fig. 14).
Many regulatory pathways, including calcineurin/NFAT2–4,25, have
both growth-promoting and -suppressing functions, depending on
cell type and context. An impact on tumour growth has been recently
attributed to calcineurin via indirect effects on angiogenesis26,27. The
intrinsic double-negative genetic pathway that we have uncovered
120

Number of proliferative
30
100 25
center per mm2

center per mm2

80 20
60 15
Ctrl
40 10
20 5
100 μm 0 0
Ctrl CsA Ctrl CsA Ctrl ATF3 Ctrl ATF3
2,500 5,000 2,500 5,000
c Quantification of histological lesions
Number Solid Tumour Solid Tumour
Cysts Cysts
of cells tumours grade tumours grade

CsA HKC + Ras + Etoh HKC + Ras + CsA

10,000
1/4 0/4 – 0/4 3/4 I-II
HKC + Ras + neo HKC + Ras + ATF3
10,000
3/4 1/4 – 0/4 4/4 I-II

d Quantification of histological lesions e Tumours from SCC13

SCC13 + Ctrl SCC13 + ATF3

Number Solid Tumour Solid Tumour

Cysts Cysts
of cells tumours grade tumours grade

500 0/4 0/4 – 0/4 2/4 II-III

1,000 1/4 0/4 I 0/4 2/4 II-III
2,000 2/4 1/4 I-II 0/4 3/4 III
50 μm
5,000 3/4 1/4 I-II 0/4 4/4 III
Ctrl 5,000 cells ATF3
f Quantification of histological lesions
g Quantification of histological lesions
HKC + Ras + Ctrl HKC + Ras + CsA HKC + Ras + ATF3 SCC13 + Ctrl SCC13 + ATF3
Number Solid Tumour Solid Tumour Solid Tumour Number Solid Tumour Solid Tumour
Cysts Cysts Cysts Cysts Cysts
of cells tumours grade tumours grade tumours grade of cells tumours grade tumours grade

1,000 0/8 0/8 – 2/4 1/4 I 2/4 1/4 I-II 1,000 0/4 0/4 – 0/4 3/4 II-III
5,000 0/8 0/8 – 2/4 1/4 I-II 0/4 3/4 I-II 5,000 1/4 0/4 I 0/4 3/4 II-III
10,000 2/8 0/8 – 0/4 3/4 II-III 0/4 4/4 II-III 10,000 2/4 0/4 I 0/4 4/4 III
50,000 3/8 0/8 – 0/4 4/4 II-III 0/4 4/4 II-III 50,000 4/4 1/4 I 0/4 4/4 III

Figure 4 | Calcineurin inhibition and increased ATF3 enhance cancer
initiating cell populations. a–c, H-rasV12-expressing HKCs sorted for high
a6 integrin and low CD71 expression (a6briCD71dim cells) were injected in
the indicated numbers into NOD/SCID-Il2rg2/2 mice22, plus/minus
subsequent CsA treatment for 4 weeks. Similar experiments were also
performed with sorted H-rasV12expressing HKCs plus/minus ATF3
overexpression. a, Histological images of nodules formed by 2,500 H-rasV12-
HKCs in mice plus/minus CsA treatment. b, Quantification of proliferative
centres and c, cysts or solid tumours formed by the indicated cell numbers.
Error bars represent mean 6 s.d. (n 5 4 nodules per condition). For low
magnification images, quantification criteria and immunofluorescence
©2010 Macmillan Publishers Limited. All rights reserved
analysis see Supplementary Fig. 12a, b. d, e, Sorted a6briCD71dim
populations of SCC13 cells plus/minus ATF3 overexpression were injected
in the indicated numbers into Scid mice. Shown are quantification of the
results and representative histological images. Experimental conditions were
the same as with CD133 sorted cells (Supplementary Fig. 13). f, g, Cells
dissociated from tumours formed by H-rasV12-HKCs in mice plus/minus
CsA treatment, or with ATF3 overexpression (f) or from tumours formed by
SCC13 cells plus/minus ATF3 overexpression (g) were injected in decreasing
numbers into secondary recipient mice (Scid), with tissue retrieval 4 weeks
later. For representative images and experimental conditions see
Supplementary Fig. 14.
371
LETTERS
here is based on calcineurin/NFAT suppression of ATF3, a negative
regulator of p53. The resulting impact on keratinocyte cancer cell
senescence versus growth is likely of clinical significance for the many
patients under treatment with calcineurin inhibitors as immunosup-
pressants1, and may be of relevance for other cancer types where
altered calcium signalling has a role28.
METHODS SUMMARY
Multistep skin carcinogenesis assays were performed with mice with keratino-
cyte-specific deletion of the CnB1 gene (CnB1loxP/loxP 3 K5-CrePR1)3 in parallel
with Cre-negative controls (CnB1loxP/loxP). Keratinocyte grafting assays were
performed as described previously29. Intradermal tumorigenicity assays were
adapted from hair follicle reconstitution assays30. Detailed conditions for these
assays as well as chromatin immunoprecipitation, immunoblotting, immuno-
fluorescence, senescence b-galactosidase staining, biotinylated DNA pull down
assays, and sorting can be found in the Method section and Supplementary
Figure legends.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 19 June 2009; accepted 8 March 2010.
1. Euvrard, S., Kanitakis, J. & Claudy, A. Skin cancers after organ transplantation. N.
Engl. J. Med. 348, 1681–1691 (2003).
2. Horsley, V., Aliprantis, A. O., Polak, L., Glimcher, L. H. & Fuchs, E. NFATc1 balances
quiescence and proliferation of skin stem cells. Cell 132, 299–310 (2008).
3. Mammucari, C. et al. Integration of Notch 1 and calcineurin/NFAT signaling
pathways in keratinocyte growth and differentiation control. Dev. Cell 8, 665–676
(2005).
4. Santini, M. P., Talora, C., Seki, T., Bolgan, L. & Dotto, G. P. Cross talk among
calcineurin, Sp1/Sp3, and NFAT in control of p21WAF1/CIP1 expression in
keratinocyte differentiation. Proc. Natl Acad. Sci. USA 98, 9575–9580 (2001).
5. Aramburu, J. et al. Affinity-driven peptide selection of an NFAT inhibitor more
selective than cyclosporin A. Science 285, 2129–2133 (1999).
6. Boukamp, P. Non-melanoma skin cancer: what drives tumor development and
progression? Carcinogenesis 26, 1657–1667 (2005).
7. Rheinwald, J. G. & Beckett, M. A. Tumorigenic keratinocyte lines requiring
anchorage and fibroblast support cultures from human squamous cell
carcinomas. Cancer Res. 41, 1657–1663 (1981).
8. Brash, D. E. et al. A role for sunlight in skin cancer: UV-induced p53 mutations in
squamous cell carcinoma. Proc. Natl Acad. Sci. USA 88, 10124–10128 (1991).
9. Burns, J. E. et al. Gene mutations and increased levels of p53 protein in human
squamous cell carcinomas and their cell lines. Br. J. Cancer 67, 1274–1284 (1993).
10. Kolev, V. et al. EGFR signalling as a negative regulator of Notch1 gene transcription
and function in proliferating keratinocytes and cancer. Nature Cell Biol. 10,
902–911 (2008).
11. Adorno, M. et al. A mutant-p53/Smad complex opposes p63 to empower TGFb-
induced metastasis. Cell 137, 87–98 (2009).
12. Gaiddon, C., Lokshin, M., Ahn, J., Zhang, T. & Prives, C. A subset of tumor-derived
mutant forms of p53 down-regulate p63 and p73 through a direct interaction with
the p53 core domain. Mol. Cell. Biol. 21, 1874–1887 (2001).
13. Keyes, W. M. et al. p63 deficiency activates a program of cellular senescence and
leads to accelerated aging. Genes Dev. 19, 1986–1999 (2005).
14. Nguyen, B. C. et al. Cross-regulation between Notch and p63 in keratinocyte
commitment to differentiation. Genes Dev. 20, 1028–1042 (2006).
15. Finkel, T., Serrano, M. & Blasco, M. A. The common biology of cancer and ageing.
Nature 448, 767–774 (2007).
372
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
16. Krizhanovsky, V. et al. Implications of cellular senescence in tissue damage
response, tumor suppression, and stem cell biology. Cold Spring Harb. Symp.
Quant. Biol. 73, 513–522 (2008).
17. Dimri, G. P. et al. A biomarker that identifies senescent human cells in culture and
in aging skin in vivo. Proc. Natl Acad. Sci. USA 92, 9363–9367 (1995).
18. Wang, A. et al. Epidermal hyperplasia and oral carcinoma in mice overexpressing
the transcription factor ATF3 in basal epithelial cells. Mol. Carcinog. 46, 476–487
(2007).
19. Neal, J. W. & Clipstone, N. A. A constitutively active NFATc1 mutant induces a
transformed phenotype in 3T3-L1 fibroblasts. J. Biol. Chem. 278, 17246–17254
(2003).
20. Cano, E., Canellada, A., Minami, T., Iglesias, T. & Redondo, J. M. Depolarization of
neural cells induces transcription of the Down syndrome critical region 1 isoform 4
via a calcineurin/nuclear factor of activated T cells-dependent pathway. J. Biol.
Chem. 280, 29435–29443 (2005).
21. Li, A. & Kaur, P. FACS enrichment of human keratinocyte stem cells. Methods Mol.
Biol. 289, 87–96 (2005).
22. Quintana, E. et al. Efficient tumour formation by single human melanoma cells.
Nature 456, 593–598 (2008).
23. Locke, M., Heywood, M., Fawell, S. & Mackenzie, I. C. Retention of intrinsic stem
cell hierarchies in carcinoma-derived cell lines. Cancer Res. 65, 8944–8950
(2005).
24. Prince, M. E. & Ailles, L. E. Cancer stem cells in head and neck squamous cell
cancer. J. Clin. Oncol. 26, 2871–2875 (2008).
25. Mancini, M. & Toker, A. NFAT proteins: emerging roles in cancer progression.
Nature Rev. Cancer 9, 810–820 (2009).
26. Baek, K. H. et al. Down’s syndrome suppression of tumour growth and the role of
the calcineurin inhibitor DSCR1. Nature 459, 1126–1130 (2009).
27. Ryeom, S. et al. Targeted deletion of the calcineurin inhibitor DSCR1 suppresses
tumor growth. Cancer Cell 13, 420–431 (2008).
28. Roderick, H. L. & Cook, S. J. Ca21 signalling checkpoints in cancer: remodelling
Ca21 for cancer cell proliferation and survival. Nature Rev. Cancer 8, 361–375
(2008).
29. Lefort, K. et al. Notch1 is a p53 target gene involved in human keratinocyte tumor
suppression through negative regulation of ROCK1/2 and MRCKa kinases. Genes
Dev. 21, 562–577 (2007).
30. Zheng, Y. et al. Organogenesis from dissociated cells: generation of mature
cycling hair follicles from skin-derived cells. J. Invest. Dermatol. 124, 867–876
(2005).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank P. Khavari, S. Kitajima, N. Clipstone and G. Crabtree
for gift of retroviruses, W. Austen for human skin material, C. Brisken and
C. Missero for careful reading of the manuscript, and E. Castillo for sequencing of
the ras and p53 genes. This work was supported by grants from NIH (AR054856
and AR39190), the Swiss National Foundation (311003A-122281/1), Oncosuisse
(OCS-02361-02-2009), the European Union (Epistem, Sixth Framework Program,
LSHB-CT-2005-019067) and, in part, by a grant to S.C. by the Korean Government
Foundation (KRF-2007-013-E00044) and to G.F.L.H. by the
Olga-Mayenfisch-Stiftung.
Author Contributions B-C.N., P.D, S.C, Y.B., K.L. and G.F.L.H. performed research
and analysed data; X.W. and G.P.D. designed and performed research, analysed
data and wrote the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to G.P.D.
(gian-paolo.dotto@unil.ch).
doi:10.1038/nature08996
METHODS
Cell culture, viruses and human samples. Culturing of primary human kerati-
nocytes and SCC13 cells, and infection with retroviruses expressing the H-rasV12
(ref. 31), Atf3 (ref. 32) and constitutively active Nfatc1 (ref. 19) genes, together
with corresponding controls, were as reported previously29. CsA (Sigma) was
dissolved in ethanol and stored in stock solution (20 mM) at 280 uC. The
VIVIT and VEET peptides were chemically synthesized by the peptide core facility
of the University of Lausanne, HPLC-purified (.95% purity), dissolved in
dimethylsulphoxide (DMSO) and stored in stock solution (20 mM) at 280 uC.
For knockdown experiments, cells were transfected as described33 with validated
stealth siRNAs for the human p53, CnB1 and Nfatc1 genes (Invitrogen) and for
Atf3 (GeneGlobe, Qiagen) in parallel with corresponding scrambled siRNA con-
trols, and analysed 72 h after transfection. The sequence of all siRNAs is shown in
Supplementary Table 1.
HKCs were treated with cycloheximide (10 mg ml21) or ethanol vehicle alone,
followed 2 h later by treatment with either CsA or VIVIT. Cells were collected at
subsequent times for mRNA analysis by real time RT–PCR. Preliminary experi-
ments (based on immunoblot analysis) showed .90% protein synthesis inhibi-
tion in HKCs treated with cycloheximide at these concentrations.
Human skin samples were obtained from abdominoplasty procedures at
Massachusetts General Hospital (Boston, Massachusetts, USA) with patients’
and institutional approvals and cultured as described previously10 plus/minus
treatment with CsA for 48 h. For RNA collection, skin samples were placed in
preheated PBS at 60 uC for 45 s, then chilled (on ice) in 0.1 M PBS for 1 min,
followed by mechanical separation of epidermis and dermis. The epidermis was
homogenized in TRI Reagent (Sigma) for RNA preparation.
Squamous cell carcinoma samples were obtained at the Department of
Dermatology of the Zurich University Hospital Switzerland from clinical biopsies.
Parts not needed for histological diagnosis were further processed with in-
stitutional review board approval.
Quantitative real time RT–PCR, chromatin immunoprecipitation and
immunodetection techniques. Conditions for real time RT–PCR analysis,
immunoblotting, immunofluorescence, FACS analysis and Magnetic Cell
Sorting (MACS) were as described previously24,29,34. The list of gene-specific
primers is provided in Supplementary Table 2. We used the following antibodies:
rabbit antiserum against human keratin 1 (PRB-149p) and loricrin (PRB-145p)
(Covance); mouse monoclonal against p53 (#2524), rabbit polyclonal against
c-Fos (#4384) (Cell Signaling); concentrated mouse monoclonal ((7A6)X, SC-
7294) and rabbit polyclonal against NFATC1 ((H-110)X, sc-13033), rabbit poly-
clonal against ATF-3 (C-19, sc188), mouse monoclonal against p21 (sc-187),
rabbit polyclonal against c-Jun (H-79, sc-1694) (Santa Cruz Biotechnology);
rabbit polyclonal anti-human ATF3 (LS-B329) (Lifespan Bioscience); mouse
monoclonal anti-pan-keratin (AE11AE3, ab961-6), rabbit anti-human vimentin
(ab45939), rabbit polyclonal anti-cytokeratin 5 (ab24647), rabbit monoclonal
anti-Ki67 (ab16667) (Abcam); mouse monoclonal anti-calcineurin B1 (CN-
B1), mouse monoclonal anti-c-Tubulin (GTU88) (Sigma). For FACS analysis:
mouse anti-CD71 allophycocyanin (APC)-conjugated and rat anti-integrin a6
fluorescein isothiocyanate (FITC)-conjugated (CD49f) (from BD bioscience),
and for MACS: mouse anti-CD133 and anti-mouse IgG beads (Miltenyi Biotec).
Senescence b-galactosidase staining. Senescence b-galactosidase (SA-b-Gal)
activity was assessed by use of a commercially available chromogenic assay kit
(Cell Signalling), according to the manufacturer’s recommendations. Assays for
endogenous b-galactosidase activity at a suboptimal pH (pH 6), which is
thought to reflect increased lysosomal activity in senescent cells17 were per-
formed in parallel with similar assays at optimal pH (pH 4), as positive control
for b-galactosidase activity in cells, irrespective of their growing or senescent
state.
Tissue arrays. Biopsies of cutaneous SCCs from patients under CsA treatment
and control patient population were collected for tissue array preparation, fol-
lowed by immunohistochemical analysis of ATF3 protein expression using poly-
clonal rabbit anti-human ATF3 antibody (Atlas Antibodies) and 10 M citrate
buffer at pH 6.0 as an antigen retrieval. Tumour specimens were distinguished
into in situ and invasive squamous cell carcinoma (SCC) of the skin. The num-
bers of analysed samples are: 126 invasive SCCs from immunocompetent
patients and 127 from CsA-treated patients; 18 in situ SCCs from immunocom-
petent patients and 16 from CsA-treated patients. Statistical analysis indicated
that the differences in ATF3 nuclear staining between tumours from CsA-treated
and untreated patients was highly significant for both in situ SCCs (P , 0.05) and
invasive SCCs (P , 0.01), and for the two groups combined (P , 0.001).
ATF3 pull-down assays. Conditions for these assays were as described previ-
ously35. Briefly, HKCs were treated with either CsA (5 mM) or VIVIT (2 mM) or
corresponding controls for 24 h, followed by lysis in HKMG buffer (10 mM
HEPES pH 7.9, 100 mM KCl, 5 mM MgCl2, 10% glycerol, 1 mM dithiothreitol
©2010 Macmillan Publishers Limited. All rights reserved
(DTT), 0.1% Nonidet P-40) with a mixture of protease inhibitors. Cell extracts
were incubated with 1 mg of biotinylated double-strand oligonucleotides for
16 h. Two biotinylated oligonucleotide sequences of the p53 promoter contain-
ing fully conserved ATF3 binding sites (p53-1 and p53-2, corresponding,
respectively, to nucleotide positions 2420 to 2358 and 22105 to 22039 from
the initiation codon) in parallel with a mutated oligonucleotide of the first
sequence (mp53-1), with internal nucleotide substitutions disrupting the
ATF3 binding site. The oligonucleotide sequences are provided in Sup-
plemental Table 3. DNA–protein complexes were collected by precipitation with
streptavidin-agarose beads (Pierce) for 1 h, washed three times with HKMG
buffer, followed by processing for standard immunoblot analysis with anti-
ATF3 antibody.
Tumorigenicity assays. Mice with the CnB1 gene flanked by loxP sites and a Cre
transgene driven by a keratinocyte-specific promoter (CnB1loxP/loxP 3 K5-
CrePR1)3 in parallel with transgene-negative controls (CnB1loxP/loxP) were sub-
jected to a multistep skin carcinogenesis protocol as described previously36.
Briefly, mice (8-weeks-old; 12–16 animals per group) were treated with 7,12-
dimethylbenz [a] anthracene (DMBA) (20 mg in 200 ml acetone), followed by
repeated treatments (twice a week) with 12-O-tetradecanoylphorbol 13-acetate
(TPA) (1024 M in acetone) for 2 months. Mice were examined throughout the
duration of the experiments for papilloma versus carcinoma formation by
macroscopic examination and palpation of the tumour base, to determine the
onset of an invasive pattern of growth as well as infiltration of the surrounding
tissues.
For grafting assays, HKCs were infected with a H-rasV12-transducing retro-
virus (LZRS-rasV12)31 and, 16 h after infection, grafted onto the back of immu-
nocompromised (Scid) mice (2 3 106 cells per graft) by use of dome-shaped
transplantation chambers as described previously29,37,38. Mice were subsequently
treated, three times a week, with CsA (via injection into the transplantation
chamber; 20 mg per gram body weight, in ethanol), or ethanol vehicle alone (four
mice per group). Mice were killed after 8 weeks for retrieval of the grafted tissue
and histological analysis.
For intradermal tumorigenicity assays, HKCs, SCC12 and SCC13 cells were
brought into suspension, and injected (106 cells per injection, two side flank
injections per mouse) at the dermal-epidermal junction of 8-weeks-old female
mice (NOD/SCID or NOD/SCID with interleukin-2 receptor gamma chain null
mutation (Il2rg2/2; ref. 22), using a procedure described previously for hair
reconstitution assays30. Starting 24 h later, mice were given intraperitoneal injec-
tions, every other day, of CsA (20 mg per gram body weight in ethanol) or FK506
(2 mg per gram body weight in DMSO) in parallel with vehicles alone, or the
VIVIT and VEET control peptides (10 mg per gram body weight in ethanol), and
killed at the indicated times after cell injections.
As an alternative approach to drug treatment, HKCs or SCC cells were trans-
fected with siRNAs against the indicated genes or scrambled siRNA controls
(100 nM siRNA in HiPerFect transfection reagent, Qiagen) for 12 h before infec-
tion with the H-rasV12 transducing retrovirus. Twenty hours later, cells were
trypsinized and injected into mice. In all cases, mice were killed at 10 days after
cell injections and the nodules found at the site of injections processed for
histological analysis.
For serial dilution tumorigenicity assays, sorted H-rasV12-expressing HKCs or
SCC cells were serially diluted, admixed in various numbers with a fixed amount
of HKCs (2 3 105 cells per injection), mixed with Matrigel (BD biosciences), and
injected into the dermal–epidermal junction of NOD/SCID or NOD/
SCID(Il2rg2/2)22 mice. Mice were killed 4 weeks later and nodules found at
the site of injections processed for histological analysis.
Secondary tumorigenicity assays were performed as described34. Briefly, primary
tumours formed by H-rasV12-expressing HKCs or SCC cells in mice plus/minus
CsA treatment were isolated, 6 weeks after cell injection, and dissociated into single
cells by collagenase digestion (0.35% solution in HBSS; Sigma) for 45 min at 37 uC.
After filtering of the cell suspension, recovered cells were washed with keratinocyte
medium, counted and admixed in various numbers with a fixed amount of HKCs
(2 3 105 cells) before injection at the dermal–epidermal junction of NOD/SCID
mice. Four weeks after injection, mice were killed for histological analysis. Similar
experiments were performed with freshly dissociated cells from primary tumours
formed by H-rasV12-expressing HKCs or SCC13 cells plus/minus increased ATF3
expression.
31. Lazarov, M. et al. CDK4 coexpression with Ras generates malignant human
epidermal tumorigenesis. Nature Med. 8, 1105–1114 (2002).
32. Tamura, K. et al. Stress response gene ATF3 is a target of c-myc in serum-induced
cell proliferation. EMBO J. 24, 2590–2601 (2005).
33. Nguyen, B. C. et al. Cross-regulation between Notch and p63 in keratinocyte
commitment to differentiation. Genes Dev. 20, 1028–1042 (2006).
34. Malanchi, I. et al. Cutaneous cancer stem cell maintenance is dependent on
b-catenin signalling. Nature 452, 650–653 (2008).
doi:10.1038/nature08996

