Science Magazine 5696 2004-10-22

EDITORIAL
www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004 575

A
s genomics has revolutionized the study of biology, so has it affected how drugs are
discovered. Pharmaceutical companies have also realized another major application: how
drugs are assessed for safety. The analysis of gene expression profiles is now actively used
alongside conventional toxicological assays to assess drug safety. Such toxicogenomic analysis
is being used to predict drug toxicities and to gain a more in-depth understanding of toxic
mechanisms, so that more successful drug candidates can be selected.
The U.S. Food and Drug Adminstration (FDA) sees genomics as a beneficial aid to the drug risk
assessment process primarily through its ability to identify specific patients who are either likely to benefit
from a particular drug or who may experience harm. The use of toxicogenomics also
has promise in proving hypotheses that support safe drug use in humans through a
mechanistic understanding of toxicities found in drug-treated animals. Take the case in
which rats were treated with a certain class of hypolipidemic drugs. Changes in the
expression of specific liver genes were seen that have been shown to correlate with
observed liver toxicity. However, when treated human and rat liver cells were compared,
analogous gene expression changes were not seen in the human cells.* Thus, by gaining
a better idea of the mechanisms of toxicity in an animal species, it becomes feasible to
examine species-specific effects to better assess the possible relevance of animal
findings to humans. After a number of conferences and workshops based on recent
FDA draft guidelines, there was agreement that some standards ought to be adopted for
the generation and subsequent submission of toxicogenomic data to the FDA. This
would help ensure that any regulatory decisions based on an interpretation of data can
be made in a consistent manner.
A number of groups are actively addressing the issues of standardization of toxicogenomic
data generation and exchange. The European Bioinformatics Institute (EBI) has created
Tox-MIAMExpress to incorporate toxicity and toxicogenomics data into their Array
Express database. Such efforts help identify the essential elements of a microarray
experiment that are critical to interpreting a gene expression profile in the context of an
associated toxicity. Some aspects of toxicogenomics experiments that are suitable for further standardization
include data normalization methods, statistical assessments of gene expression, gene annotation and function,
data visualization methods, and issues related to quality control of transcripts and probes.
Once a compound is selected for development by a pharmaceutical company, submission of toxicogenomic
data to the FDA as part of a risk assessment package may be needed to help address safety concerns. An
approach being taken by the Clinical Data Interchange Standards Consortium (CDISC) Pharmacogenomic
Standards nonclinical working group is to develop hypothetical cases in which the submission of toxicogenomic
data would assist in the interpretation and determination of the relevance of specific toxicity issues. In the
example of the hypolipidemic drugs mentioned above, the availability of toxicogenomic data submission
standards would offer drug companies the ability to submit data at a molecular level highlighting the species-
specific nature of an animal finding, thereby helping to address the safety concerns of regulators.
The development of standards for toxicogenomic data generation and interpretation will help establish
toxicogenomic approaches as scientifically accepted practices in the complex process of drug risk assessment.
Central to this process is the continuation of the ongoing dialogue between molecular and traditional
toxicologists from drug companies, regulatory bodies, and academia. The development of a scientific
consensus on toxicogenomic data standards and data interpretation would mean fewer safety concerns and fewer
delays in the drug approval process, resulting in improved availability of safe and effective drugs. The
development and acceptance of toxicogenomic data submission standards promise to significantly improve
the drug risk assessment process, which would benefit the pharmaceutical industry and public alike.
Peter G. Lord and Thomas Papoian
Peter G. Lord is in Pharmaceutical Research and Development at Johnson & Johnson in Raritan, NJ. Thomas Papoian is at the U.S. Food and Drug
Administration in Rockville, MD.
*J. W. Lawrence et al., J. Biol. Chem. 276, 31521 (2001). W. B. Mattes et al., Environ. Health Perspect. 112, 495 (2004).
Genomics and Drug Toxicity
C
R
E
D
I
T
S
:
P
H
O
T
O
S
.
C
O
M
;
(
O
V
E
R
L
A
Y
)

G
E
N
O
M
E
.
G
O
V
Published by AAAS
22 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org 586
C
R
E
D
I
T
S

(
T
O
P
T
O

B
O
T
T
O
M
)
:
J
U
S
T
I
N

I
D
E
/
H
A
R
V
A
R
D

N
E
W
S

O
F
F
I
C
E
;
W
.
S
.
H
W
A
N
G
NEWS
P A G E 5 9 0 5 9 2 5 9 6 5 9 9 6 0 3
Mitochondrial
link to
metabolic
syndrome
An avian flus
surprising
reach
Thi s Wee k
Scientists and ethicists gathered at a brain-
storming session last week in Washington,
D.C., to discuss voluntary limits on human
cloning and embryonic stem (ES) cell
research. The event echoed the 1975 meeting
in Asilomar, California, where an earlier gen-
eration tried to establish guidelines for genet-
ic engineering. As was the case 29 years ago,
researchers are eager to move ahead: Even as
the session at the National Academy of Sci-
ences got under way, for example, Harvard
University officials were announcing that
diabetes researcher Douglas Melton is apply-
ing for permission to use nuclear transfer
otherwise known as research cloningto
create new human cell lines in a privately
funded effort to model diseases.
In the United States, federal funds may be
spent only on government-approved human
ES lines, yet private funding is flowing to the
field, largely without regulation. This has in-
creased pressure on scientists to develop their
own rules. To get started, the academies cre-
ated an 11-member Committee on Guide-
lines for Human Embryonic Stem Cell Re-
search, co-chaired by cancer researcher
Richard O. Hynes of the Massachusetts Insti-
tute of Technology and Jonathan Moreno, di-
rector of the University of Virginia Center for
Biomedical Ethics. Last week experts made
suggestions covering all aspects of work with
human ES cells, from informed-consent pro-
cedures to distribution of cell lines.
Theres definitely a need to create stan-
dards in the field so it wont take 6 to 12
months to start work, said blood researcher
Leonard Zon of Childrens Hospital in
Boston. He ticked off a list of offices that
have to pass on any proj-
ectincluding adminis-
tration, research, finance,
legal, ethics, intellectual
property, and public af-
fairs. Researchers also
must contend with a
patchwork of state regu-
lations, noted George-
town University bioethi-
cist LeRoy Walters, who
described rules ranging
from California and
New Jerseys aggressive
pro-research policies to
nine states bans on hu-
man embryo research.
On top of this tangle of
standards are dizzying
moral questions, such as: If an embryo is be-
ing created for research, is it better to do it
through in vitro fertilization or nuclear trans-
fer? And what does it mean to accord an ear-
ly embryo respect?
Much of the political debate over stem
cell research in the United States has focused
on the Bush Administrations prohibition on
using federal funds to work with human ES
cell lines other than a handful already
in existence 3 years ago. All these
lines were established from spare
embryos created at fertility clinics.
More ethically charged are efforts
to create new stem cell lines by trans-
ferring DNA into an enucleated egg.
Many researchers are eager to get on
with such studies. But speakers
warned of public resistance, compli-
cated by the fact that stem cell re-
search is conflated with cloning in
many minds. I hate to break the news,
but there really isnt much support for
nuclear transfer, said Franco Furger,
director of the Human Biotechnology
Governance Forum at Johns Hopkins
University, who cited a variety of polls
to that effect. Michael Werner of the Biotech
Industry Organization warned that even
biotech investors dont distinguish between
stem cell research and cloning.
So far only one groupin South Korea
has successfully cloned a human embryo (Sci-
ence, 12 March, p. 1669), but more are on the
horizon. In the United Kingdom, the Interna-
tional Centre for Life in Newcastle last August
got the first license to clone human embryonic
cells for research (Science, 20 August, p.
1102). Another British researcher, Ian Wilmut,
is applying for a license to
use nuclear transfer to study
amyotropic lateral sclerosis.
At Harvard, Meltons propos-
al to use these techniques to
create embryonic stem cell
lines expressing the genes for
diabetes and Parkinsons and
Alzheimers disease is under
review, and Zon and George
Daley hope to create lines
expressing the genes for
blood diseases. China, India,
Japan, Singapore, Belgium,
and Israel have sanctioned
nuclear transfer; Sweden is
expected to do the same.
Given the qualms over
stem cell research and
cloning, some participants at last weeks
meeting thought voluntary guidelines would
be insufficient to reassure the public. Alison
Murdoch of the Newcastle center said she
holds the very strong view that really
strong regulation, with every embryo ac-
counted for, is needed. Speakers also saw a
need for oversight of the loosely regulated in
vitro fertilization industry so that any cell
line that might end up in a clinical applica-
tion can be shown to have a squeaky-clean
pedigree. As it stands now, said Michael Ma-
linowski of Louisiana State University
School of Law in Baton Rouge, much as-
sisted reproduction is human experimenta-
tion in the name of treatment.
Some called for a standing oversight body
like the Recombinant DNA Advisory Com-
mittee set up after Asilomar. Leon Kass, chair
of the Presidents Bioethics Council, noted
that Asilomar led to a voluntary gene-splicing
moratorium and called for a similar morato-
rium on nuclear transfer. This is momentous
enough that it should be decided on a national
level, he said. The committee is to come up
with proposed guidelines in February.
CONSTANCE HOLDEN
Stem Cell Researchers Mull
Ideas for Self-Regulation
B I OE THI CS
What scientists want. This colony of human ES cells
was cultivated from a blastocyst that a South Korean
group created using nuclear transfer.
Leading the way. Harvards Melton.
Published by AAAS
C
R
E
D
I
T
:
J
P
L
/
L
M
A
5 9 0 5 9 2 5 9 6 5 9 9 6 0 3
The immune
systems
regulatory cops
Airborne
threat to
Hawaii
Oscillating
climate
Foc us
After years of dashed hopes, a vaccine against
malaria has shown tantalizing results in a
clinical trial in Mozambique. In a study in-
volving 2022 children between the ages of 1
and 4, the vaccine lowered a recipients
chance of developing malaria symptoms by
30%. The results are the most promising so
far in the search for a vaccine against a dis-
ease that kills between 1 million and 3 million
people per year. Malaria has had a sense of
hopelessness and intractability around it,
says Melinda Moree of the Malaria Vaccine
Initiative, an independent nonprofit group
that helped fund the trial. These results bring
hope to us all that a vaccine may be possible.
Even so, the approach faces several hurdles,
including whether the complex vaccine would
be affordable in the poor countries most af-
fected by the disease.
A consortium led by GlaxoSmithKline
(GSK) Biologicals in Rixensart, Belgium,
developed the vaccine, called RTS,S/AS02A.
It uses several techniques to boost the immune
systems fight against the malaria parasite. Its
designers engineered a hybrid protein that
combines a protein fragment from the para-
site, Plasmodium falciparum, with a piece of a
protein from the hepatitis B virus. The Plas-
modium protein is a promising target because
it is present on the parasites surface when it is
first injected into the bloodstream by the bite
of an infected mosquito. The hepatitis B pro-
tein is included because it is particularly effec-
tive at prompting an immune response. The
vaccine also contains a powerful new adju-
vant, developed by GSK Biologicals, that in-
creases the bodys production of antibodies
and T cells.
The combination seemed to work, at least
partially. Although it didnt prevent all children
from being infected with the parasite, it did
seem to keep them from becoming sick. Chil-
dren who received the full three doses of the
vaccine were 30% less likely to develop clini-
cal malaria in the first 6 months following the
injections, a team from GSK Biologicals and
the University of Barcelona reported in the 16
October issue of The Lancet. Data suggest that
the vaccine is considerably more effective at
preventing the most dangerous form of the dis-
ease, lowering a recipients risk of severe
malaria by 58%. Among children between
ages 1 and 2, the results looked even better:
The vaccine seemed to reduce the chance of
severe malaria by 77%, although the numbers
were quite small. These are the best results
weve ever seen with a candidate malaria vac-
cine, says Pedro Alonso of the University of
Barcelona in Spain, who led the trial.
The sheer number of malaria parasites that
people in endemic areas are exposed to makes
it difficult to develop a vaccine that prevents
all infection, notes GSK scientist Joe Cohen.
In addition, Plasmodium has evolved multiple
ways to elude the human immune system.
Cohen says scientists arent sure exactly
A Complex New Vaccine Shows Promise
MAL ARI A
A design error by spacecraft contractor Lock-
heed Martin Astronautics Inc. caused engi-
neers to install critical sensors upside down in
the Genesis sample return capsule, dooming it
to slam into the Utah desert floor last month at
360 kilometers per hour, according to the chair
of the mishap investigation board. The acci-
dent, Lockheed Martins third major incident
of late, may be another reminder of an era
when space missions were underfunded, too
rushed, and undermanaged. Chances are good,
however, that an identically equipped space-
craft, Stardust, will escape a similar fate.
According to board chair Michael
Ryschkewitsch of NASAs Goddard Space
Flight Center in Greenbelt, Maryland, if the
two pairs of sensors had been installed right-
side up, they would have triggered Genesiss
parachutes. Flipped according to incorrect
drawings that assemblers were following, the
sensors spring-loaded weights were already
at the end of their range of possible motion as
the capsule hit the upper atmosphere and
began slowing, so deceleration could not
drive them through the required trigger point.
The snafu recalls two earlier mishaps
involving Denver-based Lockheed Martin as
contractor and NASAs
Jet Propulsion Laborato-
ry (JPL) as spacecraft
operator. In 1999, the
Mars Climate Orbiter
broke up as it skimmed
too close to Mars. Engi-
neers at the two organi-
zations had misunder-
stood which units of
thrustEnglish or met-
ricthe other group
was using. And in the
same year, Mars Polar
Lander crashed onto the
surface after a software error caused its retro-
rockets to shut down too far above the surface.
Before pointing fingers over Genesis, says
space policy analyst John Logsdon of George
Washington University in Washington, D.C.,
critics should consider its history. Although
Genesis launched late enough to get additional
reviews after the 1999 Mars losses, it was
designed years earlier, at the height of the
faster, cheaper, better era of NASA mission
design. Spacecraft were being designed, built,
and operated by fewer people in less time than
ever before. Gene-
sis was thus prone
to the same sorts
of problems as
the Mars probes,
although its partic-
ular problem still
should have been
caught by later re-
views, says Logs-
don. In its final re-
port, due out by
early December,
Ryschkewitschs
board hopes to doc-
ument why those reexaminations failed.
More pressing, perhaps, is the state of the
Stardust spacecrafts sensors. Also a Lockheed
Martin/JPL mission, Stardust will be depend-
ing on identical sensors to trigger its landing
sequence in January 2006 as it returns sam-
ples of comet dust. Preliminary indications
are that the design and installation of the
switches on Stardust are correct, says NASA
deputy associate administrator Orlando
Figueroa. Time will no doubt tell.
RICHARD A. KERR
P L ANE TARY S CI E NCE
Fateful reversal. Incorrect drawi ngs l ed
assemblers to install critical sensors upside down.
L
Flipped Switch Sealed the Fate of Genesis Spacecraft
Published by AAAS
how the vaccine works,
but they suspect that the
antibodies and T cells
produced may both inter-
rupt the parasites ability
to infect liver cells and
help the immune system
target infected cells for
destruction.
Alonso notes that
even partial protection
against malaria could
save thousands of lives
every year. Combined
with techniques such as
using bed nets and in-
secticides, the vaccine
could have a huge im-
pact, he says. But he
and others caution that
the vaccine still must be
tested for efficacy and
safety in younger chil-
dren, as large-scale immunization efforts in
Africa target children younger than 1 year.
It is a very exciting, encouraging result
that establishes the feasibility of developing a
malaria vaccine, says Stephen Hoffman of
Sanaria, a Rockville, Marylandbased
biotech company working on a different type
of malaria vaccine. But questions remain
about the GSK vaccine. Candidate vaccines
for other diseases have
seemed to protect young
children only to prove in-
effective in infants, he
notes. It is not yet clear
how long the protection
lasts. And the vaccine
also has to show its met-
tle in areas with more in-
tense malaria transmis-
sion than Mozambique.
Another possible
drawback is the vac-
cines cost, which Jean
Stephenne, president of
GSK Biologicals, esti-
mated at $10 to $20 per
dose; multiple doses will
likely be needed. Cohen
acknowledges that it
wont be cheap to pro-
duce. If it gets on the
market, it would be the
most complex vaccine ever developed, he
says. Hoffman notes that the vaccine is about
as effective as bed nets and other conventional
malaria prevention methods, although it
would be much more expensive. (A bed net
typically costs about $5.) Moree agrees: Any
vaccine that goes forward will have to be cost-
effective, or it will not be used.
GRETCHEN VOGEL
www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004
C
R
E
D
I
T
:
H
O
S
P
I
T
A
L

C
L
I
N
I
C

B
A
R
C
E
L
O
N
A
589
NSF Worries About Growing
Antarctic Ice
National Science Foundation (NSF) offi-
cials are hoping that an unexpected in-
crease in Antarctic sea ice wont compli-
cate plans to resupply two U.S. research
stations by ship. But just in case, they are
already pondering how else to get
23 million liters of fuel and 5 million kilo-
grams of material to the McMurdo and
South Pole stations during the busy aus-
tral summer scientific season.
For the past 4 years, the Coast Guard
has used both of its polar-class icebreak-
ers to clear paths for cargo ships through
the ice around McMurdo Sound and
channel. Although one of the behemoths
is now awaiting repairs, in July, Coast
Guard officials said that going solo would
be fine because the newly formed 1-
meter-thick ice cover was only 40 kilo-
meters. But by August it had grown to
200 kilometers. So next week the ser-
vices Polar Star will set out alone from
Seattle (a 6-week trip) to battle the ice.
Were still confident Polar Star can get
the job done, says the Coast Guards Capt.
Dennis Holland. NSF officials hope hes
right. But last week they outlined several al-
ternatives, including renting a commercial
ice breaker and offloading fuel and supplies
several kilometers up the channel.
JEFFREY MERVIS
As South Pole Research
Group Aims for Fresh Start
The international body that oversees re-
search in Antarctica hopes that an infu-
sion of funds and ideas will rescue it from
its scientific doldrums. We need to focus
on some big-issue science to raise our
profile, says Colin Summerhayes, execu-
tive director of the U.K.-based Scientific
Committee on Antarctic Research, a 32-
nation body that earlier this month
approved a new research agenda, changes
to its organizational structure, and a dues
increase.
The new research plan calls for Antarctic
scientists to focus on five interdisciplinary
themes, including Antarcticas role in global
climate and ocean systems (see
www.scar.org). The one-time dues increase
will help erase a $100,000 shortfall in the
groups $322,000 annual budget. Germany
and the United Kingdom have pledged to
double their contributions, to $28,000 an-
nually through 2008. But persuading other
countries to follow suit will be a bit of a
headache, predicts Chris Rapley, director of
the British Antarctic Survey.
FIONA PROFFITT
ScienceScope
New hope. A team member treats a
child with malaria at the Manhica
Health Research Center in Mozambique.
Martin Backs Science Academy
OTTAWACanadian Prime Minister Paul
Martin has given the green light to what one
prominent scientist calls scientific advice
on a shoestring budget.
On 5 October, Martin promised that his
budget next spring would include $27.6 mil-
lion over 10 years for a Canadian Academies
of Science (CAS). The announcement culmi-
nates a decade-long campaign by leading sci-
entists for a national organization that would
deliver independent assessments of pressing
scientific questions. But its status is dependent
on the survival of Martins minority govern-
ment, which narrowly avoided being toppled
in a procedural vote following his speech.
CAS would be run by the Royal Society
of Canada, the Canadian Academy of Engi-
neering (CAE), and the Canadian Institute of
Advanced Medicine. Officials from the
three organizations have long touted the idea
(Science, 27 October 2000, p. 685). Were
practically the only country in the world that
doesnt have such an organization, adds
CAE executive director Philip Cockshutt.
Royal Society past president William
Leiss estimates that CAS will carry out no
more than five studies a yearcompared
with the 200 or so churned out annually by
the U.S. National Academies, on which CAS
is modeledwith help from a small CAS
secretariat. A board of governors, featuring
two representatives apiece from the three
founding organizations and six members of
the public, will choose the panelists for each
study. (The board could grow if other mem-
bers join CAS.)
CAS may also do a small number of self-
initiated studies, Leiss said. But he expects
the government to provide the bulk of the
academies support. What theyll get is a
kind of definitive resolution of some really
thorny issues, says Leiss, adding that all its
reports would become public.
Leiss credits Canadas new science adviser,
Arthur Carty, with putting the campaign over
the top. Carty will serve as the gatekeeper and
conduit between the government and the new
academy, submitting formal requests for the
academy to undertake a study and receiving
its final reports. Carty says the academy will
give Canada a voice on the international
science front and a point of entry for coun-
tries seeking its input on international proj-
ects. He also plans to consult with its mem-
bers in preparing his recommendations to
government. WAYNE KONDRO
CANADA
Published by AAAS
Taking a high-stakes legal gamble that could
lengthen his 2-year prison term, former
plague researcher Thomas Butler is appeal-
ing his conviction for mishandling bacteria
samples and defrauding his university. Gov-
ernment prosecutors say they will respond
with their own request to erase a judges de-
cision that cut 7 years off a possible 9-year
prison term.
Butler is taking a huge, huge risk, says
former prosecutor Larry Cunningham, a law
professor at Texas Tech University in Lub-
bock. The judge gave him a sweet deal; this
gives the government a shot at overturning it.
Butler is willing to risk a longer sen-
tence to fight for important principles, says
Jonathan Turley, one of Butlers attorneys
and a law professor at George Washington
University in Washington, D.C. The trial
was rife with irregularities; the government
is pursuing a longer sentence because it is
embarrassed about losing its core case.
Prosecutors declined comment.
Butler, 63, captured national headlines
last year after he reported 30 vials of plague
bacteria missing from his Texas Tech labora-
tory, sparking a bioterror scare (Science,
19 December 2003, p. 2054). The govern-
ment ultimately charged him with 69 counts
of lying to investigators, moving plague bac-
teria without proper permits, tax fraud, and
stealing from his university by diverting
clinical trial payments to his own use. Last
December, a Texas jury acquitted him of the
central lying charge and most of the plague-
related allegations but convicted him on 44
financial charges and three export violations
involving a mismarked Federal Express
package containing bacteria.
Although government sentencing guide-
lines called for a 9-year sentence, federal
judge Sam Cummings reduced it to 2 years,
in part because Butlers research had led to
the salvage of millions of lives. Butler is
currently in a Texas prison.
Prosecutors were unhappy with the sen-
tence, say sources familiar with the case, but
agreed not to challenge it unless Butler filed
an appeal. He recently did just that, arguing
in an 80-page brief that his trial was marred
by the governments refusal to try him sepa-
rately on the plague and financial charges,
its use of vague university financial policies
as the basis for criminal charges, and a
judges ruling that barred Butler from gain-
ing access to university e-mails. He is asking
the appeals court to strike down the convic-
tions or at least order a new trial. Prosecu-
tors are expected to file a response later this
month, and a hearing in New Orleans,
Louisiana, could come as early as January.
Butler has rolled the legal dice before. He
rejected a pretrial government plea bargain
offer that included 6 months in jail. Turley
expects the government to ask the appeals
court to impose the full 10-year sentence
allowed by the export violations but says
that move would be a vindictive, gross
abuse of prosecutorial discretion.
If the government wins, Butler will lose
more than his argument. Because the appeal
is expected to take longer than his current
sentence, he could find himself back in
prison after spending time as a free man.
DAVID MALAKOFF
Butler Appeals Conviction, Risking Longer Sentence
B I OTE RRORI S M AND THE COURTS
Bird Flu Infected 1000, Dutch Researchers Say
AMSTERDAMAt least 1000 peoplemany
more than assumedcontracted an avian
influenza virus during a massive poultry out-
break in the Netherlands last year, according
to a new study. In another unexpected find-
ing, those who developed symptoms after be-
ing infected passed the virus on to
a whopping 59% of their house-
hold contacts, say the researchers
at the National Institute for Public
Health and the Environment
(RIVM), whose results were pub-
lished in Dutch last week.
Flu experts were cautious in
discussing the findings, which
they had not yet been able to read.
But if correct, they are another
warning signal, says Klaus Sthr,
head of the World Health Organi-
zations global influenza program.
Every time an avian virus infects a
human being, Sthr says, the risk
that it will mutate into a pandemic
strain grows.
Almost 31 million poultry were culled
in the Netherlands before the virus, a strain
called H7N7, was contained. By the end of
the outbreak, the virus had killed one vet-
erinarian, and some 450 people had report-
ed health complaints, mostly an eye infec-
tion called conjunctivitis. In a paper pub-
lished in The Lancet in February, RIVM
virologist Marion Koopmans and her col-
leagues reported that they detected the
H7N7 virususing the polymerase chain
reaction or by culturing the virusin eye
swabs of 89 of them.
To gauge the true reach of H7N7, Koop-
mans and her colleagues also tested those at
risk, such as poultry farmers and those hired
to cull and remove poultry, for antibodies
against the virus. This test provides more de-
finitive and longer-lasting proof of infection.
They used a new variation on the classic
hemagglutinin inhibition test, which the
team says is better at picking up antibodies
to avian flu in humans. (It uses red blood
cells from horses, rather than turkeys or
chickens, in a key step.)
They found antibodies in about half of
500 people who had handled infected poultry;
based on the total number of poultry workers
at risk, the team concludes that at least 1000
people must have become infected, most of
them without symptoms. Wearing a mask and
goggles did not seem to prevent infection;
taking an antiviral drug called oseltamivir
(Tamiflu) did, but a quarter of the cullers and
half of the farmers did not use the drugs.
Among 62 household contacts of conjunc-
tivitis patients, 33 became infectedanother
surprisingly high figure, Sthr says. Having a
pet bird at home increased household mem-
bers risk of becoming infected, perhaps
because the birds replicated the virus too.
Detecting antibodies to avian influenza is
tricky, and the results need to be corroborat-
ed, cautions flu specialist Maria Zambon of
the U.K. Health Protection Agency, whose lab
may retest the Dutch samples.
Human antibody tests for H5N1, the avian
flu virus currently ravaging Asian poultry, are
ongoing, Sthr says. So far, the results show
that, although far more lethal to humans, the
virus has caused few, if any, infections beyond
the known 43 patients. MARTIN ENSERINK
I NF E CTI OUS DI S E AS E S
C
R
E
D
I
T
:
M
I
C
H
A
E
L

K
O
O
R
E
N
/
R
E
U
T
E
R
S
Take your pills. Many of those exposed to infected chickens
did not take antiviral drugs, the study found.
N E W S O F T H E W E E K
Published by AAAS
C
R
E
D
I
T
:
D
A
R
R
E
L
L

K
A
P
C
Z
Y
N
S
K
I
ScienceScope
591
Fly Me to the Moon?
NASAs new research plan for sending
humans to the moon and Mars is a solid
beginning, says a National Academies
interim report released last week. But it
cautions that limited space-based stud-
ies, funding uncertainties, and NASAs
failure to adequately consider radiation
and psychological hazards to astronauts
could hamper preparation for long-
duration missions.
To gain better insight into the chal-
lenges of space living, the panelled by
anesthetist David Longnecker of the Uni-
versity of Pennsylvania in Philadelphia
recommends that NASA make greater use
of Earth-based experiments, including trials
in extreme environments such as polar
bases and underwater shelters. The panel is
due to release a final report next August,
and Guy Fogelman, head of NASAs new
human health and performance office, says
the agency is likely to adopt most or all
of the recommendations.
ANDREW LAWLER
Schwarzenegger Backs
Stem Cell Initiative
Californias popular Governator has
weighed in on Proposition 71, giving the
states $3 billion stem cell bond initiative
a boost that supporters predict will result
in a big win on election day. Republican
Governor Arnold Schwarzenegger this
week defied his party with a surprise en-
dorsement of the measure, saying that I
am very much interested in stem cell re-
search and support it 100%.
Schwarzenegger won his post with a
promise to rein in the states spiraling
budget deficit. Although opponents of
Proposition 71 predict that it would add
to the states debt, Schwarzenegger sided
with supporters who claim it would spark
a biotech boom. Initiative backers made
a compelling fiscal argument that he
bought, says stem cell researcher Evan
Snyder of the Burnham Institute in La
Jolla, California. We are ecstatic.
Schwarzeneggers decision could
cause political ripples beyond California,
says Matthew Nisbet, a communications
professor at Ohio State University in
Columbus who studies public opinion on
the stem cell issue. He says the endorse-
ment will reinvigorate the issue in the
[presidential] campaign too.
CONSTANCE HOLDEN
Just days after publishing a well-received
study in which they engineered the 1918
pandemic influenza virus to find out why it
was so deadly, researchers are catching flak
from critics who say their safety precautions
were inadequate. The lead investigator,
Yoshihiro Kawaoka, contends his team fol-
lowed federal guidelines. But critics say
these rules are out of date.
The brouhaha erupted after Kawaokas
team at the University of Wisconsin
and the University of Tokyo reported
in the 7 October issue of Nature that a
normal human flu virus containing a
gene for a coat protein from the 1918
flu strain is highly pathogenic to mice.
An article in the New York Times noted
that although the team began the stud-
ies in a stringent biosafety level 4
(BSL-4) lab in Canada, where workers
wear space suits, the University of
Wisconsins safety board approved
moving the work to its own BSL-3 lab.
That set off alarm bells with
some biosafety experts, including
Karl M. Johnson, a former Centers
for Disease Control and Prevention
(CDC) virologist now retired in
Albuquerque, New Mexico. He and
several others wrote to Promed, an Internet
e-mail forum widely read in the infectious-
disease community, arguing that the move
to a BSL-3 was dangerous.
Kawaoka responds that the critics do not
know the full extent of his teams precautions.
Among other steps, his workers get the regular
flu vaccine, which protects mice against 1918-
like flu viruses. All workers also take the anti-
flu drug Tamiflu prophylactically. Work by
Kawaokas group (in the BSL-4 facility) and
by a federal lab has shown that the antiviral
works extremely well at protecting mice
against 1918-like flu strains, Kawaoka says.
According to National Institutes of Health
(NIH) guidelines for research with recombi-
nant organisms, this puts the work in the
BSL-3 category, he notes.
Kawaokas group also beefed up its lab to
what is informally known as an enhanced
BSL-3, or BSL-3+. For example, workers
wear battery-powered air purifiers with face
shields and shower when they leave the lab.
When Kawaoka presented the work in Sep-
tember to NIHs Recombinant DNA Adviso-
ry Committee (RAC), which was reviewing
research with pathogenic viruses, no mem-
ber raised any concern, he says.
Johnson is partly assuaged. I feel a bit
better, he says, adding that BSL-3+ may be
adequate for some experiments with engi-
neered 1918 flu. But he still has reservations
about, for instance, whether the vaccine
would fully protect some individuals.
Other critics on Promed, however, such as
Ronald Voorhees of the New Mexico Depart-
ment of Health, argue that antiviral drugs may
not eliminate the risk of a worker passing the
virus to someone outside the lab, so a BSL-4
facility is needed. Biosafety expert Emmett
Barkley of Howard Hughes Medical Institute
suggests that if experts were polled, half of
them would call for [BSL-4].
Part of the confusion stems from another
set of federal guidelines, Biosafety in Biomed-
ical and Microbiological Laboratories
(BMBL). This manual says that flu viruses re-
quire only BSL-2 facilities, and there is no
mention of 1918 flu or enhanced BSL-3,
Johnson notes. The issue is important to re-
solve, as Kawaokas is not the only group
working on 1918-like flu viruses. A group
led by Mount Sinai School of Medicine in
New York City is doing so in a BSL-3+ facil-
ity at the CDC, and the University of Wash-
ington plans to study monkeys infected with
modified 1918 flu strains in a BSL-3+ lab.
Even more controversial are planned ex-
periments that would mix pathogenic avian
flu strains, such as the H5N1 strain now cir-
culating in Asia, with human flu viruses (Sci-
ence, 30 July, p. 594). CDC scientists have
opted for BSL-3+, but flu expert Robert Web-
ster of St. Jude Childrens Research Hospital
in Memphis, Tennessee, says he would do
these studies in a BSL-4 facility.
Clearer guidance may emerge in the next
version of BMBL, due out in 2005. Mean-
while, RAC expects to release points to con-
sider in December. JOCELYN KAISER
1918 Flu Experiments Spark
Concerns About Biosafety
B I OCONTAI NME NT
Safety risk? Some experts question whether BSL-3+ condi-
tions, like the air purifier worn here, would prevent engi-
neered 1918 pandemic flu from escaping.
Published by AAAS
The worlds a dragand
scientists have proved it.
By studying the dance of
two Earth-orbiting satel-
lites, Italian physicists have
detected the subtle twisting
of spacetime around a
massive, spinning object.
The measurement is the
most convincing sighting
yet of a hard-to-spot conse-
quence of Albert Einsteins
general theory of relativity,
says Neil Ashby, a physicist
at the University of Col-
orado, Boulder. There was
a lot of criticism of previ-
ous results, but this is the first reasonably ac-
curate measurement, Ashby says. Physicists
Ignazio Ciufolini of the University of Lecce,
Italy, and Erricos Pavlis of Goddard Space
Flight Center in Greenbelt, Maryland,
describe the result this week in Nature.
General relativity predicts that a spinning
mass drags the fabric of space and time
around with it, much as a restless sleeper
drags the sheets around while twisting and
turning in bed. This effect, known as the
Lense-Thirring or frame dragging effect,
is difficult to detect. To spot it, one has to
observe how a spinning body changes the
orientations of nearby gyroscopesmuch
tougher than seeing, say, how a massive
body bends light.
Ciufolini, Pavlis, and colleagues tried to
measure the effect in 1998 by using two
satellites, LAGEOS and LAGEOS II, as test
gyroscopes. The satelliteshalf-meter-wide
mirrored sphereswere launched in 1976
and 1992 as targets for laser range finders,
which can track their position within a few
centimeters. As the satellites spin around
Earth, the Lense-Thirring effect twists the
planes of their orbits slightly. The early
measurements were very rough, Ciufolini
says, because the uneven distribution of
Earths mass causes similar orbital distor-
tions thousands of times greater than those
due to the Lense-Thirring effect. [A satel-
lites] precession is about 2 meters per year
due to frame-dragging. The precession due
to the oblateness of the Earth is many thou-
sands of kilometers per year, Ashby says.
Because the mass distribution was poorly
known, Ciufolini and colleagues had to
make a few controversial estimates, includ-
ing one about how the satellites perigees
precess. As a result, their published value of
the Lense-Thirring effect had a large error
20%and even that was greeted with some
skepticism.
Now, thanks
to better gravitational
maps produced by
twin satellites known
as GRACE, as well
as improved gravita-
tional models and
other ref inements,
the perigee estimation is no longer needed.
The result is a much firmer detection with
an error of about 10%. I believe that in a
few years, more accurate gravitational field
models and a longer period of observation
will make it more and more accurate, says
Ciufolini. It will be at the level of a few
percent. By then, physicists hope, the
gyroscope-laden satellite Gravity Probe B,
which was designed to detect the
Lense-Thirring effect, will have
produced results with an error of
about 1%far lower than the two
LAGEOS satellites can achieve
(Science, 16 April, p. 385). I
think that the biggest contribution
is a validation of what, in a year or
so, will be the results from Gravity Probe
B, says Richard Matzner, a physicist at the
University of Texas, Austin.
To keep Gravity Probe B from being
the only game in town, Ciufolini and other
researchers are pushing to loft another
LAGEOS satellite into an orbit that would
completely eliminate the effects of Earths
mass distribution. If we had a third satel-
lite, we could go even below the 1% limi-
tation, says Ciufolini. Of course, getting
funding for a new satellite is quite a
dragone much easier to sense than the
Lense-Thirring effect. CHARLES SEIFE
Swiveling Satellites See Earths Relativistic Wake
GE NE RAL RE L ATI VI TY
Metabolic Defects Tied to Mitochondrial Gene
Abnormally high blood pressure is bad
enough by itself; it predisposes people to
diseases such as kidney failure, heart at-
tacks, and strokes. But for an estimated
47 million Americans with so-called meta-
bolic syndrome, high blood pressure (hyper-
tension) comes hand in hand with other car-
diovascular risk factors such as diabetes and
high blood concentrations of cholesterol
and triglycerides. Obese people often have
metabolic syndrome, but so do some
nonobese people, so excess weight
isnt the sole cause.
Findings published online
this week by Science (www.
sciencemag.org/cgi/content/
abstract/1102521) now point
to an unexpected new cul-
prit. The work, by a team led
by Richard Lifton of Yale
University School of Medi-
cine, shows that a mutation in
a mitochondrial gene causes peo-
ple to develop a constellation of
symptomshypertension, high concen-
trations of blood cholesterol, and lower-than-
normal concentrations of magnesiumsimi-
lar to those of metabolic syndrome.
The mutation likely disrupts the function
of mitochondria, subcellular structures that
provide most of a cells energy and have
their own small genome. Despite uncertainty
about how a mitochondrial DNA mutation
could lead to such diverse symptoms, hyper-
tension expert Theodore Kurtz of the Uni-
versity of California, San Francisco, says
that the finding could be of tremendous im-
portance. Previously, few cardiologists
looked to the mitochondria for insight into
hypertension and other cardiovascular risk
factors, but this, he says, could shift the in-
terest dramatically.
The new discovery grew out
of the examination of a fe-
male patient who was
suffering from low
blood magnesium. Lifton
and his colleagues had
previously discovered a
handful of genes that,
when mutated, cause this blood condition,
which is characterized by general malaise and
weakness. In the course of conversations with
the woman, she mentioned that several of her
relatives also suffered from low blood magne-
sium. Whats more, her family was a gene
hunters dream. It was extremely large and
ME DI CI NE
Seat of the problem? Mal-
function of the mitochondria
(blue) may underlie problems of
aging such as hypertension and
high blood cholesterol.
Curve balls. LAGEOSs laser-ranging
satellites revealed a twist in spacetime.
C
R
E
D
I
T
S

(
T
O
P
T
O

B
O
T
T
O
M
)
:
N
A
S
A
;
G
O
P
A
L

M
U
R
T
I
/
V
I
S
U
A
L
S

U
N
L
I
M
I
T
E
D

L
LAGEOS II
LAGEOS I
Published by AAAS
all lived close to one another, Lifton says.
Further investigation turned up 142 rela-
tives, many of whom had low magnesium,
hypertension, elevated blood cholesterol
concentrations, or some combination of
those problems. Even more intriguing, in all
cases, the traits had been inherited from the
individuals mothersa clear indication that
the gene at fault was located in the
mitochondrial genome. The genes that
Lifton has previously linked to low blood
magnesium had all been nuclear.
The mitochondrial location of the new
gene mutation was a big advantage because
that genome consists of only 16,000 base
pairs as opposed to the 3 billion in the
nuclear genome. Analysis of the mitochon-
drial genome of family members turned up
one mutation found only in affected mem-
bers and not detected in any of the thou-
sands of mitochondrial genomes previously
sequenced. This mutation altered one
basea thymidine was changed to a cyto-
sinein the gene for a mitochondrial trans-
fer RNA (tRNA), which carries amino acids
to the ribosome for protein synthesis.
Because virtually all tRNAs have a thymi-
dine at that spot, implying that its essential
for the molecules function, the swap likely
disrupts the tRNAs structure and interferes
with protein synthesis in the mitochondria.
The [thymidine] is extremely conserved,
says Carlos Moraes, an expert on mitochon-
drial genetics at the University of Miami,
Florida. That does indicate that the mutation
could cause some kind of problem.
Moraes adds that hes surprised that
people with the mutation dont suffer even
more serious problems. Previous mutations
found in mitochondrial tRNA genes have
caused, among other things, muscle and
nerve degeneration, although the extent of
the damage can vary.
A key question now is how the mutation
produces hypertension and the other symp-
toms, which seem to be independent of one
another. The low blood magnesium levels,
which appear even in children, might be due
to failure of the kidney to remove the
mineral from the urine before its excreted
a process that requires a great deal of energy.
In contrast, blood pressure and cholesterol
concentrations were normal in young indi-
viduals but began increasing at about age 30.
That suggests additional factors come into
play with age. These might be environmen-
talsay, a high-fat dietor related to the
declining mitochondrial function that some
researchers think contributes to aging.
Another crucial unknown is whether
mitochondrial dysfunction contributes to
metabolic syndrome in the general popula-
tion. Kurtz thinks it might. The implications
are much greater than this finding in one fam-
ily, he predicts. JEAN MARX
C
R
E
D
I
T
:
G
A
R
Y
W
.
M
E
E
K
/
G
E
O
R
G
I
A
T
E
C
H
An information theorist doesnt care
whether a quantum bit is stored on parti-
cles of matter or particles of light. But to
an experimentalist, its a very big deal
where your qubit is stored because slow
but long-lived chunks of matter have very
different properties from those of quick
but evanescent photons. Now, two physi-
cists have shown how they can reliably
transfer quantum information from matter
to light. The procedure may soon enable
scientists to exploit the advantages of both
matter and light in building systems for
quantum communications.
Its a breakthrough that is needed,
says Klaus Mller, a physicist at the Uni-
versity of Aarhus in Denmark. Its a
bridge between traveling qubitslight
and stationary ones.
For years, physicists have been excited
about using the properties of quantum theory
in computing and communications. In theory,
a quantum computer could solve certain
problems (such as cracking a code) much
faster than a classical computer can; a com-
munications system built upon quantum-
mechanical particles would be functionally
immune from eavesdropping.
But quantum computers and quantum
communications depend on having infor-
mation stored on quantum objects like
atoms and photons rather than classical
ones like hunks of siliconand quantum
objects are hard to handle. Quantum bits
stored on particles of light travel well
they can zip for kilometers down a fiber-
optic cablebut they are tricky to store.
Quantum bits stored on matter keep for
milliseconds or longer, but theyre usually
confined to a trap and cant be transmit-
ted from place to place.
In this issue (p. 663), Alexei Kuzmich
and Dmitri Matsukevich of the Georgia
Institute of Technology in Atlanta describe
how they store a quantum bit in a cloud of
rubidium atoms and induce the cloud to
inscribe that information, undamaged,
upon a photon. The researchers start with
two clouds of rubidium gas. By shooting a
laser through both clouds simultaneously,
they force the clouds to emit a single photon
that is quantum-mechanically entangled
with both of the clouds simultaneously.
(Quantum indeterminacy and the experi-
mental setup make it impossible to say the
photon came from one cloud or the other.)
The entanglement links the
fates of the photon and the
clouds; tweaking the pho-
tons polarization alters the
quantum state of the clouds.
By manipulating the pho-
ton, the physicists can in-
scribe a quantum bit upon
the two clouds.
A few hundred nano-
seconds later, the researchers
read out the information by
shining another laser upon
the rubidium samples. The
laser induces the clouds to
emit another photona pho-
ton whose polarization con-
tains the information that the
researchers had inscribed
upon the cloud. That laser-driven retrieval
transfers quantum information from matter to
light, Kuzmich says.
Although the procedure is beset by losses
due, in part, to the inefficiency of rubidium
clouds absorption of laser light, Kuzmich
believes that the method will lead to useful
tools for quantum communication. To be
honest, I think that this will be a practical
device eventually, he says. Indeed, he and
Matsukevich are hard at work trying to
hook up two of the matter-light devices to
create a quantum repeateran amplifier
that reverses the inevitable loss of signal
strength that occurs when light is sent down
a long stretch of fiber-optic cable. Such re-
peaters would be essential for long-distance
quantum communication.
I firmly believe that these guys will do
it, says Mller. If they do, he says, quantum
communications and large-scale quantum
computation will be considerably closer than
before. Everyone will benefit.
CHARLES SEIFE
Researchers Build Quantum Info Bank
By Writing on the Clouds
P HYS I CS
Bright bits. By routing laser beams through wisps of gas, physi-
cists shuttled information between light and matter.
Published by AAAS
Where are the women? Thats what some sci-
entists are asking after the National Institutes
of Health (NIH) picked nine men to receive
the inaugural Directors Pioneer Award for in-
novative research (Science, 8 October, p. 220).
The 5-year, $500,000-a-year awards are
part of NIHs roadmap for increasing the
payoff from the agencys $28 billion budget,
and Director Elias Zerhouni has compared
the winners to famed U.S. explorers Merri-
weather Lewis and William Clark for their
willingness to explore uncharted territory.
Within hours of the 29 September
announcement, however, some researchers
had begun to bristle at the gender imbalance
in that first class of biomedical pioneers.
It sends a message to women researchers
that they are not on an even playing field,
wrote Elizabeth Ivey, president of the Associa-
tion for Women in Science, in a 1 October let-
ter to Zerhouni. I hope that you [will] make
an effort to correct such a perception. The
American Society for Cell Biology, in a
15 October letter to Zerhouni, commended
him for creating the prize but lamented its
demoralizing effect on the community. Crit-
ics noted that men constituted 94% (60 of 64)
of the reviewers tapped to help winnow down
some 1300 applications for the award and
seven of the eight outside scientists on the
final review panel, which grilled applicants
for an hour before settling on the winners.
NIH officials estimate that women made
up about 20% of the Pioneer applicants. But
only about 13% of the 240 who made it
through the first cut were women, and only
two of the 21 finalists. (In contrast, about
25% of the applicants for NIHs bread-and-
butter R01 awards to individual investiga-
tors are women, and their success rate is
within a percentage point of that
of their male counterparts.)
With any elite award, there are
so many deserving candi-
dates that its easy to
choose only men, says
Stanford University
neuroscientist Ben
Barres, who says he
was outraged by
the gender imbal-
ance. I actually
think its more a
matter of neglect
than of sexism.
The gender of
the final applicants
did not come up dur-
ing the discussion, says
review panel member Ju-
dith Swain, professor of medi-
cine at Stanford. Swain, who called the ex-
ercise the most interesting review panel
Ive ever been involved in, says she saw
no evidence of active discrimination. But
she concurs that the demographics of the
reviewers and the winners lead to a dis-
turbing observation.
NIH officials are struggling to find the
best way to respond to the charges of gender
insensitivity. Stephen Straus, head of the Na-
tional Center for Complementary and Alter-
native Medicine and team leader for the
NIH-wide competition, told Science on the
day of the awards that we gave the gender
issue a great deal of thought, but none of the
women finalists came close to making the
pay line. A week later, in the first of a series
of e-mail exchanges with Barres, Straus re-
marked that the absence of women was
noted with some surprise by senior NIH
officials and that we know we can do bet-
ter in subsequent rounds. In a later ex-
change, however, Straus wrote, I dont be-
lieve that NIH can credibly discard its two-
level peer review system when nine
grants out of the many thou-
sands awarded this year turn
out differently than some
might wish.
NIH is evaluating
how it ran the Pio-
neer programin-
cluding how the
award was pub-
licized and the
demographics of the applicantsbe-
fore launching the next competition in Janu-
ary. A thorough review is essential, says
Arthur Kleinman, a medical anthropologist at
Harvard University and chair of the final re-
view panel, who believes NIH needs to do
more to reach several groupsminorities and
social and behavioral scientists as well as
womennot represented in the first batch of
winners. I agree that they need to be more
sensitive to diversity, he says. But at the
same time, I think Zerhouni deserves a lot of
credit for even trying something like this.
JEFFREY MERVIS
Male Sweep of New Award
Raises Questions of Bias
NATI ONAL I NS TI TUTE S OF HE ALTH
End of Cost Sharing Could Boost Competition
The National Science Foundation (NSF) is
leveling the playing field for grant seekers
by removing a mandatory cost-sharing
requirement for some programs. But the
move is expected to result in a more crowd-
ed playing field, too, as institutions that
couldnt afford the entry fee can now apply.
Last week the agencys policymaking
body, the National Science Board, voted to
eliminate cost-sharing rules that, in 2001, af-
fected roughly one-sixth of NSFs 20,000
grantsin particular large centers and major
instrumentation programsand added
$534 million to the agencys own spending on
research. In some cases, universities had to
come up with funding that matched the size
of the award.
The requirements were seen as a way to
stretch tax dollars and ensure that an institu-
tions leaders were committed to a proposal,
but university officials have long viewed
them as a hidden tax on federally funded re-
search. Two years ago, the board told NSF
that the amount a university promises to con-
tribute should not affect decisions on which
proposals get funded. The new rule goes one
step further, by banning mandatory cost-
sharing entirely except for a statutory 1% fee.
NSFs current policy represented an un-
fair burden on some institutions that couldnt
afford to enter the competition, says Mark
Wrighton, chancellor of Washington Univer-
sity in St. Louis, Missouri, and chair of the
board committee that recommended the
change. This will give schools greater flexi-
bility to invest their research dollars. The
board also asked for a report in 2 years on any
unintended consequences of the new policy.
NSF officials began cutting back on cost
sharing after the boards 2002 directive, and
this month a solicitation for its major research
instrumentation program, written before the
boards action, drops the requirement. But
theres a quid pro quo: Institutions must pick
up the full cost of maintenance and opera-
tions when a project starts rather than after
the grant expires. Even so, Dragana Brza-
kovic, who manages the instrumentation pro-
gram, expects success rates to drop from
nearly 40% to about 30% as more institutions
compete for the awards. JEFFREY MERVIS
NATI ONAL S CI E NCE F OUNDATI ON
Men at work. Nine men
won the first NIH Direc-
tors Pioneer Awards, cho-
sen by panels that included
few women.
S
O
U
R
C
E
:
N
I
H
of Applicants
of Finalists
of Winners
94% of
Evaluators
80%
90%
100%
M
E
N
Published by AAAS
From blood samples, pediatric immunolo-
gist Hans Ochs has diagnosed five infant
boys who all had the same devastating
problems. Their immune systems had gone
haywire, attacking their gastrointestinal
tracts within a few weeks of birth and
causing severe, intractable diarrhea. Way-
ward immune cells also laid siege to each
boys pancreas, producing diabetes around
3 months of age. Within months of their
births, several of the infants became de-
pleted of red and white blood cells from
the immune onslaught.
No cure exists for this rare and
frequently deadly immune disease
dubbed immune dysregulation,
polyendocrinopathy, entero-
pathy, X-linked syndrome
(IPEX). But thanks in large
part to recent work by Ochs
at the University of Washing-
ton School of Medicine in
Seattle and his colleagues, its
cause in most cases is now
known. A genetic defect se-
verely impairs, if not abolish-
es, the bodys ability to pro-
duce regulatory T cells, a mys-
terious class of immune cells ap-
parently designed to squelch dan-
gerous immune responses.
Once dismissed as artifacts
of misguided research, regulatory
T cellsoriginally called suppressor
T cellsare now white hot among immu-
nologists, thanks to a body of research that
over the past 8 years clarified their exis-
tence. Hundreds of researchers are flock-
ing to the field, which could lead to novel
treatments for an array of immune disor-
ders, such as type I diabetes, multiple scle-
rosis, graft-versus-host disease, and allergy,
that are far more common than IPEX.
Studying regulatory T cells may also pro-
vide clues to the treatment of cancer; the
cells actually seem to protect tumors
against immune attack.
All anyone is talking about these days is
regulatory T cells, says Ethan Shevach, a
cellular immunologist at the National Insti-
tute of Allergy and Infectious Diseases
(NIAID) in Bethesda, Maryland. Neverthe-
less, fundamental mysteries remain about
this newfound class of immune cells. Their
mechanism of action is almost totally
opaque. Also unclear is to what extent they
play roles in more ordinary human auto-
immune diseases that develop later in life,
such as diabetes and multiple sclerosis.
Regulatory T cell research is very intriguing
but is not yet ready for mass consumption,
Ochs says. There are still a lot of puzzles.
Even so, many researchers are optimistic
that studying regulatory T cells will lead to
new therapies. Drugs that seem to target
these cells are already being tested in people
with cancer and diabetes, and pharmaceuti-
cal companies are trying to develop drugs
that augment or suppress regulatory T cells
for other disorders. Within 5 years, some
clinical application of these cells will be
here, predicts Shimon Sakaguchi, an im-
munologist at Kyoto University in Japan and
a pioneer in the field.
From fantasy to reality
The human body makes several types of
T cells, including killer T cells, which eradi-
cate infected cells, and helper T cells, which
arouse killer T cells and various other
immune cells to fight invaders. And for
decades, researchers have kicked around the
idea that the body also makes a class of
T cells that act like a police departments in-
ternal affairs unit, keeping tabs on the im-
mune systems cellular cops and cracking
down on them if they threaten to spiral out of
control. In the early 1970s, the late Yale
immunologist Richard Gershon formally
proposed the existence of these sup-
pressor cells to explain a form of im-
mune tolerance he observed in a
mouse. The idea caught the fancy
of immunologists, and numerous
teams rushed to identify and
characterize these cells.
The entire concept of sup-
pressor T cells fell into disre-
pute, however, when no one
could verify reports of molecules
that supposedly characterized the
cells. It was a kind of fantasy not
supported by modern genetics and
biochemistry, recalls immunologist
Alexander Rudensky of the University
of Washington, Seattle.
By the mid-1980s, virtually everybody
had abandoned the idea of suppressor T cells
except Sakaguchi, who continued his quest
for the elusive cells. Extending earlier work,
Sakaguchi and his colleagues showed, in the
1980s, that removing the thymus of a mouse
on day 3 of its lifewhich depletes the ani-
mal of most of its T cellscauses various
autoimmune diseases to develop. Inoculating
the thymus-free mice with a mixed population
of T cells from another mouse prevents those
diseases, the researchers found.
Sakaguchi felt that the autoimmune dis-
eases resulted from a deficit in putative sup-
pressor T cells that were made in the thymus
on or after day 3; their absence left
unchecked any T cells that had developed
earlier. But critics contended that infections
could have triggered the autoimmune reac-
tion. Without pinpointing the suppressors,
the Kyoto team could not prove its case. C
R
E
D
I
T
:
K
.
H
A
S
E
N
K
R
U
G
A
N
D

D
.
D
O
R
W
A
R
D
In hopes of finding new remedies for ills including cancer and diabetes, scientists are following a band of elusive
immune-cell cops whose existence was once hotly debated
Policing the Immune System
News Focus
Caught. Scientists have finally nabbed the elu-
sive regulatory T cell.
Published by AAAS
Finally, in 1995, Sakaguchi and his col-
leagues reported that they had identified
suppressor T cells by the presence on them
of a newly identified cell surface protein
called CD25 as well as the more ubiqui-
tous surface protein CD4. When they in-
fused a batch of T cells devoid of ones
with these markers into mice that lacked
their own T cells, the mice developed auto-
immune disease. But if they infused the
suppressors along with the other T cells,
no autoimmune disease appeared. These
experiments convincingly showed that a
small, specif ic population of T cells
worked to dampen autoimmune reactions.
Virtually no immunologists read the paper,
however, because hardly anybody was inter-
ested in suppressor T cells anymore. But NI-
AIDs Shevach did. He was so struck by the
finding that he rushed to repeat it in his own
laboratoryand succeeded. I had a reli-
gious conversion to believe in regulatory T
cells, he says. That brought others into the
fold, as Shevach had been an outspoken
skeptic of the idea. Rudensky credits Sak-
aguchis 1995 paper as the turning point:
The field took off.
In 1998, Shevachs and Sakaguchis
groups independently developed cell culture
systems that enabled
others to study the sup-
pressive activity of the
rodent cells in dishes.
In 2001, several re-
search teams, including
Don Mason and Fiona
Powrys at the University of Oxford, plucked
out CD4
+
CD25
+
cells in human blood and
determined that they halted the proliferation
of other T cells, showing that the rodent data
had some relevance to humans.
And last year, the cells were shown to
underlie human disease. Three teams of in-
vestigators reported that Foxp3, the protein
that Ochs and others found to be missing
or defective in IPEX patients in 2001, is
specifically expressed in regulatory T cells
and is essential to their development.
Mice engineered with a defective Foxp3
gene have a deficit in CD4
+
CD25
+
T cells
and suffer from an IPEX-like disease called
scurfy that can be blocked by treatment with
regulatory T cells at 1 day old, Rudensky and
his colleagues revealed in Nature Immunolo-
gy. In the same journal, a team led by Fred
Ramsdell, formerly at Celltech Research &
Development in Bothell, Washington, report-
ed that the expression of Foxp3 by T cells in
mice correlates with their ability to suppress
immune responses. And transferring Foxp3
into nave T cells converts them into regula-
tory cells, the Sakaguchi group showed the
same year (Science, 14 February 2003,
p. 1057). Together, the studies indicated that
a deficiency of regulatory T cells in humans
can lead to severe immune dysfunction.
How suppressor T cells do their job re-
mains a mystery, however. In the test tube,
natural regulatory T cellsthe type that ex-
press CD25 and are made during immune
system developmentseem to suppress other
T cells through direct contact. In living ani-
mals, they also may release anti-inflammatory
cytokines such as interleukin-10 or trans-
forming growth factor . So-called adaptive
regulatory T cells, which become regulatory
only after being stimulated by an
antigen and respond only to im-
mune cells targeting that antigen,
seem to exert their influence
solely by means of cytokines. If
thats not complicated enough,
Shevach recently reported that
regulatory T cells may directly
kill the B cells that generate
antibodies (Science, 6 August, p. 772).
Good cop, bad cop
Despite basic gaps in their understanding of
regulatory T cells, researchers are tracking
down potential roles for the cells in human
disease. David Hafler and his team at Har-
vard Medical School in Boston reported in
April in the Journal of Experimental Medi-
cine that patients with multiple sclerosis
seem to have defective regulatory T cells.
Impotent regulatory T cells may also
play a role in allergy and asthma. Allergist
Douglas Robinson of Imperial College
London and his colleagues isolated natural
regulatory T cells from the blood of people
with and without allergies. They then ex-
posed the remaining T cells to an allergen.
In all of the samples, the allergen (from
grass pollen) triggered T cell proliferation
and a release of inflammatory molecules,
or cytokines, from immune cells. Adding
back regulatory T cells from the nonaller-
gic people completely suppressed this in-
flammatory response, whereas the regula-
tory T cells from the allergic individuals
were far less effective in doing so. The sup-
pression was weaker still from regulatory
T cells from patients with hay fever during
the pollen season, the researchers reported
in February in The Lancet.
One implication is that people who get
allergic disease do so because their regula-
tory T cells dont respond, Robinson says.
Boosting the response of these cells, he
adds, might help pre-
vent or treat their dis-
ease. Boosting the adap-
tive class of regulatory
T cells may also be im-
portant. Two years ago,
for example, Stanfords
Dale Umetsu and his col-
leagues identified adap-
tive regulatory T cells that
protect against asthma and
also inhibit allergic airway
inflammation in mice.
Although regulatory
T cells seem to be protec-
tive in autoimmune disor-
ders and allergy, they may have a darker
side. In the March issue of Immunity, a
team led by immunologist Kim Hasenkrug
of NIAIDs Rocky Mountain Laboratories
in Hamilton, Montana, suggests that some
viruses, such as those that cause hepatitis
and AIDS, may exploit regulatory T cells to
C
R
E
D
I
T
S

(
T
O
P
T
O

B
O
T
T
O
M
)
:
D
.
U
M
E
T
S
U
,
N
A
T
U
R
E
M
E
D
.
8
(
2
0
0
2
)
;
C
O
U
R
T
E
S
Y

O
F

S
H
I
M
O
N

S
A
K
A
G
U
C
H
I
Fighting asthma. Compared to the clear lungs of a normal mouse (left), the lungs of an egg-white-allergic mouse become inflamed when exposed to
the allergen (middle). Infusing such a mouse with regulatory T cells before exposure blocks the inflammation (right).
Within 5years, some
clinical application of these
cells will be here. Shimon Sakaguchi,
an immunologist at Kyoto University in Japan
Published by AAAS
dampen the bodys antiviral response and
allow chronic infections.
Similarly, regulatory T cells may pro-
tect tumors from immune attack. Re-
searchers have shown, for example, that
removing such T cells from a cancer-
afflicted mouse can cause the rodent to re-
ject a tumor. High levels of regulatory
T cells have also been found in samples
from several types of human tumors. More
recently, tumor immunologist Weiping Zou
of Tulane University Health Science Cen-
ter in New Orleans, Louisiana, and his col-
leagues linked the quantity of regulatory
T cells associated with a tumor to disease
severity in cancer patients.
Zous team isolated and counted the
T cells in tumor tissue from 104 ovarian
cancer patients and noted that the higher the
ratio of regulatory T cells to total T cells in
the tumor, the farther the cancer had pro-
gressed. Regulatory T cells were also associ-
ated with a higher risk of death: The more
tumor-associated regulatory T cells, the
worse the prognosis. Zou and his colleagues
further showed that the regulatory cells re-
covered from tumor tissue protected tumors
in a mouse model of ovarian cancer by in-
hibiting both the proliferation and potency
of tumor-attacking T cells.
Zous team also discovered that, as the
cancer progressed, a patients regulatory
T cells appeared to migrate progressively
away from their normal home in the lymph
nodes to the tumor. The investigators deter-
mined that tumor cells secrete a chemical,
dubbed CCL22, that attracts regulatory
T cells. Blocking CCL22 with an antibody
stopped regulatory T cells from migrating
to the tumor in the mouse model, the team
reported in the September 2004 issue of
Nature Medicine.
By extension, disarming these regulatory
cells or preventing their migration to the
tumor could leave the tumor vulnerable to
immune destruction. The Tulane researchers
and others are testing a regulatory
T cellkiller called Ontak in advanced can-
cer patients. The drug binds to CD25 and
kills the cells with diphtheria toxin. The re-
sults so far, Zou says, are encouraging.
Immunologist Steven Rosenberg of the
National Cancer Institute in Bethesda has
tested another regulatory T cellblocker in
patients with metastatic melanoma. The
treatmentan antibody to an essential mol-
ecule on the surface of regulatory T cells,
called cytotoxic T lympho-
cyteassociated antigen 4
induced cancer remission in three
of 14 treated patients who had
end-stage cancer, Rosenbergs
team reported last year. More
than 2 years later, the patients are
still in remission, and Rosen-
bergs team has now seen a simi-
lar remission rate among almost
100 patients.
The fact that inhibiting reg-
ulatory T cells enabled three
patients to undergo cancer re-
gression was very strong evi-
dence that regulatory T cells are
inhibiting the immune response
against tumors, says Rosenberg.
This is the first time that get-
ting rid of this brake on the im-
mune system has been shown to
have any impact in humans.
The study also suggests how tricky it
may be to interfere safely with the regulato-
ry T cell system, however. Six of the initial
14 cancer patients, including the three who
went into remission, developed autoimmune
diseases affecting the intestines, liver, skin,
or pituitary gland, although these were all
reversible with short-term steroid treatment.
Expansion plans
Even in disorders such as type I diabetes, in
which regulatory T cells have not been con-
sistently shown to be abnormal in function
or number, researchers are exploring them
as potential therapy. Traditionally, im-
munotherapy is designed to block effector
cells or their activities. Now there is the en-
tirely new possibility that we could treat the
disease by expanding suppressors, says
Ralph Steinman, an immunologist at Rocke-
feller University in New
York City.
In June, Steinmans
team and, separately, a
team led by Jeffrey
Bluestone at the Uni-
versity of California,
San Francisco (UCSF),
reported mouse studies in the Journal of
Experimental Medicine that illustrated
how this might work. Both research teams
plucked out natural regulatory T cells from
diabetes-prone mice that made only one
type of T cell, one that responds to an anti-
gen on the islet cells of the pancreas. Each
team then used different methods to ex-
pand the mouse regulatory T cells in lab
dishes and found that they could prevent or
reverse diabetes when infused into other
diabetes-prone mice.
A diabetes treatment that is thought to
boost T cell regulation has already reached
human trials. The treatment is an antibody to
CD3, a cell-surface protein tightly associat-
ed with the T cell receptor. The antibody
was first found to induce long-term remis-
sion of diabetes in mice a decade ago. That
surprising result contradicted the idea that
the CD3 antibodywhich was then used to
treat organ rejectionworked by inactivat-
ing destructive T cells, because the treat-
ments effects far outlasted the depletion of
activated T cells.
Last year, immunologists Jean-Franois
Bach and Lucienne Chatenoud and their
colleagues at Necker Hospital in Paris,
along with UCSFs Bluestone, reported in
Nature Medicine that the antibody appeared
to activate natural regulatory T cells in mice.
When diabetes-prone mice were treated with
the antibody a month after diabetes onset,
they became nondiabetic. But if the mice
were also treated with drugs that block regu-
latory T cells, the diabetes remained. Its a
nice story indicating that, in the mouse, im-
munoregulation explains the long-term ef-
fect of the antibody, Bach says.
After initial tests of this antibody ap-
proach proved encouraging in a small
number of diabetics, Kevan Herold, an en-
docrinologist at Columbia University
School of Medicine in New York City, and
his colleagues recently launched a six-cen-
ter trial of the therapy in 81 diabetic pa-
tients. Meanwhile, Chatenoud and her col-
leagues are about to unveil the results of a
multicenter, 80-patient, placebo-controlled
trial of the CD3-targeting antibody con-
ducted in Belgium and Germany.
Boosting regulatory T cell activity
might someday also induce drug-free im-
mune tolerance to donor organs. In July,
Sakaguchis team reported removing natu-
ral regulatory T cells from a normal mouse
and expanding them in cell culture with
interleukin-2, a growth promoter, along
with an antigen from a donor mouse of a
N E W S F O C U S
C
R
E
D
I
T
:
P
.
M
O
T
T
R
A
M
A
N
D
W
.
Z
O
U
/
T
U
L
A
N
E