35. Chen, C. R., Kang, Y. & Massague, J. Defective repression of c-myc in breast
cancer cells: a loss at the core of the transforming growth factor beta growth
arrest program. Proc. Natl Acad. Sci. USA 98, 992–999 (2001).
36. Topley, G. I., Okuyama, R., Gonzales, J. G., Conti, C. & Dotto, G. P. p21WAF1/Cip1
functions as a suppressor of malignant skin tumor formation and a determinant of
keratinocyte stem-cell potential. Proc. Natl Acad. Sci. USA 96, 9089–9094
(1999).
37. Dotto, G. P., Moellmann, G., Ghosh, S., Edwards, M. & Halaban, R. Transformation
of murine melanocytes by basic fibroblast growth factor cDNA and oncogenes
and selective suppression of the transformed phenotype in a reconstituted
cutaneous environment. J. Cell Biol. 109, 3115–3128 (1989).
38. Dotto, G. P., Weinberg, R. A. & Ariza, A. Malignant transformation of mouse
primary keratinocytes by Harvey sarcoma virus and its modulation by
surrounding normal cells. Proc. Natl Acad. Sci. USA 85, 6389–6393 (1988).

©2010 Macmillan Publishers Limited. All rights reserved

Vol 465 | 20 May 2010 | doi:10.1038/nature08994
Myosin II contributes to cell-scale actin network
treadmilling through network disassembly
Cyrus A. Wilson1{, Mark A. Tsuchida1, Greg M. Allen1, Erin L. Barnhart1, Kathryn T. Applegate2, Patricia T. Yam1{,
Lin Ji2, Kinneret Keren1{, Gaudenz Danuser2{ & Julie A. Theriot1,3
Crawling locomotion of eukaryotic cells is achieved by a process
dependent on the actin cytoskeleton1: protrusion of the leading edge
requires assembly of a network of actin filaments2, which must be
disassembled at the cell rear for sustained motility. Although ADF/
cofilin proteins have been shown to contribute to actin disassembly3,
it is not clear how activity of these locally acting proteins could be
coordinated over the distance scale of the whole cell. Here we show
that non-muscle myosin II has a direct role in actin network dis-
assembly in crawling cells. In fish keratocytes undergoing motility,
myosin II is concentrated in regions at the rear with high rates of
network disassembly. Activation of myosin II by ATP in detergent-
extracted cytoskeletons results in rear-localized disassembly of the
actin network. Inhibition of myosin II activity and stabilization of
actin filaments synergistically impede cell motility, suggesting the
existence of two disassembly pathways, one of which requires myo-
sin II activity. Our results establish the importance of myosin II as an
enzyme for actin network disassembly; we propose that gradual
formation and reorganization of an actomyosin network provides
an intrinsic destruction timer, enabling long-range coordination of
actin network treadmilling in motile cells.
A thorough understanding of actin polymerization-based crawling
at the whole-cell scale has long been a challenge in the field of cell
biology. A combination of biochemical and cell-biological experiments
has led to a consensus model for the mechanism by which steady-state
actin network treadmilling in the lamellipodia of motile cells contri-
butes to protrusion of the leading edge: new filaments are nucleated
near the leading edge, resulting in the assembly of a branched actin
network, which is eventually disassembled by ADF/cofilin proteins,
replenishing the pool of polymerizable actin monomers3.
However, extending this model from the micrometre scale of the
leading edge lamellipodium to the tens-of-micrometres scale of an
entire cell requires a mechanism for longer-range coordination: a
cell-scale spatial organization of network assembly and disassembly
processes, giving rise to sustained whole-cell motility. The molecular
mechanisms of actin network disassembly and its spatial regulation
in motile cells are not completely understood (Supplementary
Information).
Here we use fish epidermal keratocytes (Fig. 1a) as a model system
to investigate the spatial regulation of actin network turnover in
motile cells. These cells are fast-moving with persistent velocity and
shape4 and maintain a continuous actin network throughout the
lamellipodium5,6 (Fig. 1b), implying that the net rates of assembly
at the front and disassembly at the rear must be closely and constantly
coordinated. This property allows us to analyse network turnover
based on steady-state measurements.
1
Department of Biochemistry and Howard Hughes Medical Institute, Stanford University School of Medicine, Stanford, California 94305, USA. 2Department of Cell Biology, The
Scripps Research Institute, La Jolla, California 92037, USA. 3Department of Microbiology and Immunology, Stanford University School of Medicine, Stanford, California 94305, USA.
{Present addresses: Institute for Creative Technologies, University of Southern California, Marina del Rey, California 90292, USA (C.A.W.); Montreal Neurological Institute, Montreal,
Quebec, H3A 2B4, Canada (P.T.Y.); Department of Physics and the Russell Berrie Nanotechnology Institute, Technion – Israel Institute of Technology, Haifa 32000, Israel (K.K.);
Department of Cell Biology, Harvard Medical School, Boston, Massachusetts 02115, USA (G.D.).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
To investigate the spatial organization of actin network movement
and dynamics, we performed fluorescence speckle microscopy (FSM)
on cells moving at steady state (Fig. 1b). The direction and speed of
actin network movement was determined by speckle flow tracking7 as
a function of position within the lamellipodium (Fig. 1c). Consistent
with photoactivation experiments2, we found that the actin network
in the lamellipodium remains nearly stationary with respect to the
substrate, with minimal retrograde flow (Fig. 1c). At the cell rear, the
actin network moved forward and rapidly inward from the sides.
To analyse the movement of the actin network relative to the
boundaries of these fast-moving cells, we represented the FSM flow
field in the cell’s moving frame of reference8 (Fig. 1d and Supplemen-
tary Movie 1). In the cell frame of reference, movement of the net-
work appeared uniform and rearward in the front of the cell, and
almost completely perpendicular to the direction of motion at the
a Lamellipodium
b Lamellipodium
Cell body Cell body
10 µm
c d
0.25
µm s–1 f YFP–myosin distribution
e
(Median projection)
Assembly Disassembly
Figure 1 | Myosin II in keratocytes co-localizes with the primary sites of
actin network disassembly. a–e, Data and analysis of a single live
keratocyte. a, Phase-contrast image of the keratocyte moving upwards.
b, FSM image of the actin network labelled with a low concentration of
phalloidin. c, F-actin flow field based on speckle tracking, in the laboratory
frame of reference. Arrow length and colour both indicate the speed of actin
network flow. d, F-actin flow field in the cell frame of reference. e, Steady-
state net F-actin assembly and disassembly. f, Fluorescence image of YFP-
tagged myosin regulatory light chain in a keratocyte of similar size and shape
to that shown in a–e. Myosin II is found at low levels throughout the
lamellipodium and at the highest concentrations in two foci flanking the cell
body, which coincide with the primary sites of actin network disassembly as
shown in e.
373
LETTERS
rear sides. Under the cell body, network flow ceased without signifi-
cantly changing its direction.
The pattern of fluorescent speckle motion suggests that net actin
network assembly occurred in the front of the cell and net disassembly
in the rear. This was confirmed by calculating the spatial distribution
of net filamentous actin (F-actin) assembly and disassembly, using
actin speckle density and the divergence of the FSM flow field9 (Fig. 1e
and Supplementary Fig. 2a). Intriguingly, we found that the rear-
localized pattern of disassembly, with two foci flanking the cell body,
was strongly reminiscent of the distribution of myosin II, as visualized
by yellow fluorescent protein (YFP)-tagged myosin II regulatory light
chain (Fig. 1f).
Although the role of myosin II in the rear of motile cells is con-
ventionally associated with mechanical force generation and contrac-
tion (Supplementary Information), several lines of evidence raise the
possibility that myosin II may also have a specific role in driving actin
network disassembly. Spatially correlated contraction and actin
depolymerization in motile cells has been shown to be promoted
by a drug that stimulates myosin II activity9. In cytokinesis, where
contraction of the cleavage furrow is driven in part by myosin II
motor activity, inhibition of myosin II leads to increased accumula-
tion of F-actin in the cleavage furrow, consistent with a role for
myosin II in regulating or catalysing F-actin disassembly in the con-
tractile ring10,11. During the early steps of inner-ear morphogenesis,
myosin II activity is required for depletion of F-actin from the basal
cortex of an ectodermal sheet12. More directly, myosin II has been
shown to potentiate severing and disassembly of actin filament bundles
at the base of neuronal growth cones13. Finally, in a recent in vitro
experiment, a loss of actin filaments in solution was observed in the
presence of purified myosin II14.
Our observation of disassembly co-localized with myosin II density
suggested that such a link between myosin II activity and actin net-
work disassembly might be at work in keratocyte motility. To investi-
gate this possibility, we treated keratocytes with the myosin II ATPase
inhibitor blebbistatin15. Although blebbistatin-treated cells continued
to move with little shape change at the leading edge (Fig. 2 and
Supplementary Movie 1) and with only a moderate decrease in speed
(Fig. 3f), the pattern of disassembly was markedly altered: the two foci
of disassembly adjacent to the cell body disappeared, and disassembly
was distributed along the trailing edge (Fig. 2e and Supplementary
Information).
Continued motility of keratocytes under blebbistatin treatment
indicates that at least some disassembly activity of the actin network
was retained. We judged that inhibition of myosin II was essentially
complete under the conditions of our blebbistatin treatment (50 mM
active enantiomer, 10 min) based on two criteria. First, two aspects of
motility that are normally attributed to myosin II activity (Sup-
plementary Information) were reduced below the detection limit:
inward flow in the cell rear (Fig. 2b–d and Supplementary Movie 1)
and inward traction force (Supplementary Fig. 3 and Supplementary
Movie 2). Second, increasing the blebbistatin concentration by two-
fold did not have a further effect on cell speed (data not shown). These
observations led us to hypothesize that there may be a parallel pathway
that disassembles the actin network independent of myosin II.
We thus proceeded to examine the effect of myosin II inhibition in
a sensitized background where actin depolymerization was partly
inhibited. The F-actin stabilizing drug jasplakinolide11,16, at 1 mM,
decreased cell speed moderately, an effect comparable to that of
50 mM blebbistatin (Fig. 3f). For cells pre-incubated with 50 mM
blebbistatin, however, 1 mM jasplakinolide dramatically inhibited
motility (Fig. 3a, b, f and Supplementary Movie 3), often causing a
rapid and complete cessation of both movement and actin network
flow (Fig. 3a, b and Supplementary Movie 3). To test the specificity of
this synthetic effect, we examined the effect of latrunculin A, which
slows actin network turnover by sequestering monomer (inhibiting
polymerization) rather than by inhibiting depolymerization.
Although 50 nM latrunculin A caused keratocytes to lose polarity
374
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
Before treatment 10 min after 50 µM blebbistatin
a Lamellipodium Lamellipodium
Cell body Cell body
b 10 µm
0.25
0 µm s–1
c
d
0.25 –0.25
µm s–1 µm s–1
e
Assembly Disassembly
f
0.3
Inward perpendicular
flow (µm s–1)
0.2
0.1
0
0 µM blebbistatin 50 µM blebbistatin
Figure 2 | Inhibition of myosin II blocks inward flow and alters the pattern
of disassembly of the actin network. a–e, Analysis of actin network flow in a
single keratocyte before (left) and approximately 10 min after (right)
addition of 50 mM blebbistatin. a, Raw F-actin speckle flow measurements
(yellow arrows) in the laboratory frame of reference, superimposed on the
corresponding FSM frames. b, Resampled flow field in the laboratory frame
of reference. c, Resampled flow field in the cell frame of reference. d, Maps
showing the component of network flow perpendicular to the direction of
cell movement (‘perpendicular flow’) in the cell frame of reference. Red
indicates F-actin flow towards the right; blue, towards the left. Actin network
movement in the green regions is parallel to the direction of cell locomotion.
Blebbistatin treatment abolishes inward flow. e, Steady-state assembly/
disassembly maps. Before blebbistatin treatment, the highest rate of
disassembly is found in two foci flanking the cell body. After blebbistatin
treatment, disassembly is distributed along the rear margin. f, Inward
perpendicular flow of the actin network in untreated (white circles; n 5 23)
and blebbistatin-treated (black triangles; n 5 8) cells. Error bars indicate
average 6 s.d. over each movie (typically 4 min). Measurements were made
before and after treatment for two of the cells; blue and black arrows connect
the corresponding data points. The data points connected by the blue arrow
correspond to the cell shown in a–e. Compare with Supplementary Movie 1.
and stop moving (data not shown), 5 nM latrunculin A caused a speed
decrease comparable to sole treatment with either 50 mM blebbistatin
or 1 mM jasplakinolide (Fig. 3f). Treatment with the combination of
blebbistatin and latrunculin A, or jasplakinolide and latrunculin A,
did not have a synthetic effect on cell speed (Fig. 3f). This suggested
that the synthetic effect was not simply a result of actin monomer
depletion, but specific to disassembly inhibition. Both blebbistatin-
treated cells and jasplakinolide-treated cells showed an unusual accu-
mulation of F-actin in the cell rear, but the patterns were distinct:
in blebbistatin-treated cells, F-actin accumulated primarily along
the trailing edge and in tail-like structures (Fig. 3d, g), whereas in
jasplakinolide-treated cells F-actin accumulated under the cell body
(Fig. 3e, g). These observations are consistent with the existence of
NATURE | Vol 465 | 20 May 2010 LETTERS