U
N
I
V
E
R
S
I
T
Y
Tumor bodyguards. Regulatory T cells (red and green) inter-
act with tumor-killing T cells (blue) in ovarian tumor tissue.
All anyone is talking about
these days is regulatory T cells.
Ethan Shevach, a cellular immunologist at the National Institute
of Allergy and Infectious Diseases
Published by AAAS
different strain. This generated a population
of antigen-specific regulatory cells, which
they then infused into so-called nude mice,
which lack T cells. The regulatory cell in-
fusion enabled those rodents to accept skin
grafts from the donor strain even though
they were simultaneously infused with
killer and helper T cells. By contrast, nude
mice that received only killer and helper
T cellsbut no regulatory cellsquickly
rejected the grafts. With just a one-time
injection of regulatory T cells, we can in-
duce graft-specif ic tolerance without
drugs, Sakaguchi says.
In cases in which organ donorssuch as
living donorsare known in advance, Sak-
aguchi envisions generating antigen-specific
regulatory T cells prior to transplantation of
human organs. If the therapy works, he says,
it could replace the use of immunosuppres-
sive drugs, which come with a significant
risk of infection and cancer.
A boost from bugs
The growing understanding of regulatory
T cells may eventually shed some light on
an immunology-based theory called the
hygiene hypothesis. According to this con-
troversial idea, the rise in allergic disorders
in recent decades in developed countries
results from those countries increasing
cleanliness, which reduces childrens expo-
sure to protective microbes. A number of
researchers have shown that exposure to
parasitic worms called helminths may pro-
tect against allergy and asthma, among
other immune disorders, largely through
the induction of regulatory T cells
(Science, 9 July, p. 170).
Some strains of bacteria have also been
shown to be protectiveand again regula-
tory T cells may be involved. Immunolo-
gist Christoph Walker of the Novartis In-
stitutes for Biomedical Research in Hor-
sham, U.K., and his colleagues demon-
strated that treating mice with killed
Mycobacterium vaccae before sensitizing
them to egg-white allergen significantly
reduced the rodents inflammatory re-
sponses to the allergen, as compared to
mice that did not receive the bacteria. Reg-
ulatory T cells isolated from bacteria-
treated mice could transfer the protection
to untreated mice sensitized to the same al-
lergen, demonstrating that the cells medi-
ated the bacterias protective effects, the
team reported in 2002.
The suppressive
response was aller-
gen-specific: The reg-
ulatory T cells gener-
ated in the egg white
sensitized mice could
not dampen the re-
sponse to cockroach
allergen in mice made allergic to cockroaches.
Regulatory T cells generated by mycobacte-
ria treatment may have an essential role in
restoring the balance of the immune system
to prevent and treat allergic diseases, the au-
thors wrote in Nature Medicine. Walkers
team is now trying to mimic the bacterias ef-
fects with a chemical that stimulates the
same receptors on regulatory T cells that the
bacteria stimulate.
But some researchers note that opportu-
nities for rational drug design may be limit-
ed by the paucity of knowledge about how
regulatory T cells suppress their immune
system colleagues. Says Shevach: We
wont know how to enhance the response
until we know what it is.
Nevertheless, he, Sakaguchi, and others
have succeeded in a vital first step. Theyve
at long last convinced fellow immunologists
that regulatory T cells exist and are impor-
tant. So many people are working on regu-
latory T cells, Sakaguchi says. Its been a
pleasant surpirse. INGRID WICKELGREN
N E W S F O C U S
C
R
E
D
I
T
:
A
L
E
X
A
N
D
R

S
H
I
K
L
O
M
A
N
O
V
/
U
N
I
V
E
R
S
I
T
Y

O
F

N
E
W

H
A
M
P
S
H
I
R
E
Look more than a week or two into the fu-
ture, and none of the atmospheres detailed
behaviorthe weathercan be predicted.
Climate, on the other hand, is less capri-
cious. All of northern Europe, for example,
warms or cools for years or decades at a
time. The shifts in atmospheric circulation
behind such relative climatic stability once
seemed likely to provide a way to predict
patterns of regional climate change around
the world. But a profusion of patterns dis-
cerned by applying different analytical tech-
niques to different data setsthink dozens
of blind men and an elephantsoon threat-
ened to swamp the promising field.
Researchers are now managing to stem
the tide. When I was a graduate student in
the late 80s, there were a zillion patterns
of variability, says climate modeler John
Fyfe of the University of Victoria, British
Columbia. Of those, only a handful have
survived. Three new studies show that
almost all proposed patterns fit into one of
three or four globally prominent patterns:
the El Nio pattern tied to the tropical
Pacific Ocean; two great rings of climate
variability, each circling a pole at high lati-
tudes; and a last pattern across much of the
northern mid-latitudes. At one point, it
looked like there might be an infinite num-
ber of patterns, says meteorologist Kevin
Trenberth of the National Center for
Atmospheric Research (NCAR) in Boul-
der, Colorado. Well, actually, there are
relatively few.
This emerging simplification of meteo-
rologists view of atmospheric dynamics
provides a basis to move forward on re-
gional climate change, says Trenberth.
That ability will become increasingly im-
portant as the greenhouse intensif ies:
Policymakers want to know whats going
to happen regionally, not just on the global
average.
Until recently, natural climate varia-
tions were beginning to look as complex
and indecipherable as next months weather.
The climate dynamics literature abounds
with patterns of variability, note meteo-
rologists Roberta Quadrelli and Michael
Wallace of the University of Washington,
Seattle, in their October Journal of Cli-
A Few Good Climate Shifters
Meteorologists probing a dauntingly complex atmosphere have found patterns of
natural climate change that offer hope for making regional climate predictions
Cl i mate Change
Oops. When Siberian permafrost thaws, build-
ings can lose their footing and slowly crumble.
Regulatory T cell research is
very intriguing but is not yet ready
for mass consumption. There are still
a lot of puzzles. Hans Ochs, pediatric
immunologist, University of Washington
Published by AAAS
C
R
E
D
I
T
:
R
.
Q
U
A
D
R
E
L
L
I
A
N
D

J
.
M
.
W
A
L
L
A
C
E
,
J
O
U
R
N
A
L
O
F
C
L
I
M
A
T
E
1
7
(
2
0
0
4
)
mate paper. The Northern Hemisphere
alone has had at least 17 patterns pro-
posed for it, variously termed telecon-
nection patterns, oscillations, clusters,
seesaws, circulation regimes, and
modes. Quadrelli and Wallace believe
theyve narrowed it down to two prin-
cipal patterns of variability, or more
informally, modes.
Quadrelli and Wallace tried to get as
comprehensive a feel for the elephant
as they could. They considered the en-
tire Northern Hemisphere outside the
tropics during the 4 months of winter,
when atmospheric circulation is
strongest. They analyzed the longest,
most thoroughly vetted data set avail-
able, which runs from 1958 to 1999.
They started with atmospheric pressure
at sea level, but also included a meas-
ure of atmospheric pressure up through
the lower atmosphere as well as surface
air temperature. And they employed a
statistical method that is widely used to
identify the most common patterns of
atmospheric behavior.
The results were just two patterns in the
Northern Hemisphere. One is the previously
recognized Arctic Oscillation, now termed
the Northern Annular Mode (NAM). The
second is a pattern strongly resembling the
long-established PacificNorth American
(PNA) pattern. The NAM is an erratic pres-
sure seesaw that raises pressures alternately
over the pole and in a ring passing
over southern Alaska and central Europe
(Science, 9 April 1999, p. 241). These pres-
sure shifts in turn weaken and strengthen
the westerly winds there. The fluctuations
in the NAM can favor cold outbreaks down
through Canada into the lower 48 states,
for example.
The PNA is a pattern of alternating
centers of high and low pressure arcing
across the North Pacific and North America;
in part, it is set up by the Tibetan Plateau
and Rocky Mountains jutting into the
atmospheres westerly flow. Its oscillations
can shift warmth into Alaska and cool, wet
weather into the southeastern United
States. These two patterns subsume many
of the previously proposed leading modes
of the Northern Hemisphere, say Quadrelli
and Wallace, including the North Pacific
Index, the cold oceanwarm land pattern,
the Aleutian-Icelandic seesaw, and the
North Atlantic Oscillation.
Between them, the two modes account
for about half the variability of sea level
pressure from year to year and on longer
time scales, Quadrelli and Wallace find.
Half is a lot for meteorologists, who eagerly
pursue anything accounting for 10% or
more of atmospheric behavior. But the two
modes by themselves explain virtually
all of the broad trends over the 42-year
period, they say. Theres no need for any
others over the long haul.
In another upcoming paper in the Jour-
nal of Climate, meteorologists Monika
Rauthe of the Institute for Atmospheric
Physics in Khlungsborn, Germany, and
Heiko Paeth of the University of Bonn,
Germany, report that these two major
modes of pressure variation in turn
account for much of regional variability of
winter temperature. Shifting pressure pat-
terns produce wind shifts that pick up heat
from the oceans and car ry it to new
regions. Rauthe and Paeth find that their
rendition of the NAM and PNA account
for 30% to 75% of temperature variations
from year to year within swaths of the
Northern Hemisphere that are each several
thousand kilometers across. The regions
include northwestern North America,
northern Europe, and north-central
Siberia. Precipitation can vary just as
much, but over fewer and smaller regions.
South of northern mid-latitudes, Tren-
berth, David Stepaniak, and Lesley Smith
of NCAR find only two more modes, they
will report in Journal of Climate. In their
analysis of global, year-round variations of
atmospheric massa more fundamental
measure than pressurethere is no true
Southern Hemisphere equivalent of the
PNA. The Southern Hemisphere has noth-
ing quite like the towering Tibetan Plateau
and Rocky Mountains to create such a pat-
tern. And the Southern Annular Mode
presents a far more continual and symmet-
rical ring than its northern sibling, thanks
to a dearth of those disruptive influences
that distort the NAM.
And then there is the tropical Pa-
cifics El Nio. Meteorologists call it
the El NioSouthern Oscillation
(ENSO) to include the interaction of
atmosphere and ocean that produces
cyclic ocean warming (El Nio) and
cooling (La Nia) as well as the
atmospheric circulation changes that
accompany them. Trenberth and his
colleagues find that ENSO dominates
year-to-year variability in the tropics
and mid-latitudes around the globe. It
even seems to take such a strong
hand in North Pacif ic variability
thatat least on year-to-year and
longer time scalesit dominates
Quadrelli and Wallaces PNA-like
pattern. Over years, decades, and pre-
sumably centuries, that would make
ENSO and the two annular modes the
rulers of the climate change roost. A
very large fraction of large-scale
atmospheric variation can be ex-
plained by a few basic patterns, says
meteorologist Timothy Palmer of the
European Center for Medium-Range
Weather Forecasts in Reading, U.K.
Climate researchers would now like to
use these basic patterns to help predict
how regional climate will change under
the intensifying greenhouse. In their paper,
Rauthe and Paeth report that hot spots of
particularly intense warming seen in cli-
mate model simulations of greenhouse
warming are due in large part to the circu-
lation changes of modes. If the greenhouse
caused modes to shift their behavior
spending more time at one extreme of a
pressure seesaw than the otherthat
wouldnt by itself amplify global warming.
However, the resulting circulation changes
might redistribute heat, intensifying warm-
ing in some places and moderating it else-
where, or it could shift storm tracks and
redistribute precipitation.
Raut he and Paet h f i nd t hat i n an
ensemble of seven models, changes in the
intensity of modes under rising green-
house gases account for almost 60% of
t emperat ure changes over Nor t her n
Hemisphere land. Northwestern North
America and northern Siberia would
experience the most added warming,
whereas northern Africa, the southeastern
Uni t ed St at es, and far nor t heast er n
Canada/western Greenland would warm
less than the global average. Smaller
regions would see precipitation changes,
notably enhanced dyring across southern
Europe. Given the apparent utility of
modes, there will be more pressure on
modelers to look at things from this
standpoint, says Trenberth. This is the
wave of the future.
RICHARD A. KERR
Change in the air. Wind shiftsperhaps induced by rising
greenhouse gasesbrought 40 years of warming (oranges)
and cooling (blues) to high latitudes.
N E W S F O C U S
Published by AAAS
C
R
E
D
I
T
:
C
O
U
R
T
E
S
Y

O
F

N
I
M
H
When Thomas Insel took over as head of the
National Institute of Mental Health (NIMH)
in November 2002, he was seen as a reassur-
ing choice to succeed psychiatrist Steven
Hyman, who beefed up basic science and
promoted large-scale clinical trials. A for-
mer NIMH researcher, Insel had sterling
credentials, with groundbreaking work on
the neurobiology of attachment in voles. But
reassuring demeanor
aside, hes now rock-
ing the NIMH boat in
a way that has some
basic researchers send-
ing out SOS signals.
Early this month
Insel put into effect a
reorganization, in the
works for the past
6 months, intended to
move the institute
closer to the front lines
in battling mental ill-
ness through transla-
tional researchin
other words, bringing
the fruits of new
knowledge to people
with disorders such as
depression and schizo-
phrenia. Practically, it means that the agency
is lowering the priority of basic cognitive or
behavioral research unless it has a strong
disease component.
The original Basic Behavioral and Social
Science Branch has been broken up: Re-
search that can be tied with brain science is
in a new Behavioral Science and Integrative
Neuroscience branch. And some nonbiolog-
ical researchincluding studies in cogni-
tive science and social psychologyhas
been parceled out to NIMH divisions with
clinical portfolios. But the welcome mat is
no longer out for grant applications in some
areas of personality, social psychology, ani-
mal behavior, theoretical modeling, lan-
guage, and perception. When Mark Seiden-
berg of the University of Wisconsin, Madi-
son, applied for funds to continue his re-
search on models of language learning, he
says, I was told the agency no longer sup-
ported basic research on language.
NIMH has traditionally been the federal
agency that supports such research. Alan
Kraut of the American Psychological Soci-
ety (APS) guesses that perhaps $400 mil-
lion of the institutes $1.4 billion budget is
devoted to it. But with budget growth
slowing, Insel says he wants to tighten the
focus on NIMHs mission. Were one
of the disease-specif ic institutes, he
asserts, arguing that other institutes should
pick up some areas of research NIMH is
downgrading.
The move is wel-
comed by advocates for
mentally ill people. The
National Alliance for the
Mentally Ill (NAMI),
which has long
pressed for NIMH to
keep its eye on major
mental illness, is de-
lighted. Its a start in
the right direction,
says former NIMH psy-
chiatrist E. Fuller Torrey,
who runs NAMIs re-
search arm, the Stanley
Foundation. Theyre
shifting away from
studying how pigeons
think. Its a real quan-
tum leap, says James
McNulty, former NAMI head and a member
of the NIMH advisory council, applauding the
agencys transition to an applied research
institute.
But among researchers, theres a lot of
angst and anxiety, says Steven Breckler, ex-
ecutive director for science at the American
Psychological Association. Cognitive psy-
chologist Richard Shiffrin of the University
of Indiana, Bloomington, argues that transla-
tional research is all very well, but there is
still not much to translate, and gains pro-
duced by a few extra dollars for translational
research will be far outweighed by the harm
it will do to basic research.
One researcher whose grant NIMH
failed to renew this year is Mahzarin Banaji,
a Harvard University social psychologist
who examines unconscious mental pro-
cesses in stereotyping and discrimination.
She decries the timingwhen clinical
psychologists are uncovering new mental-
health uses of a scale she and her col-
leagues developed. The scale can aid in
studying phobias or probing attitudes of
people with depression, she says. Termi-
nating support for this work, says a Na-
tional Institutes of Health (NIH) official
who asked not to be quoted, means unfor-
tunately, for a topic of grave social impor-
tance, no one in the federal government
will fund it.
Some animal studies are also being de-
emphasized. Robert Seyfarth, a well-
known psychologist at the University of
Pennsylvania in Philadelphia, says his
grant almost wasnt renewed, and when it
was, funding was drastically cut, just as his
team was moving beyond basic research on
social behavior in nonhuman primates to
research linking social behavior and stress.
Some people think Im out to kill basic
behavioral science, concedes Insel, who
says that view is all wrong. Instead, he says
he wants basic behavioral scientists to be
more aware of the problems NIMH needs
to solve. For example, he says, cognitive
deficits are a major part of schizophrenia,
so if someones on the track of an impor-
tant piece of cognitive science using
healthy undergraduate [subjects], we might
work with them to begin to study people
with schizophrenia.
Insel wants to redirect some behavioral
research to institutes dealing with relevant
issues such as child development, aging,
and communication. Others agree with him
that NIMH has been carrying more than its
share of the burden. The larger issue, they
say, is which federal agency should be tak-
ing it on. The National Science Foundation
doesnt spend much on that type of re-
search, says Kraut of APS. And although
some NIH institutes have big behavioral
componentsfor example, on how to get
people to stop smokingKraut says, I
cant tell you how hard it has been to con-
vey the importance of basic behavioral sci-
ence in any sophisticated sense. Kraut and
several members of Congress are pushing
for the National Institute of General Med-
ical Sciences to take up the slack.
Basic behavioral research will not be
abandoned, Insel hopes. Last March, NIH
director Elias Zerhouni set up a working
group, chaired by sociologist Linda Waite
of the University of Chicago, which is pon-
dering the role of social and behavioral sci-
ence at NIH. And Insel himself heads an-
other group, under the White House Office
of Science and Technology Policy, that is
looking at investments in social, behavioral,
and economic research throughout the fed-
eral government. The NIH body is sched-
uled to make recommendations in Decem-
ber; Insels group will weigh in later.
CONSTANCE HOLDEN
NIMH Takes a New Tack, Upsetting
Behavioral Researchers
Basic behavioral scientists are feeling the squeeze as Thomas Insel makes a top
priority of translational research
Behavi oral Sci ence
Narrowed focus. Insel says others should
pick up research NIMH no longer funds.
Published by AAAS
C
R
E
D
I
T
S

(
T
O
P
T
O

B
O
T
T
O
M
)
:
R
O
B
E
R
T
Y
.
O
N
O
/
C
O
R
B
I
S
;
D
.
L
A
P
O
I
N
T
E
/
U
S
G
S
On 24 September, officials at the Hawaii
Department of Health (DOH) got the news
that theyd been dreading for several years:
An island bird had tested positive for West
Nile virus. Although infected birds are now
routine across the continental United States,
Hawaii has so far been spared. And it is
fighting to stay that way. Immediately after
the discovery, the health department
launched an assault; all night long a truck
fogged the Kahului Airport on Maui, where
the bird had been caught, with insecticide.
Additional crews with backpack sprayers
doused off-road sites to kill any potentially
infected mosquitoes.
State officials breathed a sigh of relief
the following week when the case turned
out to be a false positive. But they arent
letting down their guard. Should West Nile
become established on the is-
lands, virus-ridden mosquitoes
could spread the disease year
round. And many of the states
remaining endemic birds, al-
ready hammered by avian malar-
ia and pox, might go extinct.
The effects could be disas-
trous, says ornithologist Peter
Marra of the Smithsonian Envi-
ronmental Research Center in
Edgewater, Maryland.
To avert such a catastrophe,
researchers have been scrambling to
improve surveillance and eradication plans.
Observers on other Pacific islands, which
also face a similar threat, are hoping to learn
from Hawaiis efforts to stamp out the virus
as soon as it enters. Were not just throwing
our hands up in the air, says epidemiologist
Shokufeh Ramirez, who coordinates West
Nile prevention efforts for the Hawaii DOH.
On the mainland, West Nile virus has
proved unstoppable. After first appearing on
the East Coast, in New York in 1999, West
Nile virus marched steadily across the coun-
try. The virus is transmitted by mosquitoes,
which pass it on to birds and humans.
Although infection is rarely deadly to peo-
ple, it kills some bird species such as crows
with a vengeance; other infected birds re-
main healthy enough to fly and spread the
virus. Last year, it reached California.
But Hawaii has a chance, if not to keep
West Nile virus out, at least to stop it upon
arrival. Thats because researchers know
how its likely to get there. Rather than
infected humans or migratory birds, the
most probable culprits are mosquitoes in the
cargo holds of planes, concluded A. Marm
Kilpatrick of the Consortium for Conservation
Medicine at Wildlife
Trust in Palisades, New
York, and others in a paper published in
EcoHealth in May. Based on previous
research, they estimated that seven to 70 in-
fected mosquitoes probably reach Hawaii
each year. Far less is known about the risks
of introduction via shipping containers,
some 1200 of which arrive in Hawaiian har-
bors each day. The number of overseas
flightsabout 80 a dayalso makes pre-
vention difficult. Moreover, airlines have
balked at treating their cargo holds with in-
secticides that kill mosquitoes on contact.
The state has made progress on another
front: preventing infected poultry and pet
birds from entering by mail. In 2002, the
U.S. Postal Service prohibited the mailing of
most live birds to Hawaii. Quarantine regu-
lations have also been strengthened.
The health department has focused pri-
marily on monitoring 11 airports and har-
bors. In 2002, they began checking dead
birds by polymerase chain reaction (PCR)
for West Nile virus. Last year, they added
mosquito traps that are sampled each week
and also examined by PCR for the virus.
At the same time, researchers are trying
to figure out just what might happen if West
Nile virus manages to evade detection. Bird
biodiversity will probably be severely
impacted, says Jeff Burgett of the U.S. Fish
and Wildlife Service in Honolulu, who
heads an interagency task force. One reason
is that Hawaiis endemic birds have not had a
chance to build resistance to West Nile
through exposure to related viruses, such as
St. Louis encephalitis, that are not present
on the islands. Those species that survive
only in captive breeding pro-
grams, such as the Hawaiian
crow, might never be able to
return to the wild.
As a first step to gauge
the consequences, biologists
with the U.S. Geological Sur-
vey (USGS) have sent 20 na-
tive Hawaiian honeycreepers
(Hemignathus virens) to the
surveys National Wildlife
Health Center in Madison,
Wisconsin. There, veterinarian
Erik Hofmeister has injected
some of the birds with West
Nile virus and is following
their health and ability to
serve as reservoirs for the
virus. He also plans to inves-
tigate how eff iciently the
primary vector in Hawaii, the
mosquito Culex quinquefas-
ciatus, can infect these birds.
A similar experiment
should help solve a problem that hampers the
effort to spot the virus in dead birds. Hawaii
doesnt have the North American birds
crows, magpies, jaysthat provide the most
obvious warning of the virus. So Hofmeister
plans in December to examine which intro-
duced birds in Hawaii, such as minahs, might
be most susceptible to the virus. This will
assist efforts to model potential spread of the
virus. It will also tell you which species
might be amplifying the virus, and which
species you may want to control, says ecolo-
gist Dennis LaPointe of USGS.
While the health department waits for
these results, it is trying to speed its lab
testing and streamline the response plan.
Meanwhile, DOH and wildlife biologists
have their fingers crossed that Hawaiis de-
fenses will be adequate to stave off the
virusforever. Every year its going to be
knocking on Hawaiis door, says Peter
Daszak of the Consortium for Conserva-
tion Medicine at Wildlife Trust.
ERIK STOKSTAD
Hawaii Girds Itself for Arrival of
West Nile Virus
Health officials and wildlife biologists hope vigilant surveillance and rapid response
will prevent infected mosquitoes from establishing a beachhead
I nfecti ous Di seases
No barriers. Mosquitoes hitching a ride
inside airplanes could bring West Nile
virus to Hawaii, threatening honeycreep-
ers and other native birds.
Published by AAAS
Another Question for
Bush and Kerry
THE ARTICLE BUSH AND KERRY OFFER THEIR
views on science (1 Oct., p. 46) raises an
important question. Is it known for certain
that Bush and Kerry actually took the time to
read the campaign responses before they were
returned to Science, or were the responses
stock answers handled by staff? If it is not
known, then in future campaigns, it might be
wise to include a final question: Mr.
Candidate, did you personally read and
approve of the responses to our questions?
The answer, or lack thereof, might say more
about the priority of science to the candidate
than any other question you ask.
SOL MICHAEL GRUNER
Cornell University, 518 Clark Hall, Ithaca, NY
14850, USA.
Struggling to Attend
U.S. Meetings
THE FREE EXCHANGE OF KNOWLEDGE AND
ideas is a defining feature of science and the
driving force of its progress. In their Editorial
International science meetings (10 Sept., p.
1531), J. Lubchenco and G. Mehta address
some recent challenges to the freedom of
scientific communication. Perhaps one of the
most formidable for those wishing to share
information at conferences in the United
States is the visa wall. Faced by relentless
new security regulations, scientists wishing to
visit the United States to meet with colleagues
have become victims of the war on terror.
As repeatedly highlighted in recent years
(13), this problem has reached a level where
researchers from certain regions of the globe
are effectively blocked from attending U.S.
meetings, and even keynote speakers are
unable to get past the immigration bureau-
cracy, despite the lack of a reasonable expla-
nation of why several-days-long visits by
established scientists pose such a threat to the
security of the United States. Pledges (2, 4) by
renowned academics and international bodies
to ease the visa restrictions for scientists have
elicited various promises (5, 6), but have yet
to translate into tangible improvements.
Because of these circumstances, major inter-
national societies have expressed reservations
about sponsoring conferences in the United
States, and certain meetings there have been
postponed or simply cancelled (1, 7). As
things stand, the willingness of scientists or
scientific organizations to overcome political
prejudices will be of little relevance as long as
the leading scientific nation ignores the prin-
ciples of scientific universality.
BOYAN K. GARVALOV
Max Planck Institute of Neurobiology, Am
Klopferspitz 18, 82152 Martinsried, Germany
References
1. J. Schultz, J. Natl. Cancer Inst. 95, 579 (2003).
2. G. Brumfiel et al., Nature 427, 190 (2004).
3. I. Verma, Mol. Ther. 9, 767 (2004).
4. Y. Bhattacharjee, Science 304, 943 (2004).
5. Y. Bhattacharjee, Science 305, 1222 (2004).
6. Nature 431, 238 (2004).
7. J. Lubchenco, letter from the ICSU President to Dr. George
H. Atkinson, Science and Technology Advisor to the U.S.
Department of State, 10 June 2004 (available at
www.icsu.org/5_abouticsu/Visa_Restrictions.pdf).
Is Bedout an Impact
Crater? Take 1
L. BECKER ET AL.S RECENT PROPOSAL THAT
the Bedout structure off northwestern
Australia is a giant bolide impact crater of
Permian-Triassic (P-Tr) boundary age
(Bedout: a possible end-Permian impact
crater offshore of northwestern Australia,
Research Article, 4 June, p. 1469; published
online 13 May; 10.1126/science.1093925)
provides many readily testable hypotheses,
not least of which is that there should be
some evidence of the impact in the sedi-
mentary record of the surrounding area. The
Bedout structure is located in the Roebuck
Basin, which is part of a 2000-km extent of
Late Palaeozoic-Mesozoic rift basins devel-
oped from Perth to Darwin. Marine condi-
tions during the latest Permian and Early
Triassic are recorded by the Kockatea Shale
in the Perth Basin and equivalents in other
basins.
We have examined core material from the
Kockatea Shale in the Hovea-3 borehole,
located around 1000 km south of the Bedout
structure (1, 2). This reveals a P-Tr transition
in which bioturbated mudstones with a
diverse latest Permian fauna, dominated by
brachiopods, are replaced by laminated,
anoxic shales with earliest Triassic bivalves.
At no level in the core, which spans a
Wuchiapingian-to-Dienerian interval, is there
evidence for a layer of impact ejecta or a
tsunamite. A trace metal assay also failed to
find evidence for iridium enrichment. Core
material is not available from sites nearer to
Bedout, but mudlog data and wireline logs
from boreholes as close as 400 to 500 km
from Bedout also indicate a shale-on-shale
transition across the P-Tr boundary. It is
significant that the Hovea-3 core is substan-
tially closer to the proposed impact site than
the celebrated K-T impact sites in north-
eastern Mexico are to the Chicxulub Crater.
For example, the famous Mimbral site, with
its spherule layers and thick tsunamite record,
is over 1500 km distant from Chicxulub.
We suggest that either the impact did
not occur in the late Permian to Early
Triassic interval (and therefore it has no
relevance to the P-Tr mass extinction
event), or it is not an impact crater and is
more likely to be a volcanic structure.
PAUL WIGNALL,
1
BRUCE THOMAS,
2
ROBBERT WILLINK,
2
JOHN WATLING
3
1
School of Earth Sciences, University of Leeds,
Leeds LS2 9JT, UK.
2
Origin Energy Limited, 34 Colin
Street, West Perth 6005, Western Australia.
3
Department of Applied Chemistry, Curtin
University, Bentley 6102, Western Australia.
*To whom correspondence should be addressed. E-
mail: wignall@earth.leeds.ac.uk
References
1. B. M. Thomas et al., Austr. J. Earth Sci., in press.
2. B. M. Thomas, C. J. Barber, APPEA J. 44, 59 (2004).
Response
WE AGREE WITH WIGNALL ET AL. THAT THE
study of cores from Permian-Triassic (P-
Tr) sections in western Australia poten-
tially provides a valuable test of the
hypothesis that Bedout is a large P-Tr
boundary crater. What is at issue is whether
any of the existing cores or well logs,
including Hovea-3, contain a complete
record across the boundary. In many of the
onshore basins that are proximal to
Bedout, much of the Permian and Early
Triassic section is missing (up to 2 km in
some basins). Wignall et al. also state that
Hovea-3 is closer to Bedout than some K-T
impact sites (e.g., northeastern Mexico) are to
the Chicxulub crater, and, thus, Hovea-3
should have some preservation of an impact
ejecta layer. The distance between Hovea-3
Letters to the Editor
Letters (~300 words) discuss material published
in Science in the previous 6 months or issues
of general interest. They can be submitted
through the Web (www.submit2science.org)
or by regular mail (1200 New York Ave., NW,
Washington, DC 20005, USA). Letters are not
acknowledged upon receipt, nor are authors
generally consulted before publication.
Whether published in full or in part, letters are
subject to editing for clarity and space.
LETTERS
[R]esearchers from
certain regions of the
globe are effectively
blocked from attending
U.S. meetings
GARVALOV
Published by AAAS
610
Genomics
4
:
ADVERTISER DIRECTORY
Turn to page 711
Drug Discovery and
Biotechnology Trends
The following organizations
have placed ads in the
Special Advertising Section
ADVERTISER Page
21st Century
Biochemicals........................ 716
Genome Canada ............... 713
SANYO Sales &
Marketing Corporation/
SANYO Electric
Biomedical Co., Ltd. ...... 710
Takara Bio, Inc. ................... 715
Cellular Invasions
it takes
both sides of
the brain.
When the left brain collaborates with
the right brain, science merges with
art to enhance communication and
understanding of research results
illustrating concepts, depicting
phenomena, drawing conclusions.
The National Science Foundation and
Science, published by the American
Association for the Advancement of
Science, invite you to participate in
the annual Science and Engineering
Visualization Challenge. The competition
recognizes scientists, engineers, visu-
alization specialists, and artists for
producing or commissioning innova-
tive work in visual communications.
ENTRY DEADLINE:
May 31, 2005
AWARDS CATEGORIES:
Photos/Still Images, Illustrations,
Explanatory Graphics, Interactive
Media, Non-interactive media
COMPLETE INFORMATION:
www.nsf.gov/od/lpa/events/sevc
Awards in each category will be published
in the September 23, 2005 issue of
Science and Science Online and
displayed on the NSF website.
CALL F OR ENTRI ES
Science & Engineering
Visualization Challenge
Accept the challenge.
Show how youve mastered
the art of understanding.
L E T T E R S
22 OCTOBER 2004 VOL 306 SCIENCE
and Bedout (1000 km) is about the same (not
closer) as that between Chicxulub and the
northeastern Mexico sections such as El
Mimbral. These Mexican sections, and even
closer (to Chicxulub) ones in Haiti, contain
less than 1 m of ejecta; thus, even minor
erosion at the boundary could erase the record.
Wignall et al. rely on biostratrigraphy to
determine the P-Tr boundary in Hovea-3.
As shown elsewhere for the P-Tr, bio-
stratigraphy alone is not a reliable indicator
of the completeness of the boundary layer
(1, 2). However, if it can be further demon-
strated (e.g., by isotope dating) that Hovea-
3 does indeed represent a record of contin-
uous sedimentation across the P-Tr
boundary, and it contains no ejecta, then
this would argue against Bedout being a
large P-Tr boundary crater.
Finally, the absence of iridium in Hovea-3
is consistent with other P-Tr sections world-
wide. This may reflect, again, upon the
completeness of the P-Tr boundary, the type of
section (marine versus continental), or the
actual impacting body (e.g., asteroid versus
comet). Several impact tracers (see our
Research Article) occur in P-Tr boundary
sections worldwide, none of which contains
elevated iridium. Thus, we await full docu-
mentation and future investigations of impact
tracers in Hovea-3 and other onshore cores
that may include the P-Tr boundary.
L. BECKER,
1
R. J. POREDA,
2
K. O. POPE
3
1
Institute for Crustal Studies, Department of
Geological Sciences, University of California, Santa
Barbara 93106, USA.
2
Department of Earth and
Environmental Sciences, University of Rochester,
Rochester, NY, USA.
3
Geo Eco Arc Research,
Aquasco, MD 20608, USA.
References
1. S. A. Bowring et al., Science 280, 1039 (1998).
2. Y. G. Jin et al., Science 289, 432 (2000).
Is Bedout an Impact
Crater? Take 2
IN THEIR RESEARCH ARTICLE BEDOUT: A
possible end-Permian impact crater offshore of
northwestern Australia, L. Becker et al. report
having identified a buried impact structure,
which they link to the Permian-Triassic mass
extinction (4 June, p. 1469; published online
13 May; 10.1126/science.1093925). Becker et
al. have scarcely extended the suggestion made
by Australian petroleum workers (in industry
trade journals) (1). Our scrutiny of the alleged
evidence indicates that there is no substantia-
tion that this alleged structure is an impact
crater. The gravity map (fig. 11) actually high-
lights the differences between Bedout and
confirmed impact structures. There is actually
no crater defined by the geophysical data, only
a noncircular high in the seismic data, claimed
to be a central uplift. In comparison, the
Published by AAAS
central uplift feature of a large impact struc-
ture, such as the 250- to 300-km-diameter
Vredefort Structure, would reveal a signifi-
cant central positive gravity anomaly due to
the uplift of relatively denser mid- to lower
crustal material. The highly altered rocks
described by Becker et al. as impact products
strongly resemble volcanic breccias and lack
impact diagnostic textures. No true shock
features are described from any of the
samples. No mineralogical or geochemical
evidence is provided that the purported
diaplectic glass or maskelynite are indeed
glasses, and mineral chemical information is
missing. The shock features claimed to be
presented in quartz grains from ejecta hori-
zons (which remain of uncertain strati-
graphic relation either to the alleged Bedout
feature or to the end-Permian extinction) do
not show any of the characteristics of unam-
biguous shocked minerals.
The 250 Ma age inter-
preted from argon isotope data
by Becker et al., which pres-
ents the entire basis for the
sensationalistic claim of a rela-
tionship between Bedout and
the P/Tr boundary, has no objective basis.
Their data present no consistent indication of
the presence of a 250-million-year compo-
nent in the sample analyzed. Results from
only one sample, a concentrate of unknown
lithologic and stratigraphic relation to the
Bedout geophysical feature, were reported.
The data do not define a plateau, and only
two of twelve steps purportedly defining a
plateau actually encompass the ad hoc
plateau age within analytical errors. Even
allowing an exceptionally generous definition
of a plateau, the reported plateau age does
not follow from the isotope data using any
combinatoric method of which we are aware.
Yet this putative 250 Ma age (with its
alleged uncertainty of 4.5 million years
deduced by unstated and indeed cryptic
means) will inevitably be cited in the litera-
ture by uninformed nonspecialists as evi-
dence for a causal relationship to the extinc-
tion. Consequently, the report of a Bedout
impact structure of P-Tr boundary age must
be considered with utmost caution.
PAUL R. RENNE,
1
H. JAY MELOSH,
2
KENNETH A. FARLEY,
3
W. UWE REIMOLD,
4
CHRISTIAN KOEBERL,
5
MICHAEL R. RAMPINO,
6
SIMON P. KELLY,
7
BORIS A. IVANOV
8
1
Berkeley Geochronology Center, 2455 Ridge Road,
Berkeley, CA 94709, USA, and Department of Earth
and Planetary Science, University of California,
Berkeley, CA 94720, USA.
2
Lunar and Planetary
Laboratory, University of Arizona, 935 Gould/Simpson
Building, Tucson, AZ 857210092, USA.
3
Division of
Geological and Planetary Sciences, California Institute
of Technology, MS 170-25, Pasadena, CA 91125, USA.
4
Impact Cratering Research Group, School of
Geosciences, University of the Witwatersrand,
Johannesburg, Private Bag 3, P.O. Wits 205, South
Africa.
5
Institute of Geochemistry, University of
Vienna, Althanstrasse 14, Vienna A-1090, Austria.
6
Earth and Environmental Science Program, New York
University, 100 Washington Square East, New York, NY
10003, USA.
7
Department of Earth Sciences,The Open
University, Milton Keynes MK7 6AA, UK.
8
Institute for
Dynamics of Geospheres, Russian Academy of Science,
38-6 Leninsky Prospect, Moscow 11797, Russia.
Reference
1. J. Gorter, Pet. Explor. Soc. Aust. News 1996, 33
(1996).
Response
ALTHOUGH WE FEEL THAT THERE ARE ISSUES
that could use clarification in our recent
Research Article, we found the general
tone of the Renne et al. Letter to be out of
balance with the specific objections.
Renne et al. take exception to each line of
evidence we presented, but they offer no
plausible alternative expla-
nation for these data. We
considered a number of
possible explanations for the
nature of the Bedout struc-
ture, including a volcanic
origin. However, we could find no other
examples of an isolated volcano the size of
Bedout (40 to 60 km in diameter and 3 to
4 km in height) forming along a passive
continental margin. We also could find no
known terrestrial volcanic sample(s) that
exhibit the unusual melt chemistry and
shock features (maskelynite) observed in
the Bedout core. When Renne et al. claim
that we describe no true shock features,
nor present mineralogical or geochemical
evidence for glass, they ignore the petro-
graphic and chemical evidence we
provided (see tables S-1 and S-2 and figs.
S-4 to S-8 in the Supporting Online
Material in our paper). Moreover, the
shocked quartz and other impact debris
(fullerenes, meteoritic fragments, Fe-Ni
metal grains, and chromium isotopes)
from Graphite Peak (Antarctica), Frasier
Park (Australia), and Meishan (China) are
all clearly associated with end-Permian
sediments (e.g., marked by stratigraphy,
biostratigraphy, carbon isotopes,
40
Ar/
39
Ar,
and U-Pb dating), regardless of Renne et
al.s unsubstantiated and undocumented
claims to the contrary (see our paper and
references therein). We note with irony the
objection of Renne et al. to our use of
industry publications that suggested an
impact origin for the Bedout structure,
given that industry data identifying the
Chicxulub crater were essentially ignored
by academia for over a decade.
The comments by Renne et al. pertaining
to crater morphology imply that all large
impact craters should have a clear, quasicir-
cular, central gravity high, but we note that
L E T T E R S
See related Technical
Comment Abstracts on
page 613
Published by AAAS
the ~200-km Sudbury crater in Canada does
not. The gravity expression of the central
uplift at Vredefort is especially clear because
that crater is exposed by erosion and not
tectonically deformed. The Bedout structure
is deeply buried and deformed, and although
the existing geophysical data are of low reso-
lution, they do show a central uplifted core of
denser rock surrounded by an annular depres-
sion, similar to the gravity and seismic signa-
ture of Chicxulub. Better geophysical data are
needed, however, to clearly define the geom-
etry of the entire Bedout structure.
Renne et al. make the remarkable state-
ment that our
40
Ar/
39
Ar age has no objec-
tive basis. The integrated age of steps 5 to
13 of the Lagrange (LG-1) plagioclase is
250.6 4.3 Ma (1; steps 1 to 4 reflect Ar
loss, step 14 contains no radiogenic signal,
and step 15 reflects degassing of the
crucible slag). This is in agreement with
the previous independently determined K-
Ar date of 253 5 Ma [(20, 30) in our
paper] and is consistent with the strati-
graphic position of the Bedout breccia at
the top of the Permian. Clearly, there is
inhomogeneity in the
40
Ar/
39
Ar release
from LG-1, but this has been observed
previously in samples inferred to be closed
systems (1). Although there is no univer-
sally accepted definition of a plateau [(1), p.
111], we would have been better served not to
have used the term plateau, which is a
much-abused concept. For example, the
plateau method has been applied on occasion
to slowly cooled minerals that manifestly
contain heterogeneous
40
Ar/
39
Ar ages [e.g.,
(2, 3)], a view for which there is truly no
objective basis. Improved dating for Bedout
awaits future drilling of the structure or an
onshore coring project in adjacent basins
where late Permian volcanics and basement
may be preserved.
Our goal was to draw attention to the
various lines of evidencenone of which
individually are definitivethat support an
impact origin for Bedout. We stand by our
conclusion that the data we present are
consistent with an impact origin for Bedout
and that the current best estimate of the age
of the structure is within error of the age of
the P-Tr boundary.
L. BECKER,
1
* R. J. POREDA,
2
A. R. BASU,
2
K. O.
POPE,
3
T. M. HARRISON,
4
C. NICHOLSON,
1
R. IASKY
5
1
Institute for Crustal Studies, Department of
Geological Sciences, University of California, Santa
Barbara, CA 93106, USA.
2
Environmental Sciences, University of Rochester,
Rochester, NY 14627, USA.
3
Geo Eco Arc Research,
Aquasco, MD 20608, USA.
4
Australian National
University, Canberra, Australia.
5
Geological Survey
Western Australia, Perth, Australia.
mail: lbecker@crustal.ucsb.edu
References
1. I. McDougall, T. M. Harrison, Geochronology and
Thermochronology by the
40
Ar/
39
Ar Method (Oxford
Univ. Press, New York, ed. 2, 1999).
2. P. R. Renne, O. T. Tobisch, J. B. Saleeby, Geology 21, 331
(1993).
3. W. D. Sharp, O. T. Tobisch, P. R. Renne, GSA Bull. 112,
1059 (2000).
The Next Step for
Kennewick Man
THE PENULTIMATE PARAGRAPH OF C. HOLDENS
news report (News of the Week, 30 July, p.
591) on the Kennewick Man case, Court
battle ends, bones still off-limits, includes a
statement by Alan Schneider, head of the
legal team representing the Bonnichsen
plaintiffs in the case, attributing to me a
notion that is incorrect. Schneider questions
concerns that I have expressed about further
study of the Kennewick remains. He seems to
infer that I might object to any future study
because the government already has
conducted similar studies of the remains.
However, my concern, expressed to
L E T T E R S
Published by AAAS
L E T T E R S
Holden and reported in her article, about
future investigations of the Kennewick Man
skeleton is not whether additional studies may
be undertaken. That new studies will be
allowed has been determined by the outcome
of a 2-year legal struggle in federal courts.
However, it is essential that any additional
studies build on the substantial amount of
scientific investigation already conducted by
the Department of the Interior and the Corps
of Engineers as part of their efforts to resolve
the Kennewick Man case (1). Like any good
scientific investigation, new studies should
use the results of past studies, drawing fully
upon what already has been learned.
Throughout the legal contest just
concluded, the plaintiffs legal team, led by
Schneider, routinely disparaged the scientific
investigations undertaken by the government.
Now that the case has been resolved, such
legal maneuvering and tactics should be set
aside. The government agencies responsible
for the Kennewick Man skeletal remains
conducted a number of standard and detailed
scientific investigations. Over 20 scientists
from nationally recognized academic, labora-
tory, and museum departments were involved.
The research design for further study of the
Kennewick remains needs to take account of
the methods, techniques, and conclusions of
the earlier government studies.
FRANCIS P. MCMANAMON*
Archeology and Ethnography Program, National
Park Service, 1849 C Street NW, Washington, DC
20240, USA.
*Chief Archeologist for the National Park Service
and Departmental Consulting Archeologist for the
Department of the Interior
Reference
1. The government studies are available at www.cr.nps.gov/
aad/kennewick.
CORRECTIONS AND CLARIFICATIONS
Special Section on Testing Human Limits: News: A
race to the starting line by G. Vogel (30 July, p. 632).
The article stated incorrectly that [p]eople living at
high altitudes produce more EPO naturally to compen-
sate for the lower oxygen concentration in the air.The
concentration of oxygen in air does not change at high
altitudes, but the partial pressure drops.
Featured Presentations
David Baltimore PhD
President California
Institute of Technology
Biotech Has a Future
Elias A Zerhouni MD
Director National
Institutes of Health
NIH Roadmap for
Medical Research
Judah Folkman MD
Director Vascular Biology Program
Childrens Hospital and Professor
Harvard Medical School
Can the Angiogenic Switch be
Prevented in Human Cancer?
Webcast Sponsored by
View the keynote presentations and featured presentations
from the Drug Discovery Technology
World Congress
wwwdrugdisccom
TECHNICAL COMMENT ABSTRACTS
COMMENT ON Bedout: A Possible End-Permian Impact Crater Offshore
of Northwestern Australia
Andrew Glikson
A petrologicalgeochemical study of breccia cored at a depth of about 3040 meters over the Bedout basement
high identifies a hydromagmatic hyaloclastic volcanic breccia consisting of fragments of basalt and dolerite
retaining igneous textures, mineralogy, and chemistry, and set in a mafic mesostasis. Alteration by cryptocrys-
talline albite and microcrystalline albite-chlorite assemblages is widespread. No evidence is observed for impact-
induced shock metamorphism in these rocks, as suggested by Becker et al. (Research Articles, 4 June 2004, p. 1469).
Full text at www.sciencemag.org/cgi/content/full/306/5696/613b
RESPONSE TO COMMENT ON Bedout: A Possible End-Permian Impact Crater
Offshore of Northwestern Australia
L. Becker, R. J. Poreda, A. R. Basu, K. O. Pope, T. M. Harrison, C. Nicholson, R. Iasky
Many of the observations of texture, mineral chemistry, and high-silica glass composition, as well as the pres-
ence of maskelynite, in the lowermost section of the Bedout core are incompatible with an ordinary magmatic
origin. These observations are consistent, however, with an impact-induced origin for the Bedout structure.
Full text at www.sciencemag.org/cgi/content/full/306/5696/613c
Published by AAAS
Comment on Bedout: A Possible
End-Permian Impact Crater Offshore
of Northwestern Australia
Nearly a decade ago, Gorter (1) suggested
that the Bedout basement high, offshore
Western Australia, might represent an impact
structure, in view of the site_s well-defined
circular Bouguer anomaly and seismic reflec-
tion data indicating a possible ring syncline.
Recently, Becker et al. (2), as part of the
search for the cause or causes of the mass
extinction that marks the Permian-Triassic
(P-T) boundary (3), examined the P-T bound-
ary breccia cored at a depth of 3044 m in the
Bedout-1 well and with an age (using
plagioclase
40
Ar/
39
Ar dating) of 250.7 T 4.3
million years (My)within experimental
error from the age of the P-T extinction event
(251.4 T 0.4 My) and Siberian Norlisk vol-
canism E251.7 T 0.4 to 251.1 T 0.3 Ma (4)^.
Becker et al. (2) suggested the presence in
the Bedout breccia of shocked mineral grains,
diaplectic plagioclase glass (maskelynite),
and impact melt glass. Drawing an analogy
between the Bedout breccia and the suevite
melt breccia at Chicxulub and Sudbury, they
attributed the origin of the breccia to the
melting of Mg-rich sedimentary materials,
although they noted the presence of some
basalt in the target material (2).
The diagnostic hallmarks of extraterres-
trial impacts (5, 6) include (i) shocked miner-
als displaying planar deformation features
(PDFs)for example, in quartz, feldspar, and
zircon; (ii) the presence of high-pressure
mineral polymorphs such as coesite, stishov-
ite, and diamond; (iii) megascopic shock
structures (for example, shatter cones and
melt breccia); and (iv) chondritic chemical
signaturesin particular, platinum group
elements and other siderophile elements
(Ni, Co) (7). A study by the author of this
comment (8) suggested that the volcanic
breccia samples in the interval from 3035.8
to 3044.95 m in the Bedout-1 core meet none
of these criteria. The rocks consist of
metaglass-bearing hydromagmatic mafic
volcanic breccia dominated by fragments of
basalt and dolerite set in mafic pyroclastic
matrix (Fig. 1), including vesicular volcanic
lapilli, closely akin to hydromagmatic spi-
lites described in detail by Amstutz (9).
The figures presented by Becker et al. (2)
provided no suggestion of intracrystalline
PDF elements, nor are measurements of crystal
orientations reported. Instead, the principal
argument presented by Becker et al. for impact
effects hinges on the suggested existence of
diaplectic feldspar glass (maskelynite)
around and within plagioclase Efigures 6
and 8 in (2)^. No criteria allowing discrim-
ination between maskelynite and volcanic
glasses are indicated. Because the plagioclase-
to-maskelynite transformation occurs at a
pressure of 35 to 45 GPa (5, 6)after the
development of PDFs, which occurs at 10
to 35 GPaidentification of maskelynite is
closely related to the presence of PDFs,
which Becker et al. have not documented.
Nor do these authors offer any criteria for
discrimination between volcanic metaglass
and maskelynite. Any presence of pristine
unaltered glass, a highly metastable compo-
nent under hydrous conditions, requires tests
by transmission electron microscopy and
infrared spectroscopy. The intercrystaline
near-isotropic regions shown by Becker et al.
Efigure 6 in (2)^ compare well with the
cryptocrystalline chlorite/albitedominated al-
teration zones (7) common in hydromagmatic
altered spilites (9). The intracrystalline near-
isotropic regions within calcic plagioclase
Efigure 8 in (2)^ correspond to recrystallization
and alteration of internal euhedral crystal
sectors of oscillatory reverse-zoned magmatic
plagioclase, under high water pressures asso-
ciated with hydromagmatic processes (9). By
contrast, in impact-related rocks, maskelynite
irregularly overprints PDF-bearing crystalline
relics Esee, for example, figure 4.31 in (5)^.
The excellent preservation in the Bedout-1
rocks of igneous ophitic and microlitic tex-
tures in little-deformed dolerite and basalt
fragments, and of calcic plagioclase that
retains primary prismatic (euhedral) crystals
and albite twinning (Fig. 1), is hardly
consistent with the combination of deforma-
tion and the hydrothermal effects that com-
monly accompany impact (5). The presence
of euhedral chlorite pseudomorphs after
ophitically enclosed mafic phases and euhe-
TECHNICAL COMMENT
Fig. 1. (A) Ophitic-textured dolerite fragment in Bedout-1 breccia, showing lath-like calcic
plagioclase and pseudomorphs of chlorite after pyroxene (pyx) protruding into feldspar in an
ophitic texture (oph). The fragment is enveloped and injected by microbreccia (mb). Crossed
nicols. (B) Microlite-rich basalt fragment (B-mic) and dolerite fragment (dol) separated by
microbreccia (mb). Crossed nicols. (C) Plagioclase (An 52) microphenocryst in chlorite-dominated
basalt fragment. Crossed nicols. (D) Backscattered scanning electron microscope image of ophitic
dolerite fragment with plagioclase enveloping pseudomorphs of chlorite after pyroxene and
accessory rutile. White arrows denote relic ophitic crystal structure (oph) of mafic pseudomorph
within calcic plagioclase.
www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004 613b
dral Ti-magnetite and ilmenite typical of
mafic volcanic rocks (10) are all consistent
with burial metamorphism of mafic volcanic
breccia, but distinct from features diagnostic
of suevite shock-melt breccia (5). The Ries
suevite includes shocked PDF-bearing gran-
ite clasts and carbonate sediment-derived
clasts that display extensive heat-induced
plastic deformation of fragments inter-
meshed with flow-banded melt components
Efigures 5.9 to 5.15 in (5)^, clearly distinct
from the angular to subrounded breccia at
Bedout.
High degrees of impact melting may
result in magmatic-type rocks, for example
in the Mistasin and Dellen craters Efigures
6.12 to 6.21^ and in Vredefort granophyre
veins (5). However, these processes com-
monly result in coarse-grained quench crys-
tallization, not seen in the Bedout breccia.
In the absence of the unique criteria of
shock metamorphism, Becker et al. (2)
invoke novel petrologic and geochemical
arguments for an impact connection for the
Bedout breccia, including high-silica glass,
spherulitic glass, feldspar compositions,
presence of Mg-ilmenite, carbonate clasts
with fragmented ooids, and coexistence of
Ti-rich silica glass with Ti-poor aluminous
silica glass. They describe the mineral para-
geneses of the breccia as Bcompositions . . .
unknown and unlikely to exist in terrestrial
volcanic agglomerates, lava flows, and in-
trusive pipes[ (2). However, the spherulitic-
textured particles Efigure 5 and figure 7, A
and B, in (2)^ are identical to vesicular lapilli
fragments in spilitic pyroclastics Epp. 24 and
26 in (9)^. High-silica cryptocrystalline veins
form under hydromagmatic conditions, hy-
drothermal conditions, or both. Mg-ilmenite
is known in ultrabasic and alkaline xenoliths
(11) and is no criterion for impactites. Like-
wise, liquid immiscibilities, which are ob-
served in heterogeneous volcanic magmas
(12), offer no criterion for impact melting.
Becker et al. (2004) refer to the breccia in
part as product of Mg-rich sediments (e.g.,
dolomites). However, apart from the pristine
igneous textures of the breccia, the transition
element levels (chlorite in dolerite fragment
Ni 97 to 160 ppm, Co 75 to 152 ppm, Cu 69
to 204 ppm; interfragmental mesostasisNi
29 to 45 ppm, Co 18 to 52 ppm, Cu 26 to
110 ppm) are consistent with Fe-rich basalts
but exceed common abundances in carbon-
ates and marls (13).
Becker et al. (2) thus have identified no
impact effects at Bedout. The circular
gravity structure and the presence of flanking
rim synclines around the Bedout High (1)
justify drilling into the basement that under-
lies the breccia at Bedout-1 and Legrange-1
to test the origin of this important structure.
Andrew Glikson
Research School of Earth Science
Australian National University
Canberra, A.C.T. 0200, Australia
andrew.glikson@anu.edu.au
References
1. J. Gorter, PESA News October-November 1996, 32
(1996).
2. L. Becker et al., Science 304, 1469 (2004).
3. P. B. Wignall et al., in Catastrophic Events and Mass
Extinctions: Impact and Beyond, C. Koeberl, K. C.
McLeod, Eds., Geol. Soc. Am. Spec. Pap. 356, 395
(2003).
4. S. A. Bowring et al., Science 280, 1039 (1998).
5. B. M. French, Traces of Catastrophe: Lunar Planetary
Science Contribution 954 (1998).
6. D. Stoffler, F. Langenhorst, Meteoritics 29, 155 (1994).
7. F. T. Kyte, Geol. Soc. Am. Spec. Pap. 356, 21 (2002).
8. A. Y. Glikson, unpublished data.
9. G. C. Amstutz, Spilites and Spilitic Rocks (Springer-
Verlag, Berlin, 1974).
10. T. R. McGetchin, R. O. Pepin, R. J. Phillips, Basaltic
Volcanism on the Terrestrial Planets (Pergamon, New
York, 1981).
11. W. A. Deer, R. A. Howie, J. Zussman, Rock Forming
Minerals (Longman, New York, 1972).
12. J. Ferguson, K. L. Currie, Nature 235, 86 (1972).
13. K. H. Wedepohl, Handbook of Geochemistry (Springer-
Verlag, 1978).
17 May 2004; accepted 30 August 2004
T E C H N I C A L C O M M E N T
22 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org 613b
Response to Comment on
Bedout: A Possible End-Permian
Impact Crater Offshore of
Northwestern Australia
Glikson (1) suggests that any Btrue[ extra-
terrestrial impact structure should include
shocked minerals (for example, quartz with
planar deformation features, or PDFs), high-
pressure polymorphs (for example, coesite or
diamond), shatter cones, and chondritic
chemical signatures of platinum group ele-
ments (PGEs). He ignores the criteria we
present (shocked glass or maskelynite) and
later dismisses the data as being indicative of
a volcanic breccia. Moreover, his suggestion
that the identification of maskelynite in an
impact breccia necessitates the presence of
PDFs is incorrect. As stated in (2), the shock
pressures for the formation of maskelynite
(35 to 45 GPa) and silica glass (945 to 65
GPa) that characterize the Bedout core are
well above the shock pressures for preserved
PDFs. In addition, some researchers (3) now
distinguish between Bplagioclase diaplectic
glass[ and Bmaskelynite,[ with the latter
forming without being initiated in PDFs as
suggested by Glikson. Chen and El Goresy
(3) recently described maskelynite grains in
several SNC martian meteorites as smooth
with no cleavage, no contraction cracks, and
no shock-induced fractures, which is what
we see in the Bedout core Esee, for example,
figure 6 in (2)^. Thus, the notion that the
presence of preserved PDFs is required to
interpret maskelynite in a melt breccia core
especially one that is 250 million years old
and highly alteredis overstated. The petrol-
ogy and geochemistry of this Bvolcanic
breccia,[ as interpreted by Glikson, are un-
like those of any volcanic rock in the world.
It is not surprising that, as Glikson notes,
some of the clasts resemble altered basalts; as
we stated in (2), the target rocks likely con-
tained basalts, and unmelted basalt clasts are
among the more noticeable features of the
Bedout-1 core.
The evidence for impact glass that we
presented in (2) comes from the lowermost
section (3044 m) of the Bedout core, where
many of the observed features are com-
pletely at odds with a magmatic origin and
are most consistent with impact-induced
melting. At that depth, large plagioclase
crystals (An 50) have transformed to glass
Efigures 6 and 8 in (2)^; the shocked grains
are isotropic but maintain the perfect outline
of a plagioclase lath Efigure 6 in (2)^. They
do not even remotely resemble nearby grains
of crystalline plagioclase, as suggested by
Glikson. Instead, the texture indicates shock-
induced melting and quenching of the dense
melt at high pressure, which erased the
inherited shock-induced fractures but
retained the morphology of the plagioclase
lath (3). The chemistry of this isotropic
region is exactly that of plagioclase with
G0.1% TiO
2
and G1% Fe and Mg; no
volcanic glass in existence resembles that
composition.
Could the isotropic regions result from
alteration or spilitization, as suggested by
Glikson? It is unlikely that it could, and still
retain the isotropic optical character and the
exact chemical signature of An 50 plagio-
clase. The altered region within the core of
the plagioclase Etable S1, no. 21 in the
supporting online material in (2)^ shows the
changes in chemistry that occur during
alteration (loss of Na and Ca and addition
of Mg and Fe; addition of H
2
O). Conversely,
the unaltered isotropic core Etable S1, no. 4
in the supporting online material in (2)^ and
the crystalline rim of the plagioclase Etable
S1, no. 3 in the supporting online material in
(2)^ have identical Bplagioclase[ chemistry
(within error). There is no core-rim zonation
that would typify plagioclase zoning or
Bovergrowth,[ as Glikson would assert. The
only logical conclusion is that the core of the
plagioclase was shock-melted and quenched
at high pressure. Both figures 6 and 8 in (2)
show large feldspar laths (300 to 500 6m) that
have begun to alter over 250 million years of
burial, and there is clear evidence of maske-
lynite in fresh plagioclase laths Efigures S4
and S5 in the supporting online material in
(2)^. There is no visible evidence of alteration
in plane polarized light, yet the core of the
feldspars is isotropic; it is not even remotely
Bcryptocrystalline,[ as suggested by Glikson.
As noted in (2), the high-silica glass that
we described cannot be a magmatic product.
We agree with Glikson that silica does exist
in ancient volcanic rocks, precipitated as
veins during hydrothermal circulation. How-
ever, the high-silica glass in Bedout differs
from vein filling in several important
respects: (i) The rare occurrences of silica
have the shape of Brelict quartz grains[ and
do not resemble a vein. (ii) There are no
silica veins in this core; all veins examined
in the Bedout core Esee, for example, figure
S11 in the supporting online material in (2)^
are filled with lower temperature carbonates.
(iii) The high-silica glass contains substantial
amounts of TiO
2
(5%), which is not an
element commonly associated with veins of
opal or chert but which does occur as rutile
inclusions in quartz. Glikson also notes that
some of the glass that we describe in (2)
resembles volcanic lapilli. The glass photo-
micrographs in figure 7, A and B, in (2) do
resemble volcanic lapilli in gross texture,
because they both formed in explosive, high-
energy events. However, the chemistry of the
altered glass in figure 7B in (2) resembles no
known volcanic product. In particular, the
glass contains background levels of TiO
2
in a
composition that would otherwise be consid-
ered basaltic. Overall, the textures and
chemical compositions that we described in
the Bedout core require a wide range of bulk
rock compositions, from acid to ultrabasic, in
the same thin sectiona paradoxical situa-
tion that can only be explained by a process
such as impact.
Perhaps the most troubling aspect of
Glikson_s interpretation of Bedout as vol-
canic in origin is that he cites no examples or
analogs of a comparable volcanic product
nor does he attempt to explain how Bedout,
as a volcano, would have formed. We looked
at a number of possible explanations for the
nature of Bedout, including a volcanic
origin. However, we could find no explan-
ation for an isolated volcano the size of
Bedout (40 to 60 km in diameter and 3 to 4
km in height) forming along a passive
continental margin. This has been grossly
overlooked by Glikson and others (4). The
discovery of the Chicxulub crater prompted a
similar set of arguments pertaining to inter-
preting its origin. For example, the Yucatan-
6 core was originally interpreted as being
from a Bvolcanic dome[ based on the
presence of Bandesite[ in the basement
rocks, even though such a large feature was
clearly inconsistent with the region_s passive-
margin geology. More work is needed to
confirm that the Bedout structure is con-
sistent with an impact origin. For example,
we agree with Glikson that the presence of
PGEs (for example, iridium or chromium)
in the melt breccia would strengthen our
interpretation. However, we remain con-
fident that the data we present in (2) are
most consistent with an impact origin for
Bedout.
TECHNICAL COMMENT
www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004 613c
L. Becker
Institute for Crustal Studies
Department of Geological Sciences
University of California
Santa Barbara, CA 93106, USA
E-mail: lbecker@crustal.ucsb.edu
R. J. Poreda
A. R. Basu
Environmental Sciences
University of Rochester
Rochester, NY 14627, USA
K. O. Pope
Geo Eco Arc Research
Aquasco, MD 20608, USA
T. M. Harrison
Australian National University
Canberra, Australia
C. Nicholson
Institute for Crustal Studies
Department of Geological Sciences
University of California
Santa Barbara, CA 93106, USA
R. Iasky
Geological Survey Western Australia
Perth, Australia.
References
1. A. Glikson, Science 306, 613 (2004); www.sciencemag.
org/cgi/content/full/306/5696/613b.
2. L. Becker et al., Science 304, 1469 (2004).
3. M. Chen, A. El Goresy, Earth Planet. Sci. Lett. 179,
489 (2000).
4. R. A. Kerr, Science 304, 941 (2004).
10 August 2004; accepted 23 September 2004
T E C H N I C A L C O M M E N T
22 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org 613c
614
W
hen, in February 1928, the 23-
year-old German ornithologist
Ernst Mayr left Berlin for an ad-
venturous one-man expedition as a natural-
ist-explorer in New Guinea, he took only two
books with him: Hans Drieschs Philosophie
des Organischen (1899) and Henri Bergsons
LEvolution Cratrice (1911). These were
surprising choices for a young systematist,
because both authors were prominent early-
20th-century vital-
ists. More than two
years later, Mayr re-
turned from the trop-
ics having collected
thousands of speci-
mens (mostly birds,
but also mammals,
reptiles, butterflies,
and shells). He also
brought back impor-
tant observations on
the geographical vari-
ation of species, observations that in the
1940s he would use to implement the notion
of speciation by geographical isolation, or
the origin of biological diversity, as an inte-
gral part of the modern evolutionary synthe-
sis. In contrast, he was quite disappointed
with his two travel books, because the
philosophers had little to offer that would ac-
count for the phenomena of the living world.
Mayr later concluded that not only vital-
ism in particular, but traditional philosophy
in generalbased on logic, mathematics,
and the physical sciencescould not satisfy
his search for the genuine essence of biology.
The organisms he had studied in the field in
New Guinea were certainly more than mere
machines in Descartess sense. He remains
convinced that basic problems in biology
cannot be solved by either Cartesian or vital-
istic philosophy. In the 1970s, Mayr devoted
considerable attention to topics in the history
and philosophy of science, and he began to
work out his own philosophy of biology, one
based on a lifetime of empirical research in
evolutionary biology. Now, early in what is
being hailed as the century of biology, Mayr
offers his latestand, as he notes in the first
lines of his preface, finalsurvey of contro-
versial concepts in biology, What Makes
Biology Unique?
In the book, his 25th, Mayr again pre-
sents a critique of philosophys contribu-
tions to the science of biology. The book
comprises 12 essays. Four are newly writ-
ten, and the other eight chapters are revised
versions of articles formerly scattered (thus
being rather inaccessible) in journals and
symposium volumes. The collection could
be considered to serve as the authors per-
sonal festschrift in celebration of his 100th
birthday. As Mayr confessed in a recent es-
say (1), he finds evolutionary biology to be
an endless frontier where there is still
plenty to be discovered.
Those unacquainted with Mayrs think-
ing will find the book an excellent firsthand
overview of his philosophy of biology,
while those who have read previous books
and articles by Mayr will find themselves
on familiar ground. The essays present per-
spectives that are based on ideas Mayr has
expounded upon in several previous ac-
counts, including an earlier collection of es-
says (2) and an accessible introduction to
biologys place among the sciences (3). The
book covers several interrelated themes,
such as determinism and teleology (a pro-
gressive tendency toward ever-increasing
perfection), reductionism versus analysis,
populational thinking versus typology or
essentialism, emergence (the idea of sys-
tems that possess properties not present in
their individual components), the impor-
tance of species, and the differences be-
tween species concepts and the delineation
of species taxa. Although the topics are
treated in a somewhat abbreviated manner,
the essays are intended to offer a revised,
more mature version of Mayrs thoughts.
Physicists, Mayr complains, still arro-
gantly believe that there is only one true
science and that it is physics. And most
philosophers, if they do not completely ig-
nore biology, persist in viewing it as a sec-
ond physics; they are still busy with what
Kant or Hegel wrote. In contrast, Mayr the
biologist insists that his discipline, which
deals with living organisms and vital
processes, is not subsumed by physics but
remains an autonomous science. Mayr ar-
gues that none of the theories in physics, no
matter how revolutionary they may be con-
sidered, had any effect in changing biology
or how biologists view the world. He main-
tains that reductionism is an ill-fated at-
tempt to seek explanations only at the low-
est levels of organization, an attempt that,
in part, stems from confusing reductionism
with analysis as scientific methodology. He
also holds that Thomas Kuhns theory of
scientific revolutions does not apply to
changes in the theoretical framework of bi-
ology, preferring instead an evolutionary
epistemology.
Once more, Mayr compares and con-
trasts approaches in biology (e.g., historical-
narrative explanations) with those used in
physical sciences. He notes that variability
and the genetic program combine to impart
a fundamental difference between the or-
ganic and inorganic worlds. Whereas an
electron remains an electron, of Earths six
billion humans, no two are identical.
For Mayr, evolution is the most revolu-
tionary idea ever formulated, and Darwin
the greatest philosopher. But he stresses
that Darwin actually advocated five inde-
pendent theories about evolutionnot just
one, as Darwin himself insisted and many
after him have claimed. Two of the five
(transformation in time and common de-
scent) were readily accepted, while gradu-
alism, the multiplication of species, and
natural selection only gained approval in
the mid-20th century. Many controversies
have been caused by confounding these
five theories into a single composite.
The reviewer is in the Department of Malacology,
Institute of Systematic Zoology, Museum of Natural
History, Humboldt University, Invalidenstrasse 43,
D-10115 Berlin, Germany. E-mail: matthias.
glaubrecht@rz.hu-Berlin.de
P HI L OS OP HY OF B I OL OGY
A Centenarians Summary
Matthias Glaubrecht
What Makes
Biology Unique?
Considerations on
the Autonomy of a
Scientific Discipline
by Ernst Mayr
Cambridge University
Press, New York, 2004.
246 pp. $30, 25. ISBN
0-521-84114-3.
BOOKS
et al.
After productive observations but disap-
pointing reading. Mayr (right) and his field as-
sistant Sario in 1928, after two months of col-
lecting birds in the interior mountains of New
Guinea.
22 OCTOBER 2004 VOL 306 SCIENCE www.sciencemag.org
C
R
E
D
I
T
:
E
R
N
S
T