a b Actin flow-field f 0.3 n = 615 78 65 83 41 55 30

magnitude

Cell speed (µm s–1)

0.2
Maximum

50 µM blebbistatin Third quartile

0 µM jasplakinolide 0.1 Mean ± 95% CI
0.25 µm s–1 First quartile

Minimum
0.0

ed

m lin

lin

en ide

ul e
lid
si ati

nc lid
n o cu
)

cu

II)

+ pla s)

in
at

er

l
no
yo st

am o

tru no
t
re

n
n

un

fil kin
m bbi
m tru

la k i
nt

k
r

la

tin pla
t
U

tin La

la

its le
o

sp
ib B

ac as

s
+

ja

Ja
tin

es J
+
ac

ta

in
h
is
Significant increase

(in

at
rs
50 µM blebbistatin

bb

st
e

liz
st

bi
from all other groups

bi
1 µM jasplakinolide

Bl
ue

eb
0 µm s–1

ta
Significant decrease

eq

Bl

(s
(s
from all other groups
10 µm g 2.0 n = 134 108 64 4.0 (P < 0.05)

F-actin accumulation (a.u.)

Maximum
1.5 3.0
Third quartile
1.0 2.0 Mean ± 95% CI
First quartile
0 µM blebbistatin
1 µM jasplakinolide 0.5 1.0
Minimum
c d e
0.0 0.0

in

in
e

e
d

d
lid

lid
at

at
te

te
st

st
a

a
no

no
re

re
bi

bi
ki

ki
nt

nt
eb

eb
la

la
U

U
sp

sp
Bl

Bl
10 µm Untreated Blebbistatin Jasplakinolide

Ja

Figure 3 | Jasplakinolide specifically halts actin dynamics of cells in which
myosin II is inhibited. a, Raw F-actin speckle flow measurements in the cell
frame of reference (yellow arrows) superimposed on the corresponding FSM
frames, for a cell in 50 mM blebbistatin before (top) and approximately 2 min
after (middle) addition of 1 mM jasplakinolide. Bottom, a separate cell in
jasplakinolide alone. b, F-actin flow magnitude maps corresponding to
a. The combination of blebbistatin and jasplakinolide immobilizes the actin
network, an effect that is not achieved by either drug alone. c–e, Fixed,
phalloidin-labelled keratocytes, untreated (c) or treated with 50 mM
blebbistatin (d) or 1 mM jasplakinolide (e). Consistent with impaired actin-
network disassembly, blebbistatin-treated cells accumulate F-actin along the
rear margin, jasplakinolide-treated cells underneath the cell body.
f, Treatment with either 5 nM latrunculin A, 50 mM blebbistatin or
Ja
1 mM jasplakinolide can slow cells relative to the control population. The
combination of blebbistatin and latrunculin A or jasplakinolide and
latrunculin A has no significant further effect on cell speed than either drug
alone. The combination of blebbistatin and jasplakinolide significantly
(P , 0.05 by Tukey’s test) and synergistically slows cell locomotion.
g, Blebbistatin causes significant (P , 0.05 by Tukey’s test) F-actin
accumulation in the trailing ‘tails’ (but not in the cell body), whereas
jasplakinolide causes accumulation underneath the cell body (but not in the
‘tails’), relative to untreated cells. a.u., Arbitrary units (see Methods).
Compare c–e. Box-and-whisker plots (f and g) indicate the mean, 95%
confidence interval (CI), extrema and quartiles for the indicated number of
cells (n) in each treatment group. Compare with Supplementary Movie 3.

two complementary and partly redundant mechanisms for actin net-
work disassembly: one dependent on myosin II activity, the other
inhibited by jasplakinolide.
Our in vivo experiments do not distinguish between a direct or
indirect contribution of myosin II activity to actin network disassem-
bly. To investigate the specific contribution of myosin II activity, it
was essential to analyse disassembly of the rear network in isolation—
in the absence of the continuous replenishment that occurs in a live
forward-moving cell—and to exclude the action of cytosolic factors.
Accordingly, we performed experiments using isolated cytoskeletons
of Triton X-100-extracted keratocytes6. Strikingly, addition of ATP
to the extracted cytoskeletons caused not only a rapid inward con-
traction at the cell rear but also a concomitant dissolution of the actin
network in that region (Fig. 4a–d, Supplementary Movie 4 and Sup-
plementary Fig. 6). The leading edge of the cell was not significantly
affected by this process. The rear-dominant disassembly of actin (as
well as contraction) was significantly suppressed (sometimes to
undetectable levels) when cells were incubated with blebbistatin
before the detergent extraction (Fig. 4e–h, Supplementary Movie 4
and Supplementary Fig. 6). The ATP dependence, blebbistatin sensi-
tivity, rear localization and synchrony with contraction of the net-
work disassembly are consistent with myosin II as the active agent,
and recapitulation of this myosin-II-dependent disassembly in
extracted cells strongly suggests that soluble proteins are not required
for this process.
To test for the possibility that an unrelated mechanism was pro-
tecting the actin network in the front of the cell from disassembly, we
treated extracted cytoskeletons with the gelsolin-family Ca21-
dependent actin severing protein villin17,18, in the absence of ATP.
Glutathione S-transferase (GST)–villin solubilized F-actin in most
©2010 Macmillan Publishers Limited. All rights reserved
regions of the cell, including the front (Fig. 4i–l and Supplementary
Movie 4). Interestingly, however, a portion of the network in the rear
was relatively insensitive to GST–villin. This region of the network,
which lingered during prolonged villin treatment (Supplementary
Movie 4), was reminiscent of the pattern of myosin II localization
(Fig. 1f) and net actin network disassembly in intact cells (Fig. 1e) and
complementary to the pattern of actin network stability after ATP
treatment. Addition of both ATP and GST–villin, in either order, gave
nearly complete destruction of the actin cytoskeleton (Fig. 4m–t and
Supplementary Movie 4). Thus the rear-dominant pattern of ATP-
dependent disassembly (Fig. 4a–d) is not due to other regions being
impervious to disassembly per se; it probably reflects regions where
myosin II is abundant and integrated into the actin network in a
configuration that brings about maximal disassembling activity.
Our combined experimental observations point to a direct role for
myosin II in actin network disassembly in motile keratocytes of fish.
A second, unidentified disassembly mechanism appears to act in
parallel with myosin II in unperturbed cells. We found this pathway
to be sensitive to jasplakinolide, and we conjecture that it may involve
cofilin or gelsolin activity (Supplementary Information). Previous
reports showing that F-actin disassembly increases upon stimulation
of myosin II phosphorylation9 and that jasplakinolide selectively
halts actin turnover in regions of cells lacking myosin II19 are con-
sistent with these conclusions.
The observed spatial and temporal coincidence of contraction and
disassembly (Figs 1 and 4; ref. 9) is most consistent with a simple
mechanism for network disassembly by myosin II: mechanical break-
age of actin filaments20. Such an effect could be assisted by the ability
of the non-muscle myosin II motor domain to bind actin filaments
stably when under tension21, and may represent a common feature of
375
LETTERS NATURE | Vol 465 | 20 May 2010

DMSO, ATP Blebbistatin, ATP Villin, no ATP Villin, then ATP ATP, then villin
a e i m q

t=0
b f j n r
t = 420 s

c g k o s
Overlay

5 µm

d Buffer ATP h l p Buffer Villin t

Normalized intensity (a.u.)

Buffer ATP Buffer Villin Buffer ATP

Villin
1 ATP
0.8
0.6
0.4 Whole cell
Front region
0.2 Rear region
0
0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800
Time (s) Time (s) Time (s) Time (s) Time (s)

Figure 4 | Actin network disassembly in the rear of detergent-extracted arrowhead) are indicated. e–h, In a cell treated with 50 mM blebbistatin for
keratocyte cytoskeletons is ATP dependent and blebbistatin sensitive, 30 min before extraction, addition of ATP does not induce a loss of actin
consistent with a direct role for myosin II in this process. a–d, ATP triggers network. There is a slow loss of fluorescence owing to photobleaching or
an acute loss of actin network in the rear region of the cell, where myosin II is background dissociation. i–l, The F-actin severing protein villin rapidly
localized (compare with Fig. 1f). a, A detergent-extracted and phalloidin- disassembles the lamellipodial actin network, demonstrating that this part of
labelled keratocyte cytoskeleton6. b, The same cytoskeleton 7 min after the cytoskeleton is not protected against a general disassembling activity.
addition of 1 mM ATP. c, Overlay of initial frame (a, cyan) and frame at GST–villin (0.1 mM) was added instead of ATP (arrow in l). m–t, Addition of
7 min (b, yellow); regions with increase, decrease or no change in net GST–villin (arrows in p, t) in addition to ATP (arrowheads in p, t), in either
intensity appear yellow, cyan or white, respectively. d, Time evolution of order, results in complete, rapid disassembly of the actin network. Compare
fluorescence intensities (normalized at t 5 0) in the indicated regions. Time with Supplementary Movie 4.
points for a mock buffer wash (chevron) and ATP addition (black