M
A
Y
R

L
I
B
R
A
R
Y

O
F

T
H
E

M
U
S
E
U
M

O
F

C
O
M
P
A
R
A
T
I
V
E

Z
O
O
L
O
G
Y
/
H
A
R
V
A
R
D

U
N
I
V
E
R
S
I
T
Y
Published by AAAS
615
As an evolutionary biologistwho, inci-
dentally, works in the Natural History
Museum in Berlin, where Mayr began his
scientific careerI admire his clear and ele-
gant writing as well as his insights into biol-
ogy and philosophy. His book offers the bi-
ologist exactly the answers that he sought
during his exploration of New Guinea. His
earliest experience with philosophy remind-
ed me of my own frustration when, as a biol-
ogy student, I took seminars in philosophy.
I was puzzled by how completely discon-
nected from biology the philosophers were
and how at a loss they left those of us eager
to understand the living world. Therefore, I
am convinced that What Makes Biology
Unique? will be loved by those who are cu-
rious about biology and used to empirical ap-
proachesnot the least because Mayrs style
lacks the unintelligible meandering that of-
ten detracts from philosophical writing. I as-
sume, however, that many philosophers of
science will not have such a positive re-
sponse. (For one thing, Mayr holds that their
belief that their problems can be solved by
logic alone has led them to miss the impor-
tance and implications of biology.) I only
regret not having had this excellent introduc-
tion to the philosophy of biology in my pock-
et during my first research in the tropics.
References
1. E. Mayr, Science 305, 46 (2004).
2. E. Mayr, Toward a New Philosophy of Biology:
Observations of an Evolutionist (Harvard Univ. Press,
Cambridge, MA, 1988).
3. E. Mayr, This Is Biology: The Science of the Living
World (Harvard Univ. Press, Cambridge, MA, 1997).
F I CTI ON
Of Politics and
Particle Physics
Jay M. Pasachoff
H
ow wonderful to encounter a novel
that presents a scientist as a popular
and engaging principal
character. A Hole in Texas of-
fers a refreshing contrast with
the treatments of mad scientists
that are so abundant in litera-
ture and popular culture. Add to
the cauldron some significant
physics and astrophysics, mix
with an analysis of scientific
politics and the U.S. Congresss
effects on major scientific research, and
the brew is a delightful tale, one that
should be of interest to most readers of this
magazine.
Herman Wouks previous novels have
covered a wide range of topics, from life on
a mismanaged warship in The Caine Mutiny
(1951), through coming of age in New York
City and suburban Mamaroneck in Marjorie
Morningstar (1955), to the global epic of
World War II in The Winds of War (1971)
and War and Remembrance (1978). In his
most recent work, the Pulitzer Prize
winning author (now 89 years old) takes up
modern particle physics and the politics of
big science. We learn what has happened to
physicist Guy Carpenter since Congress
terminated the Superconducting Super
Collider (SSC) in 1993.
Through the stories of
Carpenter and the other
characters, Wouk has
managed to create an ex-
citing plot that involves
the Higgs boson. His sci-
entific thriller deals with
the prospective political
fallout should that elu-
sive particle be discov-
ered by the Chinese.
Could a boson bomb
result? Whose fault is it
that America has fallen
behind? The conse-
quences are assessed at
the highest levels of gov-
ernment, and we are
treated to the tense, be-
hind-the-scenes view of a
congressional hearing on
the subject.
The author gives us a
good idea of the size and
scale of the hole in
Texas (the partially ex-
cavated tunnel left behind
when Congress canned
the SSC after spending
more than $2 billion on
it) and of the devastating personal conse-
quences of so many physicists being laid
off at the same time. He suggests
that the budgetary decision a
decade ago reflected a choice be-
tween the SSC and the space sta-
tion, with the latter winning. I look
forward to seeing Hollywoods hel-
icopter views over Waxahachie,
Texas.
Readers will find many links to
headlines from our daily newspa-
pers. For example, there are bits about the
role of secrecy in government and the me-
dia backlash over the treatment of a
Chinese scientist at Los Alamos National
Laboratory. (Indeed, the beautiful physicist
who leads the Chinese program, and whom
Carpenter loved when they were in gradu-
ate school together, is named Wen Mei
Lia possible reference to Wen Ho Lee,
especially given the mention of the latters
unfortunate treatment at Los Alamos.)
In a pleasant change from the stories of
lone geniuses familiar from much popular
writing on science (fiction and nonfiction),
Wouk accurately depicts science as an often
interactive and collegial enterprise. His sto-
rytelling skill highlights the interplay
among scientists as well as researchers re-
lations with outsiders. He shows how bonds
among fellow graduate students can last
lifetimes, on both personal and scientific
bases. We hear Carpenter explain how his
career interests moved
from philosophy to
physics to astrophysics
as well as the heroines
response to his descrip-
tion: Her eyes shone at
him. A seeker after
truth, then.
I am not giving away
too much plot, I hope, to
thrill at how astrophysics
comes to the rescue.
Here, as throughout the
book, the science is pre-
sented fairly accurately
(though one almost never
sees solar flares during a
total eclipse). The novel
is especially timely given
the award of the 2004
Nobel Prize in physics
for work on quantum
chromodynamics.
One of the books he-
roes explains that the dis-
covery of the Higgs boson
goes to the fundamental
mystery of mass, and
perhaps some readers will
go on to deeper studies of
the nature of matter. In
any case, they will learn about many scientif-
ic-political choices that our society confronts.
By offering readers a sympathetic physicist
with whom they can identify, Wouk may lead
them to recall past times when the public un-
abashedly admired researchers abilities and
their scientific results.
Leon Lederman, the nonfictional
Nobelist particle physicist, has devoted
much of his recent years to outreach and
education. He has valiantly encouraged the
presentation of favorable images of scien-
tists in television and movies, and Wouks
novel provides an excellent opportunity for
developing such a script. I hope that this
engaging novel continues Wouks string of
characters who become fixed in the public
mind, and the physicist Carpenter is surely
a more pleasant person than the U.S.S.
Caines Captain Queeg.
Terminated Texas tunnel. The SSC
was abandoned after about 25% of
the tunnel for the 87-kilometer-
circumference large collider ring had
been bored.
The author is in the Department of Astronomy,
Williams College, Williamstown, MA 01267, USA. E-
mail: Jay.M.Pasachoff@williams.edu
A Hole in Texas
A Novel
by Herman Wouk
Little, Brown, New
York, 2004. 279 pp.
$25, C$35. ISBN 0-
316-52590-1.
B O O K S E T A L .
C
R
E
D
I
T
:
S
U
P
E
R
C
O
N
D
U
C
T
I
N
G

S
U
P
E
R

C
O
L
L
I
D
E
R

P
I
C
T
U
R
E

A
R
C
H
I
V
E
S
Published by AAAS
617
B
razil is planning to commission later
this year a uranium enrichment plant
that, if configured to do so, could fuel
several nuclear weapons annually. As a mem-
ber of the Nuclear Nonproliferation Treaty
(NPT), Brazil has promised not to make such
weapons and is obliged to allow the
International Atomic Energy Agency (IAEA)
to ensure this is the case. But this spring
Brazil took the extraordinary step of barring
the plants doors to the IAEAs inspectors.
Brazil and the IAEA are now negotiat-
ing how much access the IAEA will have.
The outcome will set a precedent for Iran
and any other country that builds an enrich-
ment plant while a member of the treaty.
At its announced capacity, Brazils new
facility at Resende will have the potential
to produce enough
235
U to make five to six
implosion-type warheads per year (1, 2).
By 2010, as capacity rises, it could make
enough every year for 26 to 31 (3) and by
2014 enough for 53 to 63 (4).
Brazil has pledged that to enrich uranium
to only 3.5%
235
U, the concentration required
by its two power reactors. This would be too
weak to fuel a bomb, which typically re-
quires a concentration of 90% or above. If
Brazil should change its mind, its stockpile
of uranium already enriched to 3.5 or 5%
will have received more than half the work
needed to bring it to weapon grade (5, 6).
This confers what is known as breakout ca-
pabilitythe power to make nuclear
weapons before the world can react. Such a
power is what the United States and some
European countries fear Iran is aiming at.
Iran, too, plans to field thousands of
centrifuges at a new enrichment facility at
Natanz and claims that its sole purpose is
to produce low-enriched reactor fuel. If
Brazil succeeds in denying the IAEA ac-
cess to its centrifuges, Iran can demand the
same treatment. Under the NPT, there is no
legal ground for treating the two countries
differently.
There is little evidence that Brazil actu-
ally intends to become a nuclear weapon
power. Brazils science and technology min-
ister Eduardo Campos, declared earlier this
year that the Brazilian nuclear project is in-
tended exclusively for peaceful purposes
(7). He pointed out that Brazil has joined the
Treaty of Tlatelolco, as well as the NPT,
both of which forbid Brazil to make nuclear
weapons. Brazil has also adopted a new
constitution that does the same.
These statements, however, must be seen
in light of Brazils nuclear history. During
the 1980s, Brazil ran a secret effort to build
an atomic bomb that ran in parallel with the
public program to make electricity. It was
administered by the military and hidden
from the IAEA. In 1990, the program was
openly repudiated by Brazils newly elected
president, Fernando Collor de Mello (8).
Brazil then joined the NPT and accepted in-
ternational inspection.
But now, Brazil has built a physical
screen around its centrifuges at Resende for
the express purpose of preventing inspec-
tors from seeing them. Brazil says it has
done this to protect its advanced technolo-
gy from leaking out to competitors (9). The
IAEA, however, has a long history of pro-
tecting commercial secrets. Brazil is thus a
serious challenge to the IAEAs authority.
The real effect of the screen will be to
make it harderif not impossiblefor the
IAEA to do its job. The IAEA must account
for all the enriched uranium the plant makes
and must ensure that it is used only to fuel
peaceful power reactors. Brazil contends
that the inspectors will be allowed to see
everything going into Resende and every-
thing coming out and that that should be
sufficient. But with a screen in place, it will
be difficult to be sure the centrifuges are not
hooked up to a hidden supply and outlet of
uranium. Such a hookup would allow Brazil
to stockpile enriched material while inspec-
tors believe that the facility is less efficient
than it really is. And since there is no re-
quirement that Brazil enrich a certain
amount of uranium, no one would be the
wiser. Unfortunately, the IAEA has already
allowed the Brazilian Navy to shield a group
of centrifuges for several years at a pilot
plant, where uranium was enriched. Thus,
Brazil can argue that if the IAEA could cer-
tify for years that the pilot-scale plant was
not siphoning off any uranium, and could do
so without seeing the centrifuges, the same
should be possible at Resende.
One response to this argument is that the
throughput of the plants is different.
Resende will have the capacity to enrich
enough uranium for dozens of bombs per
year. If the machines are shielded, the in-
spectors can only measure input and output
and then calculate the material unaccount-
ed for. This is the amount of uranium as-
sumed to be hung up somewhere in the sys-
tem. Every plant has some. The question is
whether the amount makes sense. At
Resende, the amount could be consider-
able, whereas the amount at the pilot plant,
given the limited number of centrifuges
there, was fairly small.
It seems unlikely that Brazil is really
concerned that the IAEA will illegally re-
veal industrial secrets. More likely, Brazil
is trying to hide the origin of the cen-
trifuges. In December 1996, Brazil arrest-
ed Karl-Heinz Schaab, a former employee
of Germanys MAN Technologie AG, a
firm that developed centrifuges for the
European enrichment consortium called
Urenco (10, 11). German authorities
wanted Schaab extradited to prosecute
him for selling centrifuge blueprints to
Iraq. There is evidence that Schaab and
other experts were helping Brazil as well
(12). It follows that, if the IAEA inspec-
tors were to see the Brazilian centrifuges,
they might discover that Urencos design
data had been transferred.
The United States has decided not to
challenge Brazils new status and instead
has tried to persuade Brazil to cooperate
with the IAEA. Its inspectors were to ar-
rive in Brazil 15 October to pursue a solu-
tion to the inspection dispute. The rest of
the world should help the United States
convince Brazil to put these concerns to
rest and to be a good nuclear citizen.
References and Notes
1. If one assumes that the plants first cascade will pro-
duce 20,000 SWU/year and that 16 kg of uranium en-
riched to 93.5% U-235 are needed for an implosion
device. For SWU capacity, see (2).
2. M. Hibbs, Nuclear Fuel, 7 July 2000.
3. If one assumes 100,000 SWU/year.
4. If one assumes 200,000 SWU/year.
5. About 3000 kg of uranium feed requires ~3500 SWU
to make one implosion bomb. The same feed needs
more than 2000 SWU to enrich to 3.5%. See (6).
6. T. B. Cochran et al., Nuclear Weapons Databook, vol. 2,
U.S. Nuclear Warhead Production (Natural Resources
Defense Council,Washington, DC, 1987),Table 5.1, p. 127.
7. Brazil refuses to let UN inspectors into nuclear facil-
ity, Agence France-Presse, 5 April 2004.
8. J. Brooke, New York Times, 9 October 1990, p. A1.
9. Brazils commitment to nonproliferation under sus-
picion, CNN.com, 4 April 2004.
10. M. Hibbs, Nucleonics Week 37 (51), 19 December
1996, p. 1.
11. M. Hibbs, Nucleonics Week 38 (12), 20 March 1997,
p. 17.
12. M. Hibbs, NuclearFuel 23 (6), 23 March 1998, p. 5.
NUCL E AR S AF E TY
Brazils Nuclear Puzzle
Liz Palmer* and Gary Milhollin
The authors are with the Wisconsin Project on Nuclear
Arms Control, Washington, DC 20006, USA.
mail: liz@wisconsinproject.org
POLICY FORUM
Published by AAAS
618
W
hen an extra electron is added to
water, a hydrated electron is
formed. First discovered in 1962
(1), this fascinating species is of funda-
mental importance in radiation chemistry
and in electron transfer processes in water,
and has therefore been studied widely (2,
3). It remains unclear, however, how the
hydrated electron moves through water
and how the water molecules are arranged
in its vicinity. Three reports in this issue
describe experimental studies of negative-
ly charged water clusters that shed light on
these questions.
In bulk water, the hydrated electron is be-
lieved to be confined in a roughly spherical
cavity with a radius of 0.22 to 0.24 nanome-
ters (1 nm = 10
9
m), and to occupy an s-type
ground electronic state. It is characterized by
a broad electronic absorption near 1.7 eV,
which can be thought of as a transition from
the s state to an excited p state (13).
Spectroscopic studies of hydrated electrons
have revealed transient absorption on time
scales of 50 femtoseconds (fs), (1 fs = 10
15
s)
200 to 300 fs, and 1 picosecond (ps) (1 ps =
10
3
fs) after excitation to the p state (4, 5).
These time scales are typical for molecular
motions. Some researchers have attributed the
50-fs process to hindered rotational motion of
water molecules in the excited state and the
200- to 300-fs process to nonradiative decay
of the p to the s state (4). Others have attrib-
uted the 50-fs process to p s decay and the
200- to 300-fs process to subsequent relax-
ation of the solvent in the s state (5). In these
scenarios, the 1-picosecond time scale corre-
sponds to long-time relaxation on the s state.
Bragg et al. (page 669) and Paik et al.
(page 672) use pump-probe photoelectron
spectroscopy to follow the dynamics of pho-
toexcited clusters containing 15 to 50 water
molecules and one excess electron (6, 7).
Hammer et al. (page 675) use vibrational
predissociation spectroscopy to elucidate the
structures of smaller clusters with just four to
six water molecules and one excess electron
(8). All three studies are motivated by the
fact that measurements
of clusters can provide
a level of detail that is
difficult to achieve in
studies of the bulk.
Bragg et al. (6) and
Paik et al. (7) both pro-
vide evidence of fast
(130 to 250 fs) dynam-
ics associated with the
decay of the p state to
the s state of their clus-
ters. Very short life-
times have previously
been reported for the
excited states of such
clusters (9). Bragg et
al. find that the excit-
ed-state lifetimes de-
crease with increasing
cluster size. They ex-
trapolate to a value of
50 fs for bulk water
and conclude that the
50-fs process observed in bulk water con-
taining hydrated electrons is due to nonra-
diative conversion from the p to the s state,
as suggested previously in (5).
Paik et al. investigate the fate of the s
state after p s decay by selecting the ener-
gy of the electron that is ejected as a result of
photoexcitation. Their experiments provide
insights into the solvent dynamics in clusters
of different sizes. The solvation dynamics
are found to occur on a time scale of 300 to
450 fs, depending on cluster size. Because
this time scale is similar to that of solvation
dynamics in bulk water, Paik et al. conclude
that the local solvent structure is critical for
electron solvation. They also observe dy-
namics on a much longer time scale of 2 to
10 ps, which they ascribe to the breakage of
hydrogen bonds followed by evaporation of a
water monomer.
These measurements (6, 7) provide
new insights into the dynamics of an ex-
cess electron interacting with hydrogen-
bonded networks. However, the relevance
of the new data for hydrated electron dy-
namics in the bulk depends on whether
the excess electron is bound to the cluster
surface or resides in its interior (10). Paik
et al. do not reach a conclusion on this is-
sue, whereas Bragg et al. argue that they
are probing an interior-bound electron. If
this is indeed the case, then Paik et al.
probably also probe an interior-bound
electron. Both studies would then be di-
rectly relevant to the dynamics of hydrat-
ed electrons.
Aside from the issue of interior versus
surface binding, there
is the question how
the water molecules
are arranged in the
vicinity of the excess
electron. This prob-
lem is addressed by
Hammer et al. (8).
With the exception
of the negatively
charged water dimer,
the geometrical struc-
tures of negatively
charged water clusters
have proven elusive.
The report by Ham-
mer et al. represents a
major advance in es-
tablishing some of
these structures (8).
By clever use of
mixed complexes of
water and argon, the
authors have been able
to synthesize the elu-
sive tetramer, either with normal or with
deuterated water, as well as the pentamer
and the hexamer. Their vibrational spectra
show conclusively that in all three clusters,
the excess electron binds in the vicinity of a
water molecule that accepts two hydrogen
bonds from adjacent molecules but does not
itself donate any hydrogen bonds to the hy-
drogen-bonding network (see the figure).
This arrangement is energetically unfavor-
able in neutral clusters. Its predominance in
the negatively charged clusters shows how
the excess electron disrupts the hydrogen-
bonding network.
The importance of the geometries with
double-acceptor waters for the binding of
excess electrons to water clusters was first
proposed by Lee et al. (11) on the basis of
electronic structure calculations for the
negatively charged hexamer. Furthermore,
Hammer et al. find that vibrational excita-
tion of the OH stretch associated with the
double-acceptor water molecule of the neg-
atively charged tetramer and pentamer
leads to rapid (50 to 300 fs) ejection of the
excess electron.
In the small clusters studied by Hammer
et al., the excess electron is surface-bound.
CHE MI S TRY
A Fresh Look
at Electron Hydration
Kenneth D. Jordan
PERSPECTIVES
The author is in the Department of Chemistry,
University of Pittsburgh, Pittsburgh, PA 15260, USA.
E-mail: jordan@pitt.edu
An excess electron binds to a cluster of
five water molecules. The water molecule
closest to the diffuse excess electron (gray
area) is in a double-acceptor hydrogen-
bonding environment. Data from electronic
structure calculations reported in (8).
Published by AAAS
619
In the interior of larger clusters and in bulk
water, the hydrated electron may not be
bound in the vicinity of double-acceptor
water molecules. However, this type of
arrangement could very well occur on ice
surfaces or at the surface of large water
clusters. The application of the vibrational
predissociation technique of Hammer et al.
to larger clusters may elucidate the location
(surface or interior) of the excess electron.
These new experimental results for neg-
atively charged water clusters are important
benchmarks for theoretical studies of the
structure and dynamics of excess electrons
in aqueous environments. Recent theoreti-
cal studies have shown that dispersion inter-
actions between the excess electron and the
electrons of the water molecules make an
important contribution to the binding ener-
gy of the former (12). Such interactions
could play a role in determining whether the
excess electron is surface- or interior-bound
and could also affect its dynamics.
References
1. E. J. Hart, J. W. Boag, J. Am. Chem. Soc. 84, 4090
(1962).
2. P. J. Rossky, J. Schnitker, J. Phys. Chem. 92, 4277
(1988).
3. L. Turi, D. Borgis, J. Chem. Phys. 117, 6186 (2002).
4. K. Yokoyama, C. Silva, D. H. Son, P. K. Walhout, P. F.
Barbara, J. Phys. Chem. A 102, 6957 (1998).
5. M. S. Pshenichnikov, A. Baltuska, D. A. Wiersma, Chem.
Phys. Lett. 389, 171 (2004).
6. A. E. Bragg, J. R. R. Verlet, A. Kammrath, O.
Cheshnovsky, D. M. Neumark, Science 306, 669
(2004); published online 16 September 2004
(10.1126/science.1103527).
7. D. H. Paik, I.-R. Lee, D.-S. Yang, J. S. Baskin, A. H. Zewail,
Science 306, 672 (2004); published online 16
September 2004 (10.1126/science.1102827).
8. N. I. Hammer et al., Science 306, 675 (2004); pub-
lished online 16 September 2004 (10.1126/
science.1102792).
9. J. M. Weber et al., Chem. Phys. Lett. 339, 337 (2001).
10. J. V. Coe et al., J. Chem. Phys. 107, 6023 (1997).
11. H. M. Lee, S. Lee, K. S. Kim, J. Chem. Phys. 119, 187
(2003).
12. F. Wang, K. D. Jordan, Annu. Rev. Phys. Chem. 54, 367
(2003).
M
anipulating mice to model human
genetic disorders has become rou-
tine since the development of meth-
ods to introduce targeted mutations by ho-
mologous recombination. Although excellent
mouse models exist for many human single-
gene disorders such as hemophilia or
Zellweger syndrome, mouse models for oth-
er diseases only partially mimic or some-
times fail to recapitulate any aspect of the hu-
man syndrome. It is therefore surprising that
some mouse models of human conditions
that are caused by chromosome-scale anom-
alies have proved valuable. Perhaps the most
ambitious of these efforts is the creation of
mouse models for Down syndrome (DS), a
developmental abnormality characterized by
trisomy of human chromosome 21. It has
been presumed that several dosage-sensitive
genes in a section of human chromosome 21
called the Down syndrome critical region
(DSCR) are responsible for many of the fea-
tures of this disease, including craniofacial
abnormalities. On page 687 of this issue,
Olson et al. (1) put this theory to the test with
their study of mice engineered to be trisomic
but only for those sections of the mouse
genome that are orthologous to the human
DSCR. In this way, the investigators hoped to
more closely model the effect of carrying
three copies of genes in this region in an in-
tact animal. Surprisingly, they discovered that
three copies of the DSCR genes are not suf-
ficient to cause the cranial anomalies charac-
teristic of Down syndrome. These findings
allow a firm refutation of the notion that tri-
somy of the DSCR is the sole cause of the
craniofacial aspect of the Down syndrome
phenotype.
Down syndrome, or trisomy 21, is the
most common genetic cause of mental re-
tardation, with a worldwide frequency of 1
in 700 births. Trisomy results from sporadic
nondisjunction of chromosome 21 leading
to three copies of the smallest human chro-
mosome. Although the vast majority of in-
dividuals with Down syndrome have three
copies of the entire chromosome (and all of
the genes it contains), rare individuals with
Down syndrome have smaller portions trip-
licated because of unbalanced transloca-
tions. Comparison of the chromosome
anomalies and physical characteristics
shared among these patients has led to the
concept of a critical region for certain fea-
tures of Down syndrome (2). Although con-
troversial, the idea of a DSCR implies that
much of Down syndrome could be caused
by extra copies of one or a small number of
genes in this region (3). The notion that a
few genes might be of critical importance in
this syndrome is particularly attractive be-
cause such a simple model would bode well
for possible therapeutic intervention.
The development of a mouse model for
Down syndrome has not been easy. Human
chromosome 21 carries about 231 defined
genes across the 33.5 million bases (Mb) of
its long arm. The orthologous genes in the
mouse are distributed across three chromo-
somes: 10, 16, and 17. Mouse chromosome
16 contains orthologs of most of the human
chromosome 21linked genes, but it also car-
ries orthologs of genes found on three other
human chromosomes. Presumably as a result
of these additional genes, mice with trisomy
16 are not viable postnatally. This has neces-
sitated the development of segmental trisomy
mouse models of Down syndrome. The
Ts65Dn mousederived by Davisson and
colleagues using translocation chromo-
somesexhibits segmental trisomy for or-
thologs of 104 human chromosome 21
linked genes, and this mouse remains viable
into adulthood (4). A second partial trisomy
mouse model, Ts1Cje, carries a smaller seg-
ment containing 81 genes in 10.3 million
bases (5). Although neither mouse perfectly
models human trisomy 21, there are substan-
tial similarities in phenotype, notably cranio-
facial changes that mimic the human condi-
tion, along with electrophysiological differ-
ences in brain activity and altered behavior.
Olson et al. (1) exploited the ability to
create defined deletions and duplications in
mouse chromosomes by introduction of
Cre recombinase recognition sequences
through homologous recombination in
mouse embryonic stem cells. Pioneered by
Bradleys group, this method enables the
generation of specific deletions and dupli-
cations spanning tens of millions of bases
(6). A particularly successful application of
this technique by Baldini and colleagues
led to the creation of mouse deletions simi-
lar to those found in human DiGeorge syn-
drome (7). However, so far, these methods
have not been widely applied to creating
mouse models of human diseases, although
this may change with the recent description
by Adams et al. of a new resource to facili-
tate manipulation of the mouse genome (8).
In the new work, Olson and co-workers
engineered mice to carry either a duplica-
tion or deletion of a 3.9-Mb segment of
mouse chromosome 16 containing the 33
orthologs of genes found in the human
DSCR. The authors analyzed the pheno-
type of these mice. They then bred animals
with the deleted chromosome segment
with those carrying existing segmental tri-
somies (the Ts65Dn and Ts1Cje mice).
This enabled analysis of animals that were
trisomic for most of the genes in the
Ts65Dn or Ts1Cje mice, but disomic for
the 33 DSCR genes. They found that three
copies of these 33 genes alone were not
GE NE TI CS
The Critical Region in Trisomy 21
David L. Nelson and Richard A. Gibbs
The authors are in the Department of Molecular and
Human Genetics and the Human Genome Sequencing
Center, Baylor College of Medicine, Houston, TX
77030, USA. E-mail: nelson@bcm.tmc.edu
P E R S P E C T I V E S
Published by AAAS
619
In the interior of larger clusters and in bulk
water, the hydrated electron may not be
bound in the vicinity of double-acceptor
water molecules. However, this type of
arrangement could very well occur on ice
surfaces or at the surface of large water
clusters. The application of the vibrational
predissociation technique of Hammer et al.
to larger clusters may elucidate the location
(surface or interior) of the excess electron.
These new experimental results for neg-
atively charged water clusters are important
benchmarks for theoretical studies of the
structure and dynamics of excess electrons
in aqueous environments. Recent theoreti-
cal studies have shown that dispersion inter-
actions between the excess electron and the
electrons of the water molecules make an
important contribution to the binding ener-
gy of the former (12). Such interactions
could play a role in determining whether the
excess electron is surface- or interior-bound
and could also affect its dynamics.
References
1. E. J. Hart, J. W. Boag, J. Am. Chem. Soc. 84, 4090
(1962).
2. P. J. Rossky, J. Schnitker, J. Phys. Chem. 92, 4277
(1988).
3. L. Turi, D. Borgis, J. Chem. Phys. 117, 6186 (2002).
5. M. S. Pshenichnikov, A. Baltuska, D. A. Wiersma, Chem.
Phys. Lett. 389, 171 (2004).
6. A. E. Bragg, J. R. R. Verlet, A. Kammrath, O.
Cheshnovsky, D. M. Neumark, Science 306, 669
(2004); published online 16 September 2004
(10.1126/science.1103527).
7. D. H. Paik, I.-R. Lee, D.-S. Yang, J. S. Baskin, A. H. Zewail,
Science 306, 672 (2004); published online 16
September 2004 (10.1126/science.1102827).
8. N. I. Hammer et al., Science 306, 675 (2004); pub-
lished online 16 September 2004 (10.1126/
science.1102792).
(2003).
12. F. Wang, K. D. Jordan, Annu. Rev. Phys. Chem. 54, 367
(2003).
M
anipulating mice to model human
genetic disorders has become rou-
tine since the development of meth-
ods to introduce targeted mutations by ho-
mologous recombination. Although excellent
mouse models exist for many human single-
gene disorders such as hemophilia or
Zellweger syndrome, mouse models for oth-
er diseases only partially mimic or some-
times fail to recapitulate any aspect of the hu-
man syndrome. It is therefore surprising that
some mouse models of human conditions
that are caused by chromosome-scale anom-
alies have proved valuable. Perhaps the most
ambitious of these efforts is the creation of
mouse models for Down syndrome (DS), a
developmental abnormality characterized by
trisomy of human chromosome 21. It has
been presumed that several dosage-sensitive
genes in a section of human chromosome 21
called the Down syndrome critical region
(DSCR) are responsible for many of the fea-
tures of this disease, including craniofacial
abnormalities. On page 687 of this issue,
Olson et al. (1) put this theory to the test with
their study of mice engineered to be trisomic
but only for those sections of the mouse
genome that are orthologous to the human
DSCR. In this way, the investigators hoped to
more closely model the effect of carrying
three copies of genes in this region in an in-
tact animal. Surprisingly, they discovered that
three copies of the DSCR genes are not suf-
ficient to cause the cranial anomalies charac-
teristic of Down syndrome. These findings
allow a firm refutation of the notion that tri-
somy of the DSCR is the sole cause of the
craniofacial aspect of the Down syndrome
phenotype.
Down syndrome, or trisomy 21, is the
most common genetic cause of mental re-
tardation, with a worldwide frequency of 1
in 700 births. Trisomy results from sporadic
nondisjunction of chromosome 21 leading
to three copies of the smallest human chro-
mosome. Although the vast majority of in-
dividuals with Down syndrome have three
copies of the entire chromosome (and all of
the genes it contains), rare individuals with
Down syndrome have smaller portions trip-
licated because of unbalanced transloca-
tions. Comparison of the chromosome
anomalies and physical characteristics
shared among these patients has led to the
concept of a critical region for certain fea-
tures of Down syndrome (2). Although con-
troversial, the idea of a DSCR implies that
much of Down syndrome could be caused
by extra copies of one or a small number of
genes in this region (3). The notion that a
few genes might be of critical importance in
this syndrome is particularly attractive be-
cause such a simple model would bode well
for possible therapeutic intervention.
The development of a mouse model for
Down syndrome has not been easy. Human
chromosome 21 carries about 231 defined
genes across the 33.5 million bases (Mb) of
its long arm. The orthologous genes in the
mouse are distributed across three chromo-
somes: 10, 16, and 17. Mouse chromosome
16 contains orthologs of most of the human
chromosome 21linked genes, but it also car-
ries orthologs of genes found on three other
human chromosomes. Presumably as a result
of these additional genes, mice with trisomy
16 are not viable postnatally. This has neces-
sitated the development of segmental trisomy
mouse models of Down syndrome. The
Ts65Dn mousederived by Davisson and
colleagues using translocation chromo-
somesexhibits segmental trisomy for or-
thologs of 104 human chromosome 21
linked genes, and this mouse remains viable
into adulthood (4). A second partial trisomy
mouse model, Ts1Cje, carries a smaller seg-
ment containing 81 genes in 10.3 million
bases (5). Although neither mouse perfectly
models human trisomy 21, there are substan-
tial similarities in phenotype, notably cranio-
facial changes that mimic the human condi-
tion, along with electrophysiological differ-
ences in brain activity and altered behavior.
Olson et al. (1) exploited the ability to
create defined deletions and duplications in
mouse chromosomes by introduction of
Cre recombinase recognition sequences
through homologous recombination in
mouse embryonic stem cells. Pioneered by
Bradleys group, this method enables the
generation of specific deletions and dupli-
cations spanning tens of millions of bases
(6). A particularly successful application of
this technique by Baldini and colleagues
led to the creation of mouse deletions simi-
lar to those found in human DiGeorge syn-
drome (7). However, so far, these methods
have not been widely applied to creating
mouse models of human diseases, although
this may change with the recent description
by Adams et al. of a new resource to facili-
tate manipulation of the mouse genome (8).
In the new work, Olson and co-workers
engineered mice to carry either a duplica-
tion or deletion of a 3.9-Mb segment of
mouse chromosome 16 containing the 33
orthologs of genes found in the human
DSCR. The authors analyzed the pheno-
type of these mice. They then bred animals
with the deleted chromosome segment
with those carrying existing segmental tri-
somies (the Ts65Dn and Ts1Cje mice).
This enabled analysis of animals that were
trisomic for most of the genes in the
Ts65Dn or Ts1Cje mice, but disomic for
the 33 DSCR genes. They found that three
copies of these 33 genes alone were not
GE NE TI CS
The Critical Region in Trisomy 21
David L. Nelson and Richard A. Gibbs
The authors are in the Department of Molecular and
Human Genetics and the Human Genome Sequencing
Center, Baylor College of Medicine, Houston, TX
77030, USA. E-mail: nelson@bcm.tmc.edu
Published by AAAS
621
sufficient to generate the cranial changes
found in mouse models of Down syndrome
with larger numbers of triplicated genes.
Furthermore, reducing trisomy of these 33
genes to disomy in the Ts65Dn mouse did
not eliminate the phenotype.
Breeding mice with the deleted chromo-
some segment with trisomy mouse models is
a particularly elegant approach to testing the
role of the DSCR segment in Down syn-
drome. It would appear that triplication of the
33 DSCR genes is not necessary at least for
the craniofacial alterations characteristic of
the disease, thus reducing the likelihood of a
contiguous critical region for this aspect of
Down syndrome. The authors favor a model
in which individual or small numbers of
genes can make a critical contribution to
Down syndrome, but where the effect is high-
ly contextual, depending on the combined ef-
fects of altering the dosage of other genes.
The generation of these new mouse
models will allow additional study of the
association of these 33 genes with other
Down syndrome abnormalities associated
with behavior, electrophysiology, and loss
of cerebellar granule cells (9). It will also be
interesting to increase the size of the dupli-
cated chromosome segments to more close-
ly mimic the human disorder, although if
Olson et al. are correct that combinations of
genes of small (or no) individual effect can
contribute to the overall phenotype, the
numbers of permutations are daunting.
Mice remain our most useful genetic rela-
tive for modeling human disorders, despite
numerous differences that complicate analy-
sis. For diseases involving mental retardation,
this is a particularly acute problem, as alter-
ations in behavior and learned tasks must suf-
fice to flag differences in mental acuity be-
tween mutant mice and their normal counter-
parts (10). Another challenge is the difference
in colinearity of the human and mouse
genomes, and the lack of conservation of
gene order. Even though we now have com-
plete genome sequences for both species,
there are still many sequences not currently
recognized as genes that could prove to be of
great importance when designing mouse
models of human disorders. Mouse models
such as those described here may offer one of
the best ways to understand whether such se-
quences contribute to phenotype.
References
1. L. E. Olson, J. T. Richtsmeier, J. Leszl, R. H. Reeves,
Science 306, 687 (2004).
2. M. K. McCormick et al., Genomics 5, 325 (1989).
3. J. R. Korenberg et al., Proc. Natl. Acad. Sci. U.S.A. 91,
4997 (1994).
4. R. H. Reeves et al., Nature Genet. 11, 177 (1995).
5. H. Sago et al., Proc. Natl. Acad. Sci. U.S.A. 95, 6256
(1998).
6. R. Ramirez-Solis et al., Nature 378, 720 (1995).
7. E. A. Lindsay et al., Nature 401, 379 (1999).
8. D. J. Adams et al., Nature Genet. 36, 867 (2004).
9. L. L. Baxter et al., Hum. Mol. Genet. 9, 195 (2000).
10. M. E. Coussons-Read, L. S. Crnic, Behav. Genet. 26, 7
(1996).
H
ow sensitive is the climate to changes
in solar irradiance, atmospheric
aerosols, greenhouse gases, and other
climate forcings? To answer this question, we
first need to know the true extent of past cli-
mate fluctuations. The changing tempera-
tures over past centuries and millennia have
been reconstructed by regressing annually re-
solved climate proxy recordsfor example,
from ice cores and tree ringsagainst recent
thermometer measurements. On page 679 of
this issue, von Storch et al. (1) investigate
whether climate changes over decades and
longer are likely to have been captured realis-
tically in such reconstructions of Northern
Hemisphere (NH) mean temperature.
The likelihood that reconstructions of
this kind represent accurate hindcasts of
past climate is usually assessed by verifi-
cation against a short period of independ-
ent thermometer data. Such verification is
only possible for short-term (annual to
decadal) climate variability, because the in-
strumental climate record is too short to
sample longer (decadal to centennial) time
scales adequately.
To overcome this limitation, von Storch
et al. use a 1000-year simulation from a
coupled ocean-atmosphere model as a test-
bed in which the (simulated) NH tempera-
ture is known. They then generate pseudo-
proxy records by sampling a small selection
of the models simulated grid-box tempera-
tures (replicating the spatial distribution of
existing proxy records) and degrading them
with statistical noise.
The authors show that most of their at-
tempts to reconstruct the models NH temper-
ature with the pseudo-proxies result in signif-
icant underestimates of the amplitude of fluc-
tuations over the last millennium. Published
temperature reconstructions for the real
world, based on similar calibration methods,
may suffer from the same limitation.
Although von Storch et al. focus their
discussion on the reconstruction method of
Mann et al. (2), their conclusions are rele-
vant to other attempts to reconstruct NH
temperature history. They demonstrate
even greater loss of long-term variations
with a simple regression-and-averaging
method [this observation was also made in
(3)]. The results may apply to all regres-
sion-based methods. Accepting von Storch
et al.s results does not mean that we must
also accept that their simulated tempera-
ture history is close to realitymerely that
it is a reasonable representation of climate
behavior for which any valid reconstruc-
tion method should perform adequately.
The underestimated long-term variability
obtained by von Storch et al. is not a result
of problems with proxy data or the ability of
the proxies to retain information on these
time scales (4), because the pseudo-proxies
were generated free from such biases.
Neither is it simply due to the usual loss of
variance associated with any regression-
based prediction (this loss already forms the
basis for published estimates of reconstruc-
tion error). This usual loss of variance is of-
ten modeled as a random error, and although
a reconstruction may not be perfect, it cannot
be scaled by a simple multiplier to achieve a
better fit (that is, the reconstruction and its
error are uncorrelated) during the calibration
period. It is clear from figure 1 of (1) that the
underestimation of long-term temperature
variability is systematic rather than random:
At these time scales, a better fit to the actual
NH temperature can be achieved by scaling
a reconstruction by a simple multiplier (>1),
because the reconstruction and its error are
correlated. Such error is not incorporated in
the uncertainties associated with any pub-
lished NH temperature reconstruction.
The source of this systematic error can be
traced to differing shapes of the variance
spectra of the NH temperature and of the
pseudo-proxy data. The authors constructed
pseudo-proxies by adding white noise to the
simulated temperatures. Doing so alters the
variance spectra and leads to a deficiency in
variance at longer time scales, even after cali-
bration (see the figure). Hence, for climate re-
constructions to be optimal on all time scales,
proxy data must have variance spectra that are
similar to those of the climate data that they
are presumed to represent. It is not only
through the noise inherent in proxy records
that this requirement may be violated. Using
CL I MATE
The Real Color
of Climate Change?
Timothy J. Osborn and Keith R. Briffa
The authors are with the Climatic Research Unit,
University of East Anglia, Norwich NR4 7TJ, UK. E-
mail: t.osborn@uea.ac.uk; k.briffa@uea.ac.uk
Published by AAAS
622
just a few proxy records to represent a well-
sampled NH average temperature, land-based
temperature proxies to reconstruct marine
temperatures, and precipitation-sensitive
proxies to reconstruct past temperatures might
all result in reconstructions that are insuffi-
ciently red (see the figure caption), that is,
lacking in variance at longer time scales.
The message of the study by von Storch
et al. is that existing reconstructions of the
NH temperature of recent centuries may
systematically underestimate the true cen-
tennial variability of climate. The factor of
2 or more suggested by their study is un-
certain because the extent of the problem
may depend on the shape of the real climate
spectrum. Of course, we do not know the
true shape of the spectrum of NH tempera-
ture for recent millennia. Nor do we know
whether the 1000-year climate simulation
used by von Storch et al. is closer to the re-
al world than any of the various proxy-
based reconstructions. Other model simu-
lations of the climate of the past 1000 years
(5) may be less red, implying that the un-
derestimation of long-term climate change
could be less pronounced.
If the true natural variability of NH tem-
perature is indeed greater than is currently
accepted, the extent to which recent warming
can be viewed as unusual would need to be
reassessed. Systematic errors in existing cli-
mate reconstructions will also complicate the
evaluation of climate model simulations of
past variability. Achieving consistency be-
tween our knowledge of past climate and
model simulations of that climate is crucial
for building confidence in our ability to sim-
ulate future climate.
The most important
ramification of the report
of von Storch et al. (1) is
that greater long-term cli-
mate variability is likely to
imply greater sensitivity of
climate to radiative forc-
ings such as greenhouse
gases. Improved climate re-
constructions, further mod-
el simulations, and a methodology that
takes account of all sources of error are
needed to determine whether the widely
cited range of a 1.5 to 4.5 K increase in av-
erage global temperature for a doubling in
CO
2
(6) is compatible with evidence from
the past. It is already clear, however, that
greater past climate variations imply
greater future climate change.
1. H. von Storch et al., Science 306, 679 (2004); published
online 30 September 2004 (10.1126/science.1096109).
2. M. E. Mann, R. S. Bradley, M. K. Hughes, Nature 392,
779 (1998).
3. T. J. Osborn, K. R. Briffa, F. H. Schweingruber, P. D. Jones,
www.cru.uea.ac.uk/~timo/p/a/.
4. K. R. Briffa et al., J. Geophys. Res. 106, 2929 (2001).
5. P. D. Jones, M. E. Mann, Rev. Geophys. 42,
2003RG000143 (2004).
6. J. T. Houghton et al., Climate Change 2001: The
Scientific Basis (Cambridge Univ. Press, Cambridge,
2001).
7. Supported by the UK Natural Environment Research
Council (NER/T/S/2002/00440).
Published online 30 September 2004;
10.1126/science.1104416
Include this information when citing this paper.
Century Decadal Annual
Actual temperature
has a red spectrum
Proxies measure
temperature plus noise
White noise
adds variance
at all time scales
T
e
m
p
e
r
a
t
u
r
e

v
a
r
i
a
b
i
l
i
t
y

s
p
e
c
t
r
u
m
Scale record to match
temperature in the
calibration period
Multiply record by a
regression constant
Good match for time
scales represented in
the calibration period
Reconstruction
is deficient at
long time scales
A B
C D
Incompatible colors of climate variability. (A) A variance spectrum expresses
the amount of variance in a time series that occurs at different frequencies or
time scales. A white spectrum has equal variance at all time scales, whereas a red
spectrum has greater variance at longer time scales than at shorter time scales. A
typical temperature record has a red variance spectrum. (B) Pseudo-proxies con-
structed by adding white noise to a simulated temperature record have variance
that is increased equally at all time scales, reducing the redness of the spectrum
(the ratio of longtime scale to shorttime scale variance). (C) Regression-based
calibration approaches scale the pseudo-proxy records by constant multipliers,
leaving their redness unchanged (and thus still lower than the redness of the ac-
tual temperature spectrum). (D) It is not possible, therefore, for any linearly scaled
proxy record to match the actual temperature spectrum at all time scales, and the
fit tends to be optimized to the time scales represented in the calibration period
(typically the last 100 years or less, and dominated by annual to decadal variabil-
ity), resulting in a deficiency in variance at longer time scales.
W
hen plants moved from water to
land 450 million years ago, they
needed to develop a sealed surface
to protect themselves against water loss in the
dry air environment. To solve this problem,
plants invented an epicuticular wax layer that
covers the entire surface of the plant that is
exposed to air. This protective wax cuticle al-
so serves a multitude of other functions. Its
elaborate micro- and nanostructure prevents
water and other particles from sticking to the
surface of leaves, keeping them clean and so
enhancing their ability to trap light for photo-
synthesis. Adhering water droplets and other
particles are washed away in a self-cleaning
process called the lotus effect (1). The wax
layer also filters out damaging ultraviolet
rays, prevents volatile chemicals and pollu-
tants from sticking to leaves and stems, and
protects plants against attack by microbes and
herbivores.
The plant cuticle is composed of a mix-
ture of cutins and polysaccharides, an intra-
cuticular wax layer, and an epicuticular sur-
face layer of wax crystals (see the figure).
The wax layer is formed from wax precursor
moleculesvery long chain fatty acids
(VLCFAs) and their derivativesthat are
synthesized by plant epidermal cells. But
how is such an elaborate structure construct-
ed on the surface of plants? How do the high-
ly hydrophobic wax precursor molecules get
to the construction site outside of the plant
cell? And what were the evolutionary steps
that led to this innovation? On page 702 of
this issue, Pighin et al. (2) provide crucial in-
P L ANT B I OL OGY
A Plant ABC Transporter
Takes the Lotus Seat
Burkhard Schulz and Wolf B. Frommer
B. Schulz is in the Department of Horticulture and
Landscape Architecture, Purdue University, West
Lafayette, IN 47907, USA. W. B. Frommer is in the
Department of Plant Biology, Carnegie Institution,
Stanford, CA 94305, USA. E-mail: wfrommer@
stanford.edu
Published by AAAS
623
formation on the mechanism by which wax
precursor molecules are exported to the plant
surface. They show that plant cells use an
ABC (ATP-binding cassette) transporter pro-
tein similar to the ABC transporters found in
mammalian cells for this purpose.
Traditionally, VLCFAs were thought to be
exported by a vesicular pathway from their
site of synthesis in the endoplasmic reticulum
to their destination at the plant surface (see the
figure). Sequestration in vesicles would keep
these potentially harmful wax precursor mol-
ecules in a hydrophobic compartment inside
the cell in the same way as plant triglycerides
are stored in oleosin-coated vesicular oil bod-
ies (3, 4). Given the difficulty in analyzing the
export of wax precursors biochemically,
Pighin et al. chose a genetic approach. They
exploited a large collection of Arabidopsis
mutants with visibly altered wax cuticles on
the surface of their inflorescence stems. In
these so-called cer or eceriferum (not wax-
carrying) mutants (5), VLCFA biosynthesis is
affected. Through careful characterization of
the cer5 Arabidopsis mutant, Pighin et al.
identified an interesting candidate protein for
VLCFA export. They show that this CER5
protein is an ABC transporter expressed in
plant epidermal cells that, when defective, re-
sults in reduced wax on the surface of the cer5
plant stem. Fluorescence imaging revealed
that CER5 resides close to or at the plasma
membrane of the plant cell. Intriguingly, the
total VLCFA content in epidermal cells from
mutant and wild-type Arabidopsis is compa-
rable. This can be explained by the intracellu-
lar accumulation of VLCFAs in trilamellar
inclusions in the mutant plant cells. The in-
clusions might be modified vesicles that con-
tain large amounts of wax precursor mole-
cules that are either destined for export or
need to be stored in a separate compartment
to protect cell membranes from accumulating
too many VLCFAs.
Members of the ABC transporter family
transport a wide variety of substrates includ-
ing hydrophilic molecules, drugs, and lipids
across the membranes of mammalian cells.
Some cancer cells are able to survive despite
treatment with multiple antitumor drugs be-
cause they are induced to express multidrug
resistance (MDR) ABC transporters that
pump out the drugs faster than they can ac-
cumulate inside the cancer cells. Transporters
of this kind also export hydrophobic sub-
strates such as platelet-activating factor
(PAF), glycerophospholipids, and sphin-
golipids. The predicted transport activity of
the plant CER5 ABC transporter resembles
that of three other ABC lipid transporters:
MsbA, a recently crystallized bacterial pro-
tein involved in export of lipophilic mole-
cules (6); TDG1, which transports fatty acids
across the chloroplast envelope (7); and
ALDP, an ABC transporter that is mutated in
the neurological disease adrenoleukodystro-
phy. ALDP and its plant counterpart
PXA1/comatose (8) are found in peroxi-
somes, and both are crucial for VLCFA
degradation inside these organelles.
Mammalian cells expressing the mutant
ALDP transporter also exhibit the trilamellar
inclusions seen in cer5 mutant plant cells (9).
A seemingly simple hypothesis is that
CER5, a half-size ABC transporter, forms a
homo- or heterodimeric pore through which
VLCFAs are actively transported across the
plasma membrane (see the figure). This hy-
pothesis seems to rule out a vesicular path-
way of export. Given the near-insolubility
of the substrate and the difficulty in setting
up an export assay for the CER5 trans-
porter, it is not easy to directly determine
transport activity and substrate specificity.
But the identification of CER5 does not ab-
solutely rule out a vesicular pathway.
Because fluorescence imaging has limited
resolution, it remains possible that CER5 is
localized in a subapical compartment in-
volved in secretion. Another possibility is
that CER5 acts as a flippase (6), flipping
VLCFAs from the inner to the outer leaflet
of the plant cell plasma membrane.
Whether bacterial MsbA acts as a pore-like
transporter or as a flippase also remains a
matter of debate (10, 11). CER5, like
MsbA, may have a side port that permits
lipids to enter or exit the pore.
No matter how the lipid transporters work,
the low solubility of the VLCFAs implicates
fatty acidbinding proteins in all of these
models. Such proteins would be analogous to
serum albumin, which binds fatty acids in
serum, and to other fatty acidbinding pro-
teins of the mammalian cell cytosol. These
proteins need to be identified to clarify further
the export pathway for wax precursors in
plants. An orphan gene in the Arabidopsis
genome encoding protein At5g58070 and ex-
tracellular lipid-transfer proteins (LTPs) are
solid candidates. Independent of the actual
C
R
E
D
I
T
:
I
L
L
U
S
T
R
A
T
I
O
N

A
D
A
P
T
E
D

B
Y

K
A
T
H
A
R
I
N
E

S
U
T
L
I
F
F
/
S
C
I
E
N
C
E
3
1`
1b
1b
1b
1a
4
2
CER5
CER5
ER
Cytoplasm
Epicuticular wax crystals
ER membrane
Plasma
membrane
Wax
layer
Polysaccharide
and cutin
Lipid transfer
protein
Elongase
FABP
A
T
P
A
T
P
A
T
P
A
T
P
ATP ATP ATP ATP
Waxing its own surface. Different models for the transport of wax precursor molecules (VLCFAs) from
their site of synthesis in the endoplasmic reticulum (ER) to their site of deposition on the outer surface
of plant cells. VLCFAs (orange) synthesized by ER-localized elongases (pink) may be transported to the
cell surface by several routes. (1) They may be picked up by fatty-acid binding proteins (FABPs; green
moons) and transported to the ABC transporter CER5 (pale blue) localized in the plasma membrane.
CER5 may actively expel VLCFAs into the cell wall space in an ATP-dependent process. (1a) In a varia-
tion on this model, VLCFAs could be transported through a side port of CER5 into the upper leaflet of
the plasma membrane bilayer. (2) Alternatively, the CER5 transporter may act as a flippase, flipping
VLCFAs from the inner to the outer leaflet of the plasma membrane. (1b) In all cases, extracellular lipid-
transfer proteins (dark pink moons) will be required to facilitate transport of VLCFAs to destinations
outside the cell. Current data (2) are also compatible with vesicular transport of VLCFAs in either (3)
oleosome bodies covered by oleosin-like proteins (purple) or (4) uncoated vesicles that contain the
VLCFAs in lipid rafts. Available CER5 localization data do not exclude the possibility that CER5 loads in-
tracellular vesicles with VLCFAs by direct transport or through a flippase mechanism.
Published by AAAS
transport mechanism, the identification of
CER5 sheds light on wax secretion in plants
and may help to elucidate how the elaborate
micro- and nanostructure of the wax layer is
constructed. How did land plants invent wax
secretion? The genomes of living land plants
contain more than 100 ABC transporter genes
(12). Because transporters seem to be sloppy
with respect to their substrate specificity (13,
14), it is feasible that when plants crept out of
the water, they turned a member of the ABC
transporter family into a lipid exporter by en-
suring that it became localized to a different
cellular compartment. Perhaps this is an ex-
ample of an evolutionary principle in which
sloppiness is transformed into flexibility.
Obviously, there is more work to be done
to identify other components of the lipid ex-
port machinery. We need to define the exact
export pathway and its components. The re-
maining Arabidopsis cer mutants provide an
outstanding resource with which to fill in
the gaps to obtain a more complete picture.
Given that the reduced-wax phenotype of
the cer5 mutant is restricted to stems, the
transporters involved in wax deposition on
leaves and pollen will need to be identified.
A comparative analysis of fatty acid trans-
port in bacteria, plants, and animals, al-
though likely to reveal variations as well as
commonalities, will cross-fertilize research
in these respective fields. Such an analysis
will help to answer crucial questions, in-
cluding whether the fatty acid substrates
are free or bound and how the trilamellar
inclusions form. The new insight provided
by Pighin and colleagues into the ABC
lipid transporter of plants has implications
beyond understanding the lotus effect
given the multifunctional role of the wax
cuticle, the new findings will be a boon to
agriculture.
References
1. W. Barthlott, C. Neinhuis, Planta 202, 1 (1997).
2. J. A. Pighin et al., Science 306, 702 (2004).
3. L. Kunst, A. L. Samuels, Prog. Lipid Res. 42, 51 (2003).
4. L. L. Listenberger et al., Proc. Natl. Acad. Sci. U.S.A.
100, 3077 (2003).
5. A. M. Rashotte et al., Planta 219, 5 (2004).
6. C. F. Higgins, K. J. Linton, Science 293, 1782 (2001).
7. C. Xu et al., EMBO J. 22, 2370 (2003).
8. B. K. Zolman et al., Plant Physiol. 127, 1266 (2001).
9. D. Valle, J. Grtner, Nature 361, 682 (1993).
10. M. A. Kol et al., J. Biol. Chem. 278, 24586 (2003).
11. W. T. Doerrler et al., J. Biol. Chem., in press; published
online 10 August 2004 (10.1074/jbc.M408106200).
12. E. Martinoia et al., Planta 214, 345 (2002).
13. G. Reuter et al., J. Biol. Chem. 278, 35193 (2003).
14. A. Schmidt et al., J. Biol. Chem., in press; published on-
line 11 August 2004 (10.1074/jbc.M405433200).
A
year has passed since the celebration
of the 50th anniversary of the
Watson-Crick model for the double-
helical structure of DNA (1). Much of the
celebration looked back at the marvelous
advances that have emerged as genetics has
come to resemble organic chemistry.
Largely overlooked,
however, is a new
frontier in organic
chemistry that has
the goal of redesigning DNA to create arti-
ficial genetic systems. These artificial
DNA-like molecules are providing deeper
insight into how DNA works and are open-
ing the door onto a new world of synthetic
biology (2). They are also proving valuable
for diagnostic testing of human diseases.
According to the first-generation model
of DNA, the DNA duplex is like a ladder,
with the upright sections composed of pen-
tose sugar molecules linked together by
negatively charged phosphate groups (see
the figure). According to the model, the
uprights constrain the length of the base
pairs that form the rungs of the ladder. This
constraint, in turn, requires that the large
purine bases, adenine (A) or guanine (G),
pair with small pyrimidine bases, thymine
(T) or cytosine (C)a phenomenon known
as size complementarity. According to the
model, hydrogen bonds between purines
and pyrimidines ensure that the correct
large bases pair with the correct small
bases. From this model arose the two prin-
cipal rules (A pairs with T, G pairs with
C) that underlie all of molecular biology.
One motivation for redesigning DNA
using organic chemistry came from a vi-
sion of therapeutic benefit. For example,
an uncharged DNA analog might be able to
pass through a cell membrane, bind to an
unwanted RNA molecule according to
Watson-Crick rules, and neutralize its ac-
tivity (3). Many dozens of DNA analogs
having uncharged scaffolds were made in
pursuit of this vision (4). Remarkably, only
one can be said to have been truly success-
ful: the polyamide-linked nucleic acid
analogs (PNA) made by Nielsen et al. (5).
We now know that the repeating nega-
tive charge of the DNA backbone is tight-
ly tied to the rule-based molecular recog-
nition needed for transmission of genetic
information (6). The repeating negative
charge keeps contacts between two com-
plementary DNA strands as far away
from the backbone as possible, enforcing
Watson-Crick pairing. Without the repeat-
ing charge, DNA analogs bend, fold, ag-
gregate, or precipitate. Even PNA does
this, given sufficient length.
The repeating charge also dominates
the physical properties of DNA. The
charge allows the individual bases to be
substituted by mutation to create new DNA
molecules that behave physically like their
parents, but carry different genetic infor-
mation. The repeating negatively charged
phosphates of the DNA and RNA back-
bone are therefore key to evolution. Hence,
a repeating charge may be a universal
structural feature of all molecules carrying
genetic information in water, perhaps even
those on alien planets circling stars in re-
mote galaxies.
Other efforts to redesign DNA have
asked simple questions about the architec-
ture of base pairing. For example, Kool
wondered how DNA might behave if one
got rid of the hydrogen bonds entirely, and
used size complementarity as the sole prin-
ciple of pairing (7). Surprisingly, certain
DNA polymerases are able to match size-
complementary species without the benefit
of hydrogen bonding. This result encour-
aged Goodman to comment that DNA has
gone on the wagon to join hydrogen
bonds anonymous (8). Schultz, Romes-
berg, and their colleagues have elaborated
on Kools general theme, generating base
analogs that contact each other through un-
usual hydrophobic interactions (9). The lat-
est products from the Kool laboratory are
fluorinated bases that also pair using size
complementarity in the absence of hydro-
gen bonds (10).
Things generally work out better, how-
ever, if the hydrogen bonds are retained.
Hydrogen bonding might be important in
size-expanded base pairs (11), something
that has been seen previously in DNA
backbones with both longer and shorter
rungs (12). Carrying the theme further,
Minakawa et al. asked what might happen
if the hydrogen-bonding pattern were to
be extended into the minor groove of the
DNA backbone (13). With four hydrogen-
bonding opportunities, we can imagine 16
different hydrogen-bonding patterns sup-
porting 32 different nucleotide letters in
an expanded genetic alphabet based on
this architecture. The expanded scaffold-
ing works well, and a new class of design-
er DNA molecules may emerge from this
architecture.
CHE MI S TRY
Redesigning Genetics
Steven A. Benner
The author is in the Department of Chemistry,
University of Florida, Gainesville, FL 32611, USA. E-
mail: benner@chem.ufl.edu
Enhanced online at
www.sciencemag.org/cgi/
content/full/306/5696/625
Published by AAAS
626
One need not extend the scaffold of the
bases into the minor groove, however, to
get extra bases (letters) into the genetic al-
phabet. More than a decade ago, Switzer et
al. (14) and Piccirilli et al. (15) found that
an additional eight letters can be added to
the DNA alphabet if one simply shuffles
the arrangement of hydrogen bond donat-
ing and accepting groups (see the figure).
The physical properties of nonstandard
bases have now been optimized. For exam-
ple, tautomerism (unwanted movement of
hydrogen atoms) that causes nonstandard
bases to be lost during repeated copying
has been fixed, undesirable acid-base prop-
erties of the artificial genetic components
have been changed, and an annoying
epimerization (unwanted change in the
geometry of the molecule) displayed by
some nonstandard nucleotides has been
corrected (16).
The architecture of this artificially ex-
panded information system is so reminis-
cent of the Watson-Crick architecture, and
its properties are so similar to those found
in standard DNA, that one may wonder
why nature has not already exploited these
extra DNA letters. Recent advances in our
understanding of how the ribose sugar
might have arisen pre-
biotically on early
Earth (17) offer a clue.
Ribose is stabilized by
minerals containing bo-
rate, which might have
allowed ribose to accu-
mulate on early Earth.
Attaching a hetero-
cyclic ring to a ribose
via a carbon-nitrogen
bond, as in standard nu-
cleotides, requires a de-
hydration event, cer-
tainly conceivable (al-
though not trivial) pre-
biotically. Attaching a
heterocyclic ring to a ri-
bose via a carbon-car-
bon bond, as in some
nonstandard nucleo-
tides, appears to be far more difficult. The
structure of our DNA may therefore reflect
the minerals that were present in ancient
deserts on early Earth.
Luckily, prebiotic chemistry does not
constrain the application of expanded ge-
netic alphabets to human health care. For
example, the U.S. Food and Drug Admin-
istration recently approved a branched
DNA assay developed by Urdea and Horn
(18) that exploits nonstandard nucleotides.
Incorporating extra letters into DNA speeds
up hybridization and allows independent
binding of two rule-based molecular sys-
tems: one based on the standard letters A, T,
G, and C, and the other based on an artifi-
cial genetic alphabet. Currently, each year
some 400,000 patients infected with the hu-
man immunodeficiency virus or the hepati-
tis B and C viruses have their care enhanced
through diagnostic assays based on an ex-
panded genetic alphabet (19). Expanded ge-
netic alphabets are working their way into
other preclinical assays that test for cystic
fibrosis, SARS, and biohazards. They are
also entering research, where nonstandard
nucleotides underlie a large number of
emerging tools for systems biology research
and genome sequencing.
So what is next on the agenda as we re-
design DNA? It is hard to say. Perhaps fore-
shadowing the future is the discussion of
recent examples where artificial genetic
systems have been copied, and the copies
copied, using engineered polymerases (20).
Although most polymerases will accept
many nonstandard nucleotides with some
degree of efficiency when given no other
choice, polymerases have evolved for bil-
lions of years to efficiently accept only A,
T, G, and C. Therefore, most polymerases
wean unnatural nucleotides from a DNA
molecule if given the chance.
Both the structure of the nucleotide and
the structure of the polymerase can be
changed to obtain a pair where this does
not happen. Polymerase engineering is in
its infancy, however, and most attempts at
site-directed mutagenesis wreak site-di-
rected damage on the enzyme. But with the
advent of selection methods for polymeras-
es (21), we can expect in the not-too-distant
future fully artificial genetic systems that
support a synthetic biologya set of artifi-
cial chemical systems that can direct their
own replication and evolve.
References
1. J. D. Watson, F. H. C. Crick, Nature 171, 737 (1953).
2. W. Gibbs, Sci. Am. 290, 74 (May 2004).
3. P. S. Miller et al., Biochemistry, 16, 1988 (1977).
4. S. M. Freier, K. H. Altmann, Nucleic Acids Res. 25, 4429
(1997).
5. P. E. Nielsen, Mol. Biotechnol. 26, 233 (2004).
6. S. A. Benner, D. Hutter, Bioorg. Chem. 30, 62 (2002).
7. S. Moran et al., Proc. Natl. Acad. Sci. U.S.A. 94, 10506
(1997).
8. M. F. Goodman, Nature Biotechnol. 17, 640 (1999).
9. M. Fa, A. Radeghieri, A. A. Henry, F. E. Romesberg, J. Am.
Chem. Soc. 126, 1748 (2004).
10. J. S. Lai, E. T. Kool, J. Am. Chem. Soc. 126, 3040 (2004).
11. H. B. Liu et al., J. Am. Chem. Soc. 126, 1102 (2004).
12. C. R. Geyer et al., Structure 11, 1485 (2003).
13. N. Minakawa et al., J. Am. Chem. Soc. 125, 9970
(2003).
14. C. Y. Switzer et al., J. Am. Chem. Soc. 111, 8322
(1989).
15. J. A. Piccirilli et al., Nature 343, 33 (1990).
16. T. A. Martinot, S. A. Benner. J. Org. Chem. 69, 3972
(2004).
17. A. Ricardo et al., Science 303, 196 (2004).
18. M. A. Collins et al. Nucleic Acids Res. 25, 2979 (1997).
19. T. Elbeik et al., J. Clin. Microbiol. 42, 563 (2004).
20. P. Ball, Nature 431, 624 (2004).
21. F. J. Ghadessy et al., Proc. Natl. Acad. Sci. U.S.A. 98,
4552 (2001).
Tinkering with DNA. The standard DNA double helix has a scaffold of re-
peating negatively charged phosphate groups (green) that link together ribose
sugars (purple). This scaffold supports size-complementary pyrimidine and
purine bases (black) that present hydrogen bond donor (pink) and acceptor
(blue) groups. Each nucleotide (sugar, phosphate, base) plays a role in trans-
mitting genetic information. Attempts are under way to redesign DNA using
organic chemistry for a variety of uses including diagnostic testing. For exam-
ple, DNA molecules have been engineered to lack negatively charged phos-
phate groups (upper right) or hydrogen-bonding groups (middle), or have
been made with an increased number of hydrogen bonds or rearrangements
of these bonds (bottom). Redesigned DNA containing rearranged hydrogen
bonds (branched DNA) enhances the medical care of about 400,000 patients
annually through its use in diagnostic tests such as those detecting human
immunodeficiency virus and human hepatitis C virus.
O
O
O
O
P
O
O
C
C
N
N
N
N
N
O
N
N
O
O
F
F
O
N
O
O
P
O N
O
N
O
O
O
P
O
N
N
O
N
H
2
C
H
2
C
N
N
H
2
C
CH
2
H
2
C
H
2
C
NH
N
C
H
2
C
NHR
P
N
O
O
O
F
F
N
H
H
H
H
P
O
F
H
H
H
H
O
N
N
N
O
N
O
O
O
P
O
O
H
H
O
O
H
3
C
Hydrogen bond donors acceptors
Charged
phosphate
backbone
Pentose
sugar
G
-
-
-
-
-
-
C
A
T
Size complementarity
No H bonds
More H bonds
Different H bonds
directs interstrand
contacts here
Repulsion between
charged phosphates
O O
NH
O
CH
2
PNA
O
O
P
O
O
O
-
O
-
F
F
F
H N
N
O
N
N
N
N
N
O
N
P
O
O
H
H
O
O
CH
3
-
-
O
P
O
O
O
-
O
P
O
O
O
-
O H
H
RO
N
N
N
N
O
N
N
N
H
O
H
N H
H
Diagnostics
assays
No
interstrand
charge
repulsion
Published by AAAS
629 www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004
G E N E S I N A C T I O N
S
P
E
C
I
A
L
S
E
C
T
I
O
N
W
ords are static images on a page unless you know what they mean, how their meanings
change depending on the context, and what the rules are for using them. Similarly, a
complex regulatory code is buried within the genome, and researchers will need to
decipher it to understand how genes are expressed, what their functions are, and how
normal instructions are altered in disease.
Throughout the magazine and online this week are features that describe different aspects of gene
expression and its control. Kosak and Groudine (p. 644) describe how genomes may be organized (linearly
and within three-dimensional space in the nucleus) for the regulation of gene expression. A News story by
Pennisi (p. 632) provides a clear introduction to the world of enhancers: regulatory elements that have
been a pivotal force in evolution. Pastinen and Hudson (p. 647) describe the pitfalls associated with analyses
of cis-acting control mechanisms governing allele-specific differences in gene expression, some of which
have been associated with disease susceptibility. At Sciences online Signal Transduction Knowledge
Environment (STKE, www.sciencemag.org/sciext/genome2004) are features describing the dance of
nuclear receptor complexes with DNA that lead to transcription (Fowler and Alaric) and approaches different
organisms use to select genes within gene families for expression (Dalgaard and Vengrova).
Aware of the magnitude of the challenge of develop-
ing a complete parts list for all of these activities, an
international consortium of scientists has begun the
ENCODE (ENCyclopedia Of DNA Elements) project
(p. 636), whose goal is to identify all of the structural
and functional elements of the human genome. In their
pilot phase, researchers are comparing multiple approaches
for detecting different elements on 30 Mb of DNA.
Some of the medical applications of this information
are tantalizing prospects, whereas others are already at
our doorstep. Hood et al. (p. 640) present a sweeping view
of how expression patterns will be combined with technological
advances to further predictive medicine. However, it is not an
entirely smooth path. For instance, as the number of gene array studies
proliferates, some researchers are finding that they dont necessarily lead to
quick diagnosis or prognosis (see News story by Marshall on p. 630). The
Science Functional Genomics Web site (www.sciencemag.org/feature/plus/sfg)
contains additional online discussion of whether microarrays have been
oversold and how they can reach their full potential. The site also has updated
links to other resources. At the Science of Aging Knowledge Environment
(SAGE KE, www.sciencemag.org/sciext/genome2004) are articles describing
how microarrays and other genome-scale technologies are being applied to
aging research (Kaeberlein; Melov and Hubbard). The Editorial by Lord and
Papoian (p. 575) explores efforts to standardize microarray data so that regulatory
agencies can use gene expression studies to evaluate drug safety.
In the course of their analysis of RNA expression for protein-coding and
nonprotein-coding sequences during the Drosophila life cycle (p. 655),
White and his group have come to see the task of assembling the functional
parts of the genome as being like a Rubiks cube. Although ~4.3 10
19
different
positions are possible, the cube can be resolved from any position by 29 or
fewer manipulations. As we begin to understand biological systems through
carefully designed experiments and analyses, the complexity we are seeing
now may begin to resolve into simpler principles.
BARBARA R. JASNY AND LESLIE ROBERTS
Solving Gene Expression
T I T L E O F S P E C I A L S E C T I O N
I NTRODUCTI ON
CONTE NTS
NE WS
630 Getting the Noise Out of Gene
Arrays
632 Searching for the Genomes
Second Code
A Fast and Furious Hunt for Gene
Regulators
VI E WP OI N T S A N D
RE V I E W
636 The ENCODE (ENCyclopedia Of
DNA Elements) Project
The ENCODE Project Consortium
640 Systems Biology and New
Technologies Enable Predictive
and Preventative Medicine
L. Hood, J. R. Heath, M. E. Phelps, B. Lin
644 Gene Order and Dynamic Domains
S. T. Kosak and M. Groudine
647 Cis-Acting Regulatory Variation
in the Human Genome
T. Pastinen and T. J. Hudson
Related STKE Articles and Protocol
and SAGE KE Articles p. 569,
Editorial p. 575, Perspective p. 625,
Research Article p. 655, Report p. 690,
and Functional Genomics Web site
C
R
E
D
I
T
:
A
D
N
A
N

L
O
T
I
A
Published by AAAS
When Margaret Cam began hunting for
genes that are turned up or down in
stressed-out pancreas cells a couple of years
ago, she wasnt looking for a scientific
breakthrough. She was shopping. As director
of a support lab at the National Institute of
Diabetes and Digestive and Kidney Diseases
(NIDDK), she wanted to test-drive manu-
factured devices called microarrays or gene
arrays that measure gene expression; she
had her eye on three different brands. These
devices are hot, as they provide panoramic
views of the genes that are active in a
particular cell or tissue at a particular time.
Gene array studies are increasingly being
used to explore biological causes and effects
and even to diagnose diseases. But array
experiments are expensive, and Cam wanted
to be sure that her colleagues would get
high-quality, repeatable, credible results.
She was taken aback by what she found.
Not only was she unable to pick a clear
winner, but she had a hard time figuring out
whether any of the arrays produced trust-
worthy results. As she delved deeper, she
found that the devices produced wildly
incompatible data, largely because they
were measuring different things. Although
the samples she tested were all the same
RNAs from a single batch of cellseach
brand identified a different set of genes as
being highly up- or down-regulated.
The disharmony appears in a striking
illustration in Cams 2003 paper in Nucleic
Acids Research. It shows a Venn diagram of
overlapping circles representing the number
of genes that were the most or least active
on each device. From a set of 185 common
genes that Cam selected, only four behaved
consistently on all three platformsvery
low concordance, she said at an August
forum in Washington, D.C., run by the
Cambridge Healthtech Institute, based in
Newton Upper Falls, Massachusetts. Using
less rigorous criteria, she found about 30%
agreementbut nev-
er more than 52%
between two brands.
It was nowhere near
what we would ex-
pect if the probes
were assaying for the
same genes.
Cams f indings
caused ones jaw
to drop, says Marc
Sal i t , a physi cal
chemist at the Na-
tional Institute of
Standards and Tech-
nology (NIST). This
was not t he f i rst
paper to highlight
such inconsistencies,
but Cams little dia-
gram had an impact:
With support from
commercial ar ray
makers and academ-
ics, Salit is now coordinating an effort at
NIST to understand exactly what is
measured by these devices.
A few ex-enthusiasts think the promise
of gene arrays may have been oversold,
especially for diagnostics. Take Richard
Klausner, the former director of the
National Cancer Institute (NCI) now at
the Bill and Melinda Gates Foundation in
Seattle, Washington. We were nave to
think that new hypotheses about disease
would emerge spontaneously from huge
files of gene-expression data, he says, or
that you could go quickly from this new
technology to a clinical tool. His own ex-
perience with arrays indicated they were
too sensitive and finicky: The more data
he gathered on kidney tumor cells, the less
significant it seemed.
But those who have persevered with
gene expression arrays attribute such prob-
lems to early growing pains. They claim
that experienced labs are already delivering
useful clinical informationsuch as
whether a breast cancer patient is likely to
require strong chemotherapyand that new
analytical methods will make it possible to
combine results from different experiments
and devices. Francis Barany of Cornell
Universitys Weill Medical College in New
York City insists that arrays work wellif
one digs deeply into the underlying biology.
Imperfections
Digging into the biology is just what Cam
did after her experiments produced reams
of discordant data. She and colleagues in
Marvin Gershengorns group at NIDDK
wanted to pick out a set of key genes active
in pancreatic tumor cells undergoing
differentiation. From there, they meant to
go on to examine how islet cells develop.
We were very surprised, she recalls,
when they couldnt cross-validate results
from studies done with Affymetrix,
Agilent, and Amersham arrays. So she
began pulling the machines apart.
Cam soon ran into a barrier: Manufac-
turers werent eager to share information
about the short DNA sequence probes
each kit uses to spot gene activity. Each
commer ci al syst em uses a si mi l ar
approach. Short bits of DNA from known
genes are printed as probes on arrays.
When an experimental sample is exposed
to the array, RNAs made by genes cling
specif ically to the probes that have a
complementary sequence, triggering a
fluorescent signal that can be read by an
optical scanner. The more RNA on a
probe, the stronger the signal. The activity
of thousands of genes can be tracked
simultaneously this way.
Although manufacturers identif ied
which genes the probes targeted, they would
not reveal the actual nucleotide sequence of
each probe. This made it difficult to know
exactly what the probes were picking up.
Recalls Zoltan Szallasi of Harvards
Childrens Hospital in Boston, For the
f irst 6 years researchers were actually
flying blind. That changed in 2001, he says,
when the companies began sharing data.
Cam says, We managed to get partial
sequences [of array probes] from Agilent,
along with full sequences from Affymetrix C
R
E
D
I
T
:
A
D
A
P
T
E
D

F
R
O
M

N
I
H
S
P
E
C
I
A
L
S
E
C
T
I
O
N
NE WS
Getting the Noise Out of Gene Arrays
Thousands of papers have reported results obtained using gene arrays,
which track the activity of multiple genes simultaneously. But are these
results reproducible?
-
-
-
-
No common standards are used
across platforms, so data are
difficult or impossible to compare
Rapid Increase in Microarray Publications
~ Number from PubMed
N
u
m
b
e
r

o
f

p
u
b
l
i
c
a
t
i
o
n
s
199596 1997 1998 1999 2000 2001 2002 2003
7 8 40 139
425
1122
2000
3000
0
500
1000
2000
3000
Hot technology. The number of studies using microarrays to analyze
genes being turned on and off in concert has exploded in the last decade.
Published by AAAS
and Amersham. Her team analyzed the
probe and gene matchup in detail and
found that supposedly different probes
were responding to pieces of the same
gene. Targeting different parts can be a
problem because genes often contain
many components that can be spliced
into variant RNA packages. The result,
several experts say, is that probes can
over- or underestimate gene activity.
Sor ting out the
confusion is tough
because the probes
have not been de-
signed to be specific
to gene-splice variants, and no one has even
created a master list of variants. Cam is
encouraged that companies are beginning to
make arrays specific to different splice vari-
ants. That should reduce the ambiguity.
Another confounding factor, Szallasi
claims, is promiscuous matches: Probes
often respond not only to gene products that
exactly fit the sequence but also to those
that cross-hybridize with near matches.
Every manufacturer claims to have avoided
this problem, but there must be a reason
why microarray probes targeting almost the
same region of a given gene give wildly
different intensity signals, he says.
Moreover, many probes just dont corre-
spond to annotated sequences in the public
database, RefSeq, Szallasi says; removing
these problematic probes signif icantly
improves study results. But the best way to
build confidence in gene array results and
novel analytical methods, he argues, is to
validate probe-gene matches using the more
rigorous and time-consuming polymerase
chain reaction methods of sequence testing.
Szallasi has been doing just that as part of
an effort to help collaborators at Harvard
and at Baylor College of Medicine in
Houston, Texas, merge their gene expression
data sets. Hes also been trying to get other
researchers in Boston to share information
on validated matches.
Emerging proof
The difficulty in comparing gene array
results, say Szallasi and others, raises ques-
tions about how much confidence to have in
the thousands of papers already published
and whether it will ever be possible to merge
existing data and find significant results.
Relatively few labs have tried to replicate
large gene expression studies, even those
using identical test devices, says NCIs
Richard Simon, a cancer clinician who
directs gene array studies. People dont want
to waste hard-to-obtain tissue on such work,
and theyd rather not spend money on repli-
cating others findings. Simon argues that the
correct test of comparability in clinical medi-
cine is not whether you come up with the
same set of genes in two studies, but
whether you come up with an accurate and
consistent prediction of patient outcomes.
He and others note that gene arrays have
already proved their mettle in two clinical
areas: breast cancer and lymphomas. Molec-
ular geneticist Todd Golub of the Broad
Institute in Cambridge, Massachusetts, says
his group has independently validated gene
expression results of Louis Staudt of NCI
and Pat Brown of Stanford University that
identify subcategories of lymphoma that
have relatively poor or good outcomes. And
Lyndsay Harris, a breast cancer researcher
at Harvards Dana-Farber Cancer Institute,
says many of her colleagues have confi-
dence in gene expression data that identify a
high-risk form of breast cancer associated
with cells in the basal epithelium, a strategy
that Charles Perou, now at the University of
North Carolina, Chapel Hill, pioneered.
In basic research as well, Golub agrees
with Simon that broad themes, not specific
genes, should be the focus of comparison
studies. He looks for a biologically robust
response in patterns of gene activitysuch
as activation of coordinated cell signalsas
a sign that two experiments have detected
the same result. Spotting a signal in the
noise is like recognizing a face, regardless
of whether youre wearing bifocals, or
sunglasses, or no glasses. Eventually,
Golub says, biostatistical methods can
probably be used to define such signa-
tures in a flexible way to recognize dif-
ferent expression patterns as alternative
forms of the same result.
Trials have begun to test some of the
newer interpretations of cancer pathology
based on gene expression, such as efforts to
profile high-risk breast cancer at the Nether-
lands Cancer Institute (Science, 19 March,
p. 1754). But many champions of gene-
expression tests contend that they are not
yet ready for prime-time clinical use.
Staudt thinks the time will come when
every cancer patient gets a microarray-
based diagnosis. But before then, we still
have to show how reproducible the results
are. He is part of an NCI-sponsored con-
sortium that is attempting to correlate re-
sults from his own group, obtained from one
type of device (spotted arrays of lengthy
cDNAs), with those from a type of mass-
produced device (printed arrays of short
oligonucleotides) made by Affymetrix.
Already, they have achieved very good
prediction of tumor type in retrospective
studies of 500 samples. Now they plan to
test the model prospectively.
Seeking harmony
Researchers have now got all the major
journals using a single format to report array
data, says Alvis Brazma of the European
Bioinformatics Institute in Hinxton, UK, a co-
founder of the Microarray Gene Expression
Data Society. But eliminating discordance in
the hardware may not be so easy, says Ernest
Kawasaki, chief of NCIs microarray facility:
If I had all the money in the world, I would
say the best thing is to start over from the
beginningpresumably with a set
of validated gene expression probes
and shared standards.
That kind of money isnt avail-
able, but Salit says NIST recently
agreed to spend $1.25 million a year
for 5 years to tackle the problem of
discordance. Salit is coordinating a
group that includes microarray
manufacturers and a coalition of
academicsthe External RNA
Control Consortiumto develop a
set of standards that can be used to
calibrate gene arrays and ensure that
results can be translated meaningfully
from one lab to another. If it
succeeds, the pie is going to get
bigger because everybodys results
will improve.
ELIOT MARSHALL
C
R
E
D
I
T
S

(
T
O
P
)
:
A
D
A
P
T
E
D

F
R
O
M

N
U
C
L
E
I
C
A
C
I
D
S
R
E
S
E
A
R
C
H
;
(
B
O
T
T
O
M
)
A
D
A
P
T
E
D

F
R
O
M

N
U
C
L
E
I
C
A
C
I
D
S
R
E
S
E
A
R
C
H
S
P
E
C
I
A
L
S
E
C
T
I
O
N
Map of discordance. An experiment at NIH found that three
commercial devices rated different genes as being turned on
(red) and turned off (green) in a single batch of pancreatic cells.
Little overlap. Three array
systems rated the activity of
185 genes differently in one test.
Amersham
93
1
4
19 39
5
24
Affymetrix
Agilent
Published by AAAS
Molecular biologists may have sequenced
the human genome, but its going to take
molecular cryptographers to crack its com-
plex code. Genes, keystones to the devel-
opment and functioning of all organisms,
cant by themselves explain what makes
cows cows and corn corn: The same genes
have turned up in organisms as different
as, say, mice and jellyfish. Instead, new
findings from a variety of researchers have
made clear that its the genomes exquisite
control of each genes activityand not
the genes per sethat matters most.
The evolution of the genetic diversity
of animal forms is really due to differences
in gene regulation, says Michael Levine, an
evolutionary biologist at the University of
California (UC), Berkeley. Turning on a
gene at a different time, or in a new place,
or under new circumstances can cause vari-
ations in, say, size, coloration, or behavior.
If the outcome of that new regulatory pat-
tern improves an organisms mating success
or ability to cope with harsh conditions,
then it sets the stage for long-term changes
and, possibly, the evolution of new species.
Unfortunately, people dont look to reg-
ulatory elements as the cause of the varia-
tion they see, says Stephen Proulx of the
University of Oregon, Eugene. These ele-
ments are a major part of the [evolution]
story thats been overlooked, Levine says.
That neglect is now being righted.
Although many biologists remain gene-
centric, an increasing
number are trying to fac-
tor in the effects of gene
regulation. Researchers
are beginning to come up
with eff icient ways to
locate the different regu-
latory regions scattered
along the strands of
DNA. Others are piecing
together the workings of
transcription factors, pro-
teins that interact with
regulatory DNA and with
each other to guide gene
activity. They are homing
in on exactly where tran-
scription factors operate
along the DNA and what
they do to ensure that a
gene turns on at the right
time and in the right amount.
Most are tackling the functions of
r egul at or y el ement s one at a t i me,
although a few are taking more global and
bioinformatics approaches (see sidebar on
p. 635). At the California Institute of
Technology (Caltech) in Pasadena, one
group is trying to identify all the regulatory
interactions in maturing embryos; their
goal is to elucidate how genes and regulatory
DNA work together to guide development
and also how those interactions change
over evolutionary time.
All this work is making clear that buried
in DNA sequence is a regulatory code akin
to the genetic code but infinitely more
complicated, says Michael Eisen, a
computational biologist at Lawrence
Berkeley National Laboratory in California.
Researchers can predict the proteins
specified by the genetic code, but, he adds,
we cant predict gene expression by
simply looking at the sequence.
Manolis Dermitzakis of the Wellcome
Trust Sanger Institute in Cambridge, U.K.,
agrees: The complexity of the genome is
much higher than we have defined for the
past 20 years. We have to change our way
of thinking.
From genes to regulation
At the Medical Research Councils Lab-
oratory of Molecular Biology in Cambridge,
U.K., Francis Crickco-discoverer of
DNAs structureSydney Brenner, and their
colleagues took the f irst steps toward
figuring out how genomes work. In 1966,
they proved that genes are written in a three-
unit code, each of which specifies a particular
amino acid. By combining these threesomes,
called codons, in different ways, the genome
encodes instructions for thousands of
proteins. This discovery focused the spot-
light on genes themselves and the coding
regions within them; for decades biologists
called the intervening DNA junk.
Consequently, the notion of gene regula-
tion languished, even when results pointed
to its importance. In the early 1970s, Allan
Wilson of UC Berkeley and his student,
Mary Claire King, demonstrated that
humans and chimps are quite similar in
their genes. The key to what makes the two
species so different, they suggested, lies in
where and when those genes are active.
But not until 2 years ago did experiments
begin to bear out this idea. Wolfgang Enard
of the Max Planck Institute for Evolutionary
Anthropology in Leipzig, Germany, and his
colleagues compared the expression of
many genes in tissues from chimps and
humans. Certain genes are far more active in
the human brain than in the chimp brain,
they reported in the 15 April 2002 issue of
Science (p. 340), a find-
ing that supports Wilson
and Kings ideas.
Enards 2002 work
came on the heels of
dozens of other studies
suggesting that gene
changes are not the be-all
and end-all of evolution-
ary innovation. Instead,
researchers increasingly
attribute innovation to a
variety of types of regula-
tory DNA, some just re-
cently detected. Certain
genes code for the pro-
teins that make up the
transcription machinery,
which binds to promot-
ers, the DNA right at the
beginning of a coding se-
NE WS
Searching for the Genomes Second Code
The genome has more than one code for specifying life. The hunt for the
various types of noncoding DNA that control gene expression is heating up
C
R
E
D
I
T
S

(
T
O
P
)
:
P
H
O
T
O

R
E
S
E
A
R
C
H
E
R
S

I
N
C
.
,
2
0
0
2
;
(
B
O
T
T
O
M
)

N
.
K
E
V
I
T
I
Y
A
G
A
L
A
/
S
C
I
E
N
C
E
S
P
E
C
I
A
L
S
E
C
T
I
O
N
Model organism. Fruit flies have played a critical role in the search for stretches of
regulatory DNA called enhancers, which control gene expression by binding to one or
more transcription factors.
Published by AAAS
C
R
E
D
I
T
S
:
(
T
O
P
)

N
.
K
E
V
I
T
I
Y
A
G
A
L
A
/
S
C
I
E
N
C
E
;
(
B
O
T
T
O
M

M
A
I
N

B
O
X
)
A
N
A
N
D

&

B
E
L
T
I
N
G
/
P
N
A
S
;
(
B
O
X
W
I
T
H
I
N
)

P
H
O
T
O

R
E
S
E
A
R
C
H
E
R
S

I
N
C
.
,
2
0
0
2

S
P
E
C
I
A
L
S
E
C
T
I
O
N
quence. Other genes code for transcription
factors that can be located anywhere in the
genome. All affect their target genes by at-
taching to regulatory DNAsometimes
called modulesthats usually near but not
next to a gene. Protein-laden modules that
stimulate gene activity are called en-
hancers; those that dampen activity are
called silencers.
As a plethora of meetings and research
reports suggests, enhancers are hot. They
are small genetic command centers,
consisting of stretches of 500 or so bases.
Those clusters in turn are peppered with
transcription factor binding sites, which
can be less than 10 bases long. The target
of a particular enhancerand its effect
depends on the spacing and order of the
binding sites within it.
Sometimes the enhancer simply contains
multiple copies of the same binding site and
therefore uses the same transcription factor
throughout its length. Other times, it has
multiple transcription factors, and slight
differences among these proteins can cause
one gene to turn on and another to turn off.
Both enhancers and silencers are
especially hard to find because they are
very small pieces of sequence and, unlike
promoters, reside at varying distances from
the gene they regulate. We have a lot of
potentially short stretches of DNA where
the action is and stretches of DNA that
dont matter, Patrick Phillips, an evolutionary
developmental biologist at the University
of Oregon, Eugene, points out. Theorists
predict that humans could have as many as
100,000 enhancers and silencers, but fewer
than 100 are known, says Levine.
Hot on the enhancer trail
To understand the role of
enhancers in development,
Levine is studying their archi-
tecture and function in the fruit
fly genome. The first challenge
he encountered was simply
f inding the elusive quarry:
Several years ago he encour-
aged his graduate student
Michele Markstein (and her
computer-savvy parents) to
write a computer program that
could begin to do just that.
The trio worked first with
a transcription factor, dorsal,
which was known to affect a
gene called zen. They already
knew that the enhancer for
zen contained four binding
sites for dorsal, packed close
together. The researchers used
that signature sequence as
a probe for f inding other
enhancers that also had
clusters of dorsal binding sites.
The method worked. Proof positive came
when the program pinpointed three previously
identified enhancers that control other genes.
It also turned up a dozen more clusters
containing three or four of dorsals binding
sites. The researchers have since shown that
at least two are definitely enhancers. Levine
is encouraged: Sometimes the clustering of
a single factors binding sites is sufficient to
find new enhancers, he says. Indeed, using a
similar strategy, Eisen identified a set of
enhancers responsible for anterior-posterior
development in the fruit fly. The groups
published their results 2 years ago.
That same year, Eisen, Levine,
and UC Berkeleys Benjamin
Berman teamed up to use this
approach, along with other
bioinformatics tools, to look for
more complex enhancers. Instead
of containing repeated binding
sites for the same transcription
factor, complex enhancers contain
binding sites for several different
factors, thereby providing precise
regulation of gene expression.
To f ind these sequences,
Berman and his colleagues looked
for pieces of DNA with binding
sites for five transcription factors.
They di dnt have a speci f i c
enhancer in mind but sought out
DNA with those binding sites
sitting close enough together to
suggest they formed an enhancer.
The five transcription factors all
affect embryonic genes. Initially,
Bermans program found several
dozen of what seem to be complex enhancers;
recently the count has more than doubled.
And in pinning down these enhancers, the
researchers uncovered almost 50 genes that
seem to be controlled by this same set of tran-
scription factors and thus are likely to work
together to guide early development
So far, the researchers have confirmed
that at least some of these newly identified
clusters really are enhancers by testing their
activity in transgenic fruit flies. They add
DNA consisting of the putative enhancer
and a marker gene. If the marker gene
shows up i n t he same pl ace as t he
enhancers target gene, then the researchers
know they have got what they want. These
dat a are showi ng t hat when several
enhancers have a similar binding site
composition, they tend to work together and
coordinate the expression of sets of genes.
Enhancer encryption
With the first enhancers in hand, Levine and
his colleagues were ready to take the next
step: to figure out how enhancers orches-
trate development and effect the changes
underlying evolution. They began to dissect
the architecture of these bits of sequence,
determining exactly where the transcription
factor dorsal attached and whether those
locations had anything to do with the
enhancers function. They also tracked down
transcription factors that interacted with the
same enhancers as dorsal.
Through these efforts, Levine and UC
Berkeley collaborator Albert Erives have
been able to decipher another layer of
code scattered in the arrangement of
binding sites within the enhancer. This code
Genome cryptography. The regulatory code is encoded in the
arrangement of an enhancers DNA binding sites (A), in the
spacing between binding sites (B), or by the loss or gain of one
or more of these sites (C).
One gene, many sizes. Enhancers from different species alter
the extent of a Hox genes expression (dark stain), variation
that leads to species-specific numbers of thoracic vertebrae.
Published by AAAS
C
R
E
D
I
T
S
:
(
T
O
P
)

P
H
I
L
L
I
P

C
O
L
L
A
,
O
C
E
A
N
L
I
G
H
T
.
C
O
M
;
(
B
O
T
T
O
M
)

J
O
S
E
P
H

D
O
U
G
H
E
R
T
Y
/
E
C
O
L
O
G
Y
.
O
R
G
S
P
E
C
I
A
L
S
E
C
T
I
O
N
is critical to directing patterns of differen-
tiation in embryonic tissue, they reported
in the 16 March Proceedings of the
National Academy of Sciences.
Dorsal, whose concentration in the
embryo is highest on the ventral side and
decreases toward the dorsal, is key in defin-
ing these regions. Gene activity varies along
the dorsal-ventral axis, leading to the differ-
entiation of tissue types. High amounts of
dorsal lead to mesoderm, a precursor to
muscle; low amounts stimulate the develop-
ment of precursor nervous system tissue.
When the researchers probed more
closely into the enhancer, they found that a
binding site for dorsal was always next to the
binding site for a transcription factor called
twist. Levine found that the close proximity
and specific order of the binding sites were
conserved in an equivalent enhancer in
mosquito and a different fruit fly: They
were not randomly arranged, Levine says.
These results suggested that the effect
of an enhancer on a gene is determined
not just by the combination of transcription
factors but also by the spacing between
the binding sites. Levine thinks that dorsal
and twist have to be quite close together
for the enhancer to work dorsally, where
concentrations of dorsal are low; when far
apart, much more dorsal is needed. Thus it
seems that the genome can use the same
subset of transcription factors to regulate
different genes simply by changing the
order or spacing of those proteins, or
where they bind along the enhancer. With
this work, Levine has gotten inside the
mind of enhancers, says Eisen.
Enhancers and evolution
Alterations in the order and spacing of
binding sites can also affect how the same
gene works across several species, new
research is finding, suggest-
ing that it might take only
a relatively simple re-
arrangement to change an
enhancer and affect evolu-
tion. At a meeting on devel-
opmental diversity held in April in Cold
Spring Harbor, New York, developmental
biologist Cooduvalli Shashikant of Penn-
sylvania State University, University Park,
described his survey of enhancer effects on a
gene called Hoxc8. This gene, found in
many organisms, helps define the number
and shape of thoracic vertebrae. Shashikant
suspected that the enhancer, rather than the
gene alone, plays a pivotal role in delineating
different vertebrae configurations among
species. To find out, he and his colleagues
analyzed the sequence of the 200-base-pair
enhancer that lays just upstream of Hoxc8 in
zebrafish, puffer fish, and mice.
After adding a reporter gene to each
enhancer so they could see where it was
active, Shashikants group inserted the com-
bination into mouse embryos. Then they
compared the pattern of expression generated
with the zebrafish and puffer fish enhancers
to that of the mouse enhancer. In all three
cases, the activity of the gene was restricted
to the back part of the developing spine.
Shashikant suspects that the subtle differ-
ences among the enhancers in each species
changes the physical boundaries of Hoxc8
expression within the embryo, thereby help-
ing explain why chickens wind up with
seven thoracic vertebrae and snakes about
200, he says. Shashikant is also looking at
the sequences of this enhancer in other
species, including whales and coelacanths,
and again he has found
changes that probably
help define each organisms
shape. Sometimes they are sim-
ple sequence changes. Other times,
missing or additional DNA alters the mix of
transcription factors involved. Through
evolution, a lot of tinkering has gone on at
the enhancer level, Shashikant says.
Evolving embryonic differences
Similarly, Eric Davidson, a developmental
gene regulation biologist at Caltech, has
found that a small change in one enhancers
structure, and likely many alterations in all
sorts of enhancers, pave the way to the
different developmental pathways that
make each species distinctive. Five years
ago, he and his colleagues embarked on an
ambitious project to map all the genetic
interactions necessary to build the embry-
onic sea urchins endomesoderm, cells that
are precursors to the skeleton, gut, and
other tissues. They worked gene by gene,
determining the expression pattern of each
and deducing their functions by knocking
them out in developing embryos. From
these data and more, they pieced together a
computer model of the 50-gene circuitry
controlling the embryos first 30 hours as
an initial glob of cells began to differentiate
into endomesoderm.
The circuit is a hairball of proteins such
as transcription factors and signaling
mol ecul es, and t hei r genes, al l
connected by means of regulatory
DNA into multiple feedback loops.
Multiple transcription factors partner
with an enhancer to control the activity
of other transcription factors. Thus,
even though the circuit involves just a
few cell types, its a very complex
network, says Davidsons Caltech
colleague Veronica Hinman.
Hinman and Davidson have now
taken the next step: elucidating the
role of gene regulation in helping to
define developmental differences in
two echinoderms. Hinman has been
working out the same genetic circuitry
in a starfish. Whereas the starfish and
the sea urchin shared a common
ancestor about 500 million years ago
and still have similar embryos, the
two species have long since
gone their separate ways.
Adult sea urchins look
like pincushions, with
rounded exoskeletons;
st arf i sh are fl at ,
with arms protruding
from a central cavity.
Hinman has focused
on the earliest moments
of the starf ishs life.
Despite the differences in
the adults, much of the
embryonic genetic circuitry
studied so far is almost identical in
both species, she reported in 2003.
Yet subtle variations have had a big
impact. For example, theres a five-gene
circuit both species share. A key gene in
this pathway is otx, and it sets off the
circuit in the sea urchin and the starfish.
Hinman has found a tiny change in this
enhancer: Between the two species, this
enhancer varies by just one binding site,
for a transcription factor called t-brain.
Enhancing differences. Sea urchins and
starfish share much of their embryonic
genetic circuitry, but their enhancers can
vary, altering developmental pathways.
Published by AAAS
C
R
E
D
I
T
:
M
A
L
A
K

S
H
O
U
K
R
Y
/
L
A
W
R
E
N
C
E

B
E
R
K
E
L
E
Y

N
A
T
I
O
N
A
L

L
A
B
O
R
A
T
O
R
Y
S
P
E
C
I
A
L
S
E
C
T
I
O
N
The starfish has this binding site; the sea
urchin does not. In the sea urchin, t-brain
works in concert with other regulatory
genes and sets off the embryos skeleton-
forming circuitry, a genetic pathway absent
in the starfish embryo. But because the otx
enhancer is missing t-brain, the sea urchin
must also rely on a different transcription
factor to get the otx gene in the five-gene
circuitry to kick in.
Meanwhile the t-brain binding site on
the starf ishs otx enhancer keeps otx
focused on genes for the incipient gut.
Davidson thinks that ancestral echino-
derms had a t-brain site on the enhancer
for otx, one that disappeared from that
enhancer in the sea urchin. This looks
like species-specif ic jury-rigging, he
points out. The evolution of body plans hap-
pens by changes in the network architecture.
Enhancing genome studies
Through these kinds of studies, researchers
are beginning to decode the regulatory
genome. If this code can be made clear, they
should be able to piece together how organ-
isms diversify, says Nipam Patel, a biologist
at UC Berkeley. For example, through
Davidson and his colleagues thorough
descriptions of the gene pathways guiding
development, they can pin down where
enhancer modules have been added or lost.
That understanding, in turn, is changing how
some researchers make sense of evolution,
adds Michael Ludwig of the University of
Chicago. Its a vision in which regulatory
elements, including enhancers and silencers,
are as important, if not more important, than
gene mutations in introducing genetic inno-
vations that may set the stage for speciation.
Changes in one type of regulatory element,
the transcription factors, can be quite detri-
mental, as each influences many genes. By
contrast, enhancer mutations work locally,
affecting just their target genes, and so are
less likely to have deleterious effects on the
rest of the genome.
Yet Ludwig and others are the first to
admit that they have not cracked this reg-
ulatory code. We need to learn what it is
and how this information is written in these
sequences, says Eisen. At this moment,
we still do not have that ability.
ELIZABETH PENNISI
A Fast and Furious Hunt for
Gene Regulators
Genes may be essential, but researchers increasingly recognize the
pivotal role that another element of the genomeregulatory DNA
plays in human disease, speciation, and evolution. In many labs, the
search to find where these regions are buried is intensifying (see main
text). While some researchers are tackling these regions one at a time,
others are experimenting with high-
speed methods to detect regulatory
regions, such as enhancers, en masse
and determine what each one does.
At the Whitehead Institute for
Biomedical Research in Cambridge,
Massachusetts, Richard Young and
Ernest Fraenkel are using microarrays to
analyze yeast in an effort to turn up all
its promoters, the sites at the beginning
of a gene that bind to activating
proteins. The team starts by treating a
cell and one of its transcription factors
so that the factor permanently sticks to
its DNA-binding sites, thereby tagging
all the promoters that use this factor.
Then they use that transcription factor
to isolate these sequences from the rest
of the genome. After filling a microarray
with pieces of yeast DNA whose positions on the genome are known,
they add the tagged DNA, which then links to its matching sequence
in the microarray, revealing the approximate location of each piece.
The computer program takes over from there, says Fraenkel,
refining the location of the binding sites using similar DNA from
other organisms as a guide. In this way the team has been able to
pinpoint the promoter-binding sites for each individual transcription
factor in yeast simultaneously. Now, Young and Fraenkel are using
this technique to find enhancers and regulatory DNA in organisms
that have more complex genomes than yeasts. Already they have
cornered enhancers on both human liver and pancreatic cells.
Others are also using comparative genomics techniques to fish
out regulatory regions. The idea is straightforward enough: Compare
two or more sequenced genomes to identify those places where DNA
outside genes is highly similar and presumably functional. The chal-
lenge lies in deciding which genomes to compare, explains postdoc
Marcelo Nobrega of Lawrence Berkeley National Laboratory (LBNL) in
California. If the animals are too closely related, their genomes will
share many noncoding sequences that have no connection to regula-
tion. If the organisms are too distant, then even functional regions,
including regulatory regions, will be too different to detect.
Nobregas LBNL colleague Len Pennacchio thinks he has the
answer. When his team compared the human genome to that of a
puffer fish, they came up with a whopping 2500 potential
enhancers. To test the effectiveness of this method, the LBNL team
has been inserting 100 of the putative
enhancers, i denti f i ed on human
chromosome 16, into transgenic mouse
embryos, which they then analyze for
signs of regulatory activity. At a meeting
in Cold Spring Harbor, New York, in May,
Pennacchio reported that 48 of the 60
enhancers tested to date were real.
But Ewan Birney and his colleagues
at the European Bioinformatics Institute
(EBI) in Hinxton, U.K., and the European
Mol ecul ar Bi ol ogy Laboratory i n
Hei del berg, Germany, worry that
comparisons alone will yield spurious
matches as well as valid ones. If you
look at conservation itself, it doesnt tell
you a great deal, says Birney.
To refine the comparative approach,
his team has written a program that
considers only short sequences that are conserved between human
and mouseand only when they are present in higher than usual
densities in front of a gene. EBIs computers divide human and mouse
genomes into small pieces, say, six or eight bases, then compare them.
In the initial experiments, the researchers asked their computers
to pick out a well-known piece of regulatory DNA called the TATA
box, which is required for the activation of many genes. The program
did just fine, says Birney, which gives him hope that it will also be
able to pin down other regulatory regions elsewhere in the genome.
Birney eventually hopes to tie these data in with forthcoming
results from other labs suggesting that gene regulation is controlled
by other aspects of the genome as wellsuch as how chromosomes
are twisted around various proteins. And that, he hopes, will enable
him to begin to address how all this is put together as a code that
researchers can use to decipher the true workings of the genome.
E.P.
Testing, testing. High-throughput screens for enhancers
can turn up false positives, which can be excluded by
looking for enhancer function in mouse embryos.
Published by AAAS
V I E W P O I N T
The ENCODE (ENCyclopedia Of
DNA Elements) Project
The ENCODE Project Consortium*.
The ENCyclopedia Of DNA Elements (ENCODE) Project aims to identify all functional
elements in the human genome sequence. The pilot phase of the Project is focused on
a specified 30 megabases (1%) of the human genome sequence and is organized as
an international consortium of computational and laboratory-based scientists
working to develop and apply high-throughput approaches for detecting all sequence
elements that confer biological function. The results of this pilot phase will guide
future efforts to analyze the entire human genome.
With the complete human genome sequence
now in hand (13), we face the enormous
challenge of interpreting it and learning how
to use that information to understand the
biology of human health and disease. The
ENCyclopedia Of DNA Elements (ENCODE)
Project is predicated on the belief that a
comprehensive catalog of the structural and
functional components encoded in the hu-
man genome sequence will be critical for
understanding human biology well enough
to address those fundamental aims of bio-
medical research. Such a complete catalog,
or Bparts list,[ would include protein-coding
genes, nonprotein-coding genes, transcrip-
tional regulatory elements, and sequences
that mediate chromosome structure and dy-
namics; undoubtedly, additional, yet-to-be-
defined types of functional sequences will
also need to be included.
To illustrate the magnitude of the chal-
lenge involved, it only needs to be pointed out
that an inventory of the best-defined func-
tional components in the human genome
the protein-coding sequencesis still incom-
plete for a number of reasons, including the
fragmented nature of human genes. Even with
essentially all of the human genome sequence
in hand, the number of protein-coding genes
can still only be estimated (currently 20,000
to 25,000) (3). Nonprotein-coding genes are
much less well defined. Some, such as the
ribosomal RNA and tRNA genes, were iden-
tified several decades ago, but more recent
approaches, such as cDNA-cloning efforts
(4, 5) and chip-based transcriptome analyses
(6, 7), have revealed the existence of many
transcribed sequences of unknown function.
As a reflection of this complexity, about 5%
of the human genome is evolutionarily
conserved with respect to rodent genomic
sequences, and therefore is inferred to be
functionally important (8, 9). Yet only about
one-third of the sequence under such selec-
tion is predicted to encode proteins (1, 2).
Our collective knowledge about putative
functional, noncoding elements, which repre-
sent the majority of the remaining functional
sequences in the human genome, is remark-
ably underdeveloped at the present time.
An added level of complexity is that many
functional genomic elements are only active
or expressed in a restricted fashionfor
example, in certain cell types or at particular
developmental stages. Thus, one could envi-
sion that a truly comprehensive inventory of
functional elements might require high-
throughput analyses of every human cell type
at all developmental stages. The path toward
executing such a comprehensive study is not
clear and, thus, a major effort to determine
how to conduct such studies is warranted.
To complement ongoing large-scale proj-
ects that are contributing to the identification
of conserved elements in the human genome
(10) (www.intlgenome.org) and to the isola-
tion and characterization of human full-
length cDNAs (11), the ENCODE Project
(www.genome.gov/ENCODE) was launched
to develop and implement high-throughput
methods for identifying functional elements
in the human genome. ENCODE is being
implemented in three phasesa pilot phase,
a technology development phase, and a
production phase. In the pilot phase, the
ENCODE Consortium (see below) is eval-
uating strategies for identifying various
types of genomic elements. The goal of
the pilot phase is to identify a set of pro-
cedures that, in combination, can be applied
cost-effectively and at high-throughput to
accurately and comprehensively character-
ize large regions of the human genome. The
pilot phase will undoubtedly reveal gaps in
the current set of tools for detecting func-
tional sequences, and may reveal that some
methods being used are inefficient or
unsuitable for large-scale utilization. Some
of these problems should be addressed in
*Affiliations for all members of the ENCODE Con-
sortium can be found on Science Online at www.
sciencemag.org/cgi/content/full/306/5696/636/DC1.
.To whom correspondence should be addressed.
E-mail: elise_feingold@nih.gov
ENCODE Project Scientific Management:
National Human Genome Research Institute (E. A.
Feingold, P. J. Good, M. S. Guyer, S. Kamholz, L. Liefer,
K. Wetterstrand, F. S. Collins).
Initial ENCODE Pilot Phase Participants:
Affymetrix, Inc. (T. R. Gingeras, D. Kampa, E. A. Sekinger,
J. Cheng, H. Hirsch, S. Ghosh, Z. Zhu, S. Patel, A.
Piccolboni, A. Yang, H. Tammana, S. Bekiranov, P.
Kapranov, R. Harrison, G. Church, K. Struhl); Ludwig
Institute for Cancer Research (B. Ren, T. H. Kim, L. O.
Barrera, C. Qu, S. Van Calcar, R. Luna, C. K. Glass, M. G.
Rosenfeld); Municipal Institute of Medical Research (R.
Guigo, S. E. Antonarakis, E. Birney, M. Brent, L. Pachter,
A. Reymond, E. T. Dermitzakis, C. Dewey, D. Keefe, F.
Denoeud, J. Lagarde, J. Ashurst, T. Hubbard, J. J.
Wesselink, R. Castelo, E. Eyras); Stanford University
(R. M. Myers, A. Sidow, S. Batzoglou, N. D. Trinklein, S. J.
Hartman, S. F. Aldred, E. Anton, D. I. Schroeder, S. S.
Marticke, L. Nguyen, J. Schmutz, J. Grimwood, M. Dickson,
G. M. Cooper, E. A. Stone, G. Asimenos, M. Brudno);
University of Virginia (A. Dutta, N. Karnani, C. M. Taylor,
H. K. Kim, G. Robins); University of Washington (G.
Stamatoyannopoulos, J. A. Stamatoyannopoulos, M.
Dorschner, P. Sabo, M. Hawrylycz, R. Humbert, J.
Wallace, M. Yu, P. A. Navas, M. McArthur, W. S. Noble);
Wellcome Trust Sanger Institute (I. Dunham, C. M. Koch,
R. M. Andrews, G. K. Clelland, S. Wilcox, J. C. Fowler, K. D.
James, P. Groth, O. M. Dovey, P. D. Ellis, V. L. Wraight, A. J.
Mungall, P. Dhami, H. Fiegler, C. F. Langford, N. P. Carter,
D. Vetrie); Yale University (M. Snyder, G. Euskirchen, A. E.
Urban, U. Nagalakshmi, J. Rinn, G. Popescu, P. Bertone, S.
Hartman, J. Rozowsky, O. Emanuelsson, T. Royce, S.
Chung, M. Gerstein, Z. Lian, J. Lian, Y. Nakayama, S.
Weissman, V. Stolc, W. Tongprasit, H. Sethi).
Additional ENCODE Pilot Phase Participants:
British Columbia Cancer Agency Genome Sciences
Centre (S. Jones, M. Marra, H. Shin, J. Schein); Broad
Institute (M. Clamp, K. Lindblad-Toh, J. Chang, D. B.
Jaffe, M. Kamal, E. S. Lander, T. S. Mikkelsen, J. Vinson, M.
C. Zody); Childrens Hospital Oakland Research Institute
(P. J. de Jong, K. Osoegawa, M. Nefedov, B. Zhu); National
Human Genome Research Institute/Computational
Genomics Unit (A. D. Baxevanis, T. G. Wolfsberg);
National Human Genome Research Institute/Molecular
Genetics Section (F. S. Collins, G. E. Crawford, J. Whittle,
I. E. Holt, T. J. Vasicek, D. Zhou, S. Luo); NIH Intramural
Sequencing Center/National Human Genome Research
Institute (E. D. Green, G. G. Bouffard, E. H. Margulies, M. E.
Portnoy, N. F. Hansen, P. J. Thomas, J. C. McDowell, B.
Maskeri, A. C. Young, J. R. Idol, R. W. Blakesley); National
Library of Medicine (G. Schuler); Pennsylvania State
University (W. Miller, R. Hardison, L. Elnitski, P. Shah);
The Institute for Genomic Research (S. L. Salzberg, M.
Pertea, W. H. Majoros); University of California, Santa
Cruz (D. Haussler, D. Thomas, K. R. Rosenbloom, H.
Clawson, A. Siepel, W. J. Kent).
ENCODE Technology Development Phase Participants:
Boston University (Z. Weng, S. Jin, A. Halees, H. Burden,
U. Karaoz, Y. Fu, Y. Yu, C. Ding, C. R. Cantor); Mas-
sachusetts General Hospital (R. E. Kingston, J. Dennis);
NimbleGen Systems, Inc. (R. D. Green, M. A. Singer, T. A.
Richmond, J. E. Norton, P. J Farnham, M. J. Oberley, D. R.
Inman); NimbleGen Systems, Inc. (M. R. McCormick, H.
Kim, C. L. Middle, M. C. Pirrung); University of California,
San Diego (X. D. Fu, Y. S. Kwon, Z. Ye); University of
Massachusetts Medical School (J. Dekker, T. M. Tabuchi,
N. Gheldof, J. Dostie, S. C. Harvey).
G E N E S I N A C T I O N G E N E S I N A C T I O N
S
P
E
C
I
A
L
S
E
C
T
I
O
N
the ENCODE technology development
phase (being executed concurrently with
the pilot phase), which aims to devise new
laboratory and computational methods that
improve our ability to identify known func-
tional sequences or to discover new func-
tional genomic elements. The results of the
first two phases will be used to determine
the best path forward for analyzing the
remaining 99% of the human genome in a
cost-effective and comprehensive produc-
tion phase.
ENCODE Targets
The defining feature of the ENCODE pilot
phase is the uniform focus on a selected 30
Mb of the human genome. Each pilot-phase
participant will study the entire set of
ENCODE targets44 discrete regions that
together encompass 1% of the human
genome. All approaches will thus be tested
on a relatively large amount of genomic se-
quence, allowing an assessment of the ability
of each to be applied at large scale. The use
of a single target set will allow the results of
different approaches to be directly compared
with one another.
The set of ENCODE targets was chosen
to represent a range of genomic features
(www.genome.gov/10005115). It was first
decided that a number of smaller regions
(0.5 to 2 Mb) distributed across many dif-
ferent chromosomes should be chosen, as
opposed to (for example) a single 30-Mb
region. To ensure that existing data sets and
knowledge would be effectively utilized,
roughly half of the 30 Mb was selected
manually. The two main criteria used for the
manual selection were as follows: (i) the
presence of extensively characterized genes
and/or other functional elements; and (ii) the
availability of a substantial amount of com-
parative sequence data. For example, the
genomic segments containing the "- and $-
globin gene clusters were chosen because of
the wealth of data available for these loci
(12). On the other hand, the region encom-
passing the CFTR (cystic fibrosis trans-
membrane conductance regulator) gene
was selected because of the extensive
amount of multispecies sequence data avail-
able (13). Once the manual selections had
been made, the remaining targets were
chosen at random by means of an algorithm
that ensured that the complete set of targets
represented the range of gene content and
level of nonexonic conservation (relative to
mouse) found in the human genome. The
locations and characteristics of the 44
ENCODE target regions (along with addi-
tional details about their selection) are
available at the UCSC ENCODE Genome
Browser (www.genome.ucsc.edu/ENCODE/
regions_build34.html).
The ENCODE Consortium
The pilot phase is being undertaken by a
group of investigators, the ENCODE Consor-
tium, who are working in a highly collabora-
tive way to implement and evaluate a set of
computational and experimental approaches
for identifying functional elements. The pilot
phase began in September 2003 with the
funding of eight projects (table S1) that in-
volve the application of existing technologies
to the large-scale identification of a variety of
functional elements in the ENCODE targets,
specifically genes, promoters, enhancers,
repressors/silencers, exons, origins of repli-
cation, sites of replication termination, tran-
scription factor binding sites, methylation
sites, deoxyribonuclease I (DNase I) hyper-
sensitive sites, chromatin modifications, and
multispecies conserved sequences of yet un-
known function (Fig. 1). Genetic variation
within the conserved sequences is also being
determined. The methodological approaches
being employed in the pilot phase include
transcript and chromatin immunoprecipitation
microarray hybridization (ChIP-chip; see
below) analyses with different microarray
platforms, computational methods for finding
genes and for identifying highly conserved
sequences, and expression reporter assays.
In addition to the initial eight, other groups
have since joined the ENCODE Consortium.
These include groups doing comparative
sequencing specifically for ENCODE, groups
coordinating databases for sequence-related
and other types of ENCODE data, and groups
conducting studies on specific sequence
elements (table S1). Beyond these current
Hypersensitive
Sites
CH
3
CO CH
3
Reporter
Assays
Epigenetic Modifications
Microarray
Hybridization
DNA Replication
Sites
Long-range regulatory elements
(enhancers, repressors/ silencers, insulators)
cis-regulatory elements
(promoters, transcription
factor binding sites)
Computational
Predictions
and RT-PCR
DNase
Digestion
Transcript
Gene
ChIP-chip other
Fig. 1. Functional genomic elements being identified by the ENCODE pilot phase. The indicated methods are being used to identify different types of
functional elements in the human genome.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
participants, the ENCODE Consortium is
open to all interested academic, government,
and private-sector investigators, as long as
they abide by established Consortium guide-
lines (www.genome.gov/10006162), which
require the commitment to work on the
entire set of ENCODE targets, to partici-
pate in all Consortium activities, to make a
significant intellectual contribution, and to
release data in accordance with the policies
specifically established for the project (see
below).
The parallel technology development
phase (table S2) is intended to expand the
tool box available for high-throughput
identification of functional genomic ele-
ments, and includes projects to develop new
methods both for more efficient identifica-
tion of known elements and for identification
of heretofore unknown elements.
Research Plans
Each group is using one or more high-
throughput approaches to detect a specific
genomic element(s). In some cases, multiple
platforms are being evaluated in comparable
experiments. For example, several types of
microarrays [e.g., oligonucleotide arrays made
by different technologies, polymerase chain
reaction (PCR)amplicon arrays] are being
used to identify transcribed regions. The pilot
project participants are primarily using a
restricted number of cell lines to identify
functional elements. This approach is being
taken for practical purposes, but it has the
limitation that not all cell types will be
surveyed, and therefore some elements with
tissue-restricted function may not be identi-
fied in the pilot phase. To facilitate compari-
son of data generated on different platforms
and by different approaches, a common set of
reagents is being included whenever appro-
priate. So far, the common reagents chosen
include two cell lines (HeLa S3, a cervical
adenocarcinoma, and GM06990, an Epstein-
Barr virustransformed B-lymphocyte) and
two antibodies [one for the general tran-
scription factor TAF
II
250 (14) and another
for the inducible transcription factor STAT-1
(15)].
The ENCODE pilot phase also includes a
component that is generating sequences of
the genomic regions that are orthologous to
the ENCODE target regions from a large set
of nonhuman vertebrates (www.nisc.nih.gov/
open_page.html?/projects/encode/index.cgi)
(16 ). This will allow ENCODE to identify
the quality and amount of comparative se-
quence data necessary to accurately identify
evolutionarily conserved elements, and to
develop more powerful computational tools
for using comparative sequence data to infer
biological function. An initial set of 10 ver-
tebrates have been selected for targeted
sequencing on the basis of multiple factors,
including phylogenetic position (Fig. 2
and table S3) and the availability of a bacteri-
al artificial chromosome (BAC) library. In
addition to this ENCODE-specific effort,
comparative sequence data are also being
captured from ongoing whole-genome se-
quencing projects, including those for mouse,
rat, dog, chicken, cow, chimpanzee, macaque,
frog, and zebrafish. Aunique RefSeq accession
number (17) is being assigned for the sequence
of each ENCODE-orthologous target region in
each species, with periodic data freezes.
A feature of the evolutionarily conserved
elements to be assayed is sequence variation.
This will be accomplished by resequencing
PCR-amplified fragments from genomic
DNA of 48 individuals, the same samples
being used by the HapMap Consortium to
determine common patterns of genetic vari-
ation (18). This will result in a quantitative
view of the evolutionary constraints on
conserved regions.
Data Management and Analysis
Capturing, storing, integrating, and display-
ing the diverse data generated will be
challenging. Data that can be directly linked
to genomic sequence will be managed at the
UCSC Genome Browser (www.genome.
ucsc.edu/ENCODE) (19). Other data types
will be stored either at available public
databases [e.g., the GEO (Gene Expression
Omnibus) (www.ncbi.nlm.nih.gov/geo) and
ArrayExpress (www.ebi.ac.uk/arrayexpress)
sites for microarray data] or on publicly
accessible Web sites specifically developed
by ENCODE Consortium participants. An
ENCODE portal will also be established to
index these data, allowing users to query
different data types regardless of location.
Access to metadata associated with each
experiment will be provided. The ENCODE
pilot phase will make use of the MAGE stan-
dard for representing microarray data (20),
and data standards for other data types will
be developed as needed.
Figure 3 uses the early pilot-phase data
for one of the ENCODE target regions
(ENr231) to illustrate how the ENCODE
data will be presented in a way that will cap-
ture the innovation of the ENCODE Projects
goal of developing a full representation of
functional genomic elements. The results of
the different analyses are presented as paral-
lel tracks aligned with the genomic sequence.
For example, the gene structures for known
genes are shown; future ENCODE data will
include the precise locations of 5 transcrip-
tion start sites, intron/exon boundaries, and 3
polyadenylation sites for each gene in the
ENCODE targets. In addition, efforts will be
made to confirm all gene predictions in these
regions. Positions of the evolutionarily con-
served regions, as detected by analyzing
sequences from several organisms, are shown
and can be correlated with results of other
studies. This presentation will allow a com-
prehensive depiction of the results from all of
the ENCODE components, enabling both
comparisons of different methods and of the
reproducibility of the data from different
laboratories. The former is illustrated by com-
parison of two different methods for localiz-
ing promoter regions, one based on reporter
constructs containing sequences around puta-
tive transcription initiation sites (21) and the
other involving chromatin immunoprecipita-
tion (ChIP) with an antibody to RNA poly-
merase (RNAP) and hybridization to DNA
microarrays (chip) (22) (so-called ChIP-chip)
to identify sequences bound by components
of the transcriptional machinery. Reproduc-
ibility is illustrated by the comparison of
Platypus
Opossum
Armadillo
Hedgehog
Shrew
Elephant
Tenrec
Bat
Bovine
Rabbit
Rat
Mouse
Dog
Galago
Cat
Guinea pig
Mouse lemur
Marmoset
Human
Chimpanzee
Baboon
Macaque
Orangutan
Colobus
Owl monkey
Dusky titi
Monotremata
Marsupialia
Edentata
Insectivora
Proboscidea
Chiroptera
Primates
Rodentia
Lagomorpha
Carnivora
Arteriodactyla
Hyracoidea
Insectivora
Fig. 2. Mammals for which genomic sequence
is being generated for regions orthologous to
the ENCODE targets. Genomic sequences of
the ENCODE targets are being generated for
the indicated mammalian species. The current
plans are to produce high-quality finished
(blue), comparative-grade finished (red), or
assembled whole-genome shotgun (green)
sequence, as indicated. High-quality finished
reflects highly accurate and contiguous se-
quence, with a best-faith effort used to resolve
all difficult regions (26). Comparative-grade
finished reflects sequence with greater than
eightfold coverage that has been subjected to
additional manual refinement to ensure accu-
rate order and orientation of all sequence
contigs (16). In the case of whole-genome
shotgun sequence, the actual coverage and
quality may vary. Other vertebrate species for
which sequences orthologous to the ENCODE
targets are being generated include chicken,
frog, and zebrafish (not shown). A complete list
of the ENCODE comparative sequencing efforts
is provided in table S3.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
studies conducted by two laboratories within
the ENCODE Consortium, which analyzed
different biological starting materials using
the ChIP-chip approach and found an 83%
concordance in the identified RNAP-binding
sites in the region (23). Representation of the
ENCODE data in this manner will afford a
synthetic and integrative view of the func-
tional structure of each of the target regions
and, ultimately, the entire human genome.
Each research group will analyze and
publish its own data to evaluate the experi-
mental methods it is using and to elucidate
new information about the biological function
of the identified sequence elements. In addi-
tion, the Consortium will organize, analyze,
and publish on all ENCODE data available on
specific subjects, such as multiple sequence
alignments, gene models, and comparison of
different technological platforms to identify
specific functional elements. At the conclusion
of the pilot phase, the ENCODE Consortium
expects to compare different methods used by
the Consortium members and to recommend a
set of methods to use for expanding this project
into a full production phase on the entire
genome.
Data Release and Accessibility
The National Human Genome Research In-
stitute (NHGRI) has identified ENCODE as
a community resource project as defined
at an international meeting held in Fort
Lauderdale in January 2003 (www.wellcome.
ac.uk/doc_WTD003208.html). Accordingly,
the ENCODE data-release policy (www.
genome.gov/ENCODE) stipulates that data,
once verified, will be deposited into public
databases and made available for all to use
without restriction.
Two concepts associated with this data-
release policy deserve additional discussion.
First, data verification refers to the assess-
ment of the reproducibility of an experi-
ment; ENCODE data will be released once
they have been experimentally shown to be
reliable. Because different types of experi-
mental data will require different means of
demonstrating such reliability, the Consor-
tium will identify a minimal verification
standard necessary for public release of each
data type. These standards will be posted
on the ENCODE Web site. Subsequently,
ENCODE pilot-phase participants will use
other experimental approaches to validate
the initial experimental conclusion. This en-
riched information will also be deposited in
the public databases.
Second, the report of the Fort Lauderdale
meeting recognized that deposition in a
public database is not equivalent to publica-
tion in a peer-reviewed journal. Thus, the
NHGRI and ENCODE participants respect-
fully request that, until ENCODE data are
published, users adhere to normal scientific
etiquette for the use of unpublished data.
Specifically, data users are requested to cite
the source of the data (referencing this
paper) and to acknowledge the ENCODE
Consortium as the data producers. Data users
are also asked to voluntarily recognize the
interests of the ENCODE Consortium and its
members to publish initial reports on the
generation and analyses of their data as pre-
viously described. Along with these publi-
cations, the complete annotations of the
functional elements in the initial ENCODE
targets will be made available at both the
UCSC ENCODE Genome Browser and the
ENSEMBL Browser (www.ensembl.org).
Conclusion
ENCODE will play a critical role in the next
phase of genomic research by defining the
best path forward for the identification of
functional elements in the human genome.
By the conclusion of the pilot phase, the 44
ENCODE targets will inevitably be the most
well-characterized regions in the human
genome and will likely be the basis of many
future genome studies. For example, other
large genomics efforts, such as the Mamma-
lian Gene Collection (MGC) program (24)
and the International HapMap Project (25),
are already coordinating their efforts to
ensure effective synergy with ENCODE ac-
tivities. Although the identification of all
functional genomic elements remains the goal
of the ENCODE project, attaining such
comprehensiveness will be challenging,
especially in the short term. For example,
not all types of elements, such as centromeres,
telomeres, and other yet-to-be defined ele-
ments, will be surveyed in the pilot project.
Nonetheless, this project will bring together
scientists with diverse interests and exper-
tise, who are now focused on assembling the
functional catalog of the human genome.
1. International Human Genome Sequencing Consor-
tium, Nature 409, 860 (2001).
2. J. C. Venter et al., Science 291, 1304 (2001).
3. International Human Genome Sequencing Consor-
tium, Nature, in press.
4. Y. Okazaki et al., Nature 420, 563 (2002).
5. T. Ota et al., Nat. Genet. 36, 40 (2004).
6. J. L. Rinn et al., Genes Dev. 17, 529 (2003).
7. P. Kapranov, V. I. Sementchenko, T. R. Gingeras, Brief
Funct. Genomic Proteomic 2, 47 (2003).
8. International Rat Sequencing Consortium, Nature
428, 493 (2004).
9. International Mouse Sequencing Consortium, Nature
420, 520 (2002).
10. D. Boffelli, M. A. Nobrega, E. M. Rubin, Nat. Rev.
Genet. 5, 456 (2004).
11. T. Imanishi et al., PLoS Biol. 2, 856 (2004).
12. T. Evans, G. Felsenfeld, M. Reitman, Annu. Rev. Cell
Biol. 6, 95 (1990).
13. J. W. Thomas et al., Nature 424, 788 (2003).
14. S. Ruppert, E. H. Wang, R. Tjian, Nature 362, 175 (1993).
Base Position
Human mRNAs
Conservation
chimp
mouse
rat
chicken
RNA Transcripts / Affymetrix
RNA Transcripts / Yale
RefSeq Genes
MGC Genes
Promoters / Stanford
ChIP-RNAP / Affymetrix
ChIP-RNAP / Ludwig
DNaseI HS / NHGRI
DNaseI HS / Regulome
DNA Rep 0-2 hr / UVa
DNA Rep 2-4 hr / UVa
148450000 148500000 148550000 148600000 148650000 148700000 148750000 148800000 148850000
Fig. 3. UCSC Genome Browser display of representative ENCODE data. The
genomic coordinates for ENCODE target ENr231 on chromosome 1 are
indicated along the top. The different tracks are labeled at the left with the
source of the data. The Conservation track shows a measure of evolutionary
conservation based on a phylogenetic hidden Markov model (phylo-HMM)
(27). Multiz (28) alignments of the human, chimpanzee, mouse, rat, and
chicken assemblies were used to generate the species tracks. RefSeq Genes,
MGC Genes indicate the mapping of mRNA transcripts from RefSeq (17)
and MGC (24) projects, respectively, whereas the track labeled Human
mRNAs represents all mRNAs in GenBank. Other tracks represent verified
data from the ENCODE Consortium (23). The track with the location of
sequences tested for promoter activity in a reporter assay is labeled as
Promoters/Stanford. The positions of transcripts identified by oligonucleo-
tide microarray hybridization (RNA Transcripts/Affymetrix and RNA Tran-
scripts/Yale) and sequences detected by ChIP/chip analysis (ChIP-RNAP/
Ludwig and ChIP-RNAP/Affymetrix) are indicated. The DNA replication
tracks show segments that are detected to replicate during specified
intervals of S phase in synchronized HeLa cells. The 0-2 hours and 2-4 hours
tracks show segments that replicate during the first and second 2 hours
periods of S phase. The DNA fragments released by DNase I cleavage were
identified either from CD4
cells (DNaseI HS/NHGRI) or from K562 cells