myosin II function in all but the most specialized organization found correlation algorithm described in ref. 7. For comparisons between cells and over
in striated muscle. time, velocity fields were binned into five subcellular regions. Time-averaged
The contribution of myosin-II-driven network disassembly to velocity fields (v) and F-actin signal (r) were used to calculate steady-state maps
of net F-actin turnover by calculating =?(rv) (similar to ref. 9).
whole-cell motility may vary in different cell types, depending on
Pharmacological treatments were performed while imaging, to capture changes
the spatial arrangement of the cytoskeletal elements. In keratocytes, as drugs took effect, by changing the buffer to one containing a new drug or drug
we find myosin II to be concentrated at the cell rear and to contribute combination. We inhibited myosin II with 100 mM (6)-blebbistatin or 50 mM
to cell-scale network treadmilling; given the similar localization of (2)-blebbistatin (the active enantiomer, whose concentration is indicated in the
myosin II to the rear in Dictyostelium22 and neutrophils23, we suggest text, figures and movies). To examine the effects on F-actin distribution, we
that its role there may be very similar. In contrast, myosin II in treated keratocytes with blebbistatin or jasplakinolide, fixed them with 4%
neurons appears to be involved in network disassembly at the base formaldehyde, permeabilized with 0.5% Triton X-100, and labelled with
of growth cones13. In fibroblasts and larger epithelial cells, the con- TRITC-phalloidin. Myosin II localization was visualized by transfection26 with
centration of myosin II in the lamella (behind the lamellipodium) YFP-tagged Xenopus laevis myosin II regulatory light chain25.
may primarily contribute to local network disassembly in the transi- Detergent-extracted cytoskeletons were obtained by the method of Svitkina et
al.6 and labelled with TRITC-phalloidin. The acute changes upon addition of
tion zone between the lamellipodial and lamellar networks19,24.
1 mM ATP or 0.1 mM GST–villin were observed by epifluorescence imaging at a
The involvement of myosin II in actin network disassembly moti- 10-s time interval. An ATP-regenerating system was used to preclude inhibition
vates an appealing model for the long-range spatial and temporal of myosin II by accumulated ADP.
coordination of disassembly in motile cells (Supplementary Fig. 1):
slow incorporation of myosin II into the network and subsequent Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
myosin-II-mediated rearrangement of F-actin occur over the same
time scale as whole cell translocation6, serving as a timing mech- Received 22 June 2007; accepted 4 March 2010.
anism. Progressively assembled myosin II bipolar filaments gradually
coalesce the initially dendritic actin network into a more parallel 1. Abercrombie, M. The Croonian lecture, 1978: the crawling movement of
metazoan cells. Proc. R. Soc. Lond. B 207, 129–147 (1980).
organization6 that allows efficient action of the myosin II motor, 2. Theriot, J. A. & Mitchison, T. J. Actin microfilament dynamics in locomoting cells.
eventually leading to disassembly. The spatiotemporal organization Nature 352, 126–131 (1991).
of myosin II incorporation and actin network rearrangement takes 3. Pollard, T. D. & Borisy, G. G. Cellular motility driven by assembly and disassembly
place at a large enough scale to serve as a basis for cell-scale tread- of actin filaments. Cell 112, 453–465 (2003).
milling of the actin cytoskeleton during steady-state motility in these 4. Euteneuer, U. & Schliwa, M. Persistent, directional motility of cells and
cytoplasmic fragments in the absence of microtubules. Nature 310, 58–61 (1984).
cells. 5. Small, J. V., Herzog, M., Häner, M. & Abei, U. Visualization of actin filaments in
keratocyte lamellipodia: negative staining compared with freeze-drying. J. Struct.
METHODS SUMMARY Biol. 113, 135–141 (1994).
Keratocytes from the Central American cichlid Hypsophrys nicaraguensis were 6. Svitkina, T. M., Verkhovsky, A. B., McQuade, K. M. & Borisy, G. G. Analysis of the
prepared for FSM by introducting Alexa Fluor 546 phalloidin by small-volume actin-myosin II system in fish epidermal keratocytes: mechanism of cell body
translocation. J. Cell Biol. 139, 397–415 (1997).
electroporation as described25. We acquired phase-contrast and epifluorescence
7. Ji, L. & Danuser, G. Tracking quasi-stationary flow of weak fluorescent signals by
images through a 360 (numerical aperture 1.4) objective at 2-s time intervals adaptive multi-frame correlation. J. Microsc. 220, 150–167 (2005).
unless otherwise noted. Using phase-contrast images, we tracked whole-cell 8. Wilson, C. A. & Theriot, J. A. A correlation-based approach to calculate rotation
movement and calculated transformations between the laboratory and cell and translation of moving cells. IEEE Trans. Image Process. 15, 1939–1951 (2006).
frames of reference, as described8. We performed F-actin flow tracking on fluor- 9. Vallotton, P., Gupton, S. L., Waterman-Storer, C. M. & Danuser, G. Simultaneous
escence images in the cell frame of reference using the adaptive multi-frame mapping of filamentous actin flow and turnover in migrating cells by quantitative
376
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010
fluorescent speckle microscopy. Proc. Natl Acad. Sci. USA 101, 9660–9665
(2004).
10. Guha, M., Zhou, M. & Wang, Y.-L. Cortical actin turnover during cytokinesis
requires myosin II. Curr. Biol. 15, 732–736 (2005).
11. Murthy, K. & Wadsworth, P. Myosin-II-dependent localization and dynamics of
F-actin during cytokinesis. Curr. Biol. 15, 724–731 (2005).
12. Sai, X. & Ladher, R. K. FGF signaling regulates cytoskeletal remodeling during
epithelial morphogenesis. Curr. Biol. 18, 976–981 (2008).
13. Medeiros, N. A., Burnette, D. T. & Forscher, P. Myosin II functions in actin-bundle
turnover in neuronal growth cones. Nature Cell Biol. 8, 215–226 (2006).
14. Haviv, L., Gillo, D., Backouche, F. & Bernheim-Groswasser, A. A cytoskeletal
demolition worker: myosin II acts as an actin depolymerization agent. J. Mol. Biol.
375, 325–330 (2008).
15. Straight, A. F. et al. Dissecting temporal and spatial control of cytokinesis with a
myosin II inhibitor. Science 299, 1743–1747 (2003).
16. Bubb, M. R., Senderowicz, A. M., Sausville, E. A., Duncan, K. L. & Korn, E. D.
Jasplakinolide, a cytotoxic natural product, induces actin polymerization and
competitively inhibits the binding of phalloidin to F-actin. J. Biol. Chem. 269,
14869–14871 (1994).
17. Bretscher, A. & Weber, K. Villin is a major protein of the microvillus cytoskeleton
which binds both G and F actin in a calcium-dependent manner. Cell 20, 839–847
(1980).
18. Glenney, J. R. Jr, Kaulfus, P. & Weber, K. F actin assembly modulated by villin:
Ca11-dependent nucleation and capping of the barbed end. Cell 24, 471–480
(1981).
19. Ponti, A., Machacek, M., Gupton, S. L., Waterman-Storer, C. M. & Danuser, G.
Two distinct actin networks drive the protrusion of migrating cells. Science 305,
1782–1786 (2004).
20. Kron, S. J. & Spudich, J. A. Fluorescent actin filaments move on myosin fixed to a
glass surface. Proc. Natl Acad. Sci. USA 83, 6272–6276 (1986).
21. Kovács, M., Thirumurugan, K., Knight, P. J. & Sellers, J. R. Load-dependent mechanism
of nonmuscle myosin 2. Proc. Natl Acad. Sci. USA 104, 9994–9999 (2007).
22. Yumura, S., Mori, H. & Fukui, Y. Localization of actin and myosin for the study of
ameboid movement in Dictyostelium using improved immunofluorescence. J. Cell
Biol. 99, 894–899 (1984).
23. Keller, H. U. & Niggli, V. Colchicine-induced stimulation of PMN motility related to
cytoskeletal changes in actin, a-actinin, and myosin. Cell Motil. Cytoskeleton 25,
10–18 (1993).
24. Vicente-Manzanares, M., Zareno, J., Whitmore, L., Choi, C. K. & Horwitz, A. F.
Regulation of protrusion, adhesion dynamics, and polarity by myosins IIA and IIB
in migrating cells. J. Cell Biol. 176, 573–580 (2007).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
25. Yam, P. T. et al. Actin–myosin network reorganization breaks symmetry at the cell
rear to spontaneously initiate polarized cell motility. J. Cell Biol. 178, 1207–1221
(2007).
26. Lacayo, C. I. et al. Emergence of large-scale cell morphology and movement from
local actin filament growth dynamics. PLoS Biol. 5, e233 (2007).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank M. J. Footer for the gift of purified GST–villin,
A. Mogilner and S. Sivaramakrishnan for discussions, and Z. Pincus, N. Dye and
T. Y.-C. Tsai for reading the manuscript. C.A.W. and J.A.T. were supported by
National Institutes of Health grant R01AI067712 (J.A.T.). M.A.T. and E.L.B. were
supported by National Institutes of Health grant T32GM007276. G.M.A. was
supported by the Stanford Medical Scientist Training Program. P.T.Y. was
supported by a Howard Hughes Medical Institute Predoctoral Fellowship,
Stanford Graduate Fellowship and a Skye International Foundation Scholarship.
K.T.A. was supported by a National Science Foundation Graduate Research
Fellowship. K.K. was a Damon Runyon Fellow supported by the Damon Runyon
Cancer Research Foundation (DRG-#1854-05). K.T.A., L.J. and G.D. were
supported by National Institutes of Health grants U54GM64346 and
U01GM67230 (G.D.). J.A.T. was supported by the Howard Hughes Medical
Institute.
Author Contributions C.A.W., M.A.T. and J.A.T. conceived and designed the
experiments. C.A.W. and P.T.Y. performed FSM on untreated motile keratocytes.
C.A.W. performed the pharmacological manipulation experiments, FSM
observation and the analysis. L.J. and G.D. developed the flow tracking algorithm
specific to the needs of this analysis. C.A.W. and P.T.Y. developed methods and
software to integrate the flow tracking algorithm with these experiments and
analysis. K.T.A., C.A.W. and G.D. developed algorithms for the F-actin turnover
analysis. E.L.B. imaged myosin II localization. G.M.A., K.K. and E.L.B. collected the
cell speed data and observed fixed cells under the different treatments; G.M.A. and
C.A.W. analysed these data. M.A.T. performed experiments on
detergent-extracted cytoskeletons and analysed the results. M.A.T., C.A.W. and
J.A.T. wrote the paper. All authors discussed the results and commented on the
manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to J.A.T.
(theriot@stanford.edu).
377
doi:10.1038/nature08994
METHODS
Keratocyte cultures and labelling. We cultured keratocytes from the scales of
the Central American cichlid Hypsophrys nicaraguensis as described in ref. 25.
Keratocytes from this species exhibit less cell-to-cell variability than those of
goldfish (Carassius auratus) and black tetra (Gymnocorymbus ternetzi). We used
cells that had been cultured for 18–36 h. Before microscopic observation, we
treated the cells with 85% PBS containing 2.5 mM EGTA for approximately
5 min to break up sheets and obtain individually crawling cells. Only persistently
polarized, fast-moving ‘coherent’ cells were analysed, as this subset of keratocytes
maintains the most consistent shape and speed over time26.
F-actin and myosin II were labelled in live keratocytes as described25 by small-
volume electroporation on the coverslip. For F-actin speckle visualization, we
introduced Alexa Fluor 546 phalloidin (Molecular Probes) into the cells at low
density. Cells were allowed to recover for approximately 15 min in complete
medium. For myosin II visualization, we transfected keratocytes26 with YFP-
tagged X. laevis non-muscle myosin II regulatory light chain (a gift of A.F.
Straight). Cells were allowed to recover for 18–24 h before treatment with EGTA.
Microscopy. We imaged cells at room temperature on a Nikon Diaphot-300
inverted microscope with a 360 (numerical aperture 1.4) objective. We acquired
phase contrast and epifluorescence images at 2-s (live cell experiments) or 10-s
(extracted cytoskeleton experiments) time intervals with a 16-bit 512 pixel 3 512
pixel cooled back-thinned charge-coupled device camera (Princeton
Instruments). One pixel corresponded to 0.112 mm. All raw images displayed
are single-image frames, except for Fig. 1f, which is a median projection from a
90-s movie (45 frames). Alexa Fluor 546 phalloidin and tetramethylrhodamine B
isothiocyanate (TRITC)-phalloidin were excited with a narrow band excitation
TRITC filter cube (Chroma Technology) to minimize light absorption by
blebbistatin, thereby limiting photoinactivation27 and phototoxicity28.
Flow tracking. We measured movement of the actin network by using the
adaptive multi-frame correlation algorithm with modifications described in
ref. 25. We used a five-frame (10-s) temporal window, and a correlation template
size adaptively adjusted between a minimum of 11 pixels 3 11 pixels and a maxi-
mum of 21 pixels 3 21 pixels. This algorithm assumes quasi-stationary network
movement within the area of the template and over the duration of the temporal
window. Motile keratocytes moved so rapidly that over the five-frame time
window the requirement for stationarity could not be fulfilled. On the other
hand, integration over five frames was necessary owing to the relatively low
signal-to-noise ratio of the FSM images. Therefore, we performed the F-actin
flow tracking in the cell frame of reference, in which the actin meshwork move-
ment was approximately constant over a local neighbourhood in space and time.
Speckle flow could then be tracked using the methods described in ref. 7. We
removed isolated or grossly incoherent vectors from the raw flow data before
resampling the flow field on a regular grid for visualization and further proces-
sing. We used the term ‘perpendicular flow’ to denote the component of flow
perpendicular to the direction of cell movement.
Frames of reference. To define the transformations between the laboratory and
cell reference frames, as well as to calculate cell speeds, we tracked whole-cell
movement in the phase-contrast image sequences using the two-stage cross-
correlation approach8. This track of cell movement approximated as rigid
motion (translation and rotation) was then used to convert image sequences
from the laboratory frame of reference (in which they were acquired) to the cell
frame of reference for flow tracking; and then to convert flow measurements
from the cell frame of reference to the laboratory frame of reference.
Region analysis. As different cells had slightly different sizes and shapes, com-
parison or averaging of F-actin velocity fields among different cells was difficult.
Instead of using the velocity fields directly, we divided them into a set of five
standard regions (illustrated in Supplementary Fig. 4b) in each cell and averaged
the measurements within each region at each time point. This yielded a set of time
series that could further be averaged over time to summarize a cell’s steady-state
actin-network movement pattern as a set of five velocity vectors. The five regions we
manually defined in each cell were the following: rear left (when facing the direction
of movement), front left, front centre, front right and rear right. Inward perpen-
dicular flow for a cell was computed by taking the average of the perpendicular flow
in the rear left region and the negated perpendicular flow in the rear right region.
Assembly and disassembly maps. To estimate the spatial pattern of net assembly
and disassembly of the actin network, we used an algorithm based on steady-state
flow field and intensity. We applied spatial smoothing to the flow field
(resampled per pixel) and normalized the fluorescence image in each frame
before computing the average over the length of the movie (30 s in Fig. 1e;
1 min in Fig. 2e). We then multiplied the average smoothed velocity field (v)
by the average smoothed intensity (r) and calculated the divergence (=?(rv))
(similar to ref. 9) to obtain a steady-state assembly/disassembly map (see
Supplementary Information for further details).
©2010 Macmillan Publishers Limited. All rights reserved
Pharmacological treatments. We performed pharmacological treatments during
observation of the cells on the microscope, such that we were able to observe the
changes that occurred as the drug took effect. The coverslip was mounted in a
chamber for live-cell imaging with 1 ml medium above the coverslip. The medium
was exchanged with one containing the desired drug or drug combination. For
myosin II inhibition we used 100 mM ( 6 )-blebbistatin (a gift of A. F. Straight) or
50 mM (2)-blebbistatin (the active enantiomer; Toronto Research Chemicals).
We refer to the concentration of the active enantiomer in the main text. To
promote myosin II activity, we used 25 mM calyculin A (Upstate
Biotechnology). The F-actin stabilizing drug jasplakinolide (Molecular Probes)
was used at 1 mM and the actin monomer sequestering drug latrunculin A
(Molecular Probes) was used at 5 nM.
Fluorescence labelling of fixed cells. To examine the effect of blebbistatin or
jasplakinolide on F-actin distribution, we fixed keratocytes with 4% formalde-
hyde for 15 min after a 15 min treatment with blebbistatin or jasplakinolide. We
permeabilized the cells with 0.5% Triton X-100 in PBS for 10 min, blocked with
3% BSA in PBS with 0.1% Triton X-100 for 10 min, and labelled with 3.5 nM
TRITC-phalloidin for 15 min before observation.
F-actin accumulation analysis. The F-actin accumulation analysis required
additional steps, because fluorescence levels between cells could vary not only
due to F-actin amount, but also due to fixation variations, phalloidin labelling
efficiency and background fluorescence on the coverslip. We coarsely defined six
regions in the phase-contrast image: front left, centre, right, and rear left, centre,
right. To refine the regions, we manually segmented the cell interior by tracing its
outline in the phase-contrast image. Note that this segmentation followed the
overall contour of the cell without including any retraction fibres or ‘tails’.
The final regions were then defined as the intersection of the coarse regions
and the cell interior.
We defined an additional background region in the phalloidin fluorescence
image, and subtracted the median value of that region from the entire phalloidin
fluorescence image, clamping to zero. We then computed the relative F-actin
amount in each region as the ratio of the median phalloidin fluorescence in the
region to the median phalloidin fluorescence in the front centre region. These
ratios for the rear centre region are shown in Fig. 3g.
The rear ‘tails’ lie outside the cell interior traced as described above. We
defined a minimum threshold for true fluorescence signal as the first centile
value of (background-subtracted) fluorescence intensity inside the cell interior.
Pixel locations that were outside the cell interior, inside the coarsely defined rear
left and rear right regions and had (background-subtracted) values above the
minimum threshold were classified as the rear ‘tail’ regions. The ratios of their
median phalloidin fluorescence to median front-centre phalloidin fluorescence
are shown in Fig. 3g.
Detergent-extracted cytoskeletons. To obtain Triton X-100-extracted keratocyte
cytoskeletons, we cultured keratocytes on 22 mm 3 40 mm coverslips. After treat-
ment with EGTA, we attached the coverslips to microscope slides in orthogonal
orientation to construct flow chambers containing the cells. Parafilm spacers were
melted onto the slide at 80 uC before chamber assembly to form a 10-mm channel.
We used silicone grease to seal the chamber and to construct an approximately
3 mm 3 3 mm inlet port, which allowed fluid (including surfactant-containing
buffers) to flow gently through the chamber by gravity, minimizing disruption of
the specimen. We treated keratocytes essentially as in ref. 6 while observing them
on the microscope. After quick rinses with Hank’s balanced salt solution and
cytoskeleton buffer6 (50 mM imidazole, pH 7.4, 50 mM KCl, 0.5 mM MgCl2,
0.1 mM EDTA, 1 mM EGTA), we exposed the cells to extraction buffer (cyto-
skeleton buffer containing 1% Triton X-100, 4% polyethylene glycol, relative
molecular mass 40,000 and 0.2 mM TRITC-phalloidin) for approximately 30 s,
then rinsed thoroughly with cytoskeleton buffer containing 0.2 mM TRITC-phal-
loidin and incubated for 5 min. Extraction buffer and all subsequent buffers added
during the experiment contained an oxygen-scavenging system (2 mM DTT,
50 U ml21 glucose oxidase (Sigma), 1,500 U ml21 catalase (Sigma), 50 mM
D-glucose) to minimize photobleaching.
To test for actin network disassembly, we perfused the chamber with assay
buffer20 (25 mM imidazole, pH 7.4, 25 mM KCl, 4 mM MgCl2, 0.2 mM CaCl2)
and started time-lapse fluorescence imaging. Immediately after the indicated
time points, we added assay buffer or assay buffer containing an ATP-regenerating
system (1 mM ATP, 5 mM creatine phosphate (Sigma), 10 mg ml21 creatine
phosphokinase (Sigma)), 0.1 mM GST–villin (a gift of M. J. Footer) or both; the
chamber was fully perfused by the new buffer before the next image acquisition
occurred. To inhibit myosin II, we treated the cells with 50 mM blebbistatin before
extraction for 30 min; in these experiments, every subsequently applied buffer
contained 50 mM blebbistatin (or, in the control, the 0.1% dimethylsulphoxide
vehicle).
For the experiments whose results are shown in Supplementary Fig. 6, imaging
was performed through a 340 (numerical aperture 0.85) objective using a 16-bit
doi:10.1038/nature08994
1024 pixel 3 1024 pixel cooled back-thinned charge-coupled device camera
(Andor iXon1 DU-888).
Deformable substrates. We prepared gelatin-coated coverslips using a protocol
derived from that of Doyle & Lee29. Heated (40 uC) 2.5% gelatin (200 ml) was
added to each coverslip, then aspirated off, leaving a thin layer. Coverslips were
cooled at 4 uC for 24 h. We then pipetted 0.5 ml media onto the gelatin and
incubated it at 4 uC for 1 h. Afterwards we placed a fish scale and 27 ml media
atop the gelatin on each coverslip and incubated at room temperature overnight.
Before transferring a coverslip to the microscope for imaging, we incubated it
briefly in a suspension of streptavidin-conjugated quantum dots (Molecular
Probes). The quantum dots stuck non-specifically to the surface of the gelatin,
©2010 Macmillan Publishers Limited. All rights reserved
acting as fiducial marks to allow observation of substrate deformation. (Note
that because the streptavidin-conjugated quantum dots stuck non-specifically,
some stuck to the dorsal surface of the cells as well.)
27. Sakamoto, T., Limouze, J., Combs, C. A., Straight, A. F. & Sellers, J. R. Blebbistatin,
a myosin II inhibitor, is photoinactivated by blue light. Biochemistry 44, 584–588
(2005).
28. Kolega, J. Phototoxicity and photoinactivation of blebbistatin in UV and visible
light. Biochem. Biophys. Res. Commun. 320, 1020–1025 (2004).
29. Doyle, A. D. & Lee, J. Cyclic changes in keratocyte speed and traction stress arise
from Ca21-dependent regulation of cell adhesiveness. J. Cell Sci. 118, 369–379
(2005).
 Vol 465 | 20 May 2010 | doi:10.1038/nature09003