(DNaseI HS/Regulome).
S
P
E
C
I
A
L
S
E
C
T
I
O
N
15. K. Shuai, C. Schindler, V. R. Prezioso, J. E. Darnell Jr.,
Science 258, 1808 (1992).
16. R. W. Blakesley et al., Genome Res. 14, 2235 (2004).
17. K. D. Pruitt, T. Tatusova, D. R. Maglott, Nucleic Acids
Res. 31, 34 (2003).
18. International HapMap Consortium, Nat. Rev. Genet.
5, 467 (2004).
19. W. J. Kent et al., Genome Res. 12, 996 (2002).
20. P. T. Spellman et al., Genome Biol 3, RESEARCH0046
(2002).
21. N. D. Trinklein et al., Genome Res. 14, 62 (2004).
22. B. Ren, B. D. Dynlacht, Methods Enzymol. 376, 304
(2004).
23. ENCODE Consortium, unpublished data.
24. Mammalian Gene Collection (MGC) Project Team,
Genome Res.14, 2121 (2004).
25. International HapMap Consortium, Nature 426, 789
(2003).
26. A. Felsenfeld, J. Peterson, J. Schloss, M. Guyer,
Genome Res. 9, 1 (1999).
27. A. Siepel, D. Haussler, in Statistical Methods in
Molecular Evolution, R. Nielsen, Ed. (Springer, New
York, in press).
28. M. Blanchette et al., Genome Res. 14, 708 (2004).
29. The Consortium thanks the ENCODE Scientific Ad-
visory Panel for their helpful advice on the project:
G. Weinstock, G. Churchill, M. Eisen, S. Elgin, S. Elledge,
J. Rine, and M. Vidal. We thank D. Leja, and M.
Cichanowski for their work in creating figures for this
paper. Supported by the National Human Genome
Research Institute, the National Library of Medicine,
the Wellcome Trust, and the Howard Hughes Medical
Institute.
Supporting Online Material
www.sciencemag.org/cgi/content/full/306/5696/636/
DC2
Tables S1 to S3
V I E W P O I N T
Systems Biology and New Technologies Enable
Predictive and Preventative Medicine
Leroy Hood,
1
* James R. Heath,
2,3
Michael E. Phelps,
3
Biaoyang Lin
1
Systems approaches to disease are grounded in the idea that disease-perturbed
protein and gene regulatory networks differ from their normal counterparts; we have
been pursuing the possibility that these differences may be reflected by multi-
parameter measurements of the blood. Such concepts are transforming current
diagnostic and therapeutic approaches to medicine and, together with new tech-
nologies, will enable a predictive and preventive medicine that will lead to per-
sonalized medicine.
Biological information is divided into the dig-
ital information of the genome and the envi-
ronmental cues that arise outside the genome.
Integration of these types of information leads
to the dynamic execution of instructions as-
sociated with the development of organisms
and their physiological responses to their en-
vironments. The digital information of the
genome is ultimately completely knowable,
implying that biology is unique among the
sciences, in that biologists start their quest
for understanding systems with a knowable
core of information. Systems biology is a sci-
entific discipline that endeavors to quantify
all of the molecular elements of a biological
system to assess their interactions and to in-
tegrate that information into graphical net-
work models (14) that serve as predictive
hypotheses to explain emergent behaviors.
The genome encodes two major types of
information: (i) genes whose proteins exe-
cute the functions of life and (ii) cis control
elements. Proteins may function alone, in
complexes, or in networks that arise from
protein interactions or from proteins that
are interconnected functionally through small
molecules (such as signal transduction or
metabolic networks). The cis control ele-
ments, together with transcription factors,
regulate the levels of expression of individual
genes. They also form the linkages and ar-
chitectures of the gene regulatory networks
that integrate dynamically changing inputs
from signal transduction pathways and pro-
vide dynamically changing outputs to the
batteries of genes mediating physiological
and developmental responses (5, 6). The
hypothesis that is beginning to revolutionize
medicine is that disease may perturb the nor-
mal network structures of a system through
genetic perturbations and/or by pathological
environmental cues, such as infectious agents
or chemical carcinogens.
Systems Approaches to Model Systems
and Implications for Disease
A model of a metabolic process (galactose
utilization) in yeast was developed from exist-
ing literature data to formulate a network hy-
pothesis that was tested and refined through a
series of genetic knockouts and environmental
perturbations (7). Messenger RNA (mRNA)
concentrations were monitored for all 6000
genes in the genome, and these data were
integrated with protein/protein and protein/
DNA interaction data from the literature by a
graphical network program (Fig. 1).
The model provided new insights into the
control of a metabolic process and its in-
teractions with other cellular processes. It
also suggested several concepts for systems
approaches to human disease. Each genet-
ic knockout strain had a distinct pattern of
perturbed gene expression, with hundreds
of mRNAs changing per knockout. About
15% of the perturbed mRNAs potentially en-
coded secreted proteins (8). If gene expres-
sion in diseased tissues also reveals patterns
characteristic of pathologic, genetic, or envi-
ronmental changes that are, in turn, reflected
in the pattern of secreted proteins in the
blood, then perhaps blood could serve as a
diagnostic window for disease analysis. Fur-
thermore, protein and gene regulatory net-
works dynamically changed upon exposure
of yeast to an environmental perturbation (9).
The dynamic progression of disease should
similarly be reflected in temporal change(s)
from the normal state to the various stages
of disease-perturbed networks.
Systems Approaches to Prostate
Cancer
Cancer arises from multiple spontaneous
and/or inherited mutations functioning in
networks that control central cellular events
(1012). It is becoming clear from our re-
search that the evolving states of prostate
cancer are reflected in dynamically changing
expression patterns of the genes and proteins
within the diseased cells.
A first step toward constructing a systems
biology network model is to build a com-
prehensive expressed-mRNA database on
the cell type of interest. We have used a
technology called multiple parallel signature
sequencing (MPSS) (13) to sequence a com-
plementary DNA (cDNA) library at a rate of
a million sequences in a single run and to
detect mRNA transcripts down to one or a
few copies per cell. A database containing
more than 20 million mRNA signatures was
constructed for normal prostate tissues and
an androgen-sensitive prostate cancer cell
line, LNCaP, in four states: androgen-starved,
1
Institute for Systems Biology, Seattle, WA, USA.
2
Department of Chemistry, California Institute of
Technology, Pasadena, CA, USA.
3
Department of
Molecular and Medical Pharmacology, The David
Geffen School of Medicine at the University of
California Los Angeles, Los Angeles, CA, USA.
*To whom correspondence should be addressed.
E-mail: lhood@systemsbiology.org
S
P
E
C
I
A
L
S
E
C
T
I
O
N
androgen-stimulated, normal conditions, and
an androgen-insensitive variant. In com-
paring the androgen-sensitive (typical of
early-stage cancer) and androgen-insensitive
(typical of late-stage cancer) stages (14, 15),
thousands of changes in mRNA expression
were identified but, out of 554 expressed
transcription factors, 112 changed between
the early- and late-stage cell lines (80% of
which were missed when
cDNA arrays were used),
and a si mi l ar number
changed between the can-
cerous cells and normal
tissue. By comparing the
prostate database with a
tissue-wide database of 58
million MPSS signatures
from 29 normal tissues from
Lynx Therapeutics, about
300 prostate-specific genes
(Fig. 2) were identified, ap-
proximately 60 of which
possessed signal peptides,
suggesting that they may
be secreted (8). Antibodies
to one of these proteins rec-
ognized, by blood analy-
ses, 5 out of 10 early and
5 out of 10 late prostate
cancers (16). In contrast,
the standard prostate cancer
blood marker, PSA, recog-
nized no early cancers but
many of the late prostate
cancers, including all of
those missed by our marker.
Thus, two markers are better
than one, and by extension a
panel of multiple markers
might recognize most early
and late prostate cancers.
Several groups have
documented the fact that
(unidentified) molecules in
blood serum, detected by
mass spectrometry, reflect
various stages of cancer
(1720). Aebersolds group
has succeeded in identifying
many of these biomarkers
through the use of a glyco-
protein capture method, cou-
pled with isotopic labeling
and analyses by mass spec-
trometry (21, 22). Molecular
diagnostics will increasingly play a key role in
providing direct measures of disease biology for
selecting and following therapeutic responses.
Given enough measurements, one can
presumably identify distinct patterns for
each of the distinct types of a particular
cancer, the various stages in the progression
of each disease type, the partition of the
disease into categories defined by critical
therapeutic targets, and the measurement of how
drugs alter the disease patterns. The fascinating
question is how many parameters need to be
measured in order to stratify and follow the
progression of various prostate cancers, or
to stratify and follow the progression of the
most frequent 20 or 30 cancers, or eventu-
ally the most common diseases. Finally,
changes in the tissue-specific markers might
identify critical points within the network. It
is the key nodal points within these per-
turbed networks that may be affected by
drugs, either to convert the diseased net-
work back toward normalcy or to permit the
specific killing of the diseased cells. Thus,
multiparameter blood measurements will not
only be invaluable for diagnostics but also for
rationalizing the discovery of appropriate
drug targets. In this scenario, molecular
diagnostics will become an invaluable tool
for molecular therapeutics.
Toward Analyses of Single Cells and
Single Molecules
The systems biology approach toward con-
structing a predictive network model of a
metabolic process in yeast required 10
5
measurements. For the
prostate cancer example,
roughly 10
8
measure-
ments were sufficient to
begin constructing a large
set of cancer markers that
could be correlated back
to the digital code of the
genome. However, for
constructing a predictive
model of human disease,
methods that can address
the heterogeneity that
characterizes biology
from the differences in
how individual cells re-
spond to environmental
perturbations, to the di-
versity of cell types and
environments within real
tissueswill be critical.
In the prostate, there
are neuroepithelial cells,
various stromal cells, en-
dothelial cells, and epi-
thelial cells (from which
95% of cancers arise),
each of which exhibits a
continuous developmen-
tal cycle. One cannot reli-
ably generate information
for networks from mixed
populations of cells. Var-
ious investigators have
used cell sorting (23),
manual dissection (24),
or laser capture micro-
dissection (LCM) (25) to
obtain relatively homo-
geneous populations of
cells. However, cell sort-
ing and LCM themselves
may cause processing-
induced changes in gene
expression (26, 27), and
manual microdissection
rarely provides complete-
ly homogeneous cell types. Furthermore,
even cells of one type typically represent dif-
ferent stages of a developmental or phys-
iological process. Biologists would like to
analyze individual cells for the key mea-
surements of systems biology, so that net-
work hypotheses could be generated from
individual cells. The mRNAs from single
cells have been analyzed after polymerase
rProtein
synth
Glycolytic
enzymes
Galactose
utilization
Amino acid,
nt synth
RNA
processing
O
2
stress
Sugar
transport
Glycogen
metabolism
Fatty acid
oxidation
Vesicular
transport
Respiration
Mating,
cell cycle
Gluconeo
-genesis
Metal
uptake
O
2
stress
Stress
Fig. 1. A network perturbation model of galactose utilization in yeast. This model
reflects the integration of mRNA levels for the 6000 yeast genes in each of 20
different genetic and environmental perturbations, as well as thousands of protein/
protein and protein/DNA interactions from the literature. The software program
Cytoscape (54) integrated these data into a network where the nodes represent
proteins (encoded by genes) and the lines represent interactions (blue straight lines,
protein/protein interactions; yellow lines with arrows, protein/DNA interactions). A
gray scale represents the levels of mRNA, with black being abundant levels and white
very low levels. The red node indicates that this network model reflects the knockout
of the corresponding gene (and protein) gal 4a key transcription factor. rProtein,
ribosomal protein; nt, nucleotide; synth, synthesis.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
chain reaction (PCR) amplification, but
there is no similar amplification technique
for proteins. Thus, techniques are needed
that are highly parallel, allow for multiple
types of measurements (genes and proteins)
and operations (such as cell sorting) to be
integrated, are miniaturized (to analyze sin-
gle cells and single molecules), and are
automated. Here we highlight just a few of
the technologies that are being driven by the
needs of systems biology.
Microfluidics has existed as a useful bio-
technology for some time (2830). How-
ever, multilayer
elastomer micro-
fluidics (Fig. 3) is
a powerful new
technology that al-
lows for the in-
tegration of many
pumps, valves, and
channels within an
easily fabricated
microchip. This
means that multiple
operations, such as
cell sorting (31,
32), DNA purifica-
tion, and single-cell
gene expression
profiling (33), can
be executed in par-
allel. This tech-
nology provides a
bri dge bet ween
biological mate-
rials and systems
biology through
large-scale multi-
parameter analysis,
with applications
ranging from mo-
lecular dissections
of single cells (for
example, from needle biopsies) and very small
cell populations to multiparameter disease di-
agnostics from cells and blood.
Nanomechanical (34) and nanoelectronic
(35, 36) devices are emerging as highly sen-
sitive, label-free, and real-time detectors of
genes, mRNAs, and proteins. To date, demon-
strations of these nanotechnologies have been
at the single- or few-device level, but the
reported detection sensitivities and dynamic
ranges (37, 38) have been spectacular. Nano-
fabrication methods for constructing large
libraries of these devices (3943) and inte-
grating nanotechnologies with elastomer mi-
crofluidics (44) are moving forward. It is
likely that within the next couple of years,
miniaturized and automated microfluidics/
nanotech platforms that integrate operations
such as cell sorting and serum purification
with measurements of 5 to 10 biomarkers
from single cells or very small fluid volumes
will emerge. New measurement types, such
as quantifying the forces associated with
protein/protein, protein/DNA, and protein/
drug interactions, are possible. Other emerg-
ing nanotechnologies include tools for the
500
450
400
350
300
250
200
150
100
50
0
A
d
r
e
n
a
l
G
la
n
d
B
la
d
d
e
r
B
o
n
e

M
a
r
r
o
w
B
r
a
in
:
A
m
y
g
d
a
la
B
r
a
in
:
C
a
u
d
a
t
e
N
u
c
le
u
s
B
r
a
in
:
C
e
r
e
b
e
llu
m
B
r
a
in
:
C
o
r
p
u
s
C
a
llo
s
u
m
B
r
a
in
:
F
e
t
a
l
B
r
a
in
:
H
y
p
o
t
h
a
la
m
u
s
B
r
a
in
:
T
h
a
la
m
u
s
H
e
a
r
t
K
id
n
e
y
L
u
n
g
M
a
m
m
a
r
y

G
la
n
d
P
a
n
c
r
e
a
s
P
it
u
it
a
r
y

G
la
n
d
P
la
c
e
n
t
a
R
e
t
in
a
S
a
liv
a
r
y

G
la
n
d
S
m
a
ll
I
n
t
e
s
t
in
e
S
p
in
a
l
C
o
r
d
S
p
le
e
n
S
t
o
m
a
c
h
T
e
s
t
is
T
h
y
m
u
s
T
h
y
r
o
id
T
r
a
c
h
e
a
U
t
e
r
u
s
P
r
o
s
t
a
t
e
Fig. 2. A prostate-specific marker identified through quantitative
profiling of all mRNAs across all 29 major organs in the human body.
The gene HOXB13 is expressed at 432 transcripts per million in the
prostate tissue but is not expressed in the other 28 normal tissues. This
method has been used to identify approximately 50 potential serum-
based protein biomarkers for prostate cancer.
Fig. 3. Microfluidic and nanotechnology platforms. (A) An integrated microfluidics environment for single-cell gene expression studies.
A single cell is introduced (i) into a 100-6m-wide channel. Before the cell is introduced, an affinity column (beads covered with
oligo dT) is loaded [dark regions in (ii)]. The orange-colored regions in (ii) are valves that separate, for example, the empty
chamber at the right from the region in which the column is being constructed. Three such valves constitute a peristaltic pump
(not shown). Data from a real-time PCR analysis of the isolated mRNA (iii) illustrate the power of this integrated microfluidics
approach. Lanes 3 and 4 correspond to one and nine cells, respectively, whereas the other lanes correspond to various controls
[adapted from (33)]. (B) Array of nanomechanical biomolecular sensors. The cantilevers are fabricated to be only a few nanometers
thick, with a molecular probe (such as single-stranded DNA) bonded to their top surface. DNA hybridization leads to steric
crowding that forces the cantilever to bend. The bending can be detected optically or electronically [adapted from (34)]. (C) An
electron micrograph showing a library of 16-nm-wide silicon nanowire biomolecular sensors. The scale bar is 200 nm, and the
structures on top of the nanowires are electrical contacts. Nanowire sensors operate by binding molecular probes (such as
antibodies) to the surface of a semiconducting nanowire. When the target protein binds to the probe, the conductivity
properties of the nanowire are altered, and so the binding event is electronically detected. Both nanocantilevers and nanowires
are capable of real-time biomolecular detection [adapted from (55)].
S
P
E
C
I
A
L
S
E
C
T
I
O
N
rapid sequence analysis of individual DNA
molecules (45) and even nanoparticle-based
in vivo cancer imaging probes (46).
These various technologies will be har-
nessed to generate preliminary network hy-
potheses for analyzing human diseases within
the next few years. Those hypotheses must
ultimately be tested in vivo. Such testing
typically means molecular imaging, which
encompasses methods ranging from bio-
luminescence and fluorescence (4750) to
positron emission tomography (PET) (4952)
and magnetic resonance imaging (MRI) (48).
The challenge is to reduce the large numbers
of elements delineated in the network ana-
lyses to one of a few targets of molecular
imaging biomarkers that can provide critical
tests of the network. For example, specific
metabolic enzymes that are selectively ex-
pressed in prostate cancer cells would con-
stitute such a target. We searched the genes
that were differentially expressed between
early- and late-stage prostate cancer cell
lines (15) and determined that L-lactate de-
hydrogenase A, which catalyzes the forma-
tion of pyruvate from (S)-lactase, was only
expressed, and at a high level, in the late-
stage cancer cells. A specific PET tracer based
on this reaction would serve to validate this
finding and might also allow the identi-
fication of prostate cancer metastases. Mo-
lecular imaging is already being aligned with
molecular therapeutics in the use of labeled
drug candidates to provide direct measure-
ments in patients by imaging pharmacoki-
netics of the drug throughout the body,
titration of drugs to their disease targets,
and measuring therapeutic effects on the
biological processes of disease (4953).
The Future
The medicine of today is reactive, with a focus
on developing therapies for preexisting dis-
eases, typically late in their progression. Over
the next 10 to 20 years, medicine will move
toward predictive and preventive modes. New
technologies will allow individuals to have the
relevant portions of their genomes sequenced,
and multiparameter informative molecular
diagnostics via blood analysis will become a
routine procedure for assessing health and dis-
ease status. During this period, there will also
be extensive correlations of genetic variations
with disease, and this combination of advances
will allow for the determination of a probabi-
listic future health history for each individual.
Preventive medicine will follow as disease-
perturbed networks can be used to identify
drug targetsfirst for therapy and later for
prevention. Pharmacological intervention will
focus on preventing disease-mediated transi-
tions, as well as reversing or terminating
those that have occurred. This will require
building a fundamental understanding of the
systems biology that underlies normal bio-
logical and pathological processes, and the
development of new technologies that will be
required to achieve this goal.
Predictive and preventative medicine will
lead naturally to a personalized medicine that
will revolutionize health care. Drug companies
will have the opportunity for more effective
means of drug discovery guided by molecu-
lar diagnostics, although the paradigm will
shift to partitioning patients with a particu-
lar disease into a series of therapeutic win-
dows, each with smaller patient populations
but higher therapeutic effectiveness. Health
care providers will move from dealing with
disease to also promoting wellness (preven-
tion). Finally, the public must be educated as
to their roles in a very different type of med-
icine, as must the physicians who practice it.
There will be enormous scientific and en-
gineering challenges to achieve this vision
far greater than those associated with the
Human Genome Project. Predictive, preven-
tive, and personalized medicine will transform
science, industry, education, and society in
ways that we are only beginning to imagine.
1. E. H. Davidson et al., Science 295, 1669 (2002).
2. E. H. Davidson, D. R. McClay, L. Hood, Proc. Natl.
Acad. Sci. U.S.A. 100, 1475 (2003).
3. H. Kitano, Science 295, 1662 (2002).
4. U. Alon, Science 301, 1866 (2003).
5. E. V. Rothenberg, E. H. Davidson, in Innate Immunity,
R. A. B. Ezekowitz, J. A. Hoffman, Eds. (Humana,
Totowa, NJ, 2003), pp. 6188.
6. D. T. Odom et al., Science 303, 1378 (2004).
7. T. Ideker et al., Science 292, 929 (2001).
8. The prediction of classical secretory proteins was
based on the existence of signal peptides and the
number and position of transmembrane domains in
a protein. The prediction of nonclassical secretory
proteins (proteins without an N-terminal signal
peptide) was based on the algorithm developed
by J. D. Bendtsen et al. (http://www.cbs.dtu.dk/
services/SecretomeP-1.0/).
9. A. D. Weston, N. S. Baliga, R. Bonneau, L. Hood, Cold
Spring Harbor Symp. Quant. Biol. 68, 345 (2003).
10. C. A. Klein, Cell Cycle 3, 29 (2004).
11. P. A. Covitz, Pharmacogenomics J. 3, 257 (2003).
12. D. Hanahan, R. A. Weinberg, Cell 100, 57 (2000).
13. S. Brenner et al., Nature Biotechnol. 18, 630 (2000).
14. B. Lin et al., in preparation.
15. B. Lin et al., in preparation.
16. A. G. Utleg et al., in preparation.
17. E. F. Petricoin 3rd et al., J. Natl. Cancer Inst. 94, 1576
(2002).
18. B. L. Adam et al., Cancer Res. 62, 3609 (2002).
19. E. P. Diamandis, Clin. Chem. 49, 1272 (2003).
20. Y. Qu et al., Clin. Chem. 48, 1835 (2002).
21. H. Zhang, X. J. Li, D. B. Martin, R. Aebersold, Nature
Biotechnol. 21, 660 (2003).
22. Using this method, Aebersolds group was able to
identify 100 peptides in sera that were strongly
associated with cancer status in a mouse model for
chemically induced skin cancer.
23. J. G. Church, E. A. Stapleton, B. D. Reilly, Cytometry
14, 271 (1993).
24. L. Whetsell, G. Maw, N. Nadon, D. P. Ringer, F. V.
Schaefer, Oncogene 7, 2355 (1992).
25. N. L. Simone, R. F. Bonner, J. W. Gillespie, M. R. Emmert-
Buck, L. A. Liotta, Trends Genet. 14, 272 (1998).
26. A. Y. Liu et al., Proc. Natl. Acad. Sci. U.S.A. 94,
10705 (1997).
27. R. A. Craven, R. E. Banks, Proteomics 1, 1200 (2001).
28. A. G. Hadd, D. E. Raymond, J. W. Halliwell, S. C.
Jacobson, J. M. Ramsey, Anal. Chem. 69, 3407 (1997).
29. I. Karube, K. Ikebukuro, Y. Murakami, K. Yokoyama,
Ann. N.Y. Acad. Sci. 750, 101 (1995).
30. L. C. Waters et al., Anal. Chem. 70, 158 (1998).
31. R. H. Carlson, C. V. Gabel, S. S. Chan, R. H. Austin, J. P.
Brody, Phys. Rev. Lett. 79, 2149 (1997).
32. A. Y. Fu, H. P. Chou, C. Spence, F. H. Arnold, S. R.
Quake, Anal. Chem. 74, 2451 (2002).
33. J . W. Hong , V. Studer, G. Hang, W. F. Anderson, S.
R. Quake, Nature Biotechnol. 22, 435 (2004).
34. J. Fritz et al., Science 288, 316 (2000).
35. Y. Cui, Q. Wei, H. Park, C. M. Lieber, Science 293,
1289 (2001).
36. R. J. Chen et al., Proc. Natl. Acad. Sci. U.S.A. 100,
4984 (2003).
37. G. Wu et al., Nature Biotechnol. 19, 856 (2001).
38. J . Hahm, C. M. Lieber, Nano Lett. 4, 51 (2004).
39. R. McKendry et al., Proc. Natl. Acad. Sci. U.S.A. 99,
9783 (2002).
40. Y. Arntz et al., Nanotechnology 14, 86 (2003).
41. N. A. Melosh et al., Science 300, 112 (2003).
42. Y. Bunimovich et al., Langmuir, in press.
43. K. B. Lee, S. J. Park, C. A. Mirkin, J. C. Smith, M. Mrksich,
Science 295, 1702 (2002).
44. R. Pantoja et al., Biosensors and Bioelectronics, in
press; published online 6 May 2004; doi.10.1016/
j.bios.2004.02.020 (http://dx.doi.org/10.1016/
j.bios.2004.02.020).
45. I. Braslavsky, B. Hebert, E. Kartalov, S. R. Quake, Proc.
Natl. Acad. Sci. U.S.A. 100, 3960 (2003).
46. S. A. Wickline, G. M. Lanza, Circulation 107, 1092
(2003).
47. G. Choy et al., Biotechniques 35, 1022 (2003).
48. R. Weissleder, U. Mahmood, Radiology 219, 316
(2001).
49. H. R. Herschman, Science 302, 605 (2003).
50. T. F. Massoud, S. S. Gambhir, Genes Dev. 17, 545
(2003).
51. M. E. Phelps, Proc. Natl. Acad. Sci. U.S.A. 97, 9226
(2000).
52. M. E. Phelps, PET: Molecular Imaging and Biological
Applications (Springer, New York, 2004).
53. K. Shah, A. Jacobs, X. O. Breakefield, R. Weissleder,
Gene Ther. 11, 1175 (2004).
54. P. Shannon et al., Genome Res. 13, 2498 (2003).
55. E. Johnston-Halperin et al., J. Appl. Phys., in press.
56. We thank all the members of the NanoSystems Biol-
ogy Alliance for their insights, expertise, and encour-
agement. We acknowledge support from the National
Cancer Institute, the Army Research Office, and the
Department of Energy.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
V I E W P O I N T
Gene Order and Dynamic Domains
Steven T. Kosak
1
and Mark Groudine
1,2
*
When considering the daunting complexity of eukaryotic genomes, some comfort
can be found in the fact that the human genome may contain only 30,000 to 40,000
genes. Moreover, growing evidence suggests that genomes may be organized in such
a way as to take advantage of space. A genes location in the linear DNA sequence
and its position in the three-dimensional nucleus can both be important in its
regulation. Contrary to prevailing notions in this postgenomic era, the bacteriophage
1, a paragon of simplicity, may still have a few things to teach us with respect to
these facets of nonrandom genomes.
Nearly 40 years have passed since Jacob and
Monod received the Nobel Prize for their
contributions to elucidating the transcription-
al regulation of the bacteriophage 4. The
paradigm of gene regulation offered by 4
stresses the importance of gene proximity (in
the form of operons), gene order, and
competitive DNA binding by regulatory
factors, largely dictated by their effective
concentrations (1). These strategies used by
the phage during the switch between lysog-
eny and lysis may seem to have little bearing
in this postgenomic age. Yet the basic
principles of phage 4 regulation appear to
be taking shape at the level of the coordi-
nated regulation of transcriptomesgene
networks dedicated to eukaryotic cell fates
such as death, division, and differentiation.
Emerging evidence indicates that eukaryotic
genes may be organized in the form of
genomic domains that allow for their coex-
pression through mechanisms both similar to
and different from 4.
Tandem Gene Arrays
Size has always seemed to matter for
molecular biologists attempting to under-
stand the coordinated regulation of eukary-
otic genes. Given the vast difference in
relative length, eukaryotic genomes have
long been thought to operate under a set of
principles different from lower organisms,
such as phage 4. However, analysis of the
growing list of sequenced eukaryotic ge-
nomes is beginning to yield commonalities
in gene regulation regardless of genome size
(2). Even before the availability of whole
genome sequences, experimental evidence
suggested that the first principle of 4
regulation, gene proximity, plays an impor-
tant role in the regulation of tandem gene
arrays, colinear series of individual genes
often formed through duplication events.
Tandem gene arrays are common features of
eukaryotic genomes, and they are both
ubiquitously expressed [e.g., small nuclear
RNA, histones, ribosomal DNA (rDNA),
major histocompatibility complex (MHC),
etc.] and tissue specific [e.g., Hox, immu-
noglobulin (Ig), T cell receptor (TCR), $-
globin, cadherin, olfactory, etc.]. The size of
tandem gene arrays variesfor example, the
mouse Ig heavy chain (IgH) contains 3 Mb,
whereas $-globin contains 120 kb (4 itself
is 48 kb)but the prevalence of these loci
suggests that proximity in the linear gene
order is important in coordinate gene regula-
tion. The actual role that linear proximity
plays in transcriptional regulation is still
being established. The rDNA arrays may
represent one potential mechanism in that a
shared linear position facilitates the nucleolar
localization of the relevant ribosomal genes
(3). Additionally, the Ig and $-globin loci, as
shown by fluorescence in situ hybridization
(FISH), are specifically positioned in the
nucleus during cellular differentiation, and
their nuclear localization has been correlated
with their state of activity (4, 5). The clus-
tering of genes in arrays may therefore allow
the simultaneous localization of genes to
compartments of the nucleus important in
their regulation.
Tandem gene arrays also demonstrate the
second principle of 4 regulation: the activa-
tion of genes according to their linear order
in the genome. With phage 4, the position of
the cl gene (the repressor) to the left of the
operator and of the cro gene (Cro) to the
right are requisite features of the genetic
switch that determines the choice between
lysogenic and lytic cell fates (1). Similarly,
the $-globin, Hox, and Ig gene arrays dem-
onstrate a sequential and developmentally
regulated activation of genes according to
their order. Not unlike phage 4, the sequen-
tial activation of these tandem gene arrays
may be due to the proximity of the early
genes to the dominant regulatory element of
a given loci. The $-globin locus undergoes
sequential activation of globin genes during
embryogenesis, with those genes closest to
the locus control region (LCR) being up-
regulated first (6). With the Ig heavy chain
locus (IgH), the genes first to undergo germ-
line transcription are those closest to the
intronic and 3 enhancers (7). For the Ig and
TCR gene arrays, this type of consecutive
activation is correlated with the sequential
events of V-D-J recombination. The verte-
brate Hox gene clusters, however, uniquely
illustrate this phenomenon with linear gene
order. In addition to the chronological acti-
vation of vertebrate Hox genes, the proteins
themselves demonstrate a corresponding ac-
tivity along the anteroposterior axis (8).
These genes are therefore uniquely linked in
their order, activation, and function.
The third property of 4 regulation is the
competitive binding of operator sequences
by either the repressor or Cro (1). The
cooperative binding of the dimerized repres-
sor protein ensures transcription of the cl
gene and continuation of the lysogenic cell
fate. In the event of external influences (such
as irradiation), the effective concentration of
the repressor is diminished, allowing for the
up-regulation of cro and the subsequent
activation of 4 early genes. Although they
have not been elucidated in such regulatory
detail, tandem gene arrays also demonstrate
sensitivity to both protein availability and
concentration, and in some cases are in-
volved in an autoregulatory feedback mech-
anism like phage 4. For example, the Hox
gene clusters encode homeodomain tran-
scription factors critical in decisions of cell
fate during metazoan development, and in
vertebrates some of these same proteins also
contribute to the regulation of the clusters
themselves (8).
With the $-globin locus, an exchange in
heterodimerization partners of a small Maf
protein, MafK, leads to the switch from
repression of the locus in committed but
undifferentiated erythroid cells to activation
of transcription in differentiated cells (9).
Before differentiation, a heterodimer com-
posed of MafK and the repressor Bach1
recruits transcriptional corepressor com-
plexes to the locus, resulting in repression
of globin gene expression. Upon induction
of erythroid differentiation, an exchange
of MafK-binding partners occurs: Bach 1
is replaced by the transcriptional activator
p45. This, in turn, leads to displacement of
corepressor complexes from the locus and
recruitment of coactivators, resulting in glo-
bin gene expression. The mechanism behind
1
Division of Basic Sciences, Fred Hutchinson Cancer
Research Center, 1100 Fairview Avenue North,
Seattle, WA 98109, USA.
2
Department of Radiation
Oncology, University of Washington School of
Medicine, Seattle, WA 98195, USA.
E-mail: markg@fhcrc.org
S
P
E
C
I
A
L
S
E
C
T
I
O
N
this exchange involves the relocation of
MafK in the nucleus (10). Before the
induction of differentiation, MafK colocal-
izes with centromeric heterochromatin,
whereas p45 is restricted to euchromatic
nuclear compartments. Terminal differentia-
tion is accompanied by the relocation of
MafK (and the $-globin locus) to euchro-
matic regions and formation of the MafK/
p45 heterodimer (10). Differentiation is also
associated with a relative increase in p45 and
a decrease in Bach1 (11). Thus, similar to
phage 4, the availability and concentration of
transcription factors play a key role in ef-
fecting a switch between repressed and acti-
vated states.
Gene Clusters
The availability of genomic sequences has
now allowed researchers to determine wheth-
er, beyond tandem gene arrays, eukaryotes
possess clusters of coexpressed genes that
are related in cellular function or fate. That
is, do genomes show a level of organization
in which genes that are part of a tran-
scriptome are also proximal in the linear
genome? Given the examples of tandem gene
arrays, it is perhaps not surprising that all
eukaryotic genomes examined to date display
a tendency for the organization of coexpressed
genes, often related in function, to be linearly
clustered (2). Therefore, as opposed to
displaying a homogenous distribution of
genes along chromosomes, genomes ap-
pear to be specifically organized for gene
regulation.
The nonrandom nature of gene distribu-
tion along chromosomes is pronounced in
vertebrate genomes. By integrating the hu-
man genomic sequence with SAGE (serial
analysis of gene expression) data for genome-
wide mRNA expression patterns from 12 tis-
sue types, a human transcriptome map has
been determined (12). The linear distribution
of expression along chromosomes depicted
in this map reveals that the human genome is
organized into regions of high and low levels
of gene activity. The highly active regions,
RIDGEs (regions of increased gene expres-
sion), are separated by large regions of low
activity, anti-RIDGEs.
RIDGEs and anti-RIDGEs coincide with
gene-dense and gene-poor chromosomal
domains, respectively. Therefore, gene ac-
tivity is inherently compartmentalized along
the chromosome, which may indicate a
higher order genomic structure. A more
recent analysis of SAGE data suggests that
RIDGEs form because they may be com-
posed of ubiquitously and highly expressed
housekeeping genes (13). Nonetheless, it is
interesting to consider these highly ex-
pressed regions in light of the synteny be-
tween the genomes of human and mouse.
The syntenic blocks of DNA, which consti-
tute 90% of both genomes, vary in length,
but half are 20 Mb, similar to the size of
RIDGEs (14). It may be that evolution deals
with a currency much larger than a single
gene; evolutionary forces may impinge on
genomic domains because of the importance
of proximity in gene regulation.
Beyond the organization at the level of
RIDGEs, genomic approaches have uncov-
ered clusters of coregulated genes encom-
passing 2 to 30 genes. In the budding yeast
Saccharomyces cerevisiae, chromosome cor-
relation maps, which plot the expression
patterns from various conditions or cell
stages along the linear gene order of the
chromosomes, reveal that genes from tran-
scriptomes dedicated to cell cycle phases,
sporulation, and the pheromone response are
found in pairs throughout the genome (15).
These pairs of genes are often functionally
related as well. Similarly, in Caenorhabditis
elegans, mRNA-tagging and microarray
approaches have uncovered clusters of two
to five genes that are coregulated in specific
cell lineages, including muscle, sperm,
oocytes, and the germ line (16). An analysis
of gene expression in Drosophila, using
microarray data determined under 80 differ-
ent experimental conditions, has revealed an
organization of coexpressed genes in groups
of 10 to 30, covering 20 to 200 kb (17). The
clustering of genes within the Drosophila
genome has also been demonstrated by
expressed sequence tag (EST) database
analyses of tissue-specific expression from
the testis, head region, and embryo. In each
cell type, coregulated genes are organized
into clusters of three or more genes, with a
trend toward large groupings (18). In hu-
mans, several metabolic pathways are also
shown to have their protein components
encoded in genes that are proximal in the
genome (19). Therefore, the clustering of
coregulated, lineage-restricted genes indi-
cates a functional organization of transcrip-
tomes that define a given cell type.
The regulatory mechanisms that govern
the expression of the tandem gene arrays
described above offer an attractive model for
the tissue-specific coregulation of gene
clusters. Like phage 4, the gene arrays use
shared regulatory elements and fluctuating
regulatory protein concentrations to ensure
their proper expression patterns. In large part
these tools are successful because of the
proximity of the tandem genes. Gene clusters
that are involved in the differentiation or
maintenance of a cell type may therefore be
selected for the same regulatory purposes. In
S. cerevisiae, for example, several of the
tandem genes observed in the chromosome
correlation maps share a common upstream
activating sequence (15). Whether all of the
gene clusters identified have a shared en-
hancer seems unlikely. However, a universal
property that proximity bestows is an in-
creased local concentration of the pools of
regulatory proteins that coexpressed genes
likely have in common. Specifically, because
of the number of binding sites for regulatory
proteins in colinear genes, gene proximity
may ensure an effective concentration of
these proteins, creating a localized transcrip-
tional center or expression hub (2). This
idea is supported by examples of definitive
nuclear subcompartments with increased
localized protein concentrations that a given
gene cluster may associate with (or, indeed,
may help to form) to direct its regulation.
The nucleolus represents a paradigm of this
phenomenon, being essentially a concentra-
tion of proteins that guarantees ribosomal
biogenesis as well as regulation of other
genes required for protein translation (3, 20).
On the basis of the evidence presented
above, it is attractive to consider lineage-
specific transcription not as the coregulation
of hundreds of disparate genes, but rather as
the coordinated regulation of various gene
domains. Beyond the examples of tandem
gene arrays and gene clusters, genomic
approaches have yielded evidence for anoth-
er feature of domain-based genomes. Analy-
sis of the histone modifications associated
with active and inactive genes indicates that
related modifications are shared by neigh-
boring genes. For example, in Drosophila, a
very tight correlation between gene expres-
sion and five different euchromatic histone
modifications was observed (21). The pat-
tern of euchromatic histone modification is
binary, with active genes showing all
such marks and inactive genes lacking such
marks. These modifications are restricted to
the transcribed region, and the degree of the
euchromatic histone marks correlates with
transcript abundance, suggesting a process
of chromatin modification that is intrinsic-
ally coupled to transcription. Regardless of
the basis of these marks, however, genes
with these histone modifications exist in
neighborhoods of two to three, similar to
coexpressed gene clusters of the same size.
A genomic analysis of repressive markers
such as heterochromatin protein 1 (HP1) has
revealed a similar association with sets of
coregulated genes (22). Therefore, the orga-
nization of the genome into expression hubs
enriched (or depleted) in such modifications
and chromatin-binding proteins may facilitate
the formation of nuclear subcompartments.
Dynamic Domains
The nucleus is a dynamic organelle, embod-
ied in its great Houdini act of mitosis, in
which we see the nucleus duplicate its chro-
mosomal content, disappear, and then reap-
pear in two places at once. In parallel, the
protein content of the interphase nucleus is
in constant flux (23). Fluorescence recovery
S
P
E
C
I
A
L
S
E
C
T
I
O
N
after photobleaching (FRAP) studies have
shown that regulatory proteins (such as
transcription factors) have rapid diffusion
rates, as do structural components of the
nucleus. HP1, for example, colocalizes with
constitutive heterochromatin and had long
been thought to statically associate at these
structures during interphase. However, FRAP
analysis indicates that most of the HP1 is in
fact highly diffusible (24).
This dynamic nature of the nucleus seems
to belie the presence of nuclear bodies (such as
Cajal bodies, interchromatin granule clusters,
PML bodies, etc.). But these bodies are
themselves dynamic
(their integral com-
ponents are constantly
exchanged), and their
structure in the nucle-
us is inseparable from
their function: If you
disturb the structure
of a body you abro-
gate its function, and
vice versa (2). Many
genes, such as the $-
globin and Ig loci, are
also localized to spe-
cific nuclear compart-
ments according to
their state of transcrip-
tional activity. More-
over, recent studies
suggest that loci have
the ability to mobi-
lize to specific lo-
cations within the
interphase nucleus
(25). Although the ex-
tent to which loci can
sample the nuclear
environment appears
to differ among orga-
nisms, evidence indi-
cates that genes have
the capacity to move
beyond their segrega-
tion during mitosis.
Dur i ng i nt er -
phase, chromosomes
are maintained as dis-
crete structures termed
chromosome territo-
ries (CTs). The majority of evidence indicates
that genes are preferentially positioned at the
territory surface, including contours and in-
vaginations, of their respective chromosomes
(2). Tandem gene arrays that are highly tran-
scribed have been shown to loop away from
their territories, as identified with whole-
chromosome paints in FISH analysis. Territo-
ry looping may indicate an association of the
looped gene array with the nuclear bodies
that are involved in transcription and are
found in the interchromatin compartment that
runs among the CTs. The MHC, for example,
has recently been shown to localize near PML
bodies and is looped from its CT when being
transcribed (26, 27). For the vertebrate Hox
cluster, territory looping has also been cor-
related with the sequential activation of par-
ticular genes (28). Studies of wild-type and
mutant $-globin loci have helped to clarify
the role of transcriptional activity in territory
looping (29). In erythroid cells, the $-globin
locus is looped away from its CT even be-
fore transcriptional induction. However, in
the absence of the LCR, which is required
for high-level globin gene transcription, the
locus is positioned at the CT surface. Further-
more, if the $-globin LCR is replaced by
elements from a B cellspecific LCR that
represses transcription of reporters in ery-
throid cells, looping is partially restored but
is now correlated with localization of the
looped locus to centromeric heterochromatin
of another CT. These results indicate that ter-
ritory looping is not simply a consequence
of activity, but may play an important role
in cell typespecific transcriptional activa-
tion or repression of a locus.
Although the basis for looping of a locus
from its chromosome territory has yet to be
established, the $-globin locus may offer
some general clues. The $-globin LCR con-
tains several binding sites for small Maf
proteins, and, as described above, the MafK/
p45 heterodimer is essential for relocation and
activation of the locus during differentiation.
Interestingly, addition of the transactivation
domain of p45 is sufficient to relocate MafK
from heterochromatic to euchromatic nuclear
compartments (10). Because the p45 activa-
tion domain interacts with coactivator com-
plexes that also bind other erythroid-specific
genes, it is tempting to
speculate that such
complexes comprise
expression hubs to
which erythroid-specific
genes may relocate (via
mass action, for exam-
ple). A similar scenario
is possible for repres-
sion of the locus by
MafK/Bach1-recruited
corepressor complexes
before erythroid differ-
entiation and in nonery-
throid cells. Moreover,
the on-off rates of
chromatin-associated
proteins, including the
small Maf heterodi-
mers, are likely influ-
enced by the pattern of
histone modifications
(30).
The observations
highlighted above have
led many to suggest
that nuclear organiza-
tion may be cell type
specific: The topo-
logical organization of
the interphase nucleus
may be involved in
the establishment and
propagation of cell-
specific patterns of
gene expression. Dy-
namic domains com-
prising both tandem
gene arrays and gene
clusters may represent the means by which
a nucleus is configured. Given that a tran-
scriptome includes hundreds of genes, it may
be expedient for transcriptional regulation to
have these domains compartmentalized to-
gether within the nucleus. Coregulated gene
clusters (including the two alleles of the same
cluster) may therefore be proximally posi-
tioned to facilitate their regulation through the
creation of expression hubs with shared con-
centrations of regulatory proteins (Fig. 1). The
combined effect of the association of domains
Interphase SKY
Theoretical magnification
Chromosome
territory
Inactive gene
Active gene
Centromeric
heterochromatin
Nuclear body
Fig. 1. What is the underlying nuclear organization of gene activity? At left is a spectral karyotype
(SKY) image of an interphase nucleus from a hematopoietic progenitor. SKY allows for the
simultaneous visualization of all chromosomes and is primarily applied to metaphase chromosomes
to detect abnormalities. Growing evidence supports the idea that a nuclear topology exists that
ensures the transcriptional program(or transcriptome) that gives rise to or maintains a given cell type.
Therefore, the SKY image may represent the appropriate nuclear topology for the progenitors
transcriptome. However, the actual chromosomal organization of an entire genome has yet to be
established. Nonetheless, tandem gene arrays and the linear clustering of genes that are coexpressed
and dedicated to a particular cell fate argue that spatial proximity is important in coordinate gene
regulation. Given the evidence for chromatin mobility and the looping of gene loci fromchromosome
territories, it is an attractive possibility to consider that coregulated gene clusters from different
genomic regions may also be proximal in the nuclear volume, as depicted in the theoretical
magnification at right. The regulation of these localized gene clusters may take advantage of protein
concentrations (or may be the basis for them), as exemplified by the various types of nuclear bodies
found in the nucleus.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
from throughout the genome would yield a
cell-specific nuclear topology, resulting in de-
finable patterns of chromosome organization.
Recent evidence has indicated that the
chromosomal organization within a nucleus
is maintained upon cell division. FRAP
analysis with histonefluorescent protein
fusions has revealed that chromosomes
appear to remain in their relative nuclear
positions (31). It is therefore possible the
nucleus of a particular cell type does have its
genome specifically organized for the ex-
pression of its relevant transcriptome. The
spectral karyotype (SKY, which allows the
simultaneous detection of all chromosomes)
(32) of an interphase nucleus in Fig. 1 would
therefore represent the topology that ensures
the overall regulation of that cell type.
Evidence supports the idea that chromo-
somes have specific positions within the
nucleus. For example, gene-dense and gene-
poor chromosomes have been shown to pre-
ferentially localize to the nuclear center and
periphery, respectively (33). Additionally, a
recent study focusing on a subset of chro-
mosomes has shown that lineage-specific
associations of certain chromosomes can
occur (34). What remains to be established,
however, is the organization of an entire
genome at the level of the chromosome.
Furthermore, whether a given nuclear topol-
ogy changes upon cell differentiation has yet
to be demonstrated.
With the idea of cell-specific nuclear
topology, we have clearly moved beyond
the realm in which bacteriophage 4 can lead
by example. The dynamic regulation of
hundreds to thousands of genes requires a
level of coordination unnecessary for a
simple phage. Still, it is important to remem-
ber that the functions of the nucleus, such as
transcription, are intertwined with its struc-
ture. If the principles of eukaryotic gene
regulation find basic parallels with a less
complicated example, then there is hope that
these principles, aided by the merging of tech-
nologies such as SKY and 3D microscopy,
may allow us to fully appreciate the dynamic
organization of a genome within its nucleus.
Note added in proof: A recent study has
provided evidence for the colocalization of
coregulated genes on the same chromosome
(35).
1. M. Ptashne, A Genetic Switch (Cold Spring Harbor
Laboratory Press, Cold Spring Harbor, NY, ed. 3,
2004).
2. S. T. Kosak, M. Groudine, Genes Dev. 18, 1371 (2004).
3. A. K. Leung, A. I. Lamond, Crit. Rev. Eukaryot. Gene
Expr. 13, 39 (2003).
4. S. T. Kosak et al., Science 296, 158 (2002).
5. C. Francastel, M. C. Walters, M. Groudine, D. I.
Martin, Cell 99, 259 (1999).
6. M. Bulger, M. Groudine, Genes Dev. 13, 2465 (1999).
7. D. Chowdhury, R. Sen, Immunol. Rev. 200, 182 (2004).
8. R. S. Mann, Bioessays 19, 661 (1997).
9. M. Brand et al., Nature Struct. Mol. Biol. 11, 73 (2004).
10. C. Francastel, W. Magis, M. Groudine, Proc. Natl.
Acad. Sci. U.S.A. 98, 12120 (2001).
11. M. Brand, M. Groudine, unpublished data.
12. H. Caron et al., Science 291, 1289 (2001).
13. M. J. Lercher, A. O. Urrutia, L. D. Hurst, Nature Genet.
31, 180 (2002).
14. R. H. Waterston et al., Nature 420, 520 (2002).
15. B. A. Cohen, R. D. Mitra, J. D. Hughes, G. M. Church,
Nature Genet. 26, 183 (2000).
16. P. J. Roy, J. M. Stuart, J. Lund, S. K. Kim, Nature 418,
975 (2002).
17. P. T. Spellman, G. M. Rubin, J. Biol. 1, 5 (2002).
18. A. M. Boutanaev, A. I. Kalmykova, Y. Y. Shevelyov,
D. I. Nurminsky, Nature 420, 666 (2002).
19. L. D. Hurst, C. Pal, M. J. Lercher, Nature Rev. Genet. 5,
299 (2004).
20. M. Thompson, R. A. Haeusler, P. D. Good, D. R. Engelke,
Science 302, 1399 (2003).
21. D. Schubeler et al., Genes Dev. 18, 1263 (2004).
22. F. Greil et al., Genes Dev. 17, 2825 (2003).
23. T. Misteli, Science 291, 843 (2001).
24. T. Cheutin et al., Science 299, 721 (2003).
25. A. Belmont, Curr. Opin. Cell Biol. 15, 304 (2003).
26. E. V. Volpi et al., J. Cell Sci. 113, 1565 (2000).
27. J. Wang et al., J. Cell Biol. 164, 515 (2004).
28. S. Chambeyron, W. A. Bickmore, Genes Dev. 18,
1119 (2004).
29. T. Ragoczy, A. Telling, T. Sawado, M. Groudine, S. T.
Kosak, Chromosome Res. 11, 513 (2003).
30. G. Felsenfeld, M. Groudine, Nature 421, 448 (2004).
31. D. Gerlich et al., Cell 112, 751 (2003).
32. E. Schrock et al., Science 273, 494 (1996).
33. S. Boyle et al., Hum. Mol. Genet. 10, 211 (2001).
34. L. A. Parada, P. G. McQueen, T. Misteli, Genome Biol.
5, R44 (2004).
35. C. S. Osborne et al., Nature Genet. 36, 1065 (2004).
R E V I E W
Cis-Acting Regulatory Variation in the
Human Genome
Tomi Pastinen and Thomas J. Hudson*
The systematic screening of the human genome for genetic variants that affect gene
regulation should advance our fundamental understanding of phenotypic diversity
and lead to the identification of alleles that modify disease risk. There are several
challenges in localizing regulatory polymorphisms, including the wide spectrum of
cis-acting regulatory mechanisms, the inconsistent effects of regulatory variants in
different tissues, and the difficulty in isolating the causal variants that are in linkage
disequilibrium with many other variants. We discuss the current state of knowledge
and technologies used for mapping and characterizing genetic variation controlling
human gene expression.
Expression profiling and genome-wide map-
ping studies have shown that strong heritable
factors govern differences in gene expression
levels within mammalian species such as the
mouse and human (1, 2). The concentration
of a given mRNA allele is controlled both by
cis-acting factors (such as DNA polymor-
phisms and methylation) in the flanking
DNA sequence of the gene and trans-acting
modulators (such as transcription factors)
that are themselves regulated by other
genetic and environmental characteristics of
the cell. Whereas heritable expression differ-
ences resulting from trans-acting mecha-
nisms appear to be quantitatively more
important, cis-acting variation may explain
up to 25 to 35% of interindividual differ-
ences in gene expression. This is likely an
underestimate, as physiological feedback
mechanisms can mask the impact of subtle
cis-acting variants on expression levels.
Evidence for the medical importance of cis-
acting polymorphisms has been provided by
recent positional cloning of susceptibility
genes that are not associated with protein
coding or splice-site polymorphisms for
common diseases such as stroke and type 2
diabetes (3, 4).
Regulatory polymorphisms are DNA ele-
ments that modify the expression level of a
transcript or its isoforms. Current techniques
can detect expression differences as low as
1.2-fold between samples or alleles; the
phenotypic consequences (if any) of such
small differences are likely to depend on the
function of the gene in question. Most
known regulatory polymorphisms are located
in gene promoter regions and function by
altering gene transcription. Coding poly-
morphisms are also known to affect the
McGill University and Genome Quebec Innovation
Centre, 740 Drive Penfield Avenue, Montreal, Quebec
H3A 1A4, Canada.
E-mail: tom.hudson@mcgill.ca
S
P
E
C
I
A
L
S
E
C
T
I
O
N
expression of alleles, as in the case of the
nonsense-mediated mRNA decay of tran-
scripts harboring early stop codons (5). The
emerging picture of regulatory sequences
distributed over long distances upstream
and downstream of a gene, including introns
as well as the 5 and 3 untranslated regions,
has led to the discovery of regulatory
variants located outside promoter regions
that can alter mRNA stability (6), mRNA
processing efficiency (7), or mRNA isoform
expression (4, 8) or induce epigenetic
changes (9). However, association of allelic
expression with heritable regulatory poly-
morphisms or epigenetic mechanisms may
not be straightforward. For example, differ-
ent mechanisms have been suggested to
underlie allelic expression of the human
TP73 gene in noncancerous cells and tissues
ranging from heritable polymorphisms (10)
to tissue-specific mono-
allelic expression (11).
Perils and Pitfalls in
the Identification of
Cis-Acting Variation
Isolating a true regulato-
ry variant is complicated
by linkage disequilibri-
um (LD) in the human
genome. This is high-
lighted by studies of the
lactase (LCT) persist-
ence phenotype, a com-
mon monogenic trait
caused by cis-acting reg-
ulatory variants. Genetic
studies in the Finnish
population have shown a
perfect correlation be-
tween LCT persistence
and the T allele in a
single-nucleotide poly-
morphism (SNP) approx-
imately 14 kb upstream
of the LCT gene (12). Whereas these
findings were supported by in vitro studies
showing functional differences between the
alleles (13), subsequent studies have identi-
fied individuals who are heterozygous for the
persistence allele but show equal expression
of LCT alleles (14). In addition, the 14-kb
SNP is not associated with LCT persistence in
some non-Caucasian populations (15); this
SNP shows very high LD with other variants
contained in a 1-Mb flanking region (16),
suggesting that it may be in LD with the
causative regulatory variant.
The modulation of gene expression
caused by epigenetic mechanisms can be
misconstrued to be due to regulatory poly-
morphisms. Classically imprinted autosomal
loci display preferential expression of a
single allele (monoallelic expression) that is
independent of sequence variation. This
unequal expression of transcripts, which is
often transmitted according to the alleles
parent of origin, is usually accompanied by
different patterns of cytosine methylation or
posttranslational histone modifications. This
phenomenon has been best studied in mice, in
which nearly 50 imprinted genes localized to
15 genomic regions have been characterized
(17). For some genes, only one allele of a
gene is arbitrarily expressed in each cell: This
process is called random monoallelic expres-
sion and is reminiscent of X-chromosome
inactivation in females (18). Other naturally
occurring epigenetic mechanisms, which do
not follow strictly parent-of-origin or random
patterns, have also been described in mam-
mals (19). Interindividual variability in levels
of allele silencing in imprinted genes can be
observed in normal controls (20), and it has
been suggested that diet may influence DNA
methylation and allelic expression of epige-
netically regulated loci (21). Epigenetic
alterations are common in neoplastic cells,
which may even be detected in peripheral
blood samples as demonstrated in colon
cancer patients showing increased biallelic
expression of insulin-like growth factor 2
(IGF2) in comparison to that of healthy
controls (22).
Detecting Allele-Specific Expression
Allele-specific expression of a transcript can
be detected by in vitro and in vivo methods
that measure the cumulative effects on a
number of cellular processes (Fig. 1).
In vitro approaches. In vitro methods
(most commonly involving transient trans-
fection assays) monitor the transcriptional
activity of a synthetic reporter construct and
are best suited for testing whether a candi-
date regulatory polymorphism affects gene
expression (23). Most current studies target
putative promoter or upstream flanking
regions; these regions are often poorly
characterized and frequently do not represent
the complete promoter that is active in the
cell line of interest. Although experimentally
validated promoters can be found in the
Eukaryotic Promoter Database (www.epd.
isb-sib.ch), these comprise less than 10% of
human genes. The initial choice of allele-
specific constructs for transfection studies can
be refined by deletion experiments to delin-
eate the most important 5 regulatory sequen-
ces. Alternatively, information from long-
range regulatory sequences can be studied to
use constructs containing proximal promoter
regions and enhancer elements (24). More
faithful reproduction of natural gene regula-
tion can be achieved by cloning whole human
genes in bacterial artifi-
cial chromosomes (25).
Such studies are techni-
cally challenging and rel-
atively uncommon; most
published studies have
used relatively small pro-
moter constructs (G1 kb)
or oligonucleotide frag-
ments studied in the con-
text of heterotypic minimal
promoters.
Transient transfection
studies may be compli-
cated by trans-acting influ-
ences on allelic expression.
Studies are often per-
formed in preexisting an-
imal or human lines, but
there are concerns about
whether the observed
data can be extrapolated
to the human tissues of
interest. Even small trans-
acting differences result-
ing from other genetic variants in the host may
be important, as shown by the quantitative
variation of allele-specific responses in
fibroblasts obtained from unrelated individ-
uals (26).
Transient transfection assays were re-
cently applied in a systematic, stringent
survey to study proximal promoter (0.5 kb)
haplotypes from 38 human genes. Significant
allele-specific expression could be repro-
duced in 13 out of 17 (75%) cases when
independent clones were used, suggesting
that 930% of proximal promoters may
harbor cis-acting variants (27).
In vivo approaches. In vivo monitoring of
allelic RNA transcripts (28) is possible in
tissues or cells of individuals heterozygous
for an informative marker within the locus.
There are several advantages to observing
relative expression of the two alleles within
Fig. 1. Cellular phenomena associated with and measured in allele-specific expression
studies. PII, RNA polymerase II.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
the same tissue sample: (i) Alleles are
expressed in their normal environment in-
cluding genomic and chromatin context; (ii)
comparison of alleles is made within rather
than between samples, maximizing the sen-
sitivity of detecting cis-acting effects; (iii)
the developmental and physiologic history of
the tissue is unlikely to be perturbed by the
presence of two low- or two high-expressing
alleles; and (iv) population-based studies
allow sampling of haplotype diversity within
each locus.
The approach has been applied in the
context of rare monogenic diseases to
demonstrate underexpression of the disease
allele (29). Similarly, underexpression of the
wild-type allele may explain variable pene-
trance in dominantly inherited Mendelian
traits (30). Evidence for cis-acting regula-
tory polymorphisms in candidate genes for
complex diseases has also been sought by
allelic expression measurements (8). Demon-
stration of parent-of-origin specific expres-
sion in tissues (or cells) (31) and monoallelic
expression in cells (32) provide commonly
used assays to establish imprinting and ran-
dom monoallelic expression, respectively.
Direct methods of visualizing allelic
expression are challenging, thus measure-
ments are commonly performed with ampli-
fied cDNA [reverse transcription polymerase
chain reaction (RT-PCR)] from tissues or
cell lines of interest and require the pres-
ence of an informative polymorphism in the
processed transcript. Increased informativity
can be achieved with the use of nuclear pre-
mRNA [heteronuclear RNA (hnRNA)] (33).
In our experience, assays performed with
hnRNA have slightly higher variability and
failure rates as compared with those that
use mRNA, likely reflecting the lower con-
centration of hnRNA in total RNA prepa-
rations. Successful hnRNA assays not only
increase the informativity of the allelic ex-
pression measurements but also provide
evidence for transcriptional causes of al-
lelic expression, because RNA processing
differences (such as alternate splicing) can
be excluded. The role of transcription in
causing allele-specific expression can also be
determined by the recently introduced poly-
merase loading assay [haplotype-specific
chromatin immunoprecipitation (Haplo-
ChIP)] (34), which is based on isolating
transcriptionally active DNA fragments by
immunoprecipitation DNA bound to RNA
polymerase II enzyme. The isolated DNA
fragments can be assessed for polymorphisms
in heterozygous samples to determine relative
transcriptional activity of the alleles as a
surrogate for relative allelic expression.
Allelic expression studies also require
quantitative genotyping assays and most
commonly primer extension methods have
been employed for relative allelic quantita-
tion. Imbalanced allelic expression is de-
tected when the heterozygote allele ratio in
RNA (cDNA) differs from the correspond-
ing 1:1 ratio in genomic DNA. The cut-off
for calling allelic imbalance would optimally
be based on the variability of cDNA hetero-
zygote ratios in samples known to express
the alleles in equal proportions; in practice,
such information is not available and thresh-
olds are commonly derived from the analysis
of variability in heterozygote genomic DNA
samples. The potential biases introduced by
the application of genomic standards are far
less serious than the artefactual allelic
imbalance caused by stochastic RT-PCR
amplification of one allele in low copy
number targets. Stochastic effects are a
particular concern in single-cell studies (32).
Recent in vivo screening studies of
relative allelic expression in normal tissues
or cell lines for hundreds of human genes
suggest that 25 to 50% of genes and 5 to
25% of heterozygotes show evidence of
unequally expressed alleles (10, 33, 35).
The abundance of allele-specific differences
in expression is notable, though variable
study designs preclude consensus estimates
of its prevalence in the human genome.
Furthermore, the allelic expression demon-
strated in informative heterozygotes has not
been correlated with total expression levels
across all genotypic groups; some allelic
differences could be compensated if the
expression of the gene were under direct
negative feedback control.
Genetic Mapping Combined with
Expression Profiling
Genome-wide expression profiling technolo-
gy has developed to a level at which even
routine clinical applications have been con-
templated. This, along with improved geno-
typing technologies, would allow large-scale
correlations of marker genotypes to gene
expression levels modeled as quantitative
traits (eQTLs). When genetic linkage of the
eQTL coincides with the genomic location of
the gene, the presence of cis-acting regulatory
variants can be deduced (1, 2). Alternatively,
panels of moderate sample size may have
sufficient power to permit whole-genome
association studies with eQTLs. The draw-
backs of both approaches are that the
detection of subtle cis-acting effects may
require large sample sizes (i.e., thousands of
RNA samples from different individuals) and
that epigenetic variation is not assessed.
Furthermore, without direct allele-specific
expression measurements, the correlation of
a marker genotype with gene expression level
does not guarantee that the effect is cis-
acting; a polymorphism in a trans-acting
regulator in linkage disequilibrium with the
marker genotype can explain the associa-
tion. This may prove to be important even
with high-density mapping, because anti-
sense transcription may be a common regula-
tory mechanism of human gene expression
(36).
Elucidating the Causal Mechanism of
an Allelic Imbalance
Heritable cis-acting effects can be demon-
strated by cosegregation of unequal allelic
expression with marker genotypes in pedi-
grees (10, 29). The lack of Mendelian in-
heritance of an allelic imbalance phenotype
may point toward epigenetic mechanisms
(33). Measurable phenomena associated with
epigenetic allele-specific expression include
replication asynchrony (32), differential meth-
ylation of the genomic loci (37), and allele-
specific posttranslational histone modification
(37). Common heritable allelic imbalance phe-
notypes can be mapped in unrelated individ-
uals to establish regulatory haplotypes (34).
For example, we demonstrated a strong reg-
ulatory haplotype in the human BTN3A2
locus, which spanned at least 15 kb flanking
the gene (33).
A tempting approach is to use existing
bioinformatic tools to identify functional
regulatory variants, but despite advances in
the field (38), these computer predictions
have relatively poor specificity. In vitro
methods may also help find the functional
SNP(s); however, their role is restricted by
the inability of plasmid constructs to mimic
the role of the natural genomic context in
establishing allele-specific expression. Most
of the transient transfection studies to date
have been corroborated with other in vitro
assays. For example, allele-specific DNA-
protein interaction assays have been used to
demonstrate that the putative regulatory
polymorphism shows allele-specific differ-
ences in recruiting nuclear transcriptional
activators or repressors. Similarly, indirect
support for the functionality of the putative
regulatory polymorphism may be sought by
correlating genotypes with altered protein
activity. Finally, the direct observation of
cis-acting effects in vivo, as demonstrated
for the human LTA gene with the use of the
HaploChIP technology, provides corrobora-
tion of transcriptional events mediating
allele-specific gene expression (34, 39).
Allelic Expression: Next Steps
The rapidly evolving data sets, technologies,
and knowledge of regulatory variation por-
tend the generation of genome-wide map-
ping and characterization of allelic variants
affecting gene expression. The large-scale in
vivo screening studies carried out to date are
generally limited to descriptions of allele-
specific differences in expression, leaving the
underlying mechanisms largely unexplored.
A more complete picture will require
genome-wide in vivo allelic expression analy-
S
P
E
C
I
A
L
S
E
C
T
I
O
N
ses followed by a systematic classification
into probable genetic or epigenetic mecha-
nisms with the use of family-based samples.
The genetic basis of allele-specific expression
phenotypes can subsequently be mapped to
regulatory haplotypes. The most limiting
factor for such a study is the lack of suitable
panels of human tissues. In the short term,
existing collections of immortalized cell lines
can provide useful starting material, al-
though these cells may express the genes of
interest at low level or under abnormal
transcriptional control.
Alternatively, genome-wide assessment
of DNA protein interaction for transcription-
ally active genes in vivo combined with
allele quantitation of the protein-bound
genomic fragments (34, 39) could be used
to determine cis-acting polymorphisms un-
derlying interindividual differences in re-
sponse to key transcriptional regulators.
The latter approach may provide a short-cut
to the identification of the causative regula-
tory polymorphism and shift the focus to
cellular processes of interest. The intersec-
tion of multidisciplinary efforts to develop
tools and characterize the regulatory compo-
nent of the human genome [such as the
ENCODE project (40)] with genome-wide
allelic expression studies and regulatory
haplotype characterization will provide a
wealth of data for understanding cis-acting
variation affecting the regulation of human
genes and its contribution to phenotypic
variance.
1. E. E. Schadt et al., Nature 422, 297 (2003).
2. M. Morley et al., Nature 430, 743 (2004).
3. Y. Horikawa et al., Nature Genet. 26, 163 (2000).
4. S. Gretarsdottir et al., Nature Genet. 35, 131 (2003).
5. M. C. Willing, S. P. Deschenes, R. L. Slayton, E. J.
Roberts, Am. J. Hum. Genet. 59, 799 (1996).
6. A. M. Carter, M. Sachchithananthan, S. Stasinopoulos,
F. Maurer, R. L. Medcalf, Thromb. Haemostasis 87,
846 (2002).
7. N. H. Gehring et al., Nature Genet. 28, 389 (2001).
8. H. Ueda et al., Nature 423, 506 (2003).
9. C. Ober et al., Am. J. Hum. Genet. 72, 1425 (2003).
10. H. Yan, W. Yuan, V. E. Velculescu, B. Vogelstein, K.
W. Kinzler, Science 297, 1143 (2002).
11. J. F. Hu et al., Biochim. Biophys. Acta 1491, 49
(2000).
12. N. S. Enattah et al., Nature Genet. 30, 233 (2002).
13. L. C. Olds, E. Sibley, Hum. Mol. Genet. 12, 2333
(2003).
14. M. Poulter et al., Ann. Hum. Genet. 67, 298 (2003).
15. C. A. Mulcare et al., Am. J. Hum. Genet. 74, 1102
(2004).
16. T. Bersaglieri et al., Am. J. Hum. Genet. 74, 1111
(2004).
17. A. Ferguson-Smith, S. P. Lin, C. E. Tsai, N. Youngson,
M. Tevendale, Semin. Cell Dev. Biol. 14, 43 (2003).
18. R. Ohlsson, B. Tycko, C. Sapienza, Trends Genet. 14,
435 (1998).
19. Z. Lin, J. Floros, Physiol. Genomics 11, 235 (2002).
20. T. Sakatani et al., Biochem. Biophys. Res. Commun.
283, 1124 (2001).
21. D. Ingrosso et al., Lancet 361, 1693 (2003).
22. H. Cui et al., Science 299, 1753 (2003).
23. M. V. Rockman, G. A. Wray, Mol. Biol. Evol. 19, 1991
(2002).
24. Y. Xu et al., J. Lipid Res. 40, 50 (1999).
25. R. Wade-Martins, Y. Saeki, E. A. Chiocca, Mol. Ther. 7,
604 (2003).
26. J. L. Rutter et al., Cancer Res. 58, 5321 (1998).
27. B. Hoogendoorn et al., Hum. Mol. Genet. 12, 2249
(2003).
28. J. Singer-Sam, J. M. LeBon, A. Dai, A. D. Riggs, PCR
Methods Appl. 1, 160 (1992).
29. D. Hewett, J. Lynch, A. Child, H. Firth, B. Sykes, Am. J.
Hum. Genet. 55, 447 (1994).
30. S. Hutchinson et al., Hum. Mol. Genet. 12, 2269
(2003).
31. J. Li et al., Proc. Natl. Acad. Sci. U.S.A. 101, 7341
(2004).
32. A. Chess, I. Simon, H. Cedar, R. Axel, Cell 78, 823
(1994).
33. T. Pastinen et al., Physiol. Genomics 16, 184 (2004).
34. J. C. Knight, B. J. Keating, K. A. Rockett, D. P.
Kwiatkowski, Nature Genet. 33, 469 (2003).
35. H. S. Lo et al., Genome Res. 13, 1855 (2003).
36. O. Rosok, M. Sioud, Nature Biotechnol. 22, 104
(2004).
37. R. Jaenisch, A. Bird, Nature Genet. 33 (Suppl), 245
(2003).
38. W. W. Wasserman, A. Sandelin, Nature Rev. Genet.
5, 276 (2004).
39. J. C. Knight, B. J. Keating, D. P. Kwiatkowski, Nature
Genet. 36, 394 (2004).
40. The ENCODE Project Consortium, Science 306, 647
(2004).
41. We thank E. Harmsen, J. Engert, and R. Sladek for
their advice. T.J.H. and T.P. are supported by Genome
Quebec and Genome Canada. T.J.H. is the recipient
of a Clinician-Scientist Award in Translational Re-
search by the Burroughs Wellcome Fund and an
Investigator Award from the Canadian Institutes of
Health Research.
S
P
E
C
I
A
L
S
E
C
T
I
O
N
A Precocial Avian Embryo from
the Lower Cretaceous of China
Zhonghe Zhou* and Fucheng Zhang
An avian embryo was collected fromLower Cre-
taceous lacustrine shale in Liaoning, in northeast
China (fig. S1). It was in the final stage of de-
velopment and represents an early precocial bird.
The embryo is enclosed in an oval 35 mm
by 20 mm. No evidence of eggshell was rec-
ognized. During incubation, bird embryos shift
into a tucking position shortly before hatching
and occupy most of the egg. The tucked posture
of the Liaoning fossil is consistent with a late-
stage embryo rather than with a hatchling, in
which case the head would have raised beyond
the vicinity of the feet.
The skeletal elements are nearly intact in
articulation, with a complete skull and nearly
complete postcranial bones (Fig. 1, A and
B). The skull is large, with toothed jaws and
an expanded braincase. The vertebral col-
umn is strongly curved. Both hindlimbs are
folded and in tight contact with the skull and
other postcranial bones. The long bones have
obvious striations and less ossified epiphy-
ses. The unguals of the pedal digits are large
and curved, indicating that the adult would
be an arboreal bird.
Extant bird embryos have a special struc-
ture on the top of the bill, the egg tooth. This
tooth is used for breaking eggs before hatching
and drops off soon after hatching. Such a
structure is absent in the Liaoning embryo,
although true teeth are still present as in most
other Mesozoic birds. The egg tooth probably
represents a novel feature that arose later.
The embryo has several enantiornithine
apomorphies such as a strutlike coracoid with
a convex lateral margin, a V-shaped furcula,
metacarpal III extending well past metacarpal
II distally, and metatarsal IV being more
slender than metatarsals II or III.
Several avian embryos have been reported
from the Upper Cretaceous, but none with
feathers (1, 2). The Liaoning embryo has a
well-ossified skeleton with well-grown feather
sheets. The jaws lie below the wing as a result
of the rotation of the skull. The distal elements
of the wing are well ossified, suggesting that it
was in the final stage of embryological
development (3, 4). The embryo also preserves
imprints of wing and tail feather anlagen
enclosed in the feather sheets (Fig. 1, B and
C). The presence of long feather sheets with
no open vanes is further evidence that the
fossil is an embryo.
Altricial birds are naked and helpless when
they hatch, whereas precocial birds are covered
with natal down and can move and feed inde-
pendently. Precocial birds hatch with relatively
large brains. The presence of well-developed
feather sheets and the large brain indicates that
the Liaoning embryo is a precocial or even a
superprecocial bird. It is generally believed that
precociality is ancient and altriciality is derived.
All exclusively precocial bird orders have been
phylogenetically arranged in the infraclass
Eoaves (5) and placed at the basal node of
the cladogram (3). The recognition of a Lower
Cretaceous precocial avian embryo supports
this phylogeny. It has been suggested that the
dinosaur Troodon developed precocially (6),
and the Late Cretaceous oviraptorid dinosaurs
may also be precocial (7). Because these are
close relatives of birds, the precociality of birds
might represent a feature derived from their
dinosaurian ancestor.
1. E. Elzanowski, Palaeont. Polon. 42, 148 (1981).
2. H. Schweitzer et al., J. Vert. Paleont. 2, 191 (2002).
3. J. M. Starck, in Current Ornithology, D. M. Power, Ed.
(Plenum, New York, 1993), vol. 10, pp. 275366.
4. R. E. Ricklefs, J. M. Starck, in Avian Growth and
Development, J. M. Starck, R. E. Ricklefs, Eds. (Oxford
Univ. Press, Oxford, 1998), pp. 3158.
5. C. G. Sibley, J. E. Ahlquist, Phylogeny and Classifica-
tion of Birds (Yale Univ. Press, New Haven, CT, 1990).
6. J. R. Horner, D. B. Weishampel, Nature 383, 103 (1996).
7. M. A. Norell, J. M. Clark, L. M. Chiappe, Am. Mus.
Novit. 3315, 1 (2001).
8. We thank X. Wang and D. Miao for help. Supported by
the National Natural Science Foundation of China (grant
no. 40121202) and Special Funds for Major State Basic
Research Projects of China (grant no. TG2000077700).
DC1
Fig. S1
BREVIA
Institute of Vertebrate Paleontology and Paleoan-
thropology, Chinese Academy of Sciences, Post Office
Box 643, Beijing 100044, China.
E-mail: zhonghe@yeah.net
Fig. 1. (A) Photograph of a Lower Cretaceous avian embryo from Liaoning, China (Institute of Vertebrate
Paleontology and Paleoanthropology no. V14238). (B) A line drawing of the fossil shown in (A). Red lines
indicate wing feather sheets, and the green line shows imprints of tail feather anlagen. (C) A close-up view
of long wing feather sheets near the pedal claws. Scale bars, 5 mm. AD, alular digit; C, coracoid; CE, cervical
vertebra; CV, caudal vertebra; D, dentary; F, frontal; FE, femur; FF, wing feather sheets; FU, furcula; H,
humerus; IL, ilium; IS, ischium; J, jugal; L, lachrymal; M, maxilla; M-II, metacarpal II; M-III, metacarpal III; N,
nasal; PD, pedal digits; P, premaxilla; PU, pubis; P-1, first phalanx of the major digit; P-2, second phalanx of
the major digit; R, radius; S; scapula; SC, semilunate carpal; SP, splenial; T, tibia; TF, tail feather sheets; TR,
thoracic rib; U, ulna; 3 and 4, pedal digits III and IV; and I to IV, metatarsals I to IV.
A Gene Expression Map
for the Euchromatic Genome of
Drosophila melanogaster
Viktor Stolc,
1,5
* Zareen Gauhar,
1,2
* Christopher Mason,
2
*
Gabor Halasz,
7
Marinus F. van Batenburg,
7,9
Scott A. Rifkin,
2,3
Sujun Hua,
2
Tine Herreman,
2
Waraporn Tongprasit,
6
Paolo Emilio Barbano,
2,4
Harmen J. Bussemaker,
7,8
Kevin P. White
2,3
.
We used a maskless photolithography method to produce DNA oligonucle-
otide microarrays with unique probe sequences tiled throughout the genome
of Drosophila melanogaster and across predicted splice junctions. RNA ex-
pression of protein coding and nonprotein coding sequences was determined
for each major stage of the life cycle, including adult males and females. We
detected transcriptional activity for 93% of annotated genes and RNA ex-
pression for 41% of the probes in intronic and intergenic sequences. Com-
parison to genome-wide RNA interference data and to gene annotations
revealed distinguishable levels of expression for different classes of genes and
higher levels of expression for genes with essential cellular functions. Differ-
ential splicing was observed in about 40% of predicted genes, and 5440
previously unknown splice forms were detected. Genes within conserved re-
gions of synteny with D. pseudoobscura had highly correlated expression;
these regions ranged in length from 10 to 900 kilobase pairs. The expressed
intergenic and intronic sequences are more likely to be evolutionarily con-
served than nonexpressed ones, and about 15% of them appear to be de-
velopmentally regulated. Our results provide a draft expression map for the
entire nonrepetitive genome, which reveals a much more extensive and di-
verse set of expressed sequences than was previously predicted.
Characterization of the complete expressed set
of RNA sequences is central to the functional
interpretation of each genome. For almost 3
decades, the analysis of the Drosophila ge-
nome has served as an important model for
studying the relationship between gene ex-
pression and development. In recent years,
Drosophila provided the initial demonstration
that DNA microarrays could be used to study
gene expression during development (1), and
subsequent large-scale studies of gene expres-
sion in this and other developmental model
organisms have given new insights into how
organisms implement their developmental
plan (2). Additional studies in Drosophila have
addressed questions in diverse biological pro-
cesses, including metamorphosis, aging, innate
immune response, and sexual dimorphism, and
in evolutionary patterns of gene expression
(39). The cumulative data from such micro-
array studies are valuable for functional anno-
tation of the genome. However, all microarray
studies of gene expression in Drosophila have
been restricted to fragments of genes from
predicted gene sets and thus are subject to
the limitations of gene prediction algorithms.
Large-scale expressed sequence tag (EST)
sequencing has also been extremely valuable
for annotation of the Drosophila genome
(10), but this approach is limited by biases
due to 5 or 3 end DNA sequencing, the
extent to which transcripts are represented in
cDNA libraries, and the number of EST
sequences generated. A nonbiased approach
is necessary for the determination of the entire
catalog of expressed sequences in the genome.
Genomic Btiling[ DNA arrays, which use oli-
gonucleotides or polymerase chain reaction
(PCR) products corresponding to chromo-
somal sequences as probes, have been used to
create transcript activity catalogs for portions
of the human genome and for Arabidopsis
(1113). Microarrays have also recently been used
to characterize the great diversity of RNA tran-
scripts brought about by differential splicing in
human tissues (14). We used both types of
approaches to characterize the Drosophila genome.
Experimental design. To determine the
expressed portion of the Drosophila genome,
we designed high-density oligonucleotide
microarrays with probes for each predicted
exon and probes tiled throughout the pre-
dicted intronic and intergenic regions of the
genome. We used maskless array synthe-
sizer (MAS) technology (15, 16) to synthe-
size custom microarrays containing 179,972
unique 36-nucleotide (nt) probes (17). Of
these, 61,371 exon probes (EPs) assayed
52,888 exons from 13,197 predicted genes,
87,814 nonexon probes (NEPs) assayed ex-
pression from intronic and intergenic regions,
and 30,787 splice junction probes (SJPs)
assayed potential exon junctions for a test
subset of 3955 genes. For the SJPs, we used
36-nt probes spanning each predicted splice
junction, with 18 nt corresponding to each
exon (14). RNA from six developmental
stages during the Drosophila life cycle (early
embryos, late embryos, larvae, pupae, and male
and female adults) was isolated and reverse-
transcribed in the presence of oligodeothy-
midine and random hexamers, and the labeled
cDNA was hybridized to these arrays. The
stages were chosen to maximize the num-
ber of transcripts that would be differentially
expressed between samples on the basis of
previous results (3, 7). Each sample was hy-
bridized four times, twice with Cy5 labeling
and twice with Cy3 labeling (fig. S1).
Genomic and chromosomal expression
patterns. We determined which exon or
nonexon probes correspond to genomic re-
gions that are transcribed at any stage during
development (18). We used a negative
control probe (NCP) distribution (fig. S3) to
score the statistical significance of the EP or
NEP signal intensities for each of the 24
unique combinations of stage, dye, and array,
correcting for probe sequence bias (17, 19).
These results were combined into a single
expression-level estimate (19), a threshold for
which was determined by requiring a false
discovery rate of 5% (20). This threshold
shows 47,419 of 61,371 EPs (77%) and
35,985 out of 87,814 NEPs (41%) were
significantly expressed at some point during
the fly life cycle. Significantly expressed EPs
correspond to 79% (41,559/52,888) of all
exons probed and 93% (12,305/13,197) of all
probed gene annotations. Our results con-
firmed 2426 annotated genes not yet validated
through an EST sequence (Fig. 1A). Out of
10,280 genes represented by EST sequences,
RESEARCH ARTICLE
1
Department of Molecular, Cellular, and Develop-
mental Biology,
2
Department of Genetics,
3
Depart-
ment of Ecol ogy and Evoluti onary Bi ol ogy,
4
Department of Mathematics, Yale University, New
Haven, CT 06520, USA.
5
Genome Research Facility,
NASA Ames Research Center, Mail Stop 239-11,
Moffett Field, CA 94035, USA.
6
Eloret Corporation,
Sunnyvale, CA 94087, USA.
7
Department of Biolog-
ical Sciences,
8
Center for Computational Biology
and Bioinformatics, Columbia University, New York, NY
10027, USA.
9
Bioinformatics Laboratory, Academic
Medical Center, University of Amsterdam, 1100 DE
Amsterdam, Netherlands.
*These authors contributed equally to this work.
E-mail: kevin.white@yale.edu
only 401 (3.0%) were not detected in these
microarray experiments. Our finding that a
large fraction of intergenic and intronic re-
gions (NEPs) is expressed in D. melanogaster
mirrors similar observations for chromosomes
21 and 22 in humans (16) and for Arabidopsis
(14). These results support the conclusion
that extensive expression of intergenic and
intronic sequences occurs in the major evo-
lutionary lineages of animals (deuterostomes
and protostomes) and in plants.
We noted that mRNA expression levels for
protein-encoding genes varied with the protein
function assigned in the Drosophila Gene
Ontology (fig. S2) (21). For example, genes
encoding G protein receptors were expressed
at relatively low levels, whereas genes encod-
ing ribosomal proteins were highly expressed.
A genes expression level was also associated
with cellular compartmentalization and the
biological process it mediates (fig. S2). For
example, genes encoding cytosolic and cyto-
skeletal factors were more highly expressed
than those predicted to function within organ-
elles such as the endoplasmic reticulum, Golgi,
and peroxisome. To determine whether a high
level of gene expression was associated with
essential genetic functions, we examined the
expression levels of genes recently shown to
be required for cell viability (Fig. 1B) in a
genome-wide RNA interference (RNAi) screen
in Drosophila (22). Compared to the rest of the
genome, the genes identified as essential by
RNAi showed a significant increase in expres-
sion during all stages of development (P 0
0.0009, t test), even when the highly expressed
ribosomal protein genes were omitted (P 0
0.0005, t test). This result is also consistent
with the observation that genes with mutant
phenotypes from the 3-Mbase Adh genomic
region are overrepresented in EST libraries
(23). High levels of essential gene expression
may in part reflect widespread expression in
cells throughout the animal, and the relative
RNA expression level may serve as a rough
predictor of essential cellular function.
We also examined changes in gene ex-
pression during the fly life cycle to deter-
mine what fraction of the entire genome is
differentially expressed between develop-
mental stages. Figure 2A shows the expres-
sion signal intensities of transcripts from a
typical 50kilobase pair (kbp) region of the
Drosophila genome during each major de-
velopmental stage. Stage-specific variation
in expression is observed not only for exon
probes, as expected, but also for intergenic
and intronic probes. We used analysis of vari-
ance (ANOVA) (24) to systematically iden-
tify probes as differentially expressed at a
false discovery rate of 5% (16). As expected,
the majority of probes detecting differentially
expressed sequences are also expressed above
background noise level (89% of EPs and 81%
of NEPs) (17) (Table 1). We found 27,176
EPs to be differentially expressed, correspond-
ing to 76% of annotated genes, and even
more when we applied a less conservative
background model (fig. S4). The fact that the
majority of genes are developmentally reg-
ulated is consistent with previous results
obtained with cDNA microarrays (3). In in-
tergenic and intronic regions, we detected dif-
ferential expression for 15% (5508/35,985) of
significantly expressed NEPs, indicating that
many of the putative noncoding expressed
transcripts are also developmentally regulated.
By examining gene expression levels along
chromosomes, we discovered that genes that
have remained linked during evolution are
correlated with one another in expression
levels. We considered the genes inside 827
syntenic blocks identified between D. mela-
nogaster and D. pseudoobscura (25), whose
expression varies substantially during devel-
opment (Fig. 2A). We computed the average
pairwise expression correlation between genes
within each syntenic block and compared with
the correlations among nonsyntenic genes.
Average correlations among genes within
729 (88%) of these syntenic blocks showed a
positive value (Fig. 2B), and genes within 369
syntenic blocks (45%) across all chromosomes
showed positive correlations that were signif-
icant (P G 0.05) (Fig. 2C). Larger blocks show
higher significance of correlated gene expres-
sion levels than smaller ones (Fig. 2D).
Expression correlations for 104 out of 108
syntenic blocks greater than 300 kbp in length
were statistically significant, and all expression
level correlations among genes in blocks
greater than 650 kbp were significant (fig. S5).
These results indicate that chromatin domains
A B
4%
3%
18%
75%
microarray expression; matching EST
microarray expression; no matching EST
not detected; matching EST
not detected; no matching EST
9 10 11 12
Early
Embryo
Late
Embryo
Larva
Pupa
Male
Female
expression level (log 2)
Cell Essential of Unk. Function (104)
Cell Essential w/out Rp (254)
Cell Essential (293)
Total Genome (13,192)
Fig. 1. Characterization of microarray probes by global expression levels can predict biological
function. (A) Annotation confirmation. Each probe was compared to the Drosophila genome
annotation version 3.1 (v3.1) and to cDNA/EST sequences produced by the Berkeley Drosophila
Genome Project. 74.9% of v3.1 genes were confirmed by cDNA/EST sequencing and by this
microarray analysis, 18.4% were confirmed solely by this study, 3.0% were confirmed solely by
cDNA/EST sequencing, and 3.7% were unconfirmed by either method. (B) Cell-essential genes are
expressed at higher-than-average levels. Compared to the average genome-wide, the 293 known
genes identified as essential by Boutros et al. (23) showed significantly higher levels of expression
during all stages of development (P 0 0.0009, t test). A similar result was obtained when the highly
expressed ribosomal protein (Rp) genes are omitted (P 0 0.0005, t test) or when only the 104 cell-
essential genes of unknown function (P 0 0.004, t test) were examined. Error bars indicate T1 SE.
Table 1. Genome-wide statistics for expressed probes. We compiled lists of probes that show significant
expression on the basis of two distinct criteria: (i) absolute probe expression above background (PEAB)
noise level in one or more stages based on comparison with negative control probes and (ii) differential
expression between stages based on ANOVA (16). FDR, false discovery rate.
Probe type Total probes
Significant probes (FDR 0 0.05)
Overlap: ANOVA and PEAB
PEAB ANOVA
NEPs 87,814 35,985 6789 5508 (81%)
EPs 61,371 47,419 27,176 24,062 (89%)
R E S E A R C H A R T I C L E
or long-range enhancers act across wide
genomic regions and that genomic rearrange-
ments in these regions have been constrained
during the evolutionary lineages leading to these
two species, thereby maintaining these blocks of
synteny. Natural selection may therefore act to
maintain genes within neighborhoods where
expression is coordinately regulated.
Exon expression and splicing. Within
each gene, expression levels of exon pairs were
typically highly correlated, and the distribution
of these correlations was significantly shifted
relative to that of exon pairs from different
genes (Fig. 3A). A pattern separation algo-
rithm was used to determine the patterns of
expression for exons within each gene during
development (17). We found three major
trends (Fig. 3, B to D): 53% of genes showed
uniform or highly correlated (r 9 0.8) expres-
sion (Fig. 3B), 46% of genes showed multiple
patterns of exon expression suggesting alter-
native promoter usage or splicing (Fig. 3C),
and 1%of genes showed multiple patterns with
at least one exon pair showing strong anti-
correlation during the life cycle (Fig. 3D).
These strong anticorrelations suggest exclusiv-
ity in the use of one exon or another for this
small subset of genes. Together, this initial
analysis indicated that a vast amount of gene
expression variation is missed in previous mi-
croarray studies that have used cDNAs or that
assay only a subset of exons from each gene.
However, although these initial analyses
allowed identification of exons that are differ-
entially expressed during development, they
did not reveal precisely which exons are
spliced to one another during posttranscrip-
tional RNA processing. To estimate the extent
of splicing in the Drosophila genome during
development, we used SJPs to directly assay
spliced exons and differential use of splice
isoforms. We focused on a subset of 3955
predicted genes that included genes with as few
as 2 and as many as 54 exons. For these genes,
we designed SJPs for all theoretically possible
splicing combinations. Because SJPs could
potentially hybridize to exons that were not
spliced directly together, we used a specialized
set of wrong junction probes (WJPs, splice
junctions formed from exon segments but that
do not match any possible transcript) as
negative controls (fig. S6). These WJPs show
higher hybridization signal than the NCPs we
used as a reference distribution to detect
absolute expression because of partial hybrid-
ization. We find that 28% (8732) of the 30,787
SJPs are expressed at a level above the
background level defined by the WJPs.
By examining the ratio of significantly
expressed, sequentially spliced exons and sig-
nificantly expressed exons that are noncontig-
uous, we were able to determine the proportion
of exons and genes that are alternatively
spliced during development. The ratio of non-
contiguous to contiguous splicing (NC/C ratio)
for exons use of downstream exons holds near
a constant 0.40 (Fig. 4A), indicating an average
of 2.5 contiguous splice events for every exon
skipping (i.e., alternative) splice form. Also,
these SJP data show that 53% (1374 of 2606)
of expressed Drosophila genes from multi-
exon genes exhibit exon skipping (18).
To determine the extent of splicing that is
captured by large-scale EST sequencing, we
used BLASTn to align sequences from the
junctions probes to known EST and cDNA
databases (Fig. 4B). We identified 3292 splice
junctions in the databases that matched our
positive hybridization results, whereas 3464
junction probes did not detect hybridization
(EST only sequences) but did match the
expression databases, providing a very high
false negative rate of 46%. Nevertheless,
5440 new splice variants were identified
3R
(192)
0 100
X
(187)
0 100
3L
(133)
0 100
syntenic block
2R
(190)
0 100
2L
(125)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0 100
e
u
l
a
v
-
p
13
12
13
12
13
12
13
12
13
12
13
12
M
F
L
P
LE
EE
5590k 5600k 5620k 5630k 5640k 5650k 5610k 5660k
gene annotation
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
0 100 200 300 400 500 600 700 800
Syntenic Block Rank
Genes within Syntenic Blocks Genes Randomly Chosen
A
C
D
0
0.05
0.1
0.15
0.2
0.25
0.3
<100 100-
200
200-
300
300-
400
400-
500
500-
600
700-
800
>800
Syntenic Block Size (kb)
M
e
a
n

p
-
v
a
l
u
e
B
M
e
a
n

C
o
r
r
e
l
a
t
i
o
n

C
o
e
f
f
i
c
i
e
n
t
Fig. 2. Gene activity in conserved chromosomal domains. (A) Expression
plots along chromosome 2L in the region: 5,600,000 to 5,650,000 show
signal intensity of probes along the chromosome arm in six developmen-
tal stages [early embryo (EE), late embryo (LE), larva (L), pupa (P), adult
male (M), and adult female (F)]. The shaded region shows a syntenic block
where expression levels of transcripts encoded by genes within the block
are highly correlated across development. (B) Distribution of Pearson
correlation coefficients. Most syntenic blocks show positive correlations
among expression levels for genes within them (blue), and the distribution
of correlations differs from a control set of randomly selected gene blocks
(yellow). Most correlations in the control set are near zero, as expected.
(C) Significance of gene expression correlations within syntenic blocks.
The distributions of P values are shown by chromosome, showing the
significance of the correlations compared to the control set, for each
syntenic block. (D) Significance of gene expression correlations increases
with syntenic block size. The larger the size of syntenic regions, the
stronger the bias for genes within the block to be significantly correlated
in expression. Error bars indicate T1 SE.
(Fig. 4B). Most of these (4564) were alter-
native splice junctions, indicating that EST
sequencing missed the vast majority of al-
ternatively spliced transcripts. Taken together,
the exon-specific expression patterns and the
splice junction expression patterns signifi-
cantly extend the functional annotation of
the predicted genes in the Drosophila genome.
Intergenic and intronic expression.
Lastly, we further examined the patterns of
RNA expression from nonexonic sequences.
There are several reasons to expect that a
significant fraction of the annotated non-
coding genome is expressed. First, previous
studies performed in Drosophila using reverse
Northern methodologies on chromosomal
walks have identified multiple noncoding
RNA transcripts at certain loci, such as in
the bithorax complex (2628). Second, the
algorithms and experimental methods used
for gene prediction and annotation may have
not exhaustively identified the entire gene
complement or all of the correct gene struc-
tures. Third, transcriptional analysis of a
fraction of the human genome (13) and the
Arabidopsis genome (12) has shown that about
50% of the predicted noncoding genome is
expressed. Thus, we might expect this to be
true for other organisms as well, although it
is difficult to predict to what extent.
We examined the 41% of noncoding re-
gions of the genome for which we detected
transcriptional activity above background.
Some of the NEP probes expression may have
been due to previously unannotated genes or to
additional exons of already annotated genes
from the current Drosophila genome annota-
tion (29, 30). To explore this possibility, we
examined whether the expressed NEP probes
corroborated computational exon predictions
by Genscan (31). Exon predictions not in
Flybase annotation (version 3.1) were divided
into two groups: those bordering annotated
genes and unique exons that did not. The
latter set includes potential previously unknown
genes. Genscan unique exons together
encompassed 2045 NEP probes, 1221 (60%)
of which were expressed. This represents a
significant enrichment for expressed probes
(compared with 41% of all NEP probes),
strongly suggesting that many of these predic-
tions represent previously unknown genes.
The overlapping NEP probes give sup-
porting expression data for 1155 of Genscans
predicted exons. Of these, 369 represent
additional exons bordering existing genes,
whereas the remaining 786 exons belong to
716 putative novel genes (18). There were
also several instances where Genscan pre-
dicted a longer upstream or downstream exon
boundary relative to the current annotation.
NEP probes overlapping these regions of
disagreement confirmed expression of 64%
(38/59) of 5 predictions and 81% (30/37) of
3 ones. Lastly, comparing to a recent study
that identified several hundred expressed
exons predicted with the Fgenesh algorithm
(32, 33), we found a considerable overlap of
expressed NEPs within these predicted exons
(61% or 477/777).
We next considered intronic expression
and found that 43% of introns (6717/15,770)
were expressed. There is also a relationship
between the activity of an EP and its nearest
intergenic NEP (and vice versa). We exam-
ined the correlations between these two
classes of probes by means of a G test and
a Dunn-Sidak correction for multiple testing.
The nearest NEP was located on the same or
opposite DNA strand in both the 5 and 3
direction, and for each of the 15 stage com-
parisons the differential expression levels of
the probe pair were recorded. The stage or
sex bias (e.g., transcript enrichment in males
versus females) of NEPs at the 3 end of
exons is highly correlated with the stage or
sex bias of the exon; similarly, the bias of
exons at the 5 end of NEPs is highly cor-
related with the bias of the NEP (fig. S7).
These results indicate that either these ex-
pressed sequences are contained within genes
whose end points have been misannotated or
that they encode noncoding RNAs that are
expressed in concert with nearby genes be-
cause of local chromatin or cis-regulatory
effects. At least some of these transcripts
are likely due to the second hypothesis, be-
cause extensive EST sequencing has failed to
reveal the transcribed sequences within se-
Fig. 3. Exon expression
within and between
genes. (A) Histogram
of exon pairs strongest
correlation within a
gene. There is a strong
bias for exons expres-
sion levels within a gene
(blue) to be positively
correlated, compared
with exons chosen at
random (red). (B) Exon
plot of the ninaE gene
(CG4550). All exons
are expressed simi-
larly within this gene.
Exon 5 is a second
probe placed within
exon 5, because it is
a large exon. To nor-
malize the data from
the different EPs, we
show the absolute ex-
pression values minus
their mean and di-
vided by the sum of
the squares. (C) Exon
plot of gene CG8946,
an example of a gene
with differential ex-
on expression during
development. Exon 1
peaks in larva, where-
as exon 4 peaks in pu-
pa, but all other exons plateau at those same stages at a much lower relative level. (D) Two anticorrelated exons within gene CG1893.
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
0
500
1000
1500
2000
2500
Best Pair, within a Gene
Best Pair, Random Exon
Correlation
A B
C D
E
x
o
n

P
a
i
r
s
CG4550
-0.4
-0.2
0
0.2
0.4
0.6
0.8
EE LE L P M F
Stage
y
t
i
s
n
e
t
n
I

d
e
z
i
l
a
m
r
o
N
exon 5' exon 5
exon 4 exon 3
exon 2 exon 1
CG8946
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
EE LE L P M F
Stage
y
t
i
s
n
e
t
n
I

d
e
z
i
l
a
m
r
o
N
exon 6 exon 5
exon 4 exon 3
exon 2 exon 1
CG1893
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
EE LE L P M F
Stage
y
t
i
s
n
e
t
n
I

d
e
z
i
l
a
m
r
o
N
exon 1
exon 4
quenced exons and because Northern blot
analysis (fig. S8) shows that the noncoding
sequences we have identified are often con-
tained within several major and many minor
polyA-RNA products that are inconsistent
with expected protein-coding gene sizes.
However, our analysis of the Genscan pre-
dictions suggests that many of the expressed
sequences that correlate in expression with
nearby annotated exons do in fact simply
correspond to unannotated exons.
The extent to which these putative non-
coding RNAs are functionally relevant awaits
strategies for systematic characterization, but
this genome-wide scan indicates that they
are both abundant and developmentally reg-
ulated. The function of such extensively
regulated noncoding gene expression dur-
ing development is unknown. To determine
whether these expressed sequences are func-
tionally constrained at the sequence level, we
used an alignment of the genomic sequences
of D. melanogaster and D. pseudoobscura
(25). Sequences corresponding to expressed
NEPs were indeed more likely to be con-
served than those corresponding to nonex-
pressed NEPs: The fraction of aligned base
pairs was 68.7% for expressed probes and
59.0% for nonexpressed probes (Fig. 5) (P 0
10
j192
, t test), whereas 63.7% (expressed) and
60.9% (nonexpressed) of the base pairs in fully
aligned probes were conserved between these
species (34) (P 0 10
j44
, t test). This small but
highly significant difference in conservation
between expressed and nonexpressed NEPs
indicates that there are evolutionary constraints
on the expressed noncoding portion of the
genome. The functional relevance of these
sequences could potentially be further tested
through methods such as large-scale RNAi
screening (22) and systematic mutational
analysis or through comparative expression
analysis with additional species of Drosophila.
Summary. Ideally, to create a finished
and fully annotated expression map of the
genome, each stage and tissue would be as-
sayed by multiple methods. Confidence mea-
sures of expression levels can have their
basis in negative controls and cross-checking
between data sets, such as we have presented
here. It is clear that our past understanding
of genome-wide RNA transcription has been
very limited, because a large proportion of
exons show dynamic patterns of differential
splicing and noncoding activity is ubiquitous.
Taken together, our results also indicate that
there are thousands of uncharacterized and
unannotated transcripts expressed in a devel-
opmentally coordinated manner. Systematic
genetic approaches will likely be required to
determine the functions of the large class
of newly identified noncoding expressed se-
quences, which are slightly more conserved
than other noncoding sequence. Additionally,
the existence of evolutionarily conserved chro-
mosomal domains of correlated gene expres-
sion indicates that these domains are also
functionally important. However, the mecha-
nisms responsible for these expression
domains remain to be elucidated. This draft
expression map of the Drosophila genome
shows that there is considerably more com-
plexity in gene and transcript regulation than
was previously known, and it represents
an initial step in identifying all the functional
elements that ultimately control the develop-
mental program of this organism.
1. K. P. White, S. A. Rifkin, P. Hurban, D. S. Hogness,
Science 286, 2179 (1999).
2. V. Reinke, K. P. White, Annu. Rev. Genom. Hum.
Genet. 3, 153 (2002).
3. M. N. Arbeitman et al., Science 297, 2270 (2002).
4. E. De Gregorio, P. T. Spellman, G. M. Rubin, B. Lemaitre,
Proc. Natl. Acad. Sci. U.S.A. 98, 12590 (2001).
5. S. A. Rifkin, J. Kim, K. P. White, Nat. Genet. 33, 138
(2003).
6. T. R. Li, K. P. White, Dev. Cell 5, 59 (2003).
7. J. De Vos et al., Oncogene 21, 6848 (2002).
8. M. Parisi et al., Science 299, 697 (2003); published
online 2 January 2003 (10.1126/science.1079190).
9. S. D. Pletcher et al., Curr. Biol. 12, 712 (2002).
10. G. M. Rubin et al., Science 287, 2222 (2000).
11. J. L. Rinn et al., Genes Dev. 17, 529 (2003).
12. K. Yamada et al., Science 302, 842 (2003).
13. P. Kapranov et al., Science 296, 916 (2002).
14. J. M. Johnson et al., Science 302, 2141 (2003).
15. E. F. Nuwaysir et al., Genome Res. 12, 1749 (2002).
16. S. Singh-Gasson et al., Nature Biotechnol. 17, 974
(1999).
17. Materials and methods are available on Science
Online.
18. All data are available online at http://flygenome.
yale.edu. GEO submission GSE 1761.
19. R. A. Fisher, in Statistical Methods for Research Workers,
ed. 11 (Oliver & Boyd, Edinburgh,1950), section 21.1,
pp. 99.
20. Y. Benjamini, Y. Hichberg, J. R. Stat. Soc. 57, 289 (1995).
21. M. Ashburner et al., Nature Genet. 25, 25 (2000).
Fig. 5. NEP conservation between D.
melanogaster and D. pseudoobscura. Ex-
pressed NEPs show increased sequence
conservation when compared to nonex-
pressed NEP sequences (P = 10
192
). We
used the whole-genome BLASTZ align-
ment between D. melanogaster and D.
pseudoobscura of (25) to classify each
base pair within our NEP probes as
aligned (i.e., belonging to a chromo-
somal region with homology to D.
pseudoobscura) or unaligned.
80
70
60
50
40
30
20
10
0
Fig. 4. Alternative and
contiguous splicing pat-
terns of the genome.
(A) Ratios of noncon-
tiguous to contiguous
splicing. For each exon,
the percentage of
positives from non-
contiguous or contigu-
ous SJPs was calculated.
The ratio of the per-
centage of contiguous
SJP positives to the
percentage of non-
contiguous SJP posi-
tives remains near 0.4,
even in larger genes
(with six exons). Dotted line shows a linear regression across the data
(y 0 0.0039x 0.3904). (B) Exon splicing activity above background
(16) (fig. S6). Many of the expressed SJPs matched sequences in
Drosophila EST databases (27%), yet a similar fraction of SJPs found
within EST databases did not pass our intensity threshold criterion
(28%). A substantial fraction of previously unknown splice junctions
were detected (45%) as well as noncontiguous (i.e., alternative) exon
splicing (84%).
A
B
27%
28%
45%
EST & Expression
EST Only
Expression Only
16%
84%
Contiguous
Non-Contiguous
22. M. Boutros et al., Science 303, 832 (2004).
23. M. Ashburner et al., Genetics 153, 179 (1999).
24. M. K. Kerr, G. A. Churchill, Proc. Natl. Acad. Sci. U.S.A.
98, 8961 (2001).
25. S. Richards et al., in preparation.
26. W. Bender, B. Weiffenbach, F. Karch, M. Peifer, Cold
Spring Harbor Symp. Quant. Biol. 50, 173 (1985).
27. W. Bender, D. P. Fitzgerald, Development 129, 4923
(2002).
28. H. D. Lipshitz, D. A. Peattie, D. S. Hogness, Genes Dev.
1, 307 (1987).
29. M. D. Adams et al., Science 287, 2185 (2000).
30. S. Misra et al., Genome Biol. 3, RESEARCH0083
(2002).
31. C. Burge, S. Karlin, J. Mol. Biol. 268, 78 (1997).
32. A. A. Salamov, V. V. Solovyev, Genome Res. 10, 516
(2000).
33. M. Hild et al., Genome Biol. 5, R3 (2003).
34. G. Halasz et al., unpublished data.
35. V.S. was supported by NASA Center for Nanotechnol-
ogy, NASA Fundamental Biology Program, and NASA
Biomolecular Systems Research Program (contract
NAS2-99092). H.J.B. was partly supported by grant
P20LM007276-01 from NIH. M.v.B. was supported by
grant BMI 050.50.201 from the Netherlands Organiza-
tion for Scientific Research (NWO). G.H. was supported
by NIH training grant T32GM008224-17. C.M. was
supported by an NSF Graduate Student Fellowship. This
work was supported by grants to K.P.W. and H.J.B. from
the Human Frontier Science Program and to K.P.W.
from the Yale Genomics and Proteomics Initiative, the
NIH National Human Genome Research Institute, and
the W. M. Keck Foundation. Thanks to S. Schaeffer for
providing syntenic block coordinates and F. Aoki for
help with the probe design. The GSE accession number
is GSE 1761.
DC1
Materials and Methods
SOM Text
fig. S1
9 June 2004; accepted 20 September 2004
Quantum-to-Classical Transition
with Single-PhotonAdded
Coherent States of Light
Alessandro Zavatta, Silvia Viciani, Marco Bellini*
Single-photonadded coherent states are the result of the most elementary
amplification process of classical light fields by a single quantum of
excitation. Being intermediate between a single-photon Fock state (fully
quantum-mechanical) and a coherent (classical) one, these states offer the
opportunity to closely follow the smooth transition between the particle-like
and the wavelike behavior of light. We report the experimental generation of
single-photonadded coherent states and their complete characterization by
quantum tomography. Besides visualizing the evolution of the quantum-
to-classical transition, these states allow one to witness the gradual change
from the spontaneous to the stimulated regimes of light emission.
A coherent state k" is the closest analog to a
classical light field and exhibits a Poisson
photon number distribution with an average
photon number k"k
2
. Coherent states have
relatively well-defined amplitude and phase,
with minimal fluctuations permitted by the
Heisenberg uncertainty principle. On the
contrary, a Fock state kn is strictly quantum-
mechanical and contains a precisely defined
number (n) of quanta of field excitation,
hence its phase is completely undefined.
Photon-added coherent states (1) are the
result of successive elementary one-photon
excitations of a classical coherent field, and
they occupy an intermediate position between
the Fock and the coherent states. They are
obtained by repeated (m times) application of
the photon creation operator aa
.
on a coherent
state (k",m 0 k
",m
aa
.m
k", k
",m
being a
normalization constant and m an integer) and
reduce to the limit Fock or coherent states
for " Y 0 or m Y 0, respectively. Quite dif-
ferently from the so-called displaced Fock
states, where a coherent state is used to dis-
place a number state Efor example, by mixing
the two fields upon a highly reflecting beam
splitter (2)^, photon-added coherent states
can be roughly viewed as obtained from the
displacement of a coherent state operated by
a Fock state. Indeed, one easily finds that all
the kn terms with n G m are missing in the
expansion of the states k",m in the Fock basis,
and that all the elements of the corresponding
density matrix are essentially displaced toward
higher indices D
i,j
Y D
im, jm
, leaving all the
elements with i, j G m void.
In the case of a single quantum of field
excitation (m 0 1), the single-photonadded
coherent states (SPACSs) read as
k"; 1 0
aa
.
k"
1k"k
2
p 1
Unlike the operation of photon annihilation,
which maps a coherent state into another
coherent state (that is, a classical field into
another classical field), a single-photon excita-
tion of a coherent state changes it into
something quite different. In general, the
application of the creation operator a
.
changes
a completely classical coherent state into a
quantum state with a varying degree of
nonclassicality that becomes more evident
the smaller the initial amplitude of the k"
state. In the extreme case of an initial vacuum
state k0, a single excitation event transforms
it into the very nonclassical single-photon
Fock state k1, which exhibits negative values
of the Wigner function (3, 4). The Wigner
function is a quasi-probability distribution
(57), which fully describes the state of a
quantum system in phase space (either the
position-momentum space for an harmonic
oscillator or, equivalently, the space spanned
by two orthogonal quadratures of the electro-
magnetic field for a single-mode state of light,
as in this case) in the same fashion as a
probability distribution (nonnegative by defi-
nition) characterizes a classical system. The
negativity of the Wigner function is indeed a
good indication of the highly nonclassical
character of the state (Fig. 1).
We report the experimental generation of
SPACSs and their complete tomographic
analysis, which unveils the nonclassical fea-
tures associated with the excitation of a clas-
sical coherent field by a single light quantum.
Parametric down-conversion in a nonlinear
crystal is the basis for the production of the
desired states (Fig. 2). Here one high-energy
pump photon can annihilate into two photons
that obey the global energy and momentum
conservation laws and thus have lower
energies and are normally emitted into sym-
metrically oriented directions, also called the
signal and idler modes. When no other field
is injected in the crystal, spontaneous para-
metric down-conversion takes place, starting
from the input vacuum field, and pairs of
entangled photons with random (but mutual-
ly correlated) phases are produced. In order
to generate SPACSs, one has to inject a seed
Istituto Nazionale di Ottica Applicata, Largo Enrico
Fermi, 6, I-50125, Florence, Italy; European Labora-
tory for Nonlinear Spectroscopy (LENS) and Depart-
ment of Physics, University of Florence, I-50019
Sesto Fiorentino, Florence, Italy.
E-mail: bellini@ino.it
REPORTS
coherent field k" into the signal mode of the
parametric amplifier, and the conditional
preparation of the target state takes place
every time that a single photon is detected in
the correlated idler mode. If the parametric
gain is kept sufficiently low, which is always
the case in our experimental situation, the
final output state can be approximated as
kB,1g aa
.
s
aa
.
i
k"
s
k0
i
0 k"
s
k0
i
g aa
.
s
k"
s
k1
i
2
(where g is a gain constant with kgk1), and
the output signal mode will contain the orig-
inal coherent state most of the times, except
for the few cases when the state k1
i
is detected
in the idler output mode; these relatively rare
detection events project the signal state into the
SPACS k",1
s
(Fig. 2B), which corresponds to
the stimulated emission of one photon in the
same mode of k". The absence of a seed
coherent field leaves with the usual expression
for the spontaneous process, so that, by studying
the evolution of the quantum state while the
amplitude " gradually increases from zero, one
can actually witness the smooth transition from
the spontaneous to the stimulated regimes of
light emission, with the transformation of an
initial purely quantum state (the single-photon
Fock state) into a classical coherent one. This is
accompanied by the birth of a well-defined
phase and can be described in more visual terms
as the transition from the particlelike to the
wavelike behaviors of the electromagnetic field.
The same state as described by Eq. 2 has
recently been generated (8) to produce arbi-
trary superpositions of zero- and one-photon
states. In that case, however, the conditioning
was performed upon the detection of a single
photon in the same mode k1
s
of the input
coherent state, so that the final state was
completely different fromthe ones investigated
here and, for low " values, of the form ("k0
i

gk1
i
). The injection of a single photon instead
of a coherent state as a seed for conditional
parametric amplification has also been inves-
tigated (9) and experimentally demonstrated
(10, 11) with the amplification to the k2
s
Fock
state in the context of quantum cloning.
The primary light source for the experi-
ment is a mode-locked laser whose pulses
are frequency-doubled to become the pump
for degenerate parametric down-conversion
in a type-I beta-barium borate (BBO) crystal.
In this configuration, the output photons share
the same linear polarization and exactly the
same wavelength, corresponding to twice that
of the pump. In order to nonlocally select a
pure state on the signal channel, idler
photons undergo narrow spatial and frequency
filtering before being detected by a single-
photon counting module (1215). The weak
seed coherent state k" is obtained by con-
trolled attenuation of a small portion of the
laser emission, which is fed into the signal
mode of the nonlinear crystal and is then di-
rected to a 50% beam splitter. Here it is over-
lapped with a second (intense) coherent state
(again obtained from a portion of the origi-
nal laser pulses), which is spatially and tem-
porally matched to the conditionally prepared
SPACS and serves as the local oscillator
(LO) for homodyne measurements, which are
performed with a recently developed high-
frequency time-domain technique (4, 16, 17).
Balanced homodyne detection (1820)
allows the measurement of the electric field
quadratures of an unknown state as a function
of the relative phase K imposed between such a
state and the reference LO. By performing a
series of homodyne measurements on equally
prepared states, it is possible to obtain the
probability distributions p(x,K) of the quadrature
operator xx
K
0
1
2
aa
.
e
iK
aae
jiK

and, given a suf-
ficient number of quadrature distributions for
different values of K, one is able to reconstruct
the density matrix elements and the Wigner
function of the field state under study (7, 17).
Figure 3 shows a sequence of reconstructed
Wigner functions for increasing values of the
seed coherent field amplitude ". The first one
(Fig. 3A), obtained with a blocked input,
corresponds to the single-photon Fock state
(3, 4). Because of the finite efficiency (mea-
sured to be about 59% in the present set of
measurements) in the detection apparatus, a
significant mixture with the vacuum state is
present, which clearly allows classically im-
possible negative values to be reached around
the center of the circularly symmetric (be-
cause of the undefined value of the phase)
Fig. 1. Theoretical
Wigner functions for
some of the quantum
states of light discussed
in the text. Upper sur-
face, SPACS k", 1;
wire-frame surface,
original unexcited co-
herent state k"; lower
surface, single-photon
Fock state k1. The
horizontal plane coor-
dinates represent two
orthogonal quadra-
tures of the field. The
single-photon Wigner
function is centered
at the origin of the
phase space. A value
of k"k
2
0 1 is used.
Fig. 2. Experimental apparatus and conceptual
scheme for the conditional SPACS preparation.
(A) Picosecond-duration pulses at 786 nm and
at a repetition rate of 82 MHz from a mode-
locked Ti:sapphire laser are split by a high-
transmission (HT-BS) and a 50% (BS) beam
splitter to serve as (i) the pump for spontaneous
parametric down-conversion in a 3-mm-thick,
type-I BBO crystal after frequency doubling in
lithium triborate (LBO) crystal; (ii) the seed
coherent field k", after proper attenuation by
variable filters (VF); (iii) the local oscillator
field for balanced homodyne detection (B.H.D.)
after mixing with the investigated states in
another 50% beam splitter (BS-H). F is a com-
bination of spectral and spatial filters consti-
tuted by a pair of etalon interference filters
with a narrow (50 GHz) spectral width, and by
a single-spatialmode optical fiber directly
connected to a single-photon counting module
(SPCM). PZT is a piezoelectric transducer used
to vary the relative phase between the SPACS
and the LO. Additional optics and computer-
controlled optical delay lines to adjust the
synchronization of the different pulses are not
shown here for the sake of clarity. (B) The
conditional preparation of a SPACS takes place
whenever a click is registered on the single-
photon detector placed in the output idler
mode. Each of these detection events triggers the acquisition in the balanced homodyne detector
analyzing the output signal mode (17).
R E P O R T S
distribution. When the coherent seed is initially
switched on at very low intensity (k"k
2
, 0.09;
that is, an average of less than one photon
every 10 pulses), the Wigner function starts to
lose its circular symmetry while moving away
from the origin because of the gradual
appearance of a defined phase, but it still
exhibits an evident nonclassical nature, as
indicated by its partial negativity (Fig. 3B).
For increasing seed amplitudes, negativity is
gradually lost (Fig. 3C) and the ringlike wings
in the distribution start to disappear, making it
more and more similar to the Gaussian typical
of a classical coherent field (Fig. 3, C and D).
Even at relatively high input amplitude, the
Wigner distribution for the SPACS k",1 clearly
shows the effect of the one-photon excitation as
compared to the corresponding, slightly dis-
placed, unexcited k" state (Fig. 3D).
Although the SPACS Wigner function
eventually becomes entirely positive for suffi-
ciently high values of the seed amplitude, the
excitation of an otherwise classical coherent
state by a single photon leaves a measurable
mark of nonclassicality in the field quadrature
statistics (1). In particular, whereas the origi-
nal coherent state has equal fluctuations in the
different quadratures independently from its
amplitude, the one-photonexcited state
exhibits a smaller uncertainty (a squeezing)
in one of the quadratures and larger fluctua-
tions in the orthogonal one. One can
interpret this as a reduction in the intensity
noise associated with the excitation by a
perfectly defined number of quanta, which
also increases the phase noise because of the
lack of phase information intrinsic to Fock
states. Whereas Fig. 4 clearly shows the
amount of maximum quadrature squeezing
for different values of the coherent seed
amplitude, the effect on the intensity and
phase noise is also evident from Figs. 1 and
3C. Here the reduced intensity fluctuations
appear in the decreased width along the radial
direction, whereas the increase in the phase
noise is indicated by the appearance of the
ringlike wings along the tangential direction
of the Wigner distribution.
The ability to experimentally investigate
the elementary action of the bosonic creation
operator on a classical state is of interest both
as a tool to take a closer look at such fundamen-
tal events in quantum physics and as a starting
point for the investigation of the fuzzy border
that separates the quantum and classical
regimes of light behavior, with natural exten-
sions toward even more Bexotic[ quantum
entities, such as Schroedinger_s cat states (21).
1. G. S. Agarwal, K. Tara, Phys. Rev. A 43, 492 (1991).
2. A. I. Lvovsky, S. A. Babichev, Phys. Rev. A 66, 011801
(2002).
3. A. I. Lvovsky et al., Phys. Rev. Lett. 87, 050402 (2001).
4. A. Zavatta, S. Viciani, M. Bellini, Phys. Rev. A, in press
(e-print available at http://arxiv.org/abs/quant-ph/
0406090).
5. L. Mandel, E. Wolf, Optical Coherence and Quantum
Optics (Cambridge Univ. Press, Cambridge, 1995).
6. D. F. Walls, G. J. Milburn, Quantum Optics (Springer-
Verlag, Berlin, 1994).
7. U. Leonhardt, Measuring the Quantum State of Light
(Cambridge Univ. Press, Cambridge, 1997).
8. K. J. Resch, J. S. Lundeen, A. M. Steinberg, Phys. Rev.
Lett. 88, 113601 (2002).
9. Z. Y. Ou, L. J. Wang, L. Mandel, J. Opt. Soc. Am. B 7,
211 (1990).
10. F. De Martini, V. Mussi, F. Bovino, Opt. Commun.
179, 581 (2000).
11. A. Lamas-Linares, C. Simon, J. C. Howell, D. Bouwmeester,
Science 296, 712 (2002).
12. Z. Y. Ou, Quantum Semiclass. Opt. 9, 599 (1997).
13. T. Aichele, A. I. Lvovsky, S. Schiller, Eur. Phys. J. D 18,
237 (2002).
14. M. Bellini, F. Marin, S. Viciani, A. Zavatta, F. T.
Arecchi, Phys. Rev. Lett. 90, 043602 (2003).
15. S. Viciani, A. Zavatta, M. Bellini, Phys. Rev. A 69,
053801 (2004).
16. A. Zavatta, M. Bellini, P. L. Ramazza, F. Marin, F. T.
Arecchi, J. Opt. Soc. Am. B 19, 1189 (2002).
17. Materials and methods are available as supporting
material on Science Online.
18. S. Reynaud, A. Heidmann, E. Giacobino, C. Fabre, in
Progress in Optics, E. Wolf, Ed. (Elsevier, Amsterdam,
1992), vol. 30, pp. 185.
19. G. Breitenbach, S. Schiller, J. Mlynek, Nature 387, 471
(1997).
20. D. T. Smithey, M. Beck, M. G. Raymer, A. Faridani,
Phys. Rev. Lett. 70, 1244 (1993).
21. A. Montina, F. T. Arecchi, Phys. Rev. A 58, 3472
(1998).
22. We thank F. T. Arecchi for helpful comments and for
a critical reading of the manuscript. This work was
performed in the frame of the Spettroscopia laser e
ottica quantistica project of the Department of
Physics of the University of Florence, with the
support of the Italian Ministry of University and
Scientific Research (MIUR) under the FIRB contract
Microdispositivi fotonici in Niobato di Litio.
DC1
References
23 July 2004; accepted 15 September 2004
Fig. 3. Experimental
Wigner functions for
the SPACS. (A) Re-
constructed Wigner
function for the single-
photon Fock state
obtained without in-
j ecti on. (B to D)
Same, but with an
input coherent field
of increasing ampli-
tude. In (D), the re-
constructed Wigner
function for both the
SPACS and the un-
excited seed coherent
state (wire-frame sur-
face) are shown.
Fig. 4. Experimental squeezing data. Data points represent the variance in the SPACS field quad-
rature x
K
, which exhibits the maximum squeezing factor (about 15% for k"k between 1.5 and 2)
normalized to the fluctuations of the coherent state k" (horizontal solid line). The dashed curve
represents the theoretical prediction for a pure SPACS, whereas the solid one is calculated by
taking into account the experimental detection efficiency. The measured variances for the
different quadratures as a function of the phase K (controlled by the PZT voltage) are shown in the
inset for the states k",1 (solid circles) and k" (open circles) with k"k , 1.8.
R E P O R T S
Quantum State Transfer Between
Matter and Light
D. N. Matsukevich and A. Kuzmich*
We report on the coherent quantum state transfer from a two-level atomic
system to a single photon. Entanglement between a single photon (signal) and a
two-component ensemble of cold rubidium atoms is used to project the
quantum memory element (the atomic ensemble) onto any desired state by
measuring the signal in a suitable basis. The atomic qubit is read out by
stimulating directional emission of a single photon (idler) from the (entangled)
collective state of the ensemble. Faithful atomic memory preparation and
readout are verified by the observed correlations between the signal and the idler
photons. These results enable implementation of distributed quantumnetworking.
The ability to coherently transfer quantum
information between photonic- and material-
based quantum systems is a prerequisite for all
practical distributed quantum computation and
scalable quantum communication protocols
(1). The importance of this process is rooted
in the fact that matter-based quantum systems
provide excellent long-term quantum memory
storage, whereas long-distance communication
of quantum information will most certainly
be accomplished by coherent propagation of
light, often in the form of single photon pulses.
In the microwave domain, coherent quan-
tum control has been obtained with single
Rydberg atoms and single photons (2); im-
portant advances have also been made in ion
trapping information processing (35). Par-
ticularly, an entangled state of an ion and a
photon has been produced (6); however, to
convert a single ion (atom) qubit state into a
photonic state, strong coupling to a single
cavity mode is required. Trapped atoms or
ions localized inside high-finesse cavities of-
fer a natural paradigm for coherent, reversible
matter-light interactions (7, 8), although
technical challenges make these systems dif-
ficult to realize in practice.
Optically thick atomic ensembles have
emerged recently as an alternative for the
light-matter interface (9, 10). Duan, Lukin,
Cirac, and Zoller (DLCZ) (11) have made a
theoretical proposal aimed at long-distance
quantum communication that uses the quan-
tum memory capability of atomic ensem-
bles. Important initial steps toward realization
of the DLCZ protocol have been made in
which nonclassical radiation has been pro-
duced from an atomic ensemble, thereby
demonstrating the collective enhancement
(1215).
Here, we report on the experimental real-
ization of coherent quantum state transfer
from a matter qubit onto a photonic qubit,
using an optically thick cold atomic cloud.
Our experiment involves three steps: (i) An
entangled state between a single photon
(signal) and a single collective excitation dis-
tributed over many atoms in two distinct
optically thick atomic samples is generated.
(ii) Measurement of the signal photon pro-
jects the atomic ensembles into a desired
state, conditioned on the choice of the basis
and the outcome of the measurement. This
atomic state is a nearly maximally entangled
state between two distinct atomic ensembles.
(iii) This nearly maximally entangled atomic
state is converted into a single photon (idler)
emitted into a well-defined mode, without
using a high-finesse cavity. These three in-
gredients constitute a complete set of tools
required to build an arbitrary large-scale
quantum network (11).
As illustrated in Fig. 1A, the classical
laser pulses used in the generation and
verification procedures define the two dis-
tinct pencil-shape components of the atomic
ensemble that form our memory qubit, L
and R. Figure 1B indicates schematically
the structure of the four atomic levels in-
volved, ka,kb, kc, and kd. The experimental
sequence starts with all of the atoms pre-
pared in state ka. A write pulse tuned to the
ka Y kc transition is split into two beams by
a polarizing beam splitter (PBS1) and passed
through the atomic sample. The light induces
spontaneous Raman scattering on the kc Y
kb transition. The classical write pulse is so
weak that less than one photon is scattered in
this manner into the forward direction mode
for each pulse in either L or R. The forward
scattered mode is dominantly correlated with
a distinct collective atomic state (11). In the
first order of perturbation theory in the atom-
light coupling 2, the atom-light state is
kA ka
1
Ika
NL NR
k0
p
L
k0
p
R
2 kL
a
k1
p
L
k0
p
R
kR
a
k0
p
L
k1
p
1
We have defined two effective states of the
atomic ensembles
kL
a
0
X
NL
i 0 1
g
i
ka
1
Ikb
i
Ika
NL
Ika
NL NR
kR
a
0
X
NL NR
j 0 NL 1
g
j
ka
1
Ika
NL
Ikb
j
Ika
NL NR
2
with the weights g
i
and g
j
determined by
the write field intensity distribution,
P
NL
i 0 1
kg
i
k
2
0 1,
P
NL NR
j 0 N
L 1
kg
j
k
2
0 1 (16, 17).
kL
a
and kR
a
have properties of a two-level
system (qubit): bL
a
kL
a
0 1, bR
a
kR
a
0 1, and
bL
a
kR
a
0 0. Although the interaction of the
light with the atoms is nonsymmetric with
respect to permutation of atoms, the second
term in Eq. 1 in fact describes a strongly
entangled atom-photon state in the sense of
(17). Using PBS4 and a half-wave plate
inserted into one of the channels, we map the
two spatial modes associated with the two
ensembles into a single spatial mode with
polarization encoding of the light_s origin:
k1
p
L
Y kH
s
; k1
p
R
Y kV
s
, where H and V
indicate horizontal and vertical polarization,
respectively, and s denotes signal. Next, the
light is passed through an arbitrary polariza-
tion state transformer R
s
(K
s
,f
s
) and a polar-
izer PBS5, so that the state at the output of
PBS5 is
kH 0 cosK
s
e
if
s
kH
s
sinK
s
kV
s
3
and is directed onto a single-photon detector
D1. When D1 detects a photon, the joint state
in Eq. 1 is projected into the desired atomic
state
kA
a
0 cosK
s
e
jif
s
kL
a
sinK
s
e
i(
s
kR
a
4
which is an entangled state of the two atomic
samples L and R. Phase (
s
is determined by
the difference in length of the two paths L
and R. After a variable delay time %t, we
convert the atomic excitation into a single
photon by illuminating the atomic ensemble
with a pulse of light near resonant with the
kb Y kd transition. For an optically thick
atomic sample, the photon will be emitted
with high probability into the spatial mode
determined by the write pulse (11, 16),
achieving memory read-out.
kA
a
0 cosK
s
e
jif
s
kL
a
sinK
s
e
i(
s
kR
a
Y kA
i
0 cosK
s
e
jif
s
kH
i