LETTERS
eIF5 has GDI activity necessary for translational
control by eIF2 phosphorylation
Martin D. Jennings & Graham D. Pavitt

In protein synthesis initiation, the eukaryotic translation ini- (GST–eIF5) purified from Escherichia coli (Fig. 1b and Supplemen-
tiation factor (eIF) 2 (a G protein) functions in its GTP-bound tary Figs 1 and 2). A significant and progressive stabilization of GDP
state to deliver initiator methionyl-tRNA (tRNAiMet) to the small binding to eIF2 was observed with increasing eIF5 (Koff reduced from
ribosomal subunit and is necessary for protein synthesis in all 0.12 to 0.077 min21 with equimolar eIF2 and eIF5), which
cells1,2. Phosphorylation of eIF2 [eIF2(aP)] is critical for trans- approached saturation at a 4:1 ratio. Importantly, GST alone did
lational control in diverse settings including nutrient deprivation, not have this activity (Fig. 1b). We observed the GDP stabilization
viral infection and memory formation3–5. eIF5 functions in start effect of eIF5 over a range of physiological magnesium concentra-
site selection as a GTPase accelerating protein (GAP) for the tions. The data also show that Mg21 is required for eIF5 GDI activity,
eIF2 N GTP N tRNAiMet ternary complex within the ribosome-bound as eIF5 has minimal GDI function in the absence of added Mg21.
pre-initiation complex6–8. Here we define new regulatory func- However, as Mg21 concentration is increased within the physio-
tions of eIF5 in the recycling of eIF2 from its inactive eIF2 N GDP logical range, Mg21 and eIF5 act together to stabilize GDP binding
state between successive rounds of translation initiation. First we to eIF2 (Supplementary Fig. 1). eIF5–Flag, purified from yeast,
show that eIF5 stabilizes the binding of GDP to eIF2 and is there- behaved identically to GST–eIF5 (Fig. 1c, rows 2 and 10). These
fore a bi-functional protein that acts as a GDP dissociation inhibi- experiments demonstrate that eIF5 can act as a GDI factor for
tor (GDI). We find that this activity is independent of the GAP eIF2 N GDP.
function and identify conserved residues within eIF5 that are The invariant Arg 15 residue in eIF5 is critical for eIF5-dependent
necessary for this role. Second we show that eIF5 is a critical GAP activity and the eIF5R15M mutation eliminates this function6–8.
component of the eIF2(aP) regulatory complex that inhibits the The R15M mutant retains full eIF5 GDI activity (Fig. 1c, row 3),
activity of the guanine-nucleotide exchange factor (GEF) eIF2B. demonstrating that GAP and GDI activities are independent. The
Together our studies define a new step in the translation initiation tif5-7A allele is a well-characterized eIF5 mutant in which seven
pathway, one that is critical for normal translational controls. evolutionarily-conserved residues in the carboxy-terminal domain
eIF2 N GTP is one of multiple translation initiation factors that has (CTD) are mutated to alanines15. These amino acid substitutions
an essential role in facilitating precise ribosome-bound tRNAiMet
recognition of initiator codons on mRNAs. A second factor, eIF5, a c
[3H]-GDP • elF2
binds eIF2 and accelerates GTP hydrolysis and release of both factors GDP
Control 1
before small and large ribosomal subunit joining2. Modulating the GDP
GDP
GDI eIF5
GST NTD LR CTD
GST–eIF5 2
GDP
activity of eIF2 via phosphorylation is a key control step in protein GDP Koff 1 152 240 405
Mg2+ GST–eIF5R15M 3
synthesis. eIF2a protein kinases respond to diverse cues and can R15M
mediate both general protein synthesis repression as well as trans- [3H]-GDP GST–eIF5W391F 4
W391F
39
39
lational activation of specific mRNAs, including those bearing two or GDP • elF2 GST–eIF5LR7A 5
more short upstream open reading frames (uORFs) in their leaders. LR7A
GST–eIF5NTD 6
eIF2(aP) inhibits the GEF activity of eIF2B blocking the reactivation
of eIF2 N GTP1,2. Recent observations suggested that eIF5 has addi- b 0.13 GST–eIF5NTD+LR 7
0.12 GST
tional functions in translation. Both eIF2 and eIF5 are dissociated GST–eIF5CTD 8
Koff GDP (min–1)

0.11 1:4
from ribosomes together9,10. In addition, in yeast, there is an abund- 0.10 GST–eIF5LR+CTD 9
ant cellular fraction of eIF2–eIF5 complexes that greatly exceeds the 0.09 1:1
eIF5–Flag 10
detectable levels of ternary complex11, suggesting that eIF5 may play a 0.08
1:2
part in the recycling of eIF2 N GDP. GDI proteins antagonize GEFs and 0.07 1:4 eIF57A–Flag 11
0.06
have been described for some G proteins12, although not for trans- GST–eIF5 7A
0.04
0.06
0.08
0.10
0.12
0.14
0.16

0.05
lation initiation. We therefore assessed if eIF5 possesses GDI activity. 0 50 100 150 200 250
Koff GDP (min–1)
GST protein (pmol)
Yeast eIF2abc complexes were purified using His-tagged eIF2c
(ref. 13). GDP release (Koff) was measured in a filter binding assay Figure 1 | eIF5 has GDI activity. a, Scheme for GDI activity assay.
where eIF2 N [3H]GDP complexes were dissociated in the presence of b, Increasing eIF5 stabilizes GDP-binding to eIF2. Koff GDP from 60 pmol
excess unlabelled GDP and magnesium ions (Fig. 1a). Increasing eIF2 with varying concentrations of GST–eIF5 (0–240 pmol, open circles) or
GST alone (filled circle). Molar eIF2:GST–eIF5 protein ratios are indicated.
Mg21 concentration progressively stabilized GDP binding to eIF2, c, Defining regions required for GDI activity. Mean Koff GDP (60 pmol eIF2)
as expected, because Mg21 helps coordinate the GDP-b phosphate– for indicated constructs derived from reactions with GST– or Flag–eIF5
eIF2c interaction14 (Supplementary Fig. 1). The effect of increasing proteins (120 pmol). Black bars represent a significant reduction in Koff GDP
concentrations of recombinant eIF5 on the GDP dissociation rate (P , 0.0001, unpaired Student’s t-test). Errors show standard deviation
was assessed using a glutathione S-transferase–eIF5 fusion protein (n . 3). Mg21 (2.9 mM) was used in b and c.
Faculty of Life Sciences, University of Manchester, Michael Smith Building, Oxford Road, Manchester M13 9PT, UK.
378
©2010 Macmillan Publishers Limited. All rights reserved
NATURE | Vol 465 | 20 May 2010 LETTERS

impair eIF5–eIF2 interactions11,15,16. We found that eIF57A–Flag retained eIF2-binding affinity (Fig. 2, lane 6) but eliminated GDI
eliminated GDI function, as did a single conservative substitution function (Fig. 1c, row 5). These analyses separate eIF2 binding and
within this motif at Trp 391 (eIF5W391F; Fig. 1c, rows 11 and 4). GST- GDI functions and identify conserved residues critical for GDI activity
protein affinity chromatography was performed to examine inter- within the LR and CTD.
actions between purified eIF5 and eIF2. This confirmed that As GDP binds eIF2c, we investigated whether eIF5LR1CTD could
GST–eIF5W391F reduces the affinity of eIF5 for purified eIF2, similar interact directly with eIF2c to mediate GDI function. The eIF5 CTD
to that described for the 7A mutant (Fig. 2, lane 5)15 and shows that was shown previously to interact with eIF2b (ref. 15). In addition,
the eIF2 binding domain provided by the eIF5-CTD is necessary for eIF2c was found to bind a construct containing eIF5 residues 1–279
GDI activity. but not residues 280–405 (ref. 19), that is, extending from the NTD
To delineate regions of eIF5 necessary for its GDI activity we used through our LR into the CTD. We assessed the ability of GST-fusion
sequence alignments and available structural information to define proteins to interact with eIF2 from yeast cell extracts expressing
the ‘amino-terminal domain’ (NTD, residues 1–152)17 and CTD c-Myc-63His-tagged eIF2c from either a single copy or high copy
(residues 241–405) of yeast eIF5 (ref. 18). The intervening sequence plasmid to provide a source of both eIF2abc complexes and excess
(residues 153–240) is defined here as the ‘linker region’ (LR). GST- free eIF2c. As expected, GST–eIF5 could specifically interact with
fusion proteins comprising these domains were assessed for eIF2 eIF2abc, as demonstrated by immunoblotting signals for the a and
interaction and GDI activity (Figs 1 and 2). Only GST–eIF5LR1CTD c subunits, and with the excess eIF2c when overexpressed (Fig. 3,
retained both activities, whereas GST–eIF5CTD could bind to eIF2, lanes 5–6). Consistent with Fig. 2, eIF5 constructs with an intact CTD
but did not retain GDI function. These experiments indicated that interacted with eIF2abc complexes in cell extracts similarly to full-
residues within the LR are required in addition to the CTD for GDI length wild-type eIF5, whereas the W391F mutation weakened
activity. Sequence alignments from diverse organisms identified a binding. However, among the mutants tested, only the eIF5LR1CTD
highly-conserved 20-residue stretch within the LR which we have construct containing an intact LR and CTD region stimulated bind-
called the ‘DWEAR’ motif after the five absolutely conserved residues ing to the excess eIF2c (lane 12). Thus the LR7A allele interacts with
it contains (Supplementary Fig. 3). A mutant allele was made within eIF2 complexes, but not with excess free eIF2c (lane 10). These results
full-length eIF5 in which seven DWEAR motif residues were mutated are consistent with eIF2b forming a major site for eIF5-CTD binding,
to alanines; we term this the LR7A allele (eIF5LR7A). GST–eIF5LR7A whereas conserved residues within the LR region define a new region
of eIF5 required for eIF2c interaction and GDI function.
To examine the effects of mutations on eIF5 function in vivo, the
GDT–eIF5NTD+LR

GST–eIF5LR+CTD

a
GST–eIF5W391F
GST–eIF5R15M

GST–eIF5LR7A

GST–eIF5NTD

GST–eIF5CTD

R15M, W391F and LR7A mutations were introduced into eIF5–Flag

eIF2 Input

GST–eIF5

on both single copy and high copy plasmids, then transformed into
an eIF5 deleted (tif5D) yeast strain by plasmid shuffling so that each
GST

kDa mutant became the sole source of eIF5. The R15M GAP mutant was
75 lethal, despite appropriate expression (Supplementary Fig. 4),
eIF2γ whereas W391F and LR7A mutants grew well (Fig. 4a). These data
50 demonstrate GAP activity is the essential function of eIF5 and imply
1 2 3 4 5 6 7 8 9 10 that these mutations eliminating GDI function have minimal impact
100% on GAP activity or growth under standard conditions.
75% Translation of GCN4 mRNA is highly sensitive to eIF2 activity and
50% eIF2γ is widely used as a reporter for the activity of translation initiation
factors20. Derepression of GCN4 translation can be scored by moni-
25%
toring cell growth in the presence of the HIS3 inhibitor 3-amino-
b 0%
triazole (3AT). Growth of wild-type cells on media containing 3AT is
kDa
37 eIF2α
GST–eIF5LR+CTD

GST–eIF5W391F
GST–eIF5LR7A
GST–eIF5CTD

25
eIF2 Input

GST–eIF5

1 2 3 4 5 6 7 8 9 10
100%
GST

75%
50% sc/hc eIF2γ sc hc sc hc sc hc sc hc sc hc sc hc sc hc
eIF2α
25% 75 c-Myc
(eIF2γ)
0%
c kDa 35 eIF2α
75
eIF2γ
kDa
50 65
Ponceau
45 Simply
37 35
eIF2α/β Blue

25
25
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Figure 2 | The CTD of eIF5 is critical for interaction with eIF2. a, b, Affinity Figure 3 | The linker region of eIF5 interacts with c subunit of eIF2. Affinity
chromatography assay between eIF2 (110 pmol) and the indicated chromatography as in Fig. 2 between indicated immobilized GST–eIF5
immobilized GST–eIF5 constructs. eIF2 was detected by immunoblotting constructs and total cell extracts (500 mg) expressing c-Myc–63His–eIF2c
using antibodies specific for eIF2c (a) or eIF2a (b). Representative blots are from either a single copy (sc) or high copy (hc) plasmid. Immunoblots were
shown. Signal intensity was quantified (Adobe Photoshop) and the developed with anti-c-Myc and eIF2a antibodies. Total protein in each
mean 6 standard deviation (n 5 3) are shown below. c, Total protein in each sample was stained with SimplyBlue. Inputs (lane 1) represent 5% (blots) or
sample stained with Ponceau S. Inputs (lanes 1) represent 10% of total. 1% (stain) of total.
379
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS NATURE | Vol 465 | 20 May 2010

a SD + Trp + 50 mM 3AT d SD + 10 mM Total eIF2α-P

3 days 5 days SCD-Ura 3AT
eIF2α

Unbound

Unbound
sc WT Flag-IP

Flag - IP

Flag - IP
/ input eIF2γ

GCN2

Input

Input
sc W391F eIF2α-P

sc LR7A 0 1 2 3 4 5
eIF5 Ratio (3AT/SCD)
sc WT

gcn2∆
eIF2α
sc W391F
e Release from Amino acid
sc LR7A eIF2α-P 48S ribosomal starvation
–1 –2 –3 –1 –2 –3 complex
1 10 10 10 1 10 10 10
eIF2γ α β
SD + Trp + 25 mM 3AT
b 3 days 5 days 2 5 Gcn2p 2 5
1 2 3 4 5 6
γ
sc WT GDI GDI
G
α
ε
hc WT c 2B
GCN2

SCD-Ura 3 days 5 5
hc W391F
gcn2∆ sc WT
hc LR7A α 2
sc WT 2
GEF 2B ε

2B ε
α
sc WT 2
sc W391F GEF
GCN2c
gcn2∆

hc WT
hc WT
GDP
hc W391F
hc W391F GTP
hc LR7A Pi
1 10–1 10–2 10–3 TC
1 10–1 10–2 10–3 1 10–1 10–2 10–3

Figure 4 | GDI activity antagonizes eIF2B and affects GCN4 activation in
vivo. a–c, Strains expressing single (sc) or high copy (hc) eIF5 plasmids as
the source of eIF5 and co-transformed with plasmids expressing GCN2,
vector alone (gcn2D) (a, b) or the constitutively active mutant
GCN2M788V,E1591K (GCN2c) (c) were grown as stated. WT, wild type. d, Left,
immunoblots following Flag–eIF5 immune precipitation of protein
complexes from cells grown in nutrient sufficient conditions (SCD) and
following starvation (SD13AT). Right, quantification 6 standard deviation
(n 5 3). e) Model for recycling and regulation of eIF2, incorporating eIF5
GDI activity. Dashed grey arrows indicate steps that limit eIF2 recycling.