sinK
s
e
i(
i
(
s
kV
i
5
That is, the polarization state of the idler
photon i is uniquely determined by the ob-
served state of the signal photon. Alterna-
tively, one could store the signal in a fiber
School of Physics, Georgia Institute of Technology,
Atlanta, GA 30332, USA.
E-mail: alex.kuzmich@physics.gatech.edu.
R E P O R T S
until after the readout. In that case, the two-
photon signal-idler state would be a maxi-
mally entangled state:
kA
M
0
1
2
p kH
s
kH
i

e
i(
s
(
i
kV
s
kV
i
6
A magneto-optical trap (MOT) of
85
Rb is used
to provide an optically thick atomic cloud for
our experiment (Fig. 1). The ground states
Aka;kbZ correspond to the 5S
1/2
,F 0 A3,2Z
levels of
85
Rb, while the excited states Akc;kdZ
represent the A5P
3/2
,F 0 3;5P
1/2
,F 0 2Z
levels of the AD
2
, D
1
Z lines at A780;795Z
nm, respectively. The experimental sequence
starts with all of the atoms prepared in state
ka by optical pumping, after shutting off the
trapping and cooling light.
A 140-ns-long write pulse tuned to the
ka Y kc transition is split into two beams by
a polarizing beam splitter PBS1 and focused
into two regions of the MOT about 1 mm
apart, with Gaussian waists of about 50 6m.
PBS2 and PBS3 separate the horizontally po-
larized component of the forward scattered
light from the vertically polarized classical
pulse. After being mixed by PBS4, the light
goes through the quarter- and the half-wave
plates that provide the state transformation
R
s
(K
s
,f
s
). The light continues to another po-
larizer, PBS5, and is directed to a single
photon detector D1. Detection of one photon
by D1 prepares the atomic ensemble in any
desired state in the basis of kL
a
, kR
a
deter-
mined by R
s
(K
s
,f
s
), and thereby concludes
the preparation of the quantum memory
qubit.
Following memory state preparation, the
read-out stage is performed. After a user-
programmable delay, %t, a 115-ns-long read
pulse tuned to the kb Y kd transition il-
luminates the two atomic ensembles. This
accomplishes a transfer of the memory state
onto the single photon (idler) emitted by the
kd Y ka transition. After passing through
the state transformer R
i
(K
i
,f
i
) and PBS6,
the two polarization components are directed
onto single-photon detectors (D2, D3), thus
accomplishing measurement of the idler pho-
ton, and hence the memory qubit, on a con-
trollable arbitrary basis.
As in any real experiment, various imper-
fections prevent the read-out of the quantum
memory (idler photon) from being identical
to the state that we intended to write into the
memory. To quantify the degree to which we
faithfully prepare and read out the quantum
memory, we measure the polarization corre-
lations between the signal and idler photons.
The observed correlations allow us to char-
acterize the extent to which our procedures
are working. To investigate the storage ca-
pabilities of our memory qubit quantitatively,
we use time-resolved detection of the signal
and idler photons for two values of delay %t
between the application of the write and read
pulses, 100 ns and 200 ns. The electronic
pulses from the detectors are gated, with
250-ns and 140-ns windows centered on the
time determined by the write and read light
pulses, respectively. Afterward, the electron-
ic pulses are fed into a time-interval analyzer
(with & 0 2 ns time resolution). To measure
the correlation between the photons produced
by the write and read pulses, the output of D1
is fed into the Bstart[ input of a time-interval
analyzer, and the outputs of D2 and D3 are
fed into two Bstop[ inputs. A coincidence
window imposed by the data acquisition
software selects a time interval between
the arrival of the idler and signal of (0,80)
ns for %t 0 100 ns and (25,145) ns for %t 0
200 ns.
We first measure the conditional proba-
bilities of detecting a certain state of the idler
(hence, of the quantum memory state) in the
basis of kH
i
and kV
i
, given the observed
state of the signal photon. Varying the angle
K
s
produces the correlation patterns shown in
Fig. 1. (A) Schematic of experimental setup. PBS1
to PBS6, polarizing beam splitters; 4/2, halfwave
plate; R
s
(K
s
, f
s
) and R
i
(K
i
, f
i
), polarization state trans-
formers; (D1, D2, D3), single photon detectors; DM,
dichroic mirror. The inset illustrates the timing of the
write and read pulses. (B) The relevant atomic level
structure.
R E P O R T S
Fig. 2A for %t 0 100 ns. Conditional proba-
bilities at the point of maximum correlation
are shown in Fig. 2B and the first line of
Table 1. To verify faithful memory prepara-
tion and readout, we repeat the correlation
measurement in a different basis, that of
states kH
i
T kV
i
=
2
p
, by choosing K
i
0
45-, f
i
0 0-, and f
s
0 j((
s
(
i
) in the state
transformers R
s
and R
i
. We vary K
s
, with the
measured interference fringes displayed in
Fig. 3A. Table 1 (second line) and Fig. 3B
show the conditional probabilities at the
point of maximum correlations. These prob-
abilities are different from 1/2 only when the
phase coherence between the two states of
the atomic qubit is preserved in the matter-
to-light quantum state mapping.
From these measured correlations, we de-
termine the fidelity of the reconstruction of
our intended quantum memory state kA
I
in
the idler, kbA
I
kA
i
k
2
. The fidelity is given by
the value of the corresponding conditional
probability at the point of maximum corre-
lation, presented in Table 1 (we choose the
lower of the two values as the lower bound).
For states in the K
i
0 0- basis, we find F
0
0
0.88 T 0.03, clearly exceeding the classical
boundary of 2/3 (18). For the K
i
0 45- basis,
we found F
45
0 0.75 T 0.02, again sub-
stantially violating the classical limit. These
fidelities give a lower bound for both the
fidelities of the memory preparation and the
read-out steps, which we do not measure
separately.
Another way to quantify the performance
of our quantum state transfer is to calculate
the fidelity of entanglement between the sig-
nal and idler photons F
si
. The lower bound
on F
si
is given by the overlap of the mea-
sured density matrix, with the maximally
entangled state we seek to achieve, kA
M
,
given by Eq. 6: F
si
0 bA
M
k D
si
kA
M
(19).
We calculated F
si
0 0.67 T 0.02, substan-
tially greater than the classical limit of 1/2
(6, 19).
At a longer delay of 200 ns, the fidelities
in the K
i
0 0- and K
i
0 45- bases are F
0
0
0.79 T 0.04 and F
45
0 0.74 T 0.04, while
fidelity of entanglement is F
si
0 0.63 T 0.03.
For both values of %t, we analyze the fidelity
of entanglement as a function of the delay
between the detections of the signal and
the idler. We split the full coincidence win-
dow into four equal intervals and calcu-
lated entanglement of formation for each one
(Fig. 4). From these results, we conclude
that our quantum memory has a useful op-
erational time of about 150 ns. The lifetime
of coherence between levels ka and kb de-
termines the lifetime of the quantum mem-
ory and is limited by the magnetic field of
the trapping quadrupole field of the MOT
(12).
Nonzero coincidence counts in the min-
ima of Fig. 2A are due to transmission losses
and nonideal spatial correlations between
the signal and idler photons. The residual
interferometric drifts in (
s
(
i
further re-
duce the visibility of Fig. 3A compared with
Fig. 2A, resulting in a degradation of the fi-
delities. Losses also reduce the rate of en-
tanglement generation. The rate of signal
photon detections (and hence, atomic qubit
preparation) is given by R
s
0 "n
s
R ; 300 s
j1
,
where " 0 0.05 is the measured transmission
efficiency for the write beam (which
includes 0.60 detection efficiency) and R 0
4.7 10
5
s
j1
is the repetition rate of the
experiment. Therefore, the inferred average
photon number in the forward scattered
mode per pulse is n
s
; 1.4 10
j2
. The co-
incident signal-idler detection rate is R
si
0
QR
s
0 Q"n
s
R ; 0.4 s
j1
, where Q K $O ; 1.1
10
j3
. The measured transmission and de-
Fig. 3. (A) Measured condi-
tional probabilities after K
i
0
>/4 polarization rotation of
the idler photon as the func-
tion of K
s
. (B) Measured con-
ditional probabilities at the
points of highest correlation.
0
0.2
0.4
0.6
0.8
1
-200 -150 -100 -50 0 50 100 150 200
s
(degrees)
P(H
i
|H
s
)
P(V
i
|H
s
)
S
I
V
H
H
V
C
o
n
d
i
t
i
o
n
a
l

p
r
o
b
a
b
i
l
i
t
i
e
s
A B
C
o
n
d
i
t
i
o
n
a
l

p
r
o
b
a
b
l
i
l
i
t
y

P
(
S
/
I
)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Fig. 2. (A) Measured condi-
tional probabilities P(H
i
kH
s
)
and P(V
i
kH
s
) as the function
of the polarization rota-
tion K
s
of the signal pho-
ton. The full curves are fits
with the visibility as the
only adjustable parameter.
(B) Measured conditional
probabilities at the points
of highest correlation.
0
0.2
0.4
0.6
0.8
1
-200 -150 -100 -50 0 50 100 150 200
s
(degrees)
P(H
i
|H
s
)
P(V
i
|H
s
)
S
I
V
H
H
V
C
o
n
d
i
t
i
o
n
a
l

p
r
o
b
a
b
i
l
i
t
i
e
s
C
o
n
d
i
t
i
o
n
a
l

p
r
o
b
a
b
i
l
i
t
y

P
(
S
/
I
)
1.0
0.8
0.9
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
A B
R E P O R T S
tection efficiency for the read beam is $ ;
0.04, so we infer the efficiency of quantum
state transfer from the atoms onto the pho-
ton, O ; 0.03.
We have realized a quantum node by
combining the entanglement of atomic and
photonic qubits with the atom-photon quan-
tum state transfer. By implementing the
second node at a different location and
performing a joint detection of the signal
photons from the two nodes, the quantum
repeater protocol (11), as well as distant te-
leportation of an atomic qubit, may be real-
ized. Based on this work, we estimate the
rate for these protocols to be R
2
; ($O"n
s
)
2
R ;
3 10
j7
s
j1
. However, improvements in O
that are based on increasing the optical
thickness of atomic samples (16), as well as
elimination of transmission losses, could pro-
vide several orders of magnitude increase in
R
2
. Our results also demonstrate the possi-
bility of realizing quantum nodes consisting
of multiple atomic qubits by using multiple
beams of light. This approach shows prom-
ise for implementation of distributed quan-
tum computation (20, 21).
1. I. Chuang, M. Nielsen, Quantum Computation and
Quantum Information (Cambridge Univ. Press, Cam-
bridge, 2000).
2. S. Haroche, J. M. Raimond, M. Brune, in Experimental
Quantum Computation and Information, F. de Martini,
C. Monroe, Eds. (Proceedings of the International
School of Physics Enrico Fermi, course CXLVIII, IOS
Press, Amsterdam, 2002), pp. 3766.
3. C. A. Sackett et al., Nature 404, 256 (2000).
4. M. D. Barrett et al., Nature 429, 737 (2004).
5. M. Riebe et al., Nature 429, 734 (2004).
6. B. B. Blinov, D. L. Moehring, L.-M. Duan, C. Monroe,
Nature 428, 153 (2004).
7. S. Bose, P. L. Knight, M. B. Plenio, V. Vedral, Phys. Rev.
Lett. 83, 5158 (1999).
8. H. J. Kimble, Phys. Scr. 76, 127 (1998).
9. A. Kuzmich, E. S. Polzik, in Quantum Information with
Continuous Variables, S. L. Braunstein, A. K. Pati, Eds.
(Kluwer, Dordrecht, 2003).
10. M. D. Lukin, Rev. Mod. Phys. 75, 457 (2003).
11. L.-M. Duan, M. D. Lukin, I. J. Cirac, P. Zoller, Nature
414, 413 (2001).
12. A. Kuzmich et al., Nature 423, 731 (2003).
13. C. H. van der Wal et al., Science 301, 196 (2003).
14. W. Jiang, C. Han, P. Xue, L.-M. Duan, G. C. Guo, Phys.
Rev. A. 69, 043819 (2004).
15. C. W. Chou, S. V. Polyakov, A. Kuzmich, H. J. Kimble,
Phys. Rev. Lett. 92, 213601 (2004).
16. L.-M. Duan, J. I. Cirac, P. Zoller, Phys. Rev. A. 66,
023818 (2002).
17. A. Kuzmich, T. A. B. Kennedy, Phys. Rev. Lett. 92,
030407 (2004).
18. M. Horodecki, P. Horodecki, R. Horodecki, Phys. Rev.
A. 60, 1888 (1994).
19. C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, W. K.
Wooters, Phys. Rev. A. 54, 3824 (1996).
20. Y. L. Lim, A. Beige, L. C. Kwek, www.arXiv.org/quant-ph/
0408043.
21. S. D. Barrett, P. Kok, www.arXiv.org/quant-ph/0408040.
22. We acknowledge fruitful conversations with T. A. B.
Kennedy, J. A. Sauer, L. You, A. Zangwill and, par-
ticularly, M. S. Chapman and thank R. Smith and E. T.
Neumann for experimental assistance. This work was
supported by NASA and the Research Corporation.
Electric Field Effect in Atomically
Thin Carbon Films
K. S. Novoselov,
1
A. K. Geim,
1
* S. V. Morozov,
2
D. Jiang,
1
Y. Zhang,
1
S. V. Dubonos,
2
I. V. Grigorieva,
1
A. A. Firsov
2
We describe monocrystalline graphitic films, which are a few atoms thick but are
nonetheless stable under ambient conditions, metallic, and of remarkably high
quality. The films are found to be a two-dimensional semimetal with a tiny overlap
between valence and conductance bands, and they exhibit a strong ambipolar
electric field effect such that electrons and holes in concentrations up to 10
13
per
square centimeter and with room-temperature mobilities of 10,000 square
centimeters per volt-second can be induced by applying gate voltage.
The ability to control electronic properties of
a material by externally applied voltage is at
the heart of modern electronics. In many
cases, it is the electric field effect that allows
one to vary the carrier concentration in a
semiconductor device and, consequently,
change an electric current through it. As the
semiconductor industry is nearing the limits
of performance improvements for the current
technologies dominated by silicon, there is a
constant search for new, nontraditional mate-
rials whose properties can be controlled by
the electric field. The most notable recent
examples of such materials are organic
conductors (1) and carbon nanotubes (2). It
has long been tempting to extend the use of
the field effect to metals Ee.g., to develop all-
metallic transistors that could be scaled down
to much smaller sizes and would consume
less energy and operate at higher frequencies
than traditional semiconducting devices (3)^.
However, this would require atomically thin
metal films, because the electric field is
screened at extremely short distances (G1 nm)
and bulk carrier concentrations in metals are
large compared to the surface charge that can
be induced by the field effect. Films so thin
tend to be thermodynamically unstable, be-
coming discontinuous at thicknesses of sev-
eral nanometers; so far, this has proved to be
an insurmountable obstacle to metallic elec-
tronics, and no metal or semimetal has been
shown to exhibit any notable (91%) field ef-
fect (4).
We report the observation of the electric
field effect in a naturally occurring two-
dimensional (2D) material referred to as
few-layer graphene (FLG). Graphene is the
name given to a single layer of carbon atoms
densely packed into a benzene-ring struc-
ture, and is widely used to describe proper-
ties of many carbon-based materials, including
graphite, large fullerenes, nanotubes, etc. (e.g.,
carbon nanotubes are usually thought of as
graphene sheets rolled up into nanometer-sized
cylinders) (57). Planar graphene itself has
been presumed not to exist in the free state,
being unstable with respect to the formation of
curved structures such as soot, fullerenes, and
nanotubes (514).
Table 1. Conditional probabilities P(IkS) to detect the idler photon in state I given detection of the signal
photon in state S, at the point of maximum correlation for %t 0 100 ns delay between read and write
pulses; all the errors are based on counting statistics of coincidence events.
Basis P(H
i
kH
s
) P(V
i
kH
s
) P(V
i
kV
s
) P(H
i
kV
s
)
0- 0.92 T 0.02 0.08 T 0.02 0.88 T 0.03 0.12 T 0.03
45- 0.75 T 0.02 0.25 T 0.02 0.81 T 0.02 0.19 T 0.02
0
0.2
0.4
0.6
0.8
1
0 50 100 150
signal-idler delay (ns)
classical threshold
E
n
t
a
n
g
l
e
m
e
n
t

f
i
d
e
l
i
t
y
Fig. 4. Time-dependent entanglement fidelity
of the signal and the idler F
si
; circles for %t 0
100 ns, diamonds for %t 0 200 ns.
1
Department of Physics, University of Manchester,
Manchester M13 9PL, UK.
2
Institute for Microelec-
tronics Technology, 142432 Chernogolovka, Russia.
E-mail: geim@man.ac.uk
R E P O R T S
We have been able to prepare graphitic
sheets of thicknesses down to a few atomic
layers (including single-layer graphene), to
fabricate devices from them, and to study
their electronic properties. Despite being
atomically thin, the films remain of high
quality, so that 2D electronic transport is
ballistic at submicrometer distances. No
other film of similar thickness is known to
be even poorly metallic or continuous under
ambient conditions. Using FLG, we demon-
strate a metallic field-effect transistor in
which the conducting channel can be
switched between 2D electron and hole gases
by changing the gate voltage.
Our graphene films were prepared by
mechanical exfoliation (repeated peeling) of
small mesas of highly oriented pyrolytic
graphite (15). This approach was found to
be highly reliable and allowed us to prepare
FLG films up to 10 6m in size. Thicker films
(d Q 3 nm) were up to 100 6m across and
visible by the naked eye. Figure 1 shows
examples of the prepared films, including
single-layer graphene Esee also (15)^. To
study their electronic properties, we pro-
cessed the films into multiterminal Hall bar
devices placed on top of an oxidized Si
substrate so that a gate voltage V
g
could be
applied. We have studied more than 60
devices with d G 10 nm. We focus on the
electronic properties of our thinnest (FLG)
devices, which contained just one, two, or
three atomic layers (15). All FLG devices
exhibited essentially identical electronic
properties characteristic for a 2D semimetal,
which differed from a more complex (2D
plus 3D) behavior observed for thicker,
multilayer graphene (15) as well as from
the properties of 3D graphite.
In FLG, the typical dependence of its sheet
resistivity D on gate voltage V
g
(Fig. 2)
exhibits a sharp peak to a value of several
kilohms and decays to 100 ohms at high V
g
(note that 2D resistivity is given in units of
ohms rather than ohms cm as in the 3D
case). Its conductivity G 0 1/D increases
linearly with V
g
on both sides of the resistivity
peak (Fig. 2B). At the same V
g
where D has its
peak, the Hall coefficient R
H
exhibits a sharp
reversal of its sign (Fig. 2C). The observed
behavior resembles the ambipolar field effect
in semiconductors, but there is no zero-
conductance region associated with the Fermi
level being pinned inside the band gap.
Our measurements can be explained
quantitatively by a model of a 2D metal
with a small overlap &( between conductance
and valence bands (15). The gate voltage
induces a surface charge density n 0 (
0
(V
g
/te
and, accordingly, shifts the position of the
Fermi energy (
F
. Here, (
0
and ( are the
permittivities of free space and SiO
2
, respec-
tively; e is the electron charge; and t is the
thickness of our SiO
2
layer (300 nm). For
typical V
g
0 100 V, the formula yields n ,
7.2 10
12
cm
j2
. The electric field doping
transforms the shallow-overlap semimetal
into either completely electron or completely
hole conductor through a mixed state where
both electrons and holes are present (Fig. 2).
The three regions of electric field doping are
clearly seen on both experimental and
theoretical curves. For the regions with only
electrons or holes left, R
H
decreases with
increasing carrier concentration in the usual
way, as 1/ne. The resistivity also follows the
standard dependence D
j1
0 G 0 ne6 (where
6 is carrier mobility). In the mixed state, G
changes little with V
g
, indicating the substi-
tution of one type of carrier with another,
while the Hall coefficient reverses its sign,
reflecting the fact that R
H
is proportional to
Fig. 1. Graphene films. (A) Photograph (in normal white light) of a relatively large multilayer
graphene flake with thickness 3 nm on top of an oxidized Si wafer. (B) Atomic force microscope
(AFM) image of 2 6m by 2 6m area of this flake near its edge. Colors: dark brown, SiO
2
surface;
orange, 3 nm height above the SiO
2
surface. (C) AFM image of single-layer graphene. Colors: dark
brown, SiO
2
surface; brown-red (central area), 0.8 nm height; yellow-brown (bottom left), 1.2 nm;
orange (top left), 2.5 nm. Notice the folded part of the film near the bottom, which exhibits a
differential height of 0.4 nm. For details of AFM imaging of single-layer graphene, see (15). (D)
Scanning electron microscope image of one of our experimental devices prepared from FLG. (E)
Schematic view of the device in (D).
Fig. 2. Field effect in FLG. (A) Typical
dependences of FLGs resistivity D on
gate voltage for different temperatures
(T 0 5, 70, and 300 K for top to bottom
curves, respectively). (B) Example of
changes in the films conductivity G 0
1/D(V
g
) obtained by inverting the 70 K
curve (dots). (C) Hall coefficient R
H
versus V
g
for the same film; T 0 5 K. (D)
Temperature dependence of carrier
concentration n
0
in the mixed state
for the film in (A) (open circles), a
thicker FLG film (squares), and multi-
layer graphene (d , 5 nm; solid circles).
Red curves in (B) to (D) are the
dependences calculated from our mod-
el of a 2D semimetal illustrated by
insets in (C).
0
2
4
6
8
-100 -50 0 50 100
0
0.5
-100 0 100
0
3
100 300
2
4
6
D
C
B
(
k
F
A
F
R
H
(
k
T
/
)
V
g
(V)
V
g
(V)
(m
-1
)
T (K)
n
0
(T )/ n
0
(4K)
0
R E P O R T S
the difference between electron and hole
concentrations.
Without electric field doping (at zero V
g
),
FLG was found to be a hole metal, which is
seen as a shift of the peak in D to large
positive V
g
. However, this shift is attributed
to an unintentional doping of the films by
absorbed water (16, 17). Indeed, we found
that it was possible to change the position of
the peak by annealing our devices in
vacuum, which usually resulted in shifting
of the peak close to zero voltages. Exposure
of the annealed films to either water vapor or
NH
3
led to their p- and n-doping, respec-
tively (15). Therefore, we believe that intrin-
sic FLG is a mixed-carrier material.
Carrier mobilities in FLG were deter-
mined from field-effect and magnetoresist-
ance measurements as 6 0 G(V
g
)/en(V
g
) and
6 0 R
H
/D, respectively. In both cases, we
obtained the same values of 6, which varied
from sample to sample between 3000 and
10,000 cm
2
/VIs. The mobilities were practi-
cally independent of absolute temperature T,
indicating that they were still limited by
scattering on defects. For 6 , 10,000 cm
2
/VIs
and our typical n , 5 10
12
cm
j2
, the mean
free path is 0.4 6m, which is surprising
given that the 2D gas is at most a few )
away from the interfaces. However, our
findings are in agreement with equally high
6 observed for intercalated graphite (5),
where charged dopants are located next to
graphene sheets. Carbon nanotubes also exhib-
it very high 6, but this is commonly attributed
to the suppression of scattering in the 1D case.
Note that for multilayer graphene, we observed
mobilities up to 15,000 cm
2
/VIs at 300 K
and 60,000 cm
2
/VIs at 4 K.
Despite being essentially gigantic fuller-
ene molecules and unprotected from the
environment, FLG films exhibit pronounced
Shubnikovde Haas (ShdH) oscillations in
both longitudinal resistivity D
xx
and Hall re-
sistivity D
xy
(Fig. 3A), serving as another
indicator of the quality and homogeneity of
the experimental system. Studies of ShdH os-
cillations confirmed that electronic transport
in FLG was strictly 2D, as one could reason-
ably expect, and allowed us to fully charac-
terize its charge carriers. First, we carried out
the standard test and measured ShdH oscil-
lations for various angles K between the
magnetic field and the graphene films. The
oscillations depended only on the perpendic-
ular component of the magnetic field BIcos K,
as expected for a 2D system. More impor-
tant, however, we found a linear dependence
of ShdH oscillations_ frequencies B
F
on V
g
(Fig. 3B), indicating that the Fermi energies
(
F
of holes and electrons were proportional
to their concentrations n. This dependence is
qualitatively different from the 3D dependence
(
F
n
2/3
and proves the 2D nature of charge
carriers in FLG. Further analysis (15) of ShdH
oscillations showed that only a single spa-
tially quantized 2D subband was occupied up
to the maximum concentrations achieved in
our experiments (3 10
13
cm
j2
). It could
be populated either by electrons with mass
m
e
, 0.06m
0
(where m
0
is the free electron
mass) located in two equivalent valleys, or
by light and heavy holes with masses of
0.03m
0
and 0.1m
0
and the double-valley
degeneracy. These properties were found to
be the same for all FLG films studied and are
notably different from the electronic struc-
ture of both multilayer graphene (15) and
bulk graphite (57). Note that graphene is
expected (57) to have the linear energy
dispersion and carriers with zero mass, and
the reason why the observed behavior is so
well described by the simplest free-electron
model remains to be understood (15).
We also determined the band overlap &(
in FLG, which varied from 4 to 20 meV for
different samples, presumably indicating a
different number of graphene layers involved
(18). To this end, we first used a peak value
D
m
of resistivity to calculate typical carrier
concentrations in the mixed state, n
0
(e.g., at
low T for the sample in Fig. 2, A to C, with
6 , 4000 cm
2
/V and D
m
, 8 kilohms, n
0
was
2 10
11
cm
j2
). Then, &( can be estimated
as n
0
/D, where D 0 2m
e
/>I
2
is the 2D
density of electron states and I is Planck_s
constant divided by 2>. For the discussed
sample, this yields &( , 4 meV Ei.e., much
smaller than the overlap in 3D graphite
(40 meV)^. Alternatively, &( could be
calculated from the temperature dependence
of n
0
, which characterizes relative contribu-
tions of intrinsic and thermally excited car-
riers. For a 2D semimetal, n
0
(T) varies as
n
0
(0 K)IfIlnE1 exp(1/f )^, where f 0 2k
B
T/&(
and k
B
is Boltzmann_s constant; Fig. 2D
shows the best fit to this dependence, which
yields &( , 6 meV. Different FLG devices
were found to exhibit a ratio of n
0
(300 K)/
n
0
(0) between 2.5 and 7, whereas for
multilayer graphene it was only 1.5 (Fig.
2D). This clearly shows that &( decreases
with decreasing number of graphene layers.
The observed major reduction of &( is in
agreement with the fact that single-layer
graphene is in theory a zero-gap semicon-
ductor (5, 18).
Graphene may be the best possible metal for
metallic transistor applications. In addition to
the scalability to true nanometer sizes envis-
aged for metallic transistors, graphene also
offers ballistic transport, linear current-voltage
(I-V) characteristics, and huge sustainable
currents (910
8
A/cm
2
) (15). Graphene tran-
sistors show a rather modest on-off resistance
ratio (less than 30 at 300 K; limited because
Fig. 3. (A) Examples of ShdH
oscillations for one of our FLG
devices for different gate volt-
ages; T 0 3 K, and B is the mag-
netic field. As the black curve shows,
we often observed pronounced
plateau-like features in D
xy
at val-
ues close to (h/4e
2
)/8 (in this case,
(
F
matches the Landau level with
filling factor 8 0 2 at around 9 T).
Such not-fully-developed Hall
plateaus are usually viewed as an
early indication of the quantum
Hall effect in the situations where
D
xx
does not yet reach the zero-
resistance state. (B) Dependence
of the frequency of ShdH oscilla-
tions B
F
on gate voltage. Solid and
open symbols are for samples with
&( , 6 meV and 20 meV, respec-
tively. Solid lines are guides to the eye. The linear dependence B
F

V
g
indicates a constant (2D) density of states (15). The observed
slopes (solid lines) account for the entire external charge n induced
by gate voltage, confirming that there are no other types of carriers
and yielding the double-valley degeneracy for both electrons and
holes (15). The inset shows an example of the temperature
dependence of amplitude % of ShdH oscillations (circles), which is
fitted by the standard dependence T/sinh(2>
2
k
B
T/I<
c
) where <
c
is
their cyclotron frequency. The fit (solid curve) yields light holes mass
of 0.03m
0
.
2 4 6 8 10
0
0.5
1.0
1.5
1
2
3
4
-100 -50 0 50 100
0
20
40
60
80
50 100
0
10
+80V
-100V
-25V
+100V
x
x
k
(

)
B (T)
+80V
A
y
x
k
(

)
B
F
)
T
(

V
g
(V)
B
T (K)
)
-100V @ 9T
R E P O R T S
of thermally excited carriers), but this is a
fundamental limitation for any material with-
out a band gap exceeding k
B
T. Nonetheless,
such on-off ratios are considered sufficient for
logic circuits (19), and it is feasible to increase
the ratio further by, for example, using p-n
junctions, local gates (3), or the point contact
geometry. However, by analogy to carbon
nanotubes (2), other, nontransistor applications
of this atomically thin material ultimately may
prove to be the most exciting.
1. C. D. Dimitrakopoulos, D. J. Mascaro, IBM J. Res. Dev.
45, 11 (2001).
2. R. H. Baughman, A. A. Zakhidov, W. A. de Heer,
Science 297, 787 (2002).
3. S. V. Rotkin, K. Hess, Appl. Phys. Lett. 84, 3139
(2004).
4. A. V. Butenko, D. Shvarts, V. Sandomirsky, Y.
Schlesinger, J. Appl. Phys. 88, 2634 (2000).
5. M. S. Dresselhaus, G. Dresselhaus, Adv. Phys. 51, 1
(2002).
6. I. L. Spain, in Chemistry and Physics of Carbon, P. L.
Walker, P. A. Thrower, Eds. (Dekker, New York,
1981), pp. 119304.
7. O. A. Shenderova, V. V. Zhirnov, D. W. Brenner, Crit.
Rev. Solid State Mater. Sci. 27, 227 (2002).
8. A. Krishnan et al., Nature 388, 451 (1997).
9. E. Dujardin, T. Thio, H. Lezec, T. W. Ebbesen, Appl.
Phys. Lett. 79, 2474 (2001).
10. H. Shioyama, J. Mat. Sci. Lett. 20, 499 (2001).
11. Other methods of preparing thin graphitic layers
exist. The closest analogs of FLG are nanometer-sized
patches of graphene on top of pyrolytic graphite (12,
13), carbon films grown on single-crystal metal
substrates (14), and mesoscopic graphitic disks with
thickness down to 60 graphene layers (8, 9).
12. A. M. Affoune et al., Chem. Phys. Lett. 348, 17
(2001).
13. K. Harigaya, Y. Kobayashi, K. Takai, J. Ravier, T. Enoki,
J. Phys. Cond. Matter 14, L605 (2002).
14. T. A. Land, T. Michely, R. J. Behm, J. C. Hemminger,
G. Comsa, Surf. Sci. 264, 261 (1992).
15. See supporting data on Science Online.
16. J. Kong et al., Science 287, 622 (2000).
17. M. Kruger, I. Widner, T. Nussbaumer, M. Buitelaar,
C. Schonenberger, N. J. Phys. 5, 138 (2003).
18. We believe that our thinnest FLG samples (as in
Fig. 2A) are in fact zero-gap semiconductors,
because small nonzero values of &( found exper-
imentally can be attributed to inhomogeneous
doping, which smears the zero-gap state over a
small range of V
g
and leads to finite apparent &(.
19. M. R. Stan, P. D. Franzon, S. C. Goldstein, J. C. Lach,
M. M. Zeigler, Proc. IEEE 91, 1940 (2003).
20. Supported by the UK Engineering and Physical
Sciences Research Council and the Russian Acad-
emy of Sciences (S.V.M., S.V.D.). We thank L. Eaves,
E. Hill, and O. Shklyarevskii for discussions and
interest.
DC1
SOM Text
Figs. S1 to S11
Hydrated Electron Dynamics:
From Clusters to Bulk
A. E. Bragg,
1
J. R. R. Verlet,
1
A. Kammrath,
1
O. Cheshnovsky,
2
D. M. Neumark
1,3
*
The electronic relaxation dynamics of size-selected (H
2
O)
n
/(D
2
O)
n
[25 e
n e 50] clusters have been studied with time-resolved photoelectron imaging.
The excess electron (e
c
) was excited through the e

c
( p) @ e
c
(s) transition with

an ultrafast laser pulse, with subsequent evolution of the excited state
monitored with photodetachment and photoelectron imaging. All clusters
exhibited p-state population decay with concomitant s-state repopulation
(internal conversion) on time scales ranging from 180 to 130 femtoseconds
for (H
2
O)
n
and 400 to 225 femtoseconds for (D

2
O)
n
; the lifetimes decrease

with increasing cluster sizes. Our results support the nonadiabatic relaxa-
tion mechanism for the bulk hydrated electron (e
aq
), which invokes a 50-

femtosecond e
aq
( p) Y e
aq
(s
.
) internal conversion lifetime.
A free electron introduced into a polar sol-
vent, such as water (1) or ammonia (2), may
be trapped by locally oriented solvent mole-
cules. In water, an Bequilibrated[ hydrated
electron Ee
aq
(s)^ can be transiently confined

within a roughly spherical cavity defined by
six OH bonds oriented toward the negative
charge distribution in the so-called Kevan
geometry (35). The hydrated electron is an
important reagent in condensed-phase chem-
istry and molecular biology, as it participates
in radiation chemistry, electron transfer, and
charge-induced reactivity. Thus, research in-
vestigating the dynamics of this species,
whether in the presence or absence of other
reagents, has attracted considerable attention
in the theoretical and experimental physical
chemistry communities. Here, we present
time-resolved results on the electronic relax-
ation dynamics of anionic clusters of water
that lend profound insight to the elucidation of
hydrated electron dynamics in the bulk.
The electronic energetics of a hydrated
electron are characterized by three types of
states (Fig. 1A): a localized e
aq
(s) ground
state; three localized, near-degenerate e
aq
( p)
excited states; and a delocalized conduction
band (CB) characterized by a charge distri-
bution spread across hundreds of molecules
in the solvent Bnetwork.[ The visible ab-
sorption spectrum of the equilibrium hy-
drated electron, a broad band peaking at
720 nm (1), is well understood as an ex-
citation from the occupied e
aq
(s) state to the

vacant e
aq
( p) states (4, 5). The electron-

solvent dynamics subsequent to e
aq
( p) @
e
aq
(s) excitation are more controversial, how-

ever, in spite of considerable experimental
(612) and theoretical (13, 14) effort devoted
to this problem.
Transient absorption measurements made
with femtosecond (fs) laser pulses as short as 5
fs (12) show a near infrared (NIR) absorption
band beyond 900 nm developing on a 40- to
50-fs time scale after e
aq
( p) @ e
aq
(s)
excitation. This broad feature shifts back to
shorter wavelengths on a time scale of several
hundred fs, with recovery of the original e
aq
(s)
absorption spectrum largely complete within
1 picosecond (ps). Deuteration of the
solvent (8, 12) appears only to affect the
fastest measured time scale (i.e., the buildup
time of the transient NIR absorption), with
C
D2O
/C
H2O
0 1.4 to 1.6. As described by
Yokoyama et al. (8), two rather different
energy relaxation mechanisms have been
proposed to account for these observations.
In the Badiabatic solvation[ scheme (13, 14),
the infrared transient at the earliest times is
attributed to absorption of the e
aq
( p) electron,
which is solvated on the upper state within 50
fs (process x in Fig. 1A). The excited electron
then undergoes internal conversion (IC) to the
ground state (y) on a 400-fs time scale,
generating ground-state electrons that further
relax on the 1-ps time scale by dissipating
energy to the solvent. In contrast, the
Bnonadiabatic relaxation[ mechanism (7, 9,
12) invokes much more rapid IC, on a 50-fs
time scale, and attributes the transient NIR
band at early times to absorption of the
ground-state electron in a vibrationally excited
solvent environment. In this model, subsequent
dynamics are assigned to reorganization of the
local (400 fs) and extended (1 ps) solvent
network following the electronic decay.
We present an alternative and comple-
mentary approach to assessing the relaxation
dynamics of the hydrated electron through
time-resolved photoelectron imaging (TRPEI)
studies (15) of electron dynamics in size-
selected water cluster anions, (H
2
O)
n
and
(D
2
O)
n
, with 25 e n e 50. Water cluster

anions were first detected mass spectromet-
1
Department of Chemistry, University of California,
2
School of Chemistry, The
Sackler Faculty of Exact Sciences, Tel-Aviv University,
69978 Israel.
3
Chemical Sciences Division, Lawrence
Berkeley National Laboratory, Berkeley, CA 94720,
USA.
E-mail: dan@radon.cchem.berkeley.edu
R E P O R T S
of thermally excited carriers), but this is a
fundamental limitation for any material with-
out a band gap exceeding k
B
T. Nonetheless,
such on-off ratios are considered sufficient for
logic circuits (19), and it is feasible to increase
the ratio further by, for example, using p-n
junctions, local gates (3), or the point contact
geometry. However, by analogy to carbon
nanotubes (2), other, nontransistor applications
of this atomically thin material ultimately may
prove to be the most exciting.
1. C. D. Dimitrakopoulos, D. J. Mascaro, IBM J. Res. Dev.
45, 11 (2001).
2. R. H. Baughman, A. A. Zakhidov, W. A. de Heer,
Science 297, 787 (2002).
3. S. V. Rotkin, K. Hess, Appl. Phys. Lett. 84, 3139
(2004).
4. A. V. Butenko, D. Shvarts, V. Sandomirsky, Y.
Schlesinger, J. Appl. Phys. 88, 2634 (2000).
5. M. S. Dresselhaus, G. Dresselhaus, Adv. Phys. 51, 1
(2002).
6. I. L. Spain, in Chemistry and Physics of Carbon, P. L.
Walker, P. A. Thrower, Eds. (Dekker, New York,
1981), pp. 119304.
7. O. A. Shenderova, V. V. Zhirnov, D. W. Brenner, Crit.
Rev. Solid State Mater. Sci. 27, 227 (2002).
8. A. Krishnan et al., Nature 388, 451 (1997).
9. E. Dujardin, T. Thio, H. Lezec, T. W. Ebbesen, Appl.
Phys. Lett. 79, 2474 (2001).
10. H. Shioyama, J. Mat. Sci. Lett. 20, 499 (2001).
11. Other methods of preparing thin graphitic layers
exist. The closest analogs of FLG are nanometer-sized
patches of graphene on top of pyrolytic graphite (12,
13), carbon films grown on single-crystal metal
substrates (14), and mesoscopic graphitic disks with
thickness down to 60 graphene layers (8, 9).
12. A. M. Affoune et al., Chem. Phys. Lett. 348, 17
(2001).
13. K. Harigaya, Y. Kobayashi, K. Takai, J. Ravier, T. Enoki,
J. Phys. Cond. Matter 14, L605 (2002).
14. T. A. Land, T. Michely, R. J. Behm, J. C. Hemminger,
G. Comsa, Surf. Sci. 264, 261 (1992).
16. J. Kong et al., Science 287, 622 (2000).
17. M. Kruger, I. Widner, T. Nussbaumer, M. Buitelaar,
C. Schonenberger, N. J. Phys. 5, 138 (2003).
18. We believe that our thinnest FLG samples (as in
Fig. 2A) are in fact zero-gap semiconductors,
because small nonzero values of &( found exper-
imentally can be attributed to inhomogeneous
doping, which smears the zero-gap state over a
small range of V
g
and leads to finite apparent &(.
19. M. R. Stan, P. D. Franzon, S. C. Goldstein, J. C. Lach,
M. M. Zeigler, Proc. IEEE 91, 1940 (2003).
20. Supported by the UK Engineering and Physical
Sciences Research Council and the Russian Acad-
emy of Sciences (S.V.M., S.V.D.). We thank L. Eaves,
E. Hill, and O. Shklyarevskii for discussions and
interest.
DC1
SOM Text
Figs. S1 to S11
Hydrated Electron Dynamics:
From Clusters to Bulk
A. E. Bragg,
1
J. R. R. Verlet,
1
A. Kammrath,
1
O. Cheshnovsky,
2
D. M. Neumark
1,3
*
The electronic relaxation dynamics of size-selected (H
2
O)
n
/(D
2
O)
n
[25 e
n e 50] clusters have been studied with time-resolved photoelectron imaging.
The excess electron (e
c
) was excited through the e

c
( p) @ e
c
(s) transition with

an ultrafast laser pulse, with subsequent evolution of the excited state
monitored with photodetachment and photoelectron imaging. All clusters
exhibited p-state population decay with concomitant s-state repopulation
(internal conversion) on time scales ranging from 180 to 130 femtoseconds
for (H
2
O)
n
and 400 to 225 femtoseconds for (D

2
O)
n
; the lifetimes decrease

with increasing cluster sizes. Our results support the nonadiabatic relaxa-
tion mechanism for the bulk hydrated electron (e
aq
), which invokes a 50-

femtosecond e
aq
( p) Y e
aq
(s
.
) internal conversion lifetime.
A free electron introduced into a polar sol-
vent, such as water (1) or ammonia (2), may
be trapped by locally oriented solvent mole-
cules. In water, an Bequilibrated[ hydrated
electron Ee
aq
(s)^ can be transiently confined

within a roughly spherical cavity defined by
six OH bonds oriented toward the negative
charge distribution in the so-called Kevan
geometry (35). The hydrated electron is an
important reagent in condensed-phase chem-
istry and molecular biology, as it participates
in radiation chemistry, electron transfer, and
charge-induced reactivity. Thus, research in-
vestigating the dynamics of this species,
whether in the presence or absence of other
reagents, has attracted considerable attention
in the theoretical and experimental physical
chemistry communities. Here, we present
time-resolved results on the electronic relax-
ation dynamics of anionic clusters of water
that lend profound insight to the elucidation of
hydrated electron dynamics in the bulk.
The electronic energetics of a hydrated
electron are characterized by three types of
states (Fig. 1A): a localized e
aq
(s) ground
state; three localized, near-degenerate e
aq
( p)
excited states; and a delocalized conduction
band (CB) characterized by a charge distri-
bution spread across hundreds of molecules
in the solvent Bnetwork.[ The visible ab-
sorption spectrum of the equilibrium hy-
drated electron, a broad band peaking at
720 nm (1), is well understood as an ex-
citation from the occupied e
aq
(s) state to the

vacant e
aq
( p) states (4, 5). The electron-

solvent dynamics subsequent to e
aq
( p) @
e
aq
(s) excitation are more controversial, how-

ever, in spite of considerable experimental
(612) and theoretical (13, 14) effort devoted
to this problem.
Transient absorption measurements made
with femtosecond (fs) laser pulses as short as 5
fs (12) show a near infrared (NIR) absorption
band beyond 900 nm developing on a 40- to
50-fs time scale after e
aq
( p) @ e
aq
(s)
excitation. This broad feature shifts back to
shorter wavelengths on a time scale of several
hundred fs, with recovery of the original e
aq
(s)
absorption spectrum largely complete within
1 picosecond (ps). Deuteration of the
solvent (8, 12) appears only to affect the
fastest measured time scale (i.e., the buildup
time of the transient NIR absorption), with
C
D2O
/C
H2O
0 1.4 to 1.6. As described by
Yokoyama et al. (8), two rather different
energy relaxation mechanisms have been
proposed to account for these observations.
In the Badiabatic solvation[ scheme (13, 14),
the infrared transient at the earliest times is
attributed to absorption of the e
aq
( p) electron,
which is solvated on the upper state within 50
fs (process x in Fig. 1A). The excited electron
then undergoes internal conversion (IC) to the
ground state (y) on a 400-fs time scale,
generating ground-state electrons that further
relax on the 1-ps time scale by dissipating
energy to the solvent. In contrast, the
Bnonadiabatic relaxation[ mechanism (7, 9,
12) invokes much more rapid IC, on a 50-fs
time scale, and attributes the transient NIR
band at early times to absorption of the
ground-state electron in a vibrationally excited
solvent environment. In this model, subsequent
dynamics are assigned to reorganization of the
local (400 fs) and extended (1 ps) solvent
network following the electronic decay.
We present an alternative and comple-
mentary approach to assessing the relaxation
dynamics of the hydrated electron through
time-resolved photoelectron imaging (TRPEI)
studies (15) of electron dynamics in size-
selected water cluster anions, (H
2
O)
n
and
(D
2
O)
n
, with 25 e n e 50. Water cluster

anions were first detected mass spectromet-
1
Department of Chemistry, University of California,
2
School of Chemistry, The
Sackler Faculty of Exact Sciences, Tel-Aviv University,
69978 Israel.
3
Chemical Sciences Division, Lawrence
Berkeley National Laboratory, Berkeley, CA 94720,
USA.
E-mail: dan@radon.cchem.berkeley.edu
R E P O R T S
rically (16) and have since been studied with
photoelectron (1719), vibrational (20), and
electronic (21) spectroscopies. Cluster studies
offer access to structural and dynamic details
that are often obscured in the larger, more
complex bulk water environment. Because
insights gleaned from clusters are meant to
extrapolate to the bulk, there is keen interest
in determining the size at which Bbulk-like[
behavior of (H
2
O)
n
clusters becomes appar-

ent (22), with the excess electron surrounded
by water rather than localized on the cluster
surface (2325).
In our TRPEI experiment, described in more
detail elsewhere (26), water cluster anions are
generated in an ion source based on a fast,
pulsed molecular beam (27) and are mass-
selected by time of flight. The mass-selected
clusters are investigated with the pump-probe
scheme illustrated by the right-most arrows in
Fig. 1B. Here, the vertical detachment energy
(VDE) is the energy required to remove an
electron from an anion while maintaining the
anion cluster geometry. We use an 100-fs
laser pulse at 1250 nm (1.0 eV) to excite e
c
, the
excess electron in the cluster anion, from the
ground s-state to the excited p-state; this energy
lies close to the absorption maximum in the
cluster size range investigated here (21). After a
delay, %I, during which relaxation may occur,
the electron is photodetached by an ultraviolet
probe pulse (400 nm or 3.1 eV, 100 fs) and
accelerated by a dc field to an imaging detector
composed of a 70-mm-diameter microchannel
plate assembly coupled to a phosphor screen.
The state of e
c
prior to detachment is
reflected in its residual velocity afterward,
which in turn determines the position the
electron strikes on the two-dimensional detec-
tor plane. This Bvelocity map image[ (28) is
then analyzed by two- to three-dimensional
inversion, radial integration, and radius-to-
energy transformation to yield time-dependent
photoelectron kinetic energy (eKE) distribu-
tions. Thus, we obtain a time-resolved series of
snapshots revealing the dynamics of the
nonstationary states created by the pump pulse.
A photoelectron image of (D
2
O)
25
ac-
quired within the 1250 nm 400 nm tempo-
ral overlap reveals two features of interest
(Fig. 2A). The outermost ring, labeled I, corre-
sponds to the signal with the highest eKE and
results from two-color (1250 nm 400 nm)
resonant photodetachment via the e
c
( p) state
(right arrows, Fig. 1B). The next ring, II, is
from direct 400-nm detachment of the anion
ground state (left arrow, Fig. 1B). Additional
features are seen from resonant two-photon
and direct one-photon detachment at 1250 nm;
however, only features I and II of Fig. 2A
show any dependence on pump-probe delay.
The time-resolved photoelectron (TRPE)
spectra showing eKE(%I) for features I and
II (Fig. 2B) highlight the dynamics that ensue
after excitation to the e
c
( p) cluster state. TRPE

spectra of (D
2
O)
25
and all other clusters

studied here are characterized by depletion
and partial recovery in intensity of feature II,
with simultaneous growth and decay of feature
I. Feature I grows as the pump pulse transfers
electronic population to the excited state of the
cluster, whereas the depletion of feature II
results mainly from this same population
transfer. The decay of I and the partial recovery
of II are unambiguously assigned to direct
electronic relaxation, i.e., internal conversion
(IC), from the excited state(s) to the ground
state of the cluster. We note that direct one-
photon detachment at 1250 nm, the innermost
feature in Fig. 2A, is seen for all clusters
studied here, which indicates that autodetach-
ment from the pump-excited e
c
(p) state (requir-

ing no probe pulse) is energetically possible and
can, in principle, compete with internal conver-
sion as an excited-state decay pathway. How-
ever, this feature shows no time dependence,
so excited-state autodetachment appears to
be at most a minor relaxation channel.
The integrated intensity of features I
and II was plotted versus pump-probe
delay and fit to single exponential decay
functions convoluted with the instrumental
temporal-response (pump-probe temporal
cross-correlation, 150 fs) of the experiment
(Fig. 2C). These fits give a e
c
( p) Y e
c
(s
.
)
relaxation lifetime of 398 T 50 fs in (D
2
O)
25
,
where e
c
(s
.
) denotes the electron in a vibra-
tionally excited ground state. The shape of
feature I remains unchanged even as its
intensity evolves, indicating the absence of
noticeable hydration dynamics in the e
c
( p)
state before e
c
( p) Y e
c
(s
.
) relaxation.
The size and isotope dependence of the
electronic decay are evident when the decay
traces of excited (D
2
O)
25
, (D
2
O)
45
, and
(H
2
O)
45
are plotted together (Fig. 3). These

clusters exhibit excited-state lifetimes of
398 T 50 fs, 240 T 25 fs, and 132 T 10 fs,
respectively. For all clusters studied, both sets
of isotopomers exhibit shortened excited-state
lifetimes with increased cluster size (Fig. 4).
Size-dependent isotopomer lifetime ratios,
C
D2O
/C
H2O
, for n 0 25, 35, and 45 have values
of 2.1 T 0.4, 1.8 T 0.2, and 1.8 T 0.2,
respectivelysimilar in magnitude to those
Fig. 1. Energetics of an electron
in water. (A) The bulk hydrated
electron. Red arrows illustrate
the transient-absorption schema
typically used to study bulk
dynamics, in which a time-
delayed NIR laser pulse probes
the change in optical density
[%OD(%C)] of the sample after
an initial infrared excitation. (B)
Water cluster anions. Levels are
e
-
aq
(s)
x
z
y
eKE()
x
y
I
I I
A B
BULK CLUSTER
CB
e
-
aq
(p)
e
-
c
(p)
e
-
c
(s)
VDE
OD()
similar to those of (A), although the photodetachment continuum replaces the conduction band.
I
I I
I I
I
I
I I
C
Pump-probe delay (fs)
0 2000 4000 6000
I
n
t
e
g
r
a
t
e
d

i
n
t
e
n
s
i
t
y

(
a
r
b
.

u
n
i
t
s
)
3.0
2.5
2.0
1.5
0.2
0.0
A
Electronic kinetic energy (eV)
R
e
l
a
t
i
v
e

I
n
t
e
n
s
i
t
y
T
i
m
e

(
f
s
)
B
1.0 2.0 3.0
30000
3090
1345
810
475
240
73
-95
-930
Fig. 2. Time-resolved photoelectron imag-
ing of water cluster anions. (A) Photoelec-
tron image of (D
2
O)
25
collected at the
1250 nm 400 nm pump-probe temporal
overlap. Features I and II arise from the two-
color resonant and direct 400-nm photo-
detachment processes shown in Fig. 1B.
(B) Time-resolved photoelectron spectra of
(D
2
O)
25
. Features highlighted in (A) are dis-

played with pump-probe delays indicated.
(C) Features I and II. Integrated intensities
of features I and II versus pulse delay (experiment, circles; fit, black lines). The fit curves produce
a 398 T 50 fs electronic lifetime for the excited cluster.
R E P O R T S
observed for the fastest dynamics measured in
condensed-phase experiments (8, 12). To test
the dependence of the excited-state lifetime on
excitation energy, (H
2
O)
45
was investigated
at a pump energy of 1.57 eV. The lifetime,
117 T 15 fs, is very close to that found at
1.0 eV and is consistent with the upper bound
previously determined by Weber et al. (19).
These results may be compared and
contrasted with the two proposed bulk re-
laxation mechanisms discussed above. The
upper state internal conversion lifetimes de-
crease markedly with increasing cluster size,
approaching 100 fs for the largest (H
2
O)
n
clusters studied here (Fig. 4). This trend

suggests even faster IC in bulk water, thereby
supporting the nonadiabatic relaxation mech-
anism in which IC occurs on a 50-fs time
scale, rather than the adiabatic solvation mod-
el in which a 400-fs IC has been assigned.
Indeed, a plot of IC lifetime versus 1/n,
shown in Fig. 4, is linear and extrapolates
to 50 fs in the bulk limit. The significant
isotope effect observed for IC in clusters,
evident in the (D
2
O)
n
lifetimes plotted in
Fig. 4, is likewise consistent with the non-
adiabatic mechanism, in which a similar iso-
tope effect is attributed to the IC step (12). In
contrast, the adiabatic solvation mechanism
attributes the isotope effect to faster excited-
state hydration in H
2
O, whereas the process
attributed to IC is insensitive to the identity of
the isotopomer (8). Finally, the absence of
observable upper-state hydration dynamics in
our experiment, which would be observed by
changes in excited-state spectral features with
time, is also in accord with the nonadiabatic
mechanism. In this study, all observed dy-
namical characteristics support the nonadia-
batic relaxation model.
A primary motivation for undertaking
studies of water cluster anions was to resolve
ambiguities regarding the relaxation dynamics
of the hydrated electron in solution. We have
shown that these ambiguities are directly
addressed by investigating the cluster environ-
ment, because photoelectron eKE distributions
of the excited- and ground-state (H
2
O)
n
are
highly distinct (Fig. 2). Thus, internal con-
version dynamics from the upper state are
cleanly decoupled from solvent relaxation/
reorganization processes in the cluster environ-
ment, with the rates and nature of relaxation
processes free of overlapping spectral signa-
tures that complicate bulk studies. With this in
mind, our findings offer important insight into
the dynamics of the bulk hydrated electron.
The above interpretation of our results
raises the issue of whether it is appropriate to
extrapolate bulk hydrated electron dynamics
from those of modest-sized water cluster
anions. This issue is particularly important
in light of debate regarding whether the
excess electron is localized at the cluster
surface or within a solvent network cavity.
Regardless of which structure holds, it
certainly appears more reasonable to extrap-
olate the cluster relaxation rates to the faster
of the two assigned IC time scales for the
bulk hydrated electron. Nonetheless, exten-
sion of our results to the bulk dynamics is
more attractive if the clusters studied here
have interior excess electrons, because such
structures should be more akin to e
aq
. Coe
(22) has argued that n
1/3
-dependent plots of
(H
2
O)
n
VDEs, obtained from anion photo-

electron spectroscopy (n 0 11 to 69) (17), and
absorption maxima, determined from electron-
ic spectroscopy (n 0 11 to 50) (21), extrapo-
late sensibly to bulk values and support the
notion that clusters in this size range have
internal states. This interpretation disagrees
with earlier simulations by Barnett et al. (23)
that predict surface states to be more stable up
to n 0 32, with the transition to interior states
occurring by n 0 64. However, Bartels (25)
has pointed out that the model potentials used
in these simulations bind the electron too
tightly and that improved model potentials for
the e
IH
2
O interaction need to be developed to
assess the possible role of surface states.
Although our TRPEI measurements do not
settle this issue, the IC lifetimes in Fig. 4 vary
smoothly with cluster size, casting further
doubt on the existence of a structural transi-
tion in the n 0 25 to 50 range. However,
preliminary experiments in our laboratory
(29) have demonstrated that ion source
conditions can be adjusted to generate
(H
2
O)
n
(n , 50) with significantly lower