dependent on activation of Gcn2p, the eIF2a kinase that responds to
amino acid starvation (Fig. 4a, gcn2D), which inhibits eIF2B GEF
activity thus lowering ternary complex abundance. This enables ribo-
somes to bypass inhibitory uORFs in GCN4 mRNA that normally
repress GCN4 translation20. As the GDI-deficient mutants exhibit
different eIF2 binding activities in vitro (Figs 2 and 3) we anticipated
that both mutants could provide complementary tools to probe GDI
function in vivo. High copy eIF5 bypasses the requirement for
eIF2(aP) for GCN4 activation, conferring resistance to 3AT in
gcn2D cells11 (Fig. 4b, sixth row). An increase in the concentration
of an eIF2–eIF5 complex in cells, which antagonizes eIF2B GEF activity,
explains this phenotype11. The GDI mutant, high copy eIF5LR7A, fails to
grow on 3AT medium (Fig. 4b, eighth row), consistent with reduced
eIF2B antagonism in vivo. In agreement with this, immune precipita-
tion of high copy eIF5LR7A–Flag failed to enhance binding with eIF2
over that bound to single copy eIF5 (Supplementary Fig. 5a). Therefore
these data are consistent with the idea that loss of GDI function in
eIF5LR7A can diminish competition between eIF5 and eIF2B (GEF)
for eIF2.
We assessed whether eIF5 GDI was required for the normal res-
ponse to eIF2 phosphorylation in GCN2 cells and found that W391F
mutant cells fail to grow on 3AT medium, showing they cannot
induce GCN4 translation (Fig. 4a, row 2). This Gcn2 phenotype is
shared with other eIF5-CTD mutations21, and was previously
ascribed to a defect within the 48S initiation complex causing
enhanced leaky-scanning of uORF1 within the GCN4 mRNA lea-
der21. However, this explanation does not take into account the
eIF5 GDI function. Because the eIF5W391F mutation reduces levels
of the eIF2–eIF5 complex (Figs 2 and 3 and Supplementary Fig. 5)
and eliminates GDI activity, we postulated that eIF5 binding to
eIF2 N GDP is necessary to inhibit eIF2B GEF function through
eIF2(aP) and that the observed reduced affinity of eIF5W391F for
eIF2 was sufficient to alter the sensitivity of cells to eIF2(aP), thus
causing the 3AT sensitivity. To test whether eIF5W391F could affect
global translational sensitivity to eIF2(aP), we introduced constitu-
tively active GCN2c alleles that hyper-phosphorylate eIF2a to high-
levels independently of amino acid starvation and downregulate
global translation initiation, conferring severely reduced rates of
growth22 (Fig. 4c, row 2). The eIF5W391F allele reverted slow-growth
phenotypes associated with two different GCN2c alleles of varying
380
©2010 Macmillan Publishers Limited. All rights reserved
phenotypic severity (Fig. 4c, row 3 and Supplementary Fig. 5b).
Immunoblotting analysis showed that the eIF5 mutation did not alter
the ability of Gcn2p to phosphorylate eIF2a (Supplementary Fig. 5c).
The Gcn2 and GCN2c growth-reversion phenotypes reported here
cannot be explained by enhanced leaky-scanning and are identical to
those conferred by mutations in eIF2B subunits that alter the ability
of eIF2B to respond to eIF2(aP)13,23 and by the overexpression of
eIF2B or eIF2 (ref. 24). Thus in W391F mutant cells, translation is
significantly less sensitive to the inhibitory effects of eIF2(aP).
We found that overexpression of eIF5W391F restored eIF2–eIF5
complex levels to those found in wild-type cells (Supplementary
Fig. 5a). Consistent with this, high copy eIF5W391F reversed the
growth enhancement associated with GCN2c alleles (Fig. 4c, row 5)
as well as the Gcn2 phenotype (compare Fig. 4a, row 2 and Fig. 4b,
row 3) and partially reverted the Gcd2 phenotype caused by high
copy eIF5 (compare Fig. 4b, rows 6 and 7). Although we assume that
the W391F mutation, in common with other CTD mutations21, has a
minor impact on pre-initiation complex formation and GAP func-
tion, the W391F phenotypes and their reversion are consistent with
reduced eIF2–eIF5 complex formation and GDI activity promoting
eIF2 nucleotide exchange and disrupting the ability of cells to regu-
late translation appropriately in response to eIF2(aP) (Supplemen-
tary Fig. 6b).
These data indicated that the eIF2–eIF5 complex was a new critical
component of the regulatory circuit that inhibits eIF2B GEF function
and regulates translation by diverse means. To test this prediction, we
immune-precipitated eIF5–Flag from wild-type cells grown in amino
acid complete medium and following starvation when eIF2 phos-
phorylation was enhanced ,fourfold (Fig. 4d). A significantly
greater fraction of eIF2 was found associated with the precipitated
eIF5–Flag following amino acid starvation (2–2.5-fold), indicating
that the affinity of eIF5 for eIF2 is enhanced, or that the release of eIF5
from eIF2 is retarded, when the latter is phosphorylated. However,
we do not suggest that eIF5 interacts directly with eIF2a because the
increase in precipitated eIF2(aP) observed following amino acid
starvation is in proportion with its higher total abundance (Fig. 4d).
Taken together, our data demonstrate that eIF5 GDI is a novel
component of the inhibitory eIF2(aP) pathway that restricts the
recycling of eIF2 to ternary complex and acts in addition to the well
defined inhibition of eIF2B whereby eIF2a phosphorylation promotes
NATURE | Vol 465 | 20 May 2010
its association with the eIF2Babd subunits25, to impair eIF2Be
subunit-mediated nucleotide exchange26. Thus our data are consistent
with a model where eIF5 is a multifunctional protein with separate
GDI and GAP activities, both necessary for controlled translation
initiation in eukaryotes and we propose an altered scheme (Fig. 4e,
Supplementary Fig. 6) to account for eIF5 GDI activity and the
eIF2N GDP–eIF5 complex in modulating the recycling of eIF2 in
response to eIF2 phosphorylation between rounds of translation
initiation.
METHODS SUMMARY
Plasmid and strain construction used standard methods27,28. Plasmid, oligonu-
cleotide and yeast construction details are given in Supplementary Tables 1–3.
eIF2 and eIF5 proteins were purified as described previously13,29. The eIF5 GDI
activity assay is a variant of an assay used to measure eIF2B GEF activity13. Full
methods are provided in the electronic version of this paper.
Full Methods and any associated references are available in the online version of
the paper at www.nature.com/nature.
Received 28 July 2009; accepted 10 March 2010.
1. Kapp, L. D. & Lorsch, J. R. The molecular mechanics of eukaryotic translation.
Annu. Rev. Biochem. 73, 657–704 (2004).
2. Sonenberg, N. & Hinnebusch, A. G. Regulation of translation initiation in
eukaryotes: mechanisms and biological targets. Cell 136, 731–745 (2009).
3. Scheuner, D. et al. Control of mRNA translation preserves endoplasmic reticulum
function in beta cells and maintains glucose homeostasis. Nature Med. 11,
757–764 (2005).
4. Costa-Mattioli, M. et al. eIF2a phosphorylation bidirectionally regulates the
switch from short- to long-term synaptic plasticity and memory. Cell 129,
195–206 (2007).
5. Mohr, I. Neutralizing innate host defenses to control viral translation in HSV-1
infected cells. Int. Rev. Immunol. 23, 199–220 (2004).
6. Algire, M. A., Maag, D. & Lorsch, J. R. Pi release from eIF2, not GTP hydrolysis, is
the step controlled by start-site selection during eukaryotic translation initiation.
Mol. Cell 20, 251–262 (2005).
7. Das, S., Ghosh, R. & Maitra, U. Eukaryotic translation initiation factor 5 functions
as a GTPase-activating protein. J. Biol. Chem. 276, 6720–6726 (2001).
8. Paulin, F. E. et al. Eukaryotic translation initiation factor 5 (eIF5) acts as a classical
GTPase-activator protein. Curr. Biol. 11, 55–59 (2001).
9. Cheung, Y. N. et al. Dissociation of eIF1 from the 40S ribosomal subunit is a key
step in start codon selection in vivo. Genes Dev. 21, 1217–1230 (2007).
10. Unbehaun, A., Borukhov, S. I., Hellen, C. U. & Pestova, T. V. Release of initiation
factors from 48S complexes during ribosomal subunit joining and the link
between establishment of codon-anticodon base-pairing and hydrolysis of eIF2-
bound GTP. Genes Dev. 18, 3078–3093 (2004).
11. Singh, C. R. et al. An eIF5/eIF2 complex antagonizes guanine nucleotide exchange
by eIF2B during translation initiation. EMBO J. 25, 4537–4546 (2006).
12. DerMardirossian, C. & Bokoch, G. M. GDIs: central regulatory molecules in Rho
GTPase activation. Trends Cell Biol. 15, 356–363 (2005).
13. Pavitt, G. D., Ramaiah, K. V., Kimball, S. R. & Hinnebusch, A. G. eIF2 independently
binds two distinct eIF2B subcomplexes that catalyze and regulate guanine-
nucleotide exchange. Genes Dev. 12, 514–526 (1998).
14. Schmitt, E., Blanquet, S. & Mechulam, Y. The large subunit of initiation factor aIF2 is
a close structural homologue of elongation factors. EMBO J. 21, 1821–1832 (2002).
©2010 Macmillan Publishers Limited. All rights reserved
LETTERS
15. Asano, K. et al. Conserved bipartite motifs in yeast eIF5 and eIF2Bepsilon,
GTPase- activating and GDP–GTP exchange factors in translation initiation,
mediate binding to their common substrate eIF2. EMBO J. 18, 1673–1688 (1999).
16. Asano, K. et al. Multiple roles for the C-terminal domain of eIF5 in translation
initiation complex assembly and GTPase activation. EMBO J. 20, 2326–2337
(2001).
17. Conte, M. R. et al. Structure of the eukaryotic initiation factor (eIF) 5 reveals a fold
common to several translation factors. Biochemistry 45, 4550–4558 (2006).
18. Wei, Z., Xue, Y., Xu, H. & Gong, W. Crystal structure of the C-terminal domain of
S. cerevisiae eIF5. J. Mol. Biol. 359, 1–9 (2006).
19. Alone, P. V. & Dever, T. E. Direct binding of translation initiation factor eIF2c-G
domain to its GTPase-activating and GDP-GTP exchange factors eIF5 and eIF2Be.
J. Biol. Chem. 281, 12636–12644 (2006).
20. Hinnebusch, A. G. Translational regulation of GCN4 and the general amino acid
control of yeast. Annu. Rev. Microbiol. 59, 407–450 (2005).
21. Singh, C. R. et al. Eukaryotic translation initiation factor 5 is critical for integrity of
the scanning preinitiation complex and accurate control of GCN4 translation. Mol.
Cell. Biol. 25, 5480–5491 (2005).
22. Ramirez, M. et al. Mutations activating the yeast eIF-2a kinase GCN2: isolation of
alleles altering the domain related to histidyl-tRNA synthetases. Mol. Cell. Biol. 12,
5801–5815 (1992).
23. Pavitt, G. D., Yang, W. & Hinnebusch, A. G. Homologous segments in three
subunits of the guanine nucleotide exchange factor eIF2B mediate translational
regulation by phosphorylation of eIF2. Mol. Cell. Biol. 17, 1298–1313 (1997).
24. Dever, T. E. et al. Modulation of tRNAiMet, eIF-2 and eIF-2B expression shows that
GCN4 translation is inversely coupled to the level of eIF-2.GTP.Met-tRNAiMet
ternary complexes. Mol. Cell. Biol. 15, 6351–6363 (1995).
25. Krishnamoorthy, T. et al. Tight binding of the phosphorylated alpha subunit of
initiation factor 2 (eIF2a) to the regulatory subunits of guanine nucleotide
exchange factor eIF2B is required for inhibition of translation initiation. Mol. Cell.
Biol. 21, 5018–5030 (2001).
26. Mohammad-Qureshi, S. S. et al. Critical contacts between the eukaryotic initiation
factor 2B (eIF2B) catalytic domain and both eIF2b and -2c mediate guanine
nucleotide exchange. Mol. Cell. Biol. 27, 5225–5234 (2007).
27. Adams, A., Gottschling, D. E., Kaiser, C. A. & Stearns, T. Methods in Yeast genetics:
A Cold Spring Harbor Laboratory Course Manual (Cold Spring Harbor Laboratory
Press, 1998).
28. Gietz, R. D. & Woods, R. A. Transformation of yeast by lithium acetate/single-
stranded carrier DNA/polyethylene glycol method. Methods Enzymol. 350, 87–96
(2002).
29. Phan, L. et al. Identification of a translation initiation factor 3 (eIF3) core complex,
conserved in yeast and mammals, that interacts with eIF5. Mol. Cell. Biol. 18,
4935–4946 (1998).
Supplementary Information is linked to the online version of the paper at
www.nature.com/nature.
Acknowledgements We thank K. Asano for plasmids and yeast strains, and
discussions, and M. Ashe and members of the Pavitt and Ashe labs for discussions.
This work was supported by grants BB/E002005/1 and BB/H010599/1 from the
BBSRC to G.D.P.
Author Contributions G.D.P. conceived the experiments, directed research,
interpreted the experiments and wrote the manuscript. M.D.J. performed the
experiments, interpreted the experiments and co-wrote the manuscript.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial interests.
Correspondence and requests for materials should be addressed to G.D.P.
(graham.pavitt@manchester.ac.uk).
381
doi:10.1038/nature09003
METHODS
Yeast genetics. Yeast strains were grown in standard media as described27.
Plasmid transformations used the lithium acetate method28 and plasmid shuff-
ling used 5-fluoro-orotic acid as described27. Plasmid, oligonucleotide and yeast
construction details are given in Supplementary Tables 1–3. Cells were grown to
exponential phase then diluted to A600 5 0.1 and serially diluted tenfold. Each
dilution (2 ml) was spotted onto the indicated medium and incubated at 30 uC.
Plasmid constructions. Individual mutants were generated by overlapping pri-
mer site-directed mutagenesis (Stratagene) or using standard PCR cloning
methods. The L7RA mutant was commercially synthesized (Mr Gene) then
sub-cloned.
SDS PAGE and immunoblotting. This was performed as described previously26
using specific antibodies for eIF5, eIF2a and 2c (custom antibodies), c-Myc–
horseradish peroxidase (9E10, Santa Cruz Biotechnology), Flag (M2, Sigma) and
eIF2a2P (9721, Cell Signaling Technology). Horseradish peroxidase secondary
chemiluminescent detection was performed as per manufacturer’s instructions
(Perkin Elmer).
Protein purification. eIF2 was purified as described previously13 using strain
GP3511. Purification of GST-tagged eIF5 wild-type and mutant constructs was
adapted from ref. 29. ArcticExpress BL21 DE3 RIL E. coli cells (Stratagene) were
transformed with recombinant pGEX-4T-1 plasmids and then grown in LB
medium to an A600 of 0.7. Expression was induced using 1 mM of isopropyl-
b-d-thiogalactopyranoside (IPTG) and cells were then grown overnight at 12 uC.
Cells were collected and lysed with BugBuster (EMDBiosciences) with protease
inhibitors added (13 complete EDTA-free protease inhibitor tablet; Roche)
before clearing by centrifugation (16,000g, 20 min, 4 uC). The soluble lysate
was then mixed for 2 h at 4 uC with glutathione-Sepharose beads (GE
Healthcare) equilibrated in 13 PBS. The beads were then washed three times,
then suspended in glutathione elution buffer (50 mM Tris-HCl pH 9.0, 500 mM
NaCl, 50 mM glutathione, 1 mM dithiothreitol (DTT), 0.1% Triton X-100) and
incubated overnight at 4 uC. Eluates were dialysed into storage buffer (30 mM
HEPES pH 7.5, 100 mM KCl, 0.1 mM MgCl2, 10% glycerol, 1 mM DTT) and
stored at 280 uC. Quantification was performed using the Bradford Protein
Assay (Bio-Rad).
To purify Flag-tagged eIF5 from yeast, strains expressing high copy eIF5–Flag
(GP5424) or eIF5-7A–Flag (GP5423) were grown in synthetic complete-Leu
medium to A600 of 3–5. Cells were collected, washed in ice-cold water and then
resuspended in high-salt lysis buffer (100 mM Tris-HCl pH 8.0, 500 mM KCl,
5 mM MgCl2, 5 mM NaF, 2.5 mM PMSF, 10% glycerol, 0.1% Triton X-100 with
1 mg ml21 pepstatin A, 1 mg ml21 leupeptin, 5 mg ml21 aprotinin, and 13 com-
plete EDTA-free protease inhibitor tablet; Roche). Lysis was performed by grind-
ing with a pestle and mortar under liquid nitrogen and lysates were then cleared
by centrifugation (16,000g, 20 min, 4 uC). The extract was then incubated for 2 h
at 4 uC with EZview Red ANTI-Flag M2 Affinity Gel (Sigma) equilibrated in
©2010 Macmillan Publishers Limited. All rights reserved
high-salt lysis buffer, washed once with high-salt lysis buffer then twice with low-
salt lysis buffer (as high-salt lysis buffer but with only 100 mM KCl). Bound
protein was then released by incubation for 30 min in the presence of
0.1 mg ml21 of 33Flag peptide (Sigma) and dialysed into storage buffer.
Quantification was performed using the Bradford Assay (Bio-Rad).
Flag immune precipitations. These were performed similarly to purifications
but using buffers containing 80 mM KCl to maintain interactions.
GDI activity. eIF2 N [3H]GDP binary complexes were formed in 75 mm 3 12 mm
soda-lime glass tubes (Fisher). Typically using eIF2 (60 pmol) and 0.25 mCi
[3H]GDP (10–15 Ci mmol21) in assay buffer (30 mM HEPES pH 7.5, 100 mM
KCl, 0.1 mM EDTA, 1 mM DTT, 2 mg ml21 creatine phosphokinase). Complex
formation was carried out for 10 min at room temperature before stabilization
by addition of MgCl2 to 2.9 mM and incubation for a further 2 min at room
temperature. Purified eIF5 (120 pmol) or sample buffer was then bound to the
eIF2 binary complex for 30 min at 10 uC. Dissociation of the binary complex was
initiated by addition of a . 100-fold excess of unlabelled GDP (2 nmol). Samples
(12 ml) were taken immediately (t 5 0) and at 2, 4, 6, 8 and 10 min. At each time
point, samples were added to 2.5 ml ice-cold Stop buffer (30 mM HEPES pH 7.5,
100 mM KCl, 0.1 mM EDTA, 5 mM MgCl2), filtered through Whatman 0.45 mm
25 mm cellulose nitrate filters using a Millipore vacuum manifold then washed
twice with 2.5 ml of ice-cold Stop buffer. Filters were dried at 65 uC then counted
by liquid scintillation in Ultima Gold F (Perkin Elmer). Experimental data were
fitted to exponential dissociation curves to obtain the dissociation rate constant
(Koff).
GST affinity chromatography. Using purified eIF2. Purified GST–eIF5 proteins
or GST alone (140 pmol) and eIF2 (110 pmol) were incubated in 400 ml of inter-
action buffer (30 mM HEPES pH 7.5, 100 mM KCl, 0.1 mM EDTA, 1 mM DTT,
2.5 mM MgCl2, 0.05% Triton X-100) with 50 ml of glutathione-Sepharose beads
for 2 h at 4 uC. The beads were then washed three times with 1 ml of interaction
buffer and the bound proteins eluted by boiling in sample buffer. Bound eIF2 was
then resolved by SDS–PAGE separation and detected by immunoblot analysis
using the indicated antibodies.
Using cell extracts. Strains GP5463 and GP5661 expressing single copy and
high copy c-Myc-tagged GCD11 (respectively) were grown in synthetic com-
plete-Leu-Trp medium to an A600 of 3–5. Cells were collected, washed in water
and then resuspended in 1 ml low-salt lysis buffer. Lysis was performed using
1.0 mm zirconia/silica beads (Thistle Scientific) and the FastPrep 24 homogenizer
at 4 uC (MP Biomedicals). Lysates were cleared by centrifugation (16,000g,
20 min, 4 uC) and proteins quantified using the Bradford assay (Bio-Rad).
Total protein lysate (500 mg) was incubated with purified GST–eIF5 (140 pmol)
or GST alone and 40 ml of glutathione-Sepharose beads in a total volume of 1 ml of
low-salt lysis buffer for 2 h at 4 uC. The resin was washed twice with 1 ml of low-salt
lysis buffer and the bound proteins eluted by boiling in sample buffer. Bound
eIF2a and c was assayed by immunoblot analysis.
NATURE|Vol 465|20 May 2010
Chang-Hwan Choi, a nanoengineer at the Stevens Institute
Q&A of Technology in Hoboken, New Jersey, received a 2010
Young Investigator Program award from the US Office of
Naval Research (ONR) for his design of anti-corrosion
surfaces that will make Navy vessels more durable.
What sparked your initial
interest in engineering?
Throughout my youth, I was
interested in designing cars
and planes. While I was at
Seoul National University,
there was a popular
television programme about
an Air Force pilot. It inspired
me to focus my interests on
planes and gain acceptance
into the aerospace
engineering department at
my university.
Did you explore
opportunities in foreign
countries?
Yes. To satisfy my military
service requirement, I
earned my masters at Seoul
University while working
for the Korea Aerospace
Research Institute. To gain
the international experience
I thought I needed to
pursue aerospace-training
opportunities abroad, I
became a lecturer in a
Korean-government overseas
volunteer programme. I
lectured on Korean languages
and computer science in
Thailand, which gave me
confidence that I would do
well in other countries.
How did you move into
nanoscience?
After my lecturing service,
I discovered that aerospace-
engineering schools in the
United States were shrinking
their programmes. So, I
decided to switch gears and
apply to schools with PhDs
in micro- and nanoscale
systems, because interest in
these areas was booming.
I gained admission to four
US schools, but I wanted to
work with Kenneth Breuer, a
fluid-mechanics researcher
who had been developing
microscale engines for
satellites. I contacted him to
express my interest. He liked
my work, and said he had
moved to Brown University
in Providence, Rhode Island,
and was looking
for a new student.
He helped me
secure admission
and full support
within a month.
What was your
most pivotal
career decision?
I chose to switch schools
halfway through my PhD. At
Brown, I was measuring flow
phenomena in microfluidic
devices I designed — an
interesting project, but
predominantly experimental.
I preferred designing and
making devices. I started
studying the work of Chang-
Jin Kim at the University
of California, Los Angeles
(UCLA), who was famous
for the design and fabrication
of micro-electromechanical
systems. When I contacted
him to ask about research
openings, he invited me to
join his lab. It was a very hard
decision. Ultimately, Breuer
said he wouldn’t prevent
me from exploring what I
really wanted to do. He asked
only that I finish my current
project and publish the
results. That paper is highly
cited and considered a classic
reference paper in microscale
fluid dynamics (C. Choi et al.
Phys. Fluids 15, 2897; 2003).
Why didn’t you do a
postdoctoral fellowship?
When I was about to finish
my PhD, I still hadn’t
published all of my major
results. But because I had
already spent two years
at Brown and almost five
years at UCLA, I worried
that one or two more years
for a postdoc was too long.
I applied for faculty jobs
even though I didn’t expect
much without postdoc
experience. Fortunately,
the Stevens Institute of
Technology was eager to
hire faculty members in
nanoscale engineering and
© 2010 Macmillan Publishers Limited. All rights reserved
saw the potential
of my work on
superhydrophobic
surfaces.
How did you
differentiate your
work from that of
your mentors?
When I came to Stevens
in 2007, I didn’t want to
compete with either of
my previous advisers, so I
decided to look for different
applications for these
superhydrophobic structures.
I discovered that there was
a need for anti-corrosive
materials in ships and planes.
I wanted to use our materials
to reduce metal corrosion by
minimizing liquids’ contact
with metal surfaces.
How did you secure
funding from the US Navy?
Initially, US funding agencies
thought my ideas were
interesting but not practical
enough to fund because I
couldn’t make nanostructures
on a large scale. This
challenge prompted me to
develop ideas for creating
nanostructures on the much
larger metre scale. The
engineering dean at Stevens
connected me with a Navy
programme manager who
helped me to secure a year’s
funding for exploratory work.
If I could prove myself, she
thought I had a strong chance
of securing more funding
through the US Navy Young
Investigator Program.
How do you think this
award will affect your
career?
I think it helps position me
for tenure because my school
holds the award in high
esteem. The ONR selects only
10–20 people from more than
10,000 applicants. It is a great
honour, especially because I
am a foreigner.
Interview by Virginia Gewin
CAREERS
IN BRIEF
Wellcome funding change
The Wellcome Trust medical-research
charity in London has unveiled an
award scheme with no need for detailed
budgets and methodologies. The scheme
is intended to encourage ambitious
proposals and boost productivity by free-
ing researchers from the time constraints
of grant renewal. Wellcome’s total funding
of £600 million (US$885 million) is the
same as last year, but this scheme allows
for longer-term projects. Early-career
and senior investigators at universities or
research institutes can have a maximum of
£425,000 a year for up to seven years. Alan
Schafer, Wellcome’s director of science
funding, says the changes are also intended
to ease burdens on grant reviewers.
Arizona boycotted
A group representing 22,000 Hispanic and
Native American scientists in the United
States has dropped Phoenix, Arizona,
from its potential 2012 conference sites
owing to the state’s new immigration
law. “We do not want to expose our
members and conference participants
to the potential for harassment by law
enforcement,” says Judit Camacho,
executive director of the Society for the
Advancement of Chicanos and Native
Americans in Science (SACNAS), based
in Santa Cruz, California. The law lets
officials query a person’s immigration
status based on a ‘reasonable suspicion’
that he or she is an illegal immigrant.
SACNAS announced its decision on 10
May. The society will hold no events in
Arizona unless the law is repealed.
Non-tenure survey mixed
Non-tenure-track academic researchers
in science, technology, engineering and
maths (STEM) in the United States had
mixed feelings about their positions,
says a survey released this month by the
Center for the Education of Women at
the University of Michigan at Ann Arbor.
Respondents liked the flexibility of their
positions and their freedom from the
tenure process. But they were concerned
about job security, generating funding,
isolation on campus, unequal treatment
compared to tenure-track faculty and lack
of transparency about their contracts,
titles and career progression. The centre
■ interviewed 343 researchers from 12
universities. More than a third of all
respondents worked in STEM.
385
CAREERS
Burdens of biodefence
Working with nature’s nastiest microbes offers a chance to help ensure public safety.
Karen Kaplan details the profession’s risks and rewards.
E
ach time he walks into his lab, Ricardo
Carrion faces a safety routine of up
to half an hour. When he leaves, even
just to grab a pen, that routine doubles
in length. And every time he goes back in or
out, he must follow the same procedures, step
by step.
Carrion, an assistant scientist at the
Southwest Foundation for Biomedical
Research (SFBR) in San Antonio, Texas, is
conducting research on microbes, but not
within the comparatively tame confines of a
lab bench. He’s one of a number of researchers
around the globe who go through these
procedures several times a day as part of their
work investigating ways to protect against
bioterrorism and combat infectious diseases.
These positions, despite their inherent dangers,
offer the opportunity to contribute to public
safety and security. And, for those with a thrill-
seeking streak, they also offer a bit of daily
excitement on top of the usual research routine.
Carrion, who manages the SFBR’s biosafety
level (BSL) 4 lab, helps to develop vaccines for
viruses that cause haemorrhagic fever. BSL-4
386
NATURE|Vol 465|20 May 2010
facilities deal with biological agents that can 2001 in the United States, policy-makers
cause serious or fatal illness in humans and deemed biodefence an esoteric sub-specialty,
Artyom KorotAyer/epA/Corbis
for which no treatment is available; there are its funding half-buried as an obscure line
about half a dozen in the United States and item in federal defence budgets. But following
roughly two dozen worldwide, in 17 nations. those strikes and the anthrax attacks a month
The daily routine can be exhausting: once later, the country started shovelling money
Carrion confirms that all gauges are working into bioterrorism research. In 2001, for
properly, he dons a set of scrubs and a headset example, the annual budget for the National
to maintain contact with an external safety Institute of Allergy and Infectious Disease
team. Next, he pumps air into (NIAID) in Bethesda,
his vinyl hazardous-materials “I don’t think you Maryland— the division of the
(hazmat) suit, making sure could do this work National Institutes of Health
that the fabric and seams are that handles biodefence and
airtight. Potential leaks are if you weren’t doing infectious-disease research —
doused with soapy water. something that you was $42 million. By 2002, it
If bubbles appear, Carrion believe is serving had ballooned to $187 million,
discards the US$2,000 suit — a 345% increase.
they last for eight months on the greater good.” Such increases have now
average — and selects another. fallen off at all US federal
Self-enclosed and pressurized, fitted with agencies and the field is redefining itself,
internal air filters and an opening for an air shifting away from its bioterrorism agenda
hose, the hazmat garment, which has three of the early 2000s that concentrated on
layers of gloves, resembles a space suit and pathogen-specific vaccines. Almost a decade
weighs about 15 pounds. later, that tight focus has been widened to
Before the terrorist attacks of 11 September include treatments for emerging diseases,
© 2010 Macmillan Publishers Limited. All rights reserved
NATURE|Vol 465|20 May 2010
says Michael Kurilla, director of the NIAID’s
Office of Biodefense Research Affairs. “Now,
rather than trying to prepare pre-event, the
focus is on prophylaxis,” Kurilla says, noting
research targets such as broad-spectrum anti-
virals, rather than specific vaccines. Gigi Kwik
Gronvall, a senior associate at the Center
for Biosecurity in Pittsburgh, Pennsylvania,
emphasizes the continued importance of
shortening the time between identifying a
disease and developing a vaccine or anti-viral.
“Right now there’s an 8–10-year timeline,” she
says. “I hope it’s not the best we can do.”
Worldwide, several BSL-4 labs are under
construction and scheduled to begin
operating within the next year or so, including
in Germany, Switzerland and the Netherlands.
In the United States, a handful of BSL-4
labs operated by various federal agencies or
universities in Kansas, Maryland and Georgia
are scheduled to be opened or commissioned
over the next couple of years. All are likely to
hire dozens of researchers and postdocs.
Postdoc positions are available in the
United States at a number of labs, including
through the National Research Council
fellowship programme at the US Army
Medical Research Institute of Infectious
Diseases (USAMRIID) in Frederick,
Maryland. The SFBR, where Carrion started
as a postdoc, also has a postdoc programme.
Aspirants should have a doctorate in a field
such as immunology, bacteriology, virology,
pathology, microbiology or physiology.
Carrion, for example, has a PhD in
immunology and microbiology and a master’s
in biology. It is also helpful, some researchers
say, to attend graduate school at a university
that operates a BSL-4 lab.
Playing it safe
The step-by-step protocols that Carrion must
follow to enter and exit his lab reflect the
risk inherent in the agents that he uses. Lab
work, even without these biohazards, can be
dangerous or even fatal (see
Nature 458, 664–665; 2009). “I wasn’t as
But biodefence researchers
routinely handle pathogens
interested in
that kill or severely sicken academia. I wanted
people, animals and plants. something closer to spills, releases, loss or theft of
Those defined by the US
government as ‘category A’
the front lines.”
bioterrorism threats include
anthrax, botulism, bubonic plague, tularemia, of, or injuries to, lab staff. “The time, the
smallpox and viral haemorrhagic fevers
such as filoviruses (Ebola and Marburg) and
arenaviruses (Lassa and Machupo).
Labs and governments have developed
an elaborate set of safety regulations and
procedures, including those that govern
entering and leaving the lab. In the United
States, the European Union and Canada,
biodefence researchers must pass government So why tolerate the regulations and laborious
or police checks and clearances — including
medical clearances — before accepting a post. several reasons: it is exciting, they feel they’re
Non-US researchers seeking a US position
must be able to obtain Federal Bureau of
Investigation clearance, which can be a
lengthy process. Next, they undergo months
of training that covers biocontainment
protocol, safety procedures, emergency
operations and use of the hazmat suit, lab
systems and engineering operations. The
trainee, typically a postdoc, learns how to
work with biosafety cabinets,
store and keep written records
of pathogens, and clean up and
decontaminate a spill.
The researcher then begins
working with live pathogens
in the lab, under the direct
supervision of a mentor. This
stage usually lasts for three
to six months, the length of
time that it takes an average
trainee to become adept at
all the necessary procedures
and operations while wearing
the hazmat suit. Then the
researcher can enter the lab on
his or her own. The process
takes up to a year, and lab staff
must be retrained annually.
“It’s definitely overwhelming
at first,” Louis Altamura says
of the protocols, procedures
and paperwork. A postdoc in
the virology division of the
USAMRIID, Altamura works
with bunyaviruses, including
certain haemorrhagic fever Ricardo Carrion (top) and
viruses such as the Crimean- Louis Altamura.
Congo. “You have to be a little
bit of a lab manager even within your own
project. I don’t want to say it’s all a distraction,
but it is something you need to account for.”
Verena Krähling, a postdoc working with
filoviruses at the Institute for Virology at the
Philipps University of Marburg, Germany,
says that she is sometimes worn out by the
extensive safety precautions, especially those
involving the weighty hazmat
suit. “We have to know all the
emergency training,” she says.
It includes learning policies
on and practices for handling
pathogens inside or outside
a biological safety cabinet
and dealing with infections
preparation — it’s kind of exhausting,” says
Krähling. The suit’s air supply dries out the
skin and respiratory system, she says, and
those who breathe it daily for more than
four hours are often susceptible to colds or
respiratory infections.
Helping humankind
trappings of the research? Investigators cite
contributing to public health and safety,
© 2010 Macmillan Publishers Limited. All rights reserved
CAREERS
and they can see the results of their work far
more quickly than is likely in, for example,
pharmaceutical research, where it takes years
to identify and develop a potential drug, let
alone to see its beneficial effects.
“I wasn’t as interested in academia: I wanted
something closer to the front lines,” says
Daniel Sanford, a senior research scientist at
Battelle, a contract research
lab based in Columbus, Ohio,
that operates BSL-3 facilities,
dealing with dangerous or
lethal pathogens for which
treatments exist. Sanford
studied basic cancer research
as a graduate student; it was
difficult, he said, to see the
direct benefits. At Battelle,
he’s helping to develop
animal models for vaccine
and therapeutic products, to
determine whether they are
safe and efficacious. “That in
itself is gratifying,” he says.
Steven Jones, head of the
Emerging Bacterial Diseases
section at the National
Microbiology Laboratory
in Manitoba, Canada — the
nation’s only BSL-4 lab —
says that he too likes the
immediacy of the work and
results. Because vaccines and
therapeutic candidates are
tested on animals, researchers
get instant feedback on
whether they work and how
safe they are, he says.
It is comforting to some to know that
they’re helping to mitigate the effects of an
attack. “I don’t think you could stay here and
do this work and deal with all the logistical
issues if you weren’t doing something that
you believe is serving the greater good,”
says Lisa Hensley, a microbiologist and
principal investigator at the USAMRIID who
researches filoviruses, arenaviruses and the
pox family of viruses.
Carrion agrees. In the suit room, in
full hazmat garb, he walks through a
decontamination chamber — lab staff don’t
need to be decontaminated on the way
in — and then an airlock. Only after the
airlock’s double doors close completely can
he open the door to the lab. Once inside,
company regulations permit him to remain
for five hours; some labs allow only four.
Each time he leaves, Carrion enters the
decontamination chamber and showers
under a disinfectant cycle with the suit, rinses
it and hangs it to dry. Then he enters a second
chamber, puts his scrubs in an autoclave
and takes a personal shower for at least five
minutes. Then, finally, at long last, he puts his
street clothes back on. ■
Karen Kaplan is the assistant editor of
Naturejobs.
387
FUTURES
Mind expeditions
Fighting the good fight.
Brenda Cooper
The bright light shining on the podium
made it impossible to see the myriad
student faces out there, but I knew what
they would look like. Earnest. Curious.
Unblooded. They’d be in jeans, with
glasses that let them go far away if I
bored them, and writing on the insides of
their arms to stare at and study for their
next class.
A blonde woman with wisps of flyaway
hair and a linen suit coat over linen shorts
over black boots introduced me. “Please
welcome Private First-Class Eleanor Prac-
tice to career day. She’ll tell you about her
first job in Continental Security.”
That was my opening. “People used to
join the forces to see the world.”
Soft laughs floated up from behind the
screening, blinding light.
“But I’ve never been out of San Diego.
Thank you for inviting me. After my talk,
some of you may want to join us, some may
decide you don’t like the idea after all.
“The — event — happened on our first
virtual mission. I thought it would feel
like a video game. The team was me, Alvar
from Mexicali and Louisa from Toronto.
We came from three countries and never
met in person.”
I knew what they looked like well
enough that I’d recognize them on the
street. But I did not know how they felt or
smelled or walked. They might not
recognize me.
“We were in the Yucatan,
trying to stop a drug ring, help
Mexico rebuild. Our weapons
were databases and wireless mesh,
data blockers and listeners.
“An agent on the ground had
thrown a hidden mesh net over a
small valley. We used it to watch
a poor family’s house. Palm roof,
pressboard walls.” I swallowed,
seeing it again. So poor. “The
father and the two boys carried
drugs for the valley kingpin, the
FUTURES
mother cooked tortillas and fish
soup. There was a little girl in a wheel-
chair. Maribel. She was the reason for the
drug running. Money for Maribel’s treat-
ment.”
I had hated the assignment then.
“Louisa and Alvar and I talked while we
watched, tried to say tough things so we
wouldn’t get sentimental. We wanted to
stop the bosses.
“We were gathering evidence. It wasn’t
390
© 2010 Macmillan Publishers Limited. All rights reserved
for us to act. We had no bodies there. Inter-
national law kept us from listening inside,
of course, but Mexico is a hot place and
many conversations happen outside.”
The audience was very quiet. I hoped
they were listening. In the wings, soft light
fell on the blonde’s face. She watched.
“We expected Americans and arrests.
The men who came were Mexican. We rec-
ognized at least one of them as a drug run-
ner. They hung in the shadows, moved like
black shadows, like the devil.” I shouldn’t
say such things. “Like special operations.
Trained. They slid against the house,
planted charges near the doors and win-
dows and then stood on a hill under a tree,
watching and smoking and laughing.
“It began to rain. I hoped the explosives
would grow too wet. It was my job to put
the father in jail; not to see him murdered.
He loved his daughter.
“When he opened the door, the house
bloomed with fire.
“He fell right there, his face black and his
clothes charred and smoking.
“The older boy ran out.
The mother followed,
carrying a small
figure, and
screaming
NATURE|Vol 465|20 May 2010
as she set the dead body of her youngest
son onto the fecund forest floor. She ran
back towards the house, tugging on her
husband’s body.
“A man on the hill emptied a rifle into
all three of them. Even virtual shots heard
through wireless in a recliner sound like
death.” I swallowed and looked out at the
crowd. “Not like a game.
“They left them, the house less than half
burned, rain falling onto the husk of it and
making a column of white steam.
“We heard a scream from inside. ‘Mar-
ibel,’ Louisa said. And then ‘Eleanor.’
“Alvar agreed. ‘You are the fastest.’
“He couldn’t know why this would be
hard for me. It didn’t matter. I went. We
blessed the wireless. Burnt wires would
have blinded us. One of the phones had
melted, but the other sat in its charger. Its
camera didn’t see her. Their own wireless
access point — provided by the criminals
they worked for — pinpointed her chair.
The chair had her vitals. She breathed,
but she had stopped screaming. I imag-
ined her burned and bleeding as well as
Jacey
paraplegic, wanted to leave her there and
let it be over for her.”
There was a tear on my cheek. I hoped
the auditorium lights weren’t picking
it up.
“But Louisa whispered in my ear and
Alvar said, ‘I’ve called help.’
“I had a team. Maybe someday Maribel
would have a team. The doorway smol-
dered and we had to pass her parents’
bodies. Her mother had cleared the space
for her before she died. We’d have to be
fast and precise. I sent signals to her chair.
Forward and back. Testing. When I knew
I could do it, I lied to the chair and told it
to expect a hill and go fast and hard.
“The chair burst through, raced too fast
down the ramp, teetered, didn’t fall. The
signal was better now, so I guided it a few
more yards and left her in shade to wait
for help.”
I paused and waited, then spoke softly.
“Maribel is alive and in school. Her par-
ents’ employers are in jail.”
I waited for the audience to react. A clap,
then another, then another.
I rolled my chair to the front of the room
to take questions. ■
Brenda Cooper lives in Seattle where she
is a technology professional, a futurist
and a science-fiction writer. Watch for her
next novel, Mayan December, from Prime
Books in 2011. Learn more at www.
brenda-cooper.com.