VDEs compared with the clusters discussed
here. These newly observed clusters may well
represent the long-sought surface states, in
which case the cluster dynamics presented in
this paper are most likely from structures in
which the excess electron has appreciable
amplitude within the cluster.
1. E. J. Hart, J. W. Boag, J. Am. Chem. Soc. 84, 4090 (1962).
2. W. Weyl, Ann. Phys. (Leipzig) 197, 601 (1863).
3. L. Kevan, Acc. Chem. Res. 14, 138 (1981).
4. P. J. Rossky, J. Schnitker, J. Phys. Chem. 92, 4277 (1988).
5. M. Boero, M. Parrinello, K. Terakura, T. Ikeshoji, C. C.
Liew, Phys. Rev. Lett. 90, 226403 (2003).
6. A. Migus, Y. Gauduel, J. L. Martin, A. Antonetti, Phys.
Rev. Lett. 58, 1559 (1987).
7. Y. Kimura, J. C. Alfano, P. K. Walhout, P. F. Barbara, J.
Phys. Chem. 98, 3450 (1994).
9. A. Hertwig, H. Hippler, A. N. Unterreiner, P. Vohringer,
Ber. Bunsenges. Phys. Chem. 102, 805 (1998).
10. M. Assel, R. Laenen, A. Laubereau, Chem. Phys. Lett.
317, 13 (2000).
11. M. J. Tauber, R. A. Mathies, J. Am. Chem. Soc. 125,
1394 (2003).
12. M. S. Pshenichnikov, A. Baltuska, D. A. Wiersma,
Chem. Phys. Lett. 389, 171 (2004).
13. B. J. Schwartz, P. J. Rossky, J. Chem. Phys. 101, 6902
(1994).
14. S. Bratos, J. C. Leicknam, D. Borgis, A. Staib, Phys.
Rev. A. 55, 7217 (1997).
15. A. Stolow, A. E. Bragg, D. M. Neumark, Chem. Rev.
104, 1719 (2004).
16. M. Armbruster, H. Haberland, H. G. Schindler, Phys.
Rev. Lett. 47, 323 (1981).
18. J. Kim, I. Becker, O. Cheshnovsky, M. A. Johnson,
Chem. Phys. Lett. 297, 90 (1998).
20. P. Ayotte et al., J. Chem. Phys. 110, 6268 (1999).
21. P. Ayotte, M. A. Johnson, J. Chem. Phys. 106, 811
(1997).
22. J. V. Coe, Int. Rev. Phys. Chem. 20, 33 (2001).
23. R. N. Barnett, U. Landman, C. L. Cleveland, J. Jortner,
J. Chem. Phys. 88, 4429 (1988).
24. A. Khan, J. Chem. Phys. 118, 1684 (2003).
25. D. M. Bartels, J. Chem. Phys. 115, 4404 (2001).
26. A. E. Bragg, R. Wester, A. V. Davis, A. Kammrath,
D. M. Neumark, Chem. Phys. Lett. 376, 767 (2003).
27. U. Even, J. Jortner, D. Noy, N. Lavie, C. Cossart-
Magos, J. Chem. Phys. 112, 8068 (2000).
28. A. T. J. B. Eppink, D. H. Parker, Rev. Sci. Instrum. 68,
3477 (1997).
29. J. R. R. Verlet, A. Kammrath, A. E. Bragg, D. M.
Neumark, unpublished data.
30. This research is supported by the National Science
Foundation under grant no. CHE-0350585. Addition-
al support from the U.S.-Israel Binational Science
Foundation is gratefully acknowledged.
2 August 2004; accepted 27 August 2004
10.1126/science.1103527
N
o
r
m
a
l
i
z
e
d

i
n
t
e
n
s
i
t
y
Pump-probe delay (fs)
cluster data fit
1.0
0.8
0.6
0.4
0.2
0.0
-500 0 500 1000 1500 2000 2500
(D
2
O)
25
-
(D
2
O)
45
-
(H
2
O)
45
-
Fig. 3. Size and isotope dependence of excited-
state electronic decay traces. Fits (solid lines) to
the integrated photoelectron intensities (sym-
bols) versus time yield excited-state lifetimes
for (D
2
O)
25
, (D
2
O)
45
, and (H
2
O)
45
of 398 T
50 fs, 240 T 25 fs, and 132 T 10 fs, respectively.
0
100
200
300
400
500

n
L
i
f
e
t
i
m
e

(
f
s
)
1/n
0.00 0.01 0.02 0.03 0.04 0.05
Fig. 4. Size-dependent lifetime trends for
(D
2
O)
n
and (H
2
O)
n
. Size-dependent decay life-

times of (D
2
O)
n
(squares) and (H
2
O)
n
(circles)
with inverse cluster size (1/n). IC lifetimes linearly
extrapolate (black dashed lines) to a bulk (1/n 0
0) decay time scale of 50 fs (70 fs) in H
2
O
(D
2
O). Investigated cluster sizes (n) are labeled.
R E P O R T S
Electrons in Finite-Sized Water
Cavities: Hydration Dynamics
Observed in Real Time
D. Hern Paik, I-Ren Lee, Ding-Shyue Yang, J. Spencer Baskin,
Ahmed H. Zewail*
We directly observed the hydration dynamics of an excess electron in the
finite-sized water clusters of (H
2
O)
n
j
with n 0 15, 20, 25, 30, and 35. We
initiated the solvent motion by exciting the hydrated electron in the cluster.
By resolving the binding energy of the excess electron in real time with
femtosecond resolution, we captured the ultrafast dynamics of the electron in
the presolvated (wet) and hydrated states and obtained, as a function of
cluster size, the subsequent relaxation times. The solvation time (300
femtoseconds) after the internal conversion [140 femtoseconds for (H
2
O)
35
j
]
was similar to that of bulk water, indicating the dominant role of the local
water structure in the dynamics of hydration. In contrast, the relaxation in
other nuclear coordinates was on a much longer time scale (2 to 10 pico-
seconds) and depended critically on cluster size.
The nature of the solvated electron, which
was first observed in liquid ammonia in
1864, continues to pose several fundamen-
tal problems. When the solvent medium is
water, the hydrated electron becomes essen-
tial to a myriad of physical, chemical, and
biological processes. In a simple picture of
an electron in a cavity, the description of the
hydrated electron state structure is analogous
to that of a hydrogen atom, with a ground
state of s-type and an excited state of p-type
character. However, the hydrated electron is
far more complex, because of the ultrafast
dynamics of structural change, solvation, and
recombination. After postulation of the
existence of the hydrated electron and the
discovery of its absorption, experimental and
theoretical efforts have focused on studies in
bulk water in which the Bcavity[ is sur-
rounded by a continuum of other water
molecules.
A key issue for understanding electron
hydration is knowledge of the time scales
involved: the motion of water molecules
toward the equilibrium structure and the life-
time of the electron in the different states it
occupies. In bulk water, early femtosecond
transient absorption studies (1, 2) resolved
electron hydration dynamics using excitation
by two ultraviolet photons to eject bound
electrons from water molecules or solute
anions. During the succeeding decade, dif-
ferent research groups have provided a vast
amount of experimental data on the time
scales of relaxation and the theoretical
underpinnings of the hydrated electron sys-
tem (312). Among these was the first three-
pulse experiment (3), in which a population of
ground-state hydrated electrons created by
an initial laser pulse was subsequently
studied using two additional pulses, the first
of which excited the electrons from the s- to
the p-state and the second of which probed
either state. More recently, studies have been
made with pulses as short as 5 fs in order to
elucidate the different relaxation pathways
(5, 9). In these bulk studies, there remain
unanswered questions, especially regarding
the microscopic molecular structure and dy-
namics of hydration.
Mesoscopic clusters (13, 14) are ideal for
forming finite-sized nanoscale water cavi-
ties for electron hydration, and because of
the charge of the electron, it is in principle
possible to select a particular size of cluster
and study its isolated structure and dynam-
ics. Results from such studies provide in-
sight into the bulk behavior. For example,
accurate spectra of neutral water dimers are
predictive of the properties of larger clusters
and bulk water (15). For electrons in water
clusters, Haberland and co-workers (16) first
reported the experimental observation of
(H
2
O)
n
j
clusters, and, in a series of compre-
hensive studies, the Johnson (1719) and
Bowen (20, 21) groups have elucidated the
size dependence of spectroscopic proper-
ties, examining the role of the core motif in
reaching bulk hydration.
The structure of these finite-sized clusters
has been studied both experimentally and
theoretically, addressing the issue of surface
and interior electron binding (18, 20, 2224).
Theoretical studies (23, 24) of small (H
2
O)
n
j
clusters, n e 14, predict that the electron lies
at the surface. Earlier calculations by the
groups of Landman and Jortner (22) indicat-
ed that for small clusters (n e 32), the
electron tends to remain on the surface,
whereas for the larger ones (n 0 64 and
128), the electron is in the interior. Recent
work for n 0 24 indicates that although three
isomers (with the electron inside or outside
at the surface) are energetically similar, the
vertical detachment energy closest to the
experimental value is that of the isomer with
the electron inside (25). Despite these
extensive studies, the only report of real-
time dynamics of water cluster anions has
been that of a preliminary p-state lifetime,
limited by laser pulse duration, for a cluster
of unspecified size (19).
We present here direct observation of the
femtosecond dynamics of electrons in water
clusters varying in size up to n 0 35. We
focused our attention on the dynamics in
systems with different solvation cavities (n 0
15, 20, 25, 30, and 35). The finite-sized
clusters were selectively intercepted by femto-
second pulses to promote the electron from
the s- to the p-state (Fig. 1A) (26). We
followed subsequent relaxation dynamics by
monitoring the evolution of the photoelectron
(PE) spectrum with kinetic energy resolution
(27). The latter proved critical, as did the
resolution of kinetic energy of ions (28), for
deciphering different pathways of dynamics.
This PE energy resolution allowed us to
address whether hydration proceeds while
the electron is in the ground (s-type) and/or
excited (p-type) states. For the cavity sizes
under study, the behavior observed can be
correlated to that of bulk-type hydration.
We generated the negatively charged wa-
ter clusters by crossing a continuous electron
beam (1 keV) with a jet of water vapor,
using nitrogen as a carrier gas at 150 to 250
kPa. After 100 6s of drift time, appli-
cation of a properly timed voltage pulse
accelerated (H
2
O)
n
j
clusters into the field-
free time-of-flight region, where the desired
size was intercepted with femtosecond laser
pulses (29). A typical cluster size distribu-
tion is shown in Fig. 1B. The laser pulses
(110 fs) were generated from a Ti:sapphire
oscillator and amplified by regenerative and
multipass amplifiers. The 800-nm laser output
was frequency doubled to generate a 400-nm
pulse. The residual 800-nm light was used as
the excitation pulse, and the 400-nm laser
pulse, delayed in time, was used as the probe
to photodetach electrons. Photoelectrons
were collected by a magnetic-bottle PE
spectrometer, and the metastable anions and
photofragments were detected by a linear
reflectron mass spectrometer. We recorded
transients by integrating the PE intensity in
selected electron kinetic energy (eKE) win-
dows as a function of the delay time.
A conceptual illustration of our experi-
ments and methodology is shown in Fig. 2.
Arthur Amos Noyes Laboratory of Chemical Physics,
Laboratory for Molecular Sciences, California Institute
of Technology, Pasadena, CA 91125, USA.
E-mail: zewail@caltech.edu
R E P O R T S
When a particular size of (H
2
O)
n
j
is selec-
tively excited by the 800-nm femtosecond
laser pulse, the excess electron is promoted
to the p-state (Fig. 2A). The coordinate-
labeled solvation represents all nuclear mo-
tions of the solvent that strongly affect the
electrostatic environment and thus the en-
ergy of the electron. As a result of the
electronic transition, the electron charge dis-
tribution is significantly changed, driving a
rearrangement of water molecules around
the electron that corresponds to displace-
ment along the solvation coordinate with
characteristic time I
p
. Similarly, when the
excited state relaxes down to the ground
state by internal conversion with character-
istic time I
ic
, displacement along the solva-
tion coordinate will reverse, and the solvent
will move back toward its original configu-
ration with characteristic time I
s
. Because
the pump photon energy, which is several
times greater than the binding energy of
surface water molecules (30), is retained in
these isolated clusters, the flow of energy
from the solvation coordinate into other nu-
clear coordinates will lead to the eventual
evaporation of water molecules (Fig. 2A,
left). The fact that 800-nm absorption ul-
timately leads to evaporation is confirmed by
our mass spectra of anionic fragments and
has been reported elsewhere (31).
The PE spectra were used to follow these
dynamics and disentangle the pathways. The
energy level diagram (Fig. 2B) shows qual-
itatively how the energy content in the sol-
vation coordinate of the anion is expected to
affect the eKE of the electron ejected upon
absorption of the 400-nm probe pulse. Franck-
Condon considerations indicate that the eKE
distribution broadens asymmetrically as the
amount of energy in the solvation coordinate
increases in a given electronic state (yellow
to red); a change in electronic state causes
the spectrum to shift position (yellow to
blue). The degree of broadening depends on
Fig. 2. (A) A schematic illustra-
tion of the solvation and evapo-
rative dissociation exhibited in
the hydrated electron clusters.
The upward arrow represents the
excitation to the p-state. The
relaxation and dissociation path-
ways are depicted and labeled by
their characteristic times as fol-
lows: I
p
, solvation in p-state, I
ic
,
internal conversion, I
s
, solvation
in s-state, I
r
, relaxation and evap-
oration. h3, photon energy. (B)
Probing of different transient
states of the solvated electron by
energy and time resolution. Left:
An energy-level diagram illus-
trating the changes in PE distri-
bution originating from anion
populations with different elec-
tronic and solvation energies.
The upward arrows indicate the
400-nmprobe pulse that detached
the electron, and the downward
arrows correspond to the kinetic
energy of the detached electron
(the eKE). Right: The resulting PE
spectra are plotted together on a
common energy axis, with a, b,
and c indicating points with differ-
ing sensitivities to the location of the transient state. The distribution shown is for a moderate weak-
ening of the force constant and displacement of minimum position between the anion and the neutral
potentials, as confirmed by Franck-Condon calculation. Asterisk indicates an excited electronic state.
Fig. 3. (A) Top: PE
spectrum of (H
2
O)
35
j
obtained by irradiation
with the 400-nm pulse
only. Bottom: Time-
dependent PE differ-
ence spectra at several
time delays of the
400-nm probe pulse
relative to the 800-nm
excitation pulse. Each
spectrum was con-
structed by subtracting
the reference at 100 ps
from the PE spectrum
at the time specified.
Regions a, b, and c
indicate the energy
windows of interest.
(B) Three-dimensional
representation of the
time-dependent PE
spectra. The intensity
trends in region a, b,
and c are indicated by
yellow, red, and blue
ribbons. (C) Femto-
second transients of
(H
2
O)
30
j
and (H
2
O)
35
j
obtained by integra-
tion of the three dif-
ferent gated regions (a,
b, and c) as a function
of delay time. The tran-
sients of region b for
(D
2
O)
30
j
and (D
2
O)
35
j
are
also shown.
Fig. 1. (A) Schematic representation of the s-
and p-states of the hydrated electron, based on
fig. 14 of (5). (B) Experimental mass spectrum
of (H
2
O)
n
j
generated by the ion source in our
apparatus. A series of O
2
j
(H
2
O)
n
peaks is seen
at low mass. amu, atomic mass units.
R E P O R T S
the relative flatness and equilibrium position
displacement of the neutrals and anions
solvation coordinate potentials. With this
picture in mind, solvation dynamics in both
the p- and s-states can be followed by mon-
itoring the dependence on probe delay time
of the production of detached electrons with
various values of eKE (e.g., the energies
labeled a, b, and c in Fig. 2B).
The PE spectrum of (H
2
O)
35
j
upon ir-
radiation by the 400-nm femtosecond pulse
confirmed ejection of the excess electron by
vertical detachment, producing the char-
acteristic eKE distribution (Fig. 3A, top).
When the clusters were irradiated by both
800-nm and 400-nm pulses, the PE spec-
trum changed with delay time, as indicated
in the difference spectra (Fig. 3A, bottom).
The same spectra are also shown in a three-
dimensional representation (Fig. 3B). It is
apparent that different regions of the PE
spectrum exhibit distinct temporal behaviors,
and we focused only on the three particular
regions labeled a, b, and c. As indicated by
the ribbons in Fig. 3B, at high eKE above the
onset of the PE spectrum (region a), a new
peak appeared at time zero and disappeared
by 0.7 ps. The PE intensity near the onset
(region b) decayed with time, whereas the
PE signal near the peak of the spectrum
(region c) displayed an abrupt drop at time
zero and a rise at positive time.
To quantify the trends shown by the rib-
bons in Fig. 3B, we obtained transients by
integrating the PE intensity as a function of
the delay time for each energy window. The
same forms of apparent temporal behaviors
were evident for both n 0 30 and n 0 35
clusters (Fig. 3C). The greater level of detail
provided by the transients revealed that the
fast decay for region a is not limited by laser
pulse duration, as the asymmetry is evident,
and there are two distinct time scales of decay
for region b. We also compared with region b
results obtained for (D
2
O)
30
j
and (D
2
O)
35
j
.
For (H
2
O)
n
j
, n 0 15, 20, and 25, only
transients for region b were measured (Fig.
4A) (32).
The temporal behavior observed in the
time-dependent PE spectra of (H
2
O)
30
j
and
(H
2
O)
35
j
elucidates the ultrafast dynamics.
The instantaneous rise and the fast decay in
region a (at high eKE) represent vertical
excitation to the p-state followed by rapid
relaxation to the s-state. As discussed earlier,
an isolated peak of a-type character should
rise within our pulse duration and decay
with the p-state lifetime. Moreover, this peak
temporally changes at an eKE shift of 1 eV,
which corresponds to the s- to p-state energy
gap (19), as predicted conceptually (Fig. 2B).
In bulk water (4, 5, 33), the inertial solvent
motions (libration) are expected to occur
before the internal conversion. In the clusters,
the decay of the a-peak gives the time scale
for population transfer to the s-state, with an
effective time constant, I
eff
, for inertial
motion and internal conversion (I
p
and I
ic
).
Single exponential fits to the region a data
give corresponding C
eff
values of 170 fs (n 0
30) and 140 fs (n 0 35) (Fig. 3C). Given the
short time scale of these motions, coherent
effects may be present (4, 34), but a kinetic
description suffices for the behavior ob-
served here.
After the internal conversion, the electron
becomes presolvated in the ground state, and
solvation of the electron (I
s
) follows, as
revealed by the short time behavior in region
b and c. Relaxation (I
r
) after solvation can be
most clearly seen in the long time behavior
in region b (Fig. 4A, right). To account for
the delayed return of population from the
p- to the s-state, a proper kinetic analysis
must also include the influence of I
eff
on the
transients of regions b and c (35). The results
(Figs. 3C and 4A) give I
s
values of 300 fs
(the range for the clusters studied was T150
fs with our current analysis) and I
r
values
ranging from 2 to 10 ps, depending on
cluster size (Fig. 4A). For the different
clusters n 0 15 to n 0 35 (Fig. 4B), the time
scale of solvation was within a factor of two
and was similar to that of electrons in bulk
water. It was also on the order of the time
scale of the diffusive rotational and transla-
tional motions of bulk water around molec-
ular probes (33). The solvent rearrangement
time we obtained here for the s-state of these
large clusters is about the same as that found
for electron hydration after excitation of a
charge transfer band of I
(H
2
O)
n
for n 0 5
and n 0 6 (36). Such weak size dependence
and the similarity between bulk and clusters
indicate that the dynamics of hydration are
in large part controlled by the local struc-
ture of water molecules in immediate contact
with the electron.
The longer relaxation times are deter-
mined by the energy content in the s-state,
the rate of intramolecular vibrational ener-
gy redistribution (IVR), and hydrogen-bond
breakage (Fig. 4, B and C). Because these
times increase with cluster size, we excluded
both time-dependent solvation and IVR as
rate-determining and considered evaporation.
Manifestation of evaporation in the transients
was expected because the process alters the
PE spectra as a result of change in cluster
size and internal energy. We calculated the
statistical rate constants for clusters under-
going evaporation by one water molecule
(1/I
evp
). Using Rice-Ramsperger-Kassel-
Marcus theory (37), we obtained rate
constants that were slower or much slower
than the experimental observation, depend-
ing on the particular values of frequencies
and the reaction barrier; for example, for
n 0 35, I
evp
ranged from 1 ns to 15 6s
versus the observed value of 10 ps. The
discrepancy would indicate a nonstatistical
behavior (40). However, there have been re-
ports that such clusters may live for much
Fig. 4. (A) Femtosecond transients of (H
2
O)
n
j
(where n 0 15, 20, 25, 30, or 35) obtained by
integration of the PE signals in region b. The short- and long-range scans are shown in the left and
right columns, respectively. (B) Time constants I
s
(squares) and I
r
(circles) versus cluster size n.
The solid curves indicate the observed trends. (C) Observed relaxation rate 1/I
r
(circles) and the
scaled RRKM rates for the hydrogen-bond breakage 1/I
evp
(triangles). The RRKM rate is scaled by
11,000 to match the observed rate for n 0 35. The solid curves indicate the observed trend, and
the dashed curve indicates the theoretical trend.
R E P O R T S
longer times (31), and because the internal
energy needs to be determined, the results
for the (H
2
O)
n
j
and (D
2
O)
n
j
systems should
be extended with variation of energy and
cluster initial temperature. The experimental
time scale for relaxation (2 to 10 ps) of the
clusters at finite temperature is not that dif-
ferent from that of hydrogen-bond making/
breaking dynamics in bulk at room temper-
ature (33).
Our observations demonstrate that solva-
tion dynamics in mesoscopic hydrated elec-
tron clusters can be probed directly in real
time. From the energy- and time-resolved
PE spectra, we were able to follow the ultra-
fast processes that occur in the presolvated
and hydrated states of the electron. With size-
selection capability, we also observed the
behavior of the rates of solvation and relax-
ation as a function of cluster size. For the
clusters studied, our time-resolved data
display solvation dynamics similar to those
of the hydrated electron in bulk water,
suggesting a local-water-structure model for
hydration, and the pathways of electronic
relaxation, solvation, and hydrogen-bond
breakage have distinctly resolvable time
scales. Mesoscopic scale clusters can thus
provide the elementary dynamics and, as such,
represent simplified model systems to study
the behavior of bulk systems.
1. A. Migus, Y. Gauduel, J. L. Martin, A. Antonetti, Phys.
Rev. Lett. 58, 1559 (1987).
2. F. H. Long, H. Lu, K. B. Eisenthal, Phys. Rev. Lett. 64,
1469 (1990).
3. J. C. Alfano, P. K. Walhout, Y. Kimura, P. F. Barbara,
J. Chem. Phys. 98, 5996 (1993).
5. A. Baltus ka, M. F. Emde, M. S. Pshenichnikov, D. A.
Wiersma, J. Phys. Chem. A 103, 10065 (1999).
6. M. Assel, R. Laenen, A. Laubereau, Chem. Phys. Lett.
317, 13 (2000).
7. N. A. Anderson, K. Hang, J. B. Asbury, T. Lian, Chem.
Phys. Lett. 329, 386 (2000).
8. A. Hertwig, H. Hippler, A.-N. Unterreiner, Phys.
Chem. Chem. Phys. 4, 4412 (2002).
9. M. S. Pshenichnikov, A. Baltus ka, D. A. Wiersma,
Chem. Phys. Lett. 389, 171 (2004).
10. E. Neria, A. Nitzan, R. N. Barnett, U. Landman, Phys.
Rev. Lett. 67, 1011 (1991).
11. C.-Y. Yang, K. F. Wong, M. S. Skaf, P. J. Rossky, J. Chem.
Phys. 114, 3598 (2001).
12. C. Nicolas, A. Boutin, B. Le vy, D. Borgis, J. Chem.
Phys. 118, 9689 (2003).
13. A. W. Castleman Jr., K. H. Bowen, J. Phys. Chem. 100,
12911 (1996).
14. M. A. Johnson, W. C. Lineberger, in Techniques for
The Study of Ion-Molecule Reactions, J. M. Farrar,
W. H. Saunders Jr., Eds. (Wiley-Interscience, New York,
1988), vol. 20, pp. 591635.
15. N. Goldman, R. J. Saykally, J. Chem. Phys. 120, 4777
(2004).
16. M. Armbruster, H. Haberland, H. G. Schindler, Phys.
Rev. Lett. 47, 323 (1981).
17. P. Ayotte, M. A. Johnson, J. Chem. Phys. 106, 811
(1997).
J. Chem. Phys. 88, 4429 (1988).
23. A. L. Sobolewski, W. Domcke, Phys. Chem. Chem.
Phys. 5, 1130 (2003).
24. J. Kim et al., J. Chem. Phys. 106, 10207 (1997).
25. A. Kahn, J. Chem. Phys. 121, 280 (2004).
26. The Fig. 1A depictions and use of the standard s and
p designations for the solvated electron states in
bulk cavities are not meant to imply that, for all
clusters, the excess electron must be bound in the
interior of the clusters.
27. A. Stolow, A. E. Bragg, D. M. Neumark, Chem. Rev.
104, 1719 (2004).
28. D. P. Zhong, A. H. Zewail, J. Phys. Chem. A 102, 4031
(1998).
29. D. H. Paik, N. J. Kim, A. H. Zewail, J. Chem. Phys. 118,
6923 (2003).
30. S. T. Arnold, R. A. Morris, A. A. Viggiano, M. A.
Johnson, J. Phys. Chem. 100, 2900 (1996).
31. P. J. Campagnola, L. A. Posey, M. A. Johnson, J. Chem.
Phys. 95, 7998 (1991).
32. Each transient of regions b and c also includes a fast
response-limited negative component due to the
vertical detachment by the 800-nm pump pulse the
photon energy of which exceeds that of the
detachment to the neutral. For small clusters (n 0
15, 20, and 25), the photodetachment process is
increasingly favored over the transition to the bound
p-state because of spectral shift (19), and this
accounts for the small amplitude of the decay
component (region b) and the fact that neither the
peak near time zero in region a nor the rise
component in region c were detected.
33. S. K. Pal, A. H. Zewail, Chem. Rev. 104, 2099 (2004).
34. A. Kummrow, M. F. Emde, A. Baltus ka, M. S.
Pshenichnikov, D. A. Wiersma, J. Phys. Chem. A
102, 4172 (1998).
35. We employed the scheme of consecutive popu-
lation flow and selective population window (PE
probing) and performed least-squares fitting. With
experimentally determined I
eff
values for the ini-
tial state decay and the rise of the first interme-
diate of n 0 30 and n 0 35, the separation in time
scales of the femtosecond and picosecond decays
allows values of I
s
and I
r
to be extracted by the
global fits. The validity of the model was sup-
ported by the pronounced dip in the region b
transients of (D
2
O)
30
j
and (D
2
O)
35
j
(Fig. 3C).
36. L. Lehr, M. T. Zanni, C. Frischkorn, R. Weinkauf, D. M.
Neumark, Science 284, 635 (1999).
37. Vibrational frequencies were estimated from those
calculated for small clusters (38, 39). Calculation for
different size clusters differed only in the number
of modes used. Thus, it is assumed that signifi-
cant structural changes, for instance transition
from the surface to interior states (20, 22), do not
occur as cluster size increases. The reaction co-
ordinate was taken to be 260 cm
j1
(39), the dis-
sociation energy to be 2700 cm
j1
(15), and the
vibrational frequencies to remain the same at the
transition state. The initial internal energy of the
cluster is assumed to be the highest thermal energy
of stable clusters (dissociation energy); the drift time
before excitation is 100 6s, but clusters with higher
internal energy may persist during this time. For
the degeneracy factor of the reaction coordinate,
we take half of the water molecules to be surface-
bound and thus subject to dissociation.
38. D. M. A. Smith, J. Smets, Y. Elkadi, L. Adamowicz, J.
Chem. Phys. 107, 5788 (1997).
39. J. Kim, S. B. Suh, K. S. Kim, J. Chem. Phys. 111, 10077
(1999).
40. E. W. G. Diau, J. L. Herek, Z. H. Kim, A. H. Zewail,
Science 279, 841 (1998).
41. We thank M. A. Johnson and K. H. Bowen for helpful
communication. Supported by the National Science
Foundation.
16 July 2004; accepted 25 August 2004
10.1126/science.1105018
How Do Small Water Clusters
Bind an Excess Electron?
Nathan I. Hammer, Joong-Won Shin, Jeffrey M. Headrick,
Eric G. Diken, Joseph R. Roscioli, Gary H. Weddle,*
Mark A. Johnson.
The arrangement of water molecules around a hydrated electron has eluded
explanation for more than 40 years. Here we report sharp vibrational bands
for small gas-phase water cluster anions, (H
2
O)
4-6
j
and (D
2
O)
4-6
j
. Analysis of
these bands reveals a detailed picture of the diffuse electron-binding site. The
electron is closely associated with a single water molecule attached to the
supporting network through a double H-bond acceptor motif. The local OH
stretching bands of this molecule are dramatically distorted in the pentamer
and smaller clusters because the excited vibrational levels are strongly
coupled to the electron continuum. The vibrationtoelectronic energy
transfer rates, as revealed by line shape analysis, are mode-specific and
remarkably fast, with the symmetric stretching mode surviving for less than
10 vibrational periods [50 fs in (H
2
O)
4
j
].
In dilute ionic solutions, the negative charge
carriers are typically closed-shell molecular
species (such as acetate) that capture the extra
electron into their molecular orbital networks.
Pure dipolar solvents such as ammonia and
water do not possess any low-lying valence
molecular orbitals to accommodate an extra
electron; nonetheless, an electron can become
trapped in a cavity within these liquids. The
resulting localized charge defect is called a
Bsolvated electron[ (15). The aqueous elec-
tron, e
aq
j
, has been subjected to intense study
since its discovery in 1962 (6), yet impor-
Sterling Chemistry Laboratory, Yale University, Post
Office Box 208107, New Haven, CT 06520, USA.
*Present address: Department of Chemistry, Fairfield
University, Fairfield, CT 06430, USA.
E-mail: mark.johnson@yale.edu
R E P O R T S
tant questions remain regarding the molecu-
lar structure of the solvating cavity.
Some of the complexity presented by e
aq
j
arises from the small spatial extent of the
excess electron wavefunction (2.4 ) radius)
(7), which is on the same order as the vol-
ume occupied by only a few water molecules
(8, 9). The relaxation dynamics of photo-
excited electrons are still controversial, with
reported values ranging from 50 to 300 fs
(10, 11). The 50-fs value is particularly
interesting because it is on the same order
as the time scale for the OH stretching
motion, introducing the possibility that the
intramolecular vibrations are intrinsically in-
volved in the relaxation. Two of the impor-
tant outstanding questions to be answered are
thus exactly how many water molecules are
in direct contact with the excess electron,
and to what extent are they perturbed by this
interaction? Some workers, for example, have
even invoked alternatives to the cavity mod-
el (that is, with rearrangements involving
the H
3
O Rydberg molecule) to explain e
aq
j
behavior (12), although such drastic per-
turbations appear to be at odds with recent
resonant Raman data (13).
One approach to resolving these contro-
versies is to focus on a simpler system first.
The study of size-selected water clusters in
the gas phase isolates local ion-solvent
interactions from the background created
by the bulk solution (14). As such, experi-
mental observables can be directly com-
pared with theoretical calculations that treat
the aggregate as a Bsupermolecule.[ Anionic
water clusters, (H
2
O)
n Q 2
j
, were first
isolated in 1984 (1517), and their reaction
chemistry (18, 19), photophysics (20), and
spectroscopy (21, 22) have been extensively
investigated. Unfortunately, it has not yet
been possible to unambiguously invert the
available spectral data to reconstruct the geom-
etry and electronic structure of the charged
clusters.
The basic problem is that the smallest
cluster for which sharp vibrational features
have been recovered is the hexamer anion,
(H
2
O)
6
j
(23). At that size, the high dimen-
sionality of the neutral cluster is already a
challenge for theory (24), and the additional
complications introduced by the diffuse
excess electron have made the calculations
too computationally intensive for ab initio
methods. Density functional theory (DFT)
calculations have recently become the meth-
od of choice to calculate (H
2
O)
n
j
behavior
(25), although this approach systematically
overbinds a diffuse excess electron and does
not offer the same convergence guarantee as
fully ab initio methods (26). Given these
theoretical difficulties, it is not surprising
that, during the past several years, the
observed vibrational bands of the hexamer
anion have been rationalized in the context
of very different network arrangements and
electron-binding motifs.
Kim and co-workers (25) have extrapo-
lated the DFT methods used for their hexa-
mer calculation to predict the energies and
vibrational spectra for various isomers of the
smaller water clusters (n e 5). Experimental
validation of these predictions would afford
a high level of confidence in the theoretical
method and associated structural assignments.
To that end, we exploited very recent ad-
vances in size-selected argon nanomatrix
spectroscopy (27), involving both synthetic
methodology (28) and wavelength coverage,
to establish the network morphologies and
intramolecular distortions that occur when
the four- to six-membered water clusters bind
an excess electron. Our results associate the
bound electron most closely with a single
water molecule, attached to the cluster through
two hydrogen bonds directed toward its O
atom. The OH stretches local to this molecule
account for a signature doublet feature, pre-
viously observed in the spectra of the anionic
clusters with n 0 6 to 9 and 11 (29).
Our experimental method probes argon-
solvated water cluster anions, for which
infrared absorptions can be readily detected
in an action mode by mass loss. This nano-
matrix spectroscopy (27) relies on the fact
that a resonantly prepared vibrational level
eventually decays, depositing its energy as
heat into the argon cluster. This, in turn,
causes the evaporation of several argon atoms
H
2
O
j
46
I Ar
n
h8 Y
H
2
O
j
46
I Ar
m
n Y m Ar 1
The lighter photofragments are easily isolated
from the mass-selected parent ions with the
use of a double-focusing, tandem time-of-
flight photofragmentation mass spectrometer
(30).
Preparation of the tetramer and, to a
lesser extent, pentamer anions has tradition-
ally been hampered by the lack of an
efficient pathway for direct electron attach-
ment onto the neutral clusters (31). In our
approach (32), (H
2
O)
4,5
j
I Ar
m
clusters are
prepared using argon-mediated sequential
condensation of water molecules onto the
easily generated dimer anion
H
2
O
j
2
I Ar
n
2H
2
O Y
H
2
O
j
4
I Ar
m
n Y m Ar 2
However, even when the species of interest
are successfully captured in an argon cluster,
sharp bands will be observed only when the
vibrationally excited states are stable with
respect to spontaneous electron detachment.
We previously determined the vertical elec-
tron detachment energy (VDE) of the tetra-
mer anion (dominant isomer) to be 350 T 20
meV (32). The OH stretching frequency of
420 meV lies above this VDE, and so the
associated vibrational transitions are expected
to be severely broadened, as reported for the
pentamer anion (29).
To circumvent the broadening problem,
we have probed the 200-meV intramolec-
ular bending vibrations rather than the OH
stretches. These bending bands lie below the
VDEs of the n 0 4 and 5 clusters and there-
fore should yield sharp features (33). As a
complementary strategy, we have prepared
the deuterated water clusters (D
2
O)
4,5
j
. The
OD stretches at 310 meV also lie below the
corresponding VDEs (34).
The argon predissociation spectra of the
(H
2
O)
4-6
j
clusters in the intramolecular bend-
ing region show sharp bands as expected
(Fig. 1). The lowest-energy feature appears in
almost the same location (1540 T 5 cm
j1
) in
all three spectra and occurs roughly 50 cm
j1
below the bend fundamental in the isolated
water molecule (arrow in Fig. 1A). This red-
shift points to a significant influence of the
extra electron, because the bending modes
have been shown to shift exclusively to higher
energy (relative to bare H
2
O) in the spectra
of the small neutral water clusters (35).
We can extract structural information from
the way in which the vibrational energies
depend on the number of donor and acceptor
H-bonds associated with water molecules in
particular local environments. For example,
molecules in single or double acceptor (A or
AA) sites have two nonbonded hydrogen
atoms and tend to have the lowest energy
1400 1500 1600 1700 1800
Photon Energy (cm
-1
)
C (H
2
O)
4
_
B (H
2
O)
5
_
P
r
e
d
i
s
s
o
c
i
a
t
i
o
n

Y
i
e
l
d
A (H
2
O)
6
_
Fig. 1. Argon predissociation spectra of the small
(H
2
O)
n
clusters in the intramolecular bending

region. Each displayed spectrum is an average of
20 scans. (A) (H
2
O)
6
I Ar
3
, (B) (H
2
O)
5
I Ar
4
,
and (C) (H
2
O)
4
I Ar
5
. The arrow in (A) marks the
bend fundamental (8
2
) for isolated H
2
O.
R E P O R T S
bending vibrations. Conversely, double donor
(DD) sites provide the largest constraint on
the bending motion and are highest in energy.
Molecules with AD and D motifs, having one
nonbonded hydrogen atom, give rise to bands
falling between the two limiting cases.
The three distinct bands in the n 0 4
spectrum (Fig. 1C) indicate that three binding
sites are in play, with the central band ap-
pearing somewhat broader, as would be ex-
pected from the contribution of two closely
related water molecules. The bottom panel
of Fig. 2 pairs this bending spectrum (from
Fig. 1C) on the left with the spectrum of the
deuterated tetramer in the OD stretching re-
gion on the right. The recent DFT calcula-
tions (25) predict that, even for the relatively
small tetramer anion, six classes of isomers
exist, with small variations available in each
class. Of these, however, only the chain
(Fig. 2, top panel) and the C
s
symmetry cyclic
(Fig. 2, middle panel) isomers are consistent
with our observed bands in the bending region.
The more symmetrical cyclic form best
explains the pattern of bending bands, as two
equivalent AD water molecules bridge the
other two molecules with DD and AA
motifs, respectively. Also, the OD stretching
data rule out the linear structure. Even minor
details in the experimental spectrum, such as
the shoulder on the 2560 cm
j1
band, are
accurately recovered by the calculated spec-
trum of the cyclic isomer (Fig. 2D). Such
agreement indicates that the cyclic isomer is
the species formed in the jet, and thus we can
rely on this analysis to establish the atomic
motions that contribute to the various bands.
This form was not calculated to be the lowest
energy isomer, but Kim et al. (25) suspected
that it might nonetheless be a dominant
species under experimental conditions. An
enlarged view of the calculated tetramer
anion structure is presented in Fig. 3, which
also includes a contour (isosurface at a value
of 0.005) for the excess electron orbital. Note
that the AA water molecule is completely
surrounded by the diffuse electron cloud.
The two strong OD stretching bands
(Fig. 2B) appear to match the signature dou-
blet feature common to the OH stretching
region of all anionic water clusters from n 0
6 to 9 and 11. In the tetramer, the bands can
be traced to the symmetric (8
s
) and asym-
metric (8
a
) stretching motions of the two
deuterium atoms on the AA water molecule.
Although these deuterium atoms are slightly
inequivalent due to the cis orientation of the
neighboring AD water molecules, the splitting
between the two normal modes (120 cm
j1
)
is quite close to that found in isolated D
2
O
(117 cm
j1
). The retention of this large
splitting in the cluster environment indicates
that, unlike hydrated anions such as Cl
2
j
(36), the diffuse electron does not compress
the HOH bond angle. The strong redshift of
these nominally Bfree[ or nonbonded hydro-
gen atoms apparently results from the AA
arrangement, wherein the acceptor H-bonds
act to cooperatively enhance the donor H-
bonds to the excess electron (37).
The structural assignment for the tetramer
anion can be used in characterizing the
evolving electron-binding motifs in the larger
clusters. The lowest energy bending transition
is nearly identical in the n 0 4 to 6 cluster
anions (Fig. 1), which suggests that the AA
binding site is preserved in larger clusters. The
OD stretching bands for these larger species
(Fig. 4) again appear as sharp features, and the
8
s
band associated with the AA water mole-
cule in the n 0 4 spectrum (Fig. 2B) is virtually
identical in the spectra of the larger clusters.
There is, however, a modest redshift of the
n 0 4 8
a
feature with increasing cluster size.
The n 0 6 spectrum (Fig. 4B) is particu-
larly useful because the corresponding spec-
trum in the OH stretching region has also
been reported (23, 29), allowing us to link
the behavior of the two isotopomers conclu-
sively. The band patterns are remarkably simi-
lar for both isotopes; as mentioned earlier, the
doublet observed in the OH stretching region
was found to persist in larger clusters. In the
OH spectra, the band splitting is smaller than
in the OD spectra (100 versus 120 cm
j1
), a
finding consistent with the 99 and 117 cm
j1
splittings measured for isolated H
2
O and D
2
O,
respectively. These observations strongly sug-
gest the continued importance of the AA
water molecule.
Using the same level of theory (DFT) that
correctly predicted the vibrational bands in
the tetramer anion, we have also made
structural and band assignments for the
pentamer and hexamer. The calculated spec-
tra (Fig. 4, top) accurately reproduce all of
the experimentally observed transition ener-
gies (Fig. 4, bottom), with the caveat that
DFT systematically underestimates the in-
tensities of the most redshifted bands.
The cluster structures giving rise to these
patterns are also included above the calcu-
Fig. 2. (Bottom) Argon
predissociation spectra
of (H
2
O)
4
I Ar
5
in the
bending region (A) and
(D
2
O)
4
I Ar
10
in the OD
stretching region (B).
(Middle) Calculated har-
monic spectra of (H
2
O)
4
in the bending region

(C) and (D
2
O)
4
in the
stretching region (D)
for the cyclic structure
shown, with A and D in-
dicating acceptor and do-
nor H-bonds, respectively.
(Top) Analogous calcu-
lated spectra for the linear
structure shown. The cal-
culated spectra [B3LYP/
6-311G**(sp)] are
scaled by 0.996 and
0.970 in the bend (E) and
stretch (F) regions, respec-
tively, where the factors
indicate the corrections
required to recover the
bands in the isolated wa-
ter isotopomers.
C
a
l
c
u
l
a
t
e
d

i
n
t
e
n
s
i
t
y
E F
AD
DD
AD
C
AA
D
1400 1500 1600 1700 1800
Photon Energy (cm
-1
)
P
r
e
d
i
s
s
o
c
i
a
t
i
o
n

Y
i
e
l
d
DD
AD A
AA
2400 2500 2600 2700 2800
OD Stretching
AA
AD
DD
//
//
//
B
//
s
a
HOH Bending
Fig. 3. A contour of the calculated orbital
describing the excess electron bound to the
cyclic water tetramer anion, (H
2
O)
4
.
R E P O R T S
lated spectra in Fig. 4. The structure calcu-
lated for the hexamer anion is the armchair
form previously assigned by Kim and co-
workers (25) on the basis of the OH stretching
spectra. In the n 0 5 and 6 clusters, the AA
water molecule, which was associated with a
pair of flanking AD molecules in the tetra-
mer, becomes closely associated with only
one nearby AD molecule as the cluster grows.
The higher energy vibrational bands
shown in Fig. 4 allow us to determine the
structures associated with water molecules
remote from the excess electron. The bands
closest to the free OD stretching frequency
(indicated by arrows in the lower traces of
Fig. 4) exhibit a discontinuous evolution, first
falling toward lower energy in the pentamer
and then returning back toward the free OD
position in the hexamer. This trend is readily
understood in the context of the proposed
structures. The n 0 5 cluster effectively
consists of a dimer subcluster mounted on a
cyclic trimer base (Fig. 4E). The dimer is
supported by three donor H-bonds from the
ring, so that the only nonbonded hydrogen
atoms are in direct contact with the diffuse
electron (38). The n 0 6 structure (Fig. 4F) is
closely related to that of the pentamer, but a
fourth water molecule is added to the trimer
base. This more open arrangement in the
hexamer leaves one nonbonded hydrogen
atom farther from the excess electron and
thus yields a band closer to the free OD
location, as is observed in the spectrum.
The emergence of the dimer-subcluster
motif is also important because the donor
hydrogen atom is calculated to be complete-
ly coupled to the stretches of the nonbonded
hydrogen atoms on the AA water molecule
in the signature doublet. The normal mode
vectors for the 8
s
vibration are indicated by
the arrows on the structures at the top of
Fig. 4. Only these three atoms undergo sig-
nificant displacement. This DAA type of vi-
bration appears to be driving the dimer toward
an OH
j
IIIH
3
O
ion pair configuration, which

should cause a dramatic increase in the di-
pole moment, and hence increased coupling
to the excess electron.
We can use our spectral assignments to
probe the dynamics of vibration-to-electronic
coupling in the clusters. In general, when a
vibrational level occurs above the electron-
binding energy of an anion, the coupling
between molecular vibrations and the back-
ground electron continuum leads to decay of
the vibrational energy via autodetachment
H
2
O
j
n
h8 YH
2
O
j
n
vYH
2
O
n
e
j
3
However, in cases where the optical absorp-
tion transition moments to the embedded
vibrational level and the electron continuum
are comparable, the bands become strongly
distorted into so-called BFano[ line shapes as
a result of interference between the two
absorption pathways (39).
This phenomenon is observed in the OH
stretching region of the (H
2
O)
4,5
j
clusters.
The electron photodetachment spectra of
both the (H
2
O)
4
j
I Ar
6
cluster and the
corresponding pentamer (29) (Fig. 5) display
the classic derivative-type profiles, indicating
very strong coupling to the background elec-
tron continuum. These line shapes are known
to follow the form (40)
G 88 0
q (
2
1 (
2
4
where the q parameter controls the relative
absorption strengths of the vibrational level
and background continuum, and ( contains
information on both the decay rate (I) and
frequency 3
0
of the embedded resonance
through the relation
( 0 4 I > c 88 j 88
0
5
To fit the distorted Fano profiles, we
iteratively convoluted the independent con-
tributions of the AA symmetric and asym-
metric stretching bands, with (, q, and 3
0
as
adjustable parameters. The unperturbed loca-
tions of the embedded vibrational levels are
Fig. 4. (Bottom) Argon
predissociation spectra of
(D
2
O)
5
I Ar
6
(A) and
(D
2
O)
6
I Ar
7
(B), with ar-
rows indicating 8
OD
in
free D
2
O. (Top) Calculated
spectra for the corre-
sponding pentamer and
hexamer structures shown
above (E and F), with nor-
mal mode displacement
vectors included for the
lowest-energy symmetric
stretching vibrations (8
s
)
on each structure.
C D
E F
s
s
Photon Energy (cm

-1
)
B A
C
a
l
c
u
l
a
t
e
d

i
n
t
e
n
s
i
t
y
P
r
e
d
i
s
s
o
c
i
a
t
i
o
n

Y
i
e
l
d
(D
2
O)
5
_
(D
2
O)
6
_
2400 2500 2600 2700 2800 2400 2500 2600 2700 2800
3200 3300 3400 3500
Photon Energy (cm
-1
)
(H
2
O)
5
_
(H
2
O)
4
_
B
P
h
o
t
o
e
l
e
c
t
r
o
n

Y
i
e
l
d
A

s a

s
a
Fig. 5. (A) Electron photodetachment spectrum
of (H
2
O)
4
I Ar
6
. (B) Electron photodetachment
spectrum of (H
2
O)
5
, reproduced from (29);

the sawtooth modulation is an artifact of laser
power stabilization. The solid lines are fits as
described in the text. The solid arrows indicate
the unperturbed locations of the two stretch-
ing modes (labeled 8
s
and 8
a
) of the AA water
molecule from the fit, whereas the dotted
arrows denote the respective energies extrap-
olated from the sharper OD bands (Fig. 4).
R E P O R T S
indicated by the solid arrows in Fig. 5. The
lifetimes extracted from the fits (solid lines
in Fig. 5) are remarkably short, with the
tetramer symmetric stretch decaying in only
about 50 fs and the asymmetric stretch in
approximately 150 fs. The pentamer life-
times are approximately twice as long, with
a similar ratio of symmetric to asymmetric
decay rates.
This mode-specific relaxation behavior
may well be relevant in explaining the
ultrashort time scales of the absorption
transients observed in the bulk hydrated
electron (41, 42). For example, Wiersma
and co-workers (11) have already noted that
their 50-fs electronic relaxation time could
only be explained with a model in which the
relaxation is mediated by coupling to the
vibrational modes of the water molecules.
Such a local interaction has also been
observed through the resonant Raman spec-
tra of hydrated electrons in the bulk, which
reveal that at least one OH bond forms an
exceptionally strong donor H-bond either to
the electron or to oxygen (13). Extension of
the ultrafast experiments to the larger
clusters will further explore whether the
local binding motif uncovered in this work
quantitatively recovers the dynamics at work
in the bulk environment.
1. W. Weyl, Ann. Phys. (Leipzig) 197, 601 (1864).
2. C. A. Kraus, J. Am. Chem. Soc. 30, 1323 (1908).
3. R. A. Ogg Jr., J. Chem. Phys. 14, 114 (1946).
J. Chem. Phys. 88, 4421 (1988).
5. B. J. Schwartz, P. J. Rossky, J. Chem. Phys. 101, 6917
(1994).
6. E. J. Hart, J. W. Boag, J. Am. Chem. Soc. 84, 4090 (1962).
7. S. Golden, T. R. Tuttle Jr., J. Chem. Soc. Faraday
Trans. II 75, 474 (1979).
8. D. M. Bartels, J. Chem. Phys. 115, 4404 (2001).
9. P. Ayotte, thesis, Yale University (1999).
10. P. Kambhampati, D. H. Son, T. W. Kee, P. F. Barbara,
J. Phys. Chem. A 106, 2374 (2002).
11. M. S. Pshenichnikov, A. Baltus ka, D. A. Wiersma,
Chem. Phys. Lett. 389, 171 (2004).
12. A. L. Sobolewski, W. Domcke, Phys. Chem. Chem.
Phys. 4, 4 (2002).
13. M. J. Tauber, R. A. Mathies, J. Am. Chem. Soc. 125,
1394 (2003).
14. W. H. Robertson, M. A. Johnson, Annu. Rev. Phys.
Chem. 54, 173 (2003).
15. H. Haberland, H.-G. Schindler, D. R. Worsnop, Ber.
Bunsenges. Phys. Chem. 88, 271 (1984).
16. H. Haberland, H. Langosh, H.-G. Schindler, D. R.
Worsnop, J. Phys. Chem. 88, 3903 (1984).
17. H. Haberland, C. Ludewigt, H.-G. Schindler, D. R.
Worsnop, J. Chem. Phys. 81, 3742 (1984).
18. S. T. Arnold, R. A. Morris, A. A. Viggiano, M. A.
Johnson, J. Phys. Chem. 100, 2900 (1996).
19. M. K. Beyer, B. S. Fox, B. M. Reinhard, V. E. Bondybey,
J. Chem. Phys. 115, 9288 (2001).
20. P. J. Campagnola, D. J. Lavrich, M. J. Deluca, M. A.
Johnson, J. Chem. Phys. 94, 5240 (1991).
21. C. G. Bailey, J. Kim, M. A. Johnson, J. Phys. Chem.
100, 16782 (1996).
23. P. Ayotte, C. G. Bailey, J. Kim, M. A. Johnson, J. Chem.
Phys. 108, 444 (1998).
24. C. J. Tsai, K. D. Jordan, Chem. Phys. Lett. 213, 181 (1993).
(2003).
26. F. Wang, K. D. Jordan, Annu. Rev. Phys. Chem. 54,
367 (2003).
27. S. A. Corcelli, J. A. Kelley, J. C. Tully, M. A. Johnson,
J. Phys. Chem. A 106, 4872 (2002).
28. W. H. Robertson, J. A. Kelley, M. A. Johnson, Rev. Sci.
Instrum. 71, 4431 (2000).
30. L. A. Posey, M. A. Johnson, J. Chem. Phys. 89, 4807
(1988).
31. J. Kim, I. Becker, O. Cheshnovsky, M. A. Johnson,
Chem. Phys. Lett. 297, 90 (1998).
32. J.-W. Shin, N. I. Hammer, J. M. Headrick, M. A.
Johnson, Chem. Phys. Lett., in press.
33. Accessing the lower wavelength region of the
bending modes (1500 cm
j1
) required modification
of our pulsed infrared laser (LaserVision) by addition
of a third nonlinear mixing stage based on a AgGaSe
2
crystal. This arrangement provides about 100 6J of
energy in the bending region.
34. Because of the near mass degeneracy between Ar
and (D
2
O)
2
, we verified that the dimer and trimer
clusters exhibit no photofragmentation or sharp
bands, consistent with their low VDEs. Thus, the
(D
2
O)
4
j
I Ar
m
spectrum could be selectively moni-
tored by detecting the photofragment anions that
arise only from this species.
35. J. B. Paul et al., J. Phys. Chem. A 103, 2972 (1999).
36. E. A. Price, N. I. Hammer, M. A. Johnson, J. Phys.
Chem. A 108, 3910 (2004).
37. W. H. Robertson, K. Karapetian, P. Ayotte, K. D. Jordan,
M. A. Johnson, J. Chem. Phys. 116, 4853 (2002).
38. T. Tsurusawa, S. Iwata, Chem. Phys. Lett. 315, 433 (1999).
39. U. Fano, Phys. Rev. 124, 1866 (1961).
40. A. R. P. Rau, Phys. Scrip. 69, C10 (2004).
41. C. Silva, P. K. Walhout, K. Yokoyama, P. F. Barbara,
Phys. Rev. Lett. 80, 1086 (1998).
42. A. Baltus ka, M. F. Emde, M. S. Pshenichnikov, D. A.
Wiersma, J. Phys. Chem. A 103, 10065 (1999).
43. We thank the U.S. Department of Energy (grant DR-
FG02-00ER15066) for support of this research
direction, as well as NSF, which provided the
equipment used in this study. We also thank K. D.
Jordan for valuable discussions regarding the for-
mation mechanisms at work in these clusters and
T. Zwier and J. Stearns for help with the wavelength
extension of the laser.
15 July 2004; accepted 20 August 2004
Reconstructing Past Climate
from Noisy Data
Hans von Storch,
1
* Eduardo Zorita,
1
Julie M. Jones,
1
Yegor Dimitriev,
1
Fidel Gonza lez-Rouco,
2
Simon F. B. Tett
3
Empirical reconstructions of the Northern Hemisphere (NH) temperature in
the past millennium based on multiproxy records depict small-amplitude vari-
ations followed by a clear warming trend in the past two centuries. We use
a coupled atmosphere-ocean model simulation of the past 1000 years as a
surrogate climate to test the skill of these methods, particularly at multi-
decadal and centennial time scales. Idealized proxy records are represented by
simulated grid-point temperature, degraded with statistical noise. The cen-
tennial variability of the NH temperature is underestimated by the regression-
based methods applied here, suggesting that past variations may have been
at least a factor of 2 larger than indicated by empirical reconstructions.
Reconstruction of past climate from palaeo-
climate proxy data is important for the de-
tection of anthropogenic climate change. A
number of studies have attempted to recon-
struct variations in global or NH temperature
within the past millennium by regressing
proxy indicators and early instrumental time
series on recent instrumental climate variables
with high spatial resolution (13). Regression
models are developed during the period of
common instrumental and proxy data and are
then applied to longer proxy records to re-
construct past climates. Similar methods
have been applied for the reconstruction of
atmospheric circulation indices, such as the
North Atlantic Oscillation (4) or the Ant-
arctic Oscillation (5). A number of reconstruc-
tions show that the temperatures in the last
millennium were characterized by geograph-
ically varying warm values in the 11th and
12th centuries, followed by a secular cooling
trend punctuated by decadal-scale colder pe-
riods in the mid-16th, early 17th, and early
19th centuries (6). These cooler intervals were
followed by the marked warming experienced
until today. Although the amplitude of these
preindustrial variations is still debated, ac-
cording to the most quoted NH temperature
reconstruction EMann, Bradley, Hughes, 1998
(MBH98) (1) and Mann, Bradley, Hughes,
1999 (MBH99) (2)^ and the most recent In-
tergovernmental Panel on Climate Change
(IPCC) report (7), these variations were of
small amplitude. However, recent studies
with general circulation models suggest that
these centennial variations may have been
larger (810). We used a coupled atmosphere-
ocean model simulation of the past 1000
years as surrogate climate to test whether
1
Institute for Coastal Research, GKSS Research Cen-
tre, Geesthacht 21502, Germany.
2
Department of
Astrophysics and Atmospheric Physics, Universidad
Complutense, Madrid 28040, Spain.
3
UK Meteorol-
ogical Office, Hadley Centre (Reading Unit), Mete-
orology Building, University of Reading, Reading, RG6
6BB, UK.
E-mail: storch@gkss.de
R E P O R T S
the reconstruction method of MBH98 and
MBH99 and a much simpler regression meth-
od can yield realistic estimates of the multi-
decadal and secular temperature variations.
A number of modeling studies of the
evolution of the climate in the past centuries
(1114) pose some questions about the re-
liability of empirical reconstructions based on
regression methods. For instance, concerning
the cooling around 1700, the reconstructions
by MBH98 have agreed with the results ob-
tained with the the Goddard Institute for Space
Studies model only in the model version with
a low climate sensitivity (0.4K/W/m
2
) (10).
Similarly, the agreement between an energy-
balance model (11) and reconstructions has
been achieved by prescribing a model sensi-
tivity to changes in radiative forcing of 0.5K/
W/m
2
. These values of climate sensitivity are
at the low end of the range of the models
included in the IPCC analysis (7). Other re-
constructions that indicate markedly stronger
cooling in the 16th to 18th centuries are, for
instance, the result of empirical methods that
explicitly aim to preserve low-frequency
variability (15) and the borehole-based re-
constructions (16). Interestingly, the latter is
not based on empirical regression methods.
This apparent discrepancy poses a question as
to whether model simulations overestimate
secular climate variability or regression-based
reconstructions underestimate it.
The reliability of these empirical meth-
ods at centennial time scales can be tested in
the surrogate climate simulated by three-
dimensional climate models driven by plau-
sibly estimated historical external forcing.
The results of the reconstruction process
can then be validated against the climate
fields simulated by the climate model. The
data representing the proxy records are cli-
mate variables simulated at grid-box resolu-
tion, which can be degraded with statistical
noise to mimic more realistic data Eso-called
pseudoproxies (17)^. Here, we follow this
strategy with the output of a climate simu-
lation of the past millennium with the coupled
atmosphere-ocean general circulation model
European Centre Hamburg 4Hamburg Ocean
Primitive EquationG (ECHO-G) (18), driven
by estimations of historical climate forcing.
This simulation provides a data set in which
the potential nonstationarity of the covari-
ances and the length of the time series are
similar to those found in applications of the
empirical reconstruction methods.
This simulation reproduces warming
around 1100 and extended coolings over the
SpPrer, Maunder, and Dalton Minima as near-
global events, as well as the recent anthropo-
genic warming. Compared with the recon-
structions of MBH99, however, the variations
are stronger. For the purpose of this paper, it is
not critical if the simulation is not absolutely
realistic because of model limitations (e.g.,
coarse resolution or deficient representation of
processes) or uncertainties in external forcings.
The crucial point is that the model simulates a
reasonable, internally consistent climate, and
the external forcing lies within the envelope of
possible values. In this case, it will be used
as a virtual world to determine the skill of
regression-based reconstruction methods like
MBH98 to estimate its temperature variations.
Here, we focus on the reconstruction of
NH temperature. For this analysis, we applied
as realistically as possible the statistical
method of MBH98. However, when arbitrary
choices were required, the a priori most fa-
vorable for the statistical method were im-
plemented, thus probably minimizing the
loss of variance in the statistical reconstruc-
tions. For instance, the proxy network in
MBH98 gets coarser backward in time, but
the pseudoproxy network in this study was
not decimated to avoid loss of skill. Also,
only temperature pseudoproxies were used
to reconstruct the temperature evolution. The
pseudoproxies were generated by adding a
statistical white noise to the simulated tem-
peratures in grid points colocated to the
MBH98 proxy network (1). Several tests with
varying amounts of noise were carried out.
The loss of variance resulting from a
regression-type method may be simply con-
ceptualized (18). The proxy data P are
thought to blend local temperature T
l
and
unrelated variability (: P 0 "T
l
(. The tem-
perature variations are essentially estimated
as T
l
* 0 $P, where $ 0 DG
T
/G
P
G 1 and D is
the correlation between T
l
and P. Therefore,
Variance(T*) 0 D
2
Variance(T
l
) G Variance(T
l
).
In case of proxy data, the correlation D is
mostly on the order of 0.4 to 0.7 (8), re-
sulting in a leakage or variance on the order
of 50 to 80%. In particular, if T
l
has a red
spectrum and ( has a white spectrum, T* will
underestimate the low-frequency variance of
T
l
. In the case of MBH98 and other recon-
structions, the methodological process is more
sophisticated, but the fundamental problem
of the loss of variance resulting from noisy
proxy data may also exist in these studies
(19). This loss of variance, also known in
areas such as regionalization and long-term
forecasting (20), is sometimes accounted for
by artificially inflating the parameter D. For
paleoclimate reconstructions, D should be
made time-scale dependent, and this depen-
dency is unknown.
To implement the method of MBH98, we
selected model grid boxes colocated with
their proxy data network (Fig. 1, red pixels)
and added white noise (21) to the grid-point
temperatures T
g
, so that the pseudoproxy
data are P 0 T
g
(. The variance of ( varies
between m 0 0 and m 0 4 Var(T
g
). The
correlation between the T
g
and P is then (1
m)
1/2
. Thus, with m 0 0, the local P variance
described by T
g
is 100%; for m 0 1, when
noise with the same variance as that of the
local temperature is added, the percentage
of described variance is 50%. For m 0 4,
the described variance is only 20%. Ideally,
the reconstructions would coincide with the
simulated NH temperature, but actually they
do not, even for m 0 0 (Fig. 2A, illustrating
the loss of variance induced by the method
alone). The short-term variations are reason-
ably reproduced, at least for m G 4. For
instance, on an interannual time scale, the fit
between simulated and reconstructed NH
temperature is good, with a calibration
reduction-of-error statistics of 0.7 for perfect
pseudoproxies and 0.30 for pseudoproxies
with ( 0 0.5. The substantial underestimation
of low-frequency temperature variations is
evident from Fig. 2B. For example, only 20%
of the 100-year variability is recovered when
the noise level is 50%. For time scales of 20
years, about three-quarters of the variability is
lost. Similar results are obtained with a simu-
lation with the third Hadley Centre coupled
model (HadCM3), demonstrating that the
results obtained here are not dependent on
the particular climate characteristics of the
ECHO-G simulation (18). Also a spatially
varying level of noise does not essentially
modify these conclusions (18).
Our setup allowed the test of a number of
hypotheses. The first hypothesis is that the
inclusion of more instrumental data would
improve the estimate, as the multiproxy data
used by MBH99 contained a number of long
instrumental temperature data, which start
Fig. 1. Grid boxes in the
ECHO-G model, from
which simulated tem-
peratures are used to
estimate NH tempera-
tures. The red pixels are
used for the basic re-
construction; the blue
pixels are added in a
test of whether a bet-
ter spatial coverage
would improve the
reconstruction.
Location of pseudo-proxies in the ECHO-G grid
R E P O R T S
typically at the end of the 18th century. To test
the influence of such instrumental data, we
included grid-box temperatures without add-
ing noise (18). The effect of this modification
on the hemispheric temperature was small;
the differences in the reconstructed tempera-
ture anomalies were within a range of 2%.
This can be explained by the relatively low
number of perfect pseudoproxies included in
comparison with the total number of pseudo-
proxies and by the built-in robustness against
local influences of the inverse regression
method used by MBH98, because the signal
is extracted nonlocally from the whole proxy
data set simultaneously (18). Other direct
regression methods, aimed at more regionally
limited temperature reconstructions, do show
an improvement when instrumental records
are included in the proxy network (22).
The second hypothesis is related to the
sparseness of the proxy locations (Fig. 1, red
pixels). The proxy data set was enlarged by
adding 18 locations in Asia and Africa (Fig. 1,
blue pixels) to increase the spatial coverage.
This leads to a minor improvement in the NH
temperature reconstruction (Fig. 3), which was
largely independent of the inclusion or exclu-
sion of Southern Hemisphere pseudoproxies.
Finally, we tested whether the range of
variability present in the instrumental period is
sufficient to reconstruct the climate of past
centuries. To test this hypothesis, 40 years were
taken from the Late Maunder Minimum (1680
to 1720) and 40 years from the early part of the
20th century (1900 to 1940) to calibrate the
statistical model, thereby expanding the range
of temperature variability present in the
pseudoproxies. When the proxies are free of
noise, the reconstruction of the simulated NH
temperature is greatly improved (Fig. 3). With
50% local noise included, the reconstruction
is also improved, although the loss of low-
frequency variance is still large. Therefore,
augmenting the variability in the calibration
period improves the skill, but obviously this is
limited by the available observational record.
A further question is whether the limita-
tions we have found are common to regression
methods in general. Thus, two further ap-
proaches were tested. In the first, local temper-
atures were estimated by a linear regression
from pseudoproxies, and the local temperature
estimations were spatially averaged to derive
the NH temperature. This method mimics the
situation in which, for example, local dendro-
chronologies calibrated in terms of local tem-
perature are just arithmetically averaged. In
the second approach, the pseudoproxies at
the various locations were directly simply
averaged. This is more similar to the borehole
methodology (16). For the first method, we
found qualitatively similar but quantitatively
even worse problems than with the MBH98
method; that is, the underestimation of low-
frequency variability for a given amount of
noise is greater than for MBH98, whereas
the second method returns good estimates of
NH temperature, with very little loss of var-
iance with 75% variance noise (Fig. 4). This
result is not surprising given that the first
method suffers from the variance loss related
to regression, whereas in the second, the
noise contributions are simply averaged out.
Hints of the underestimation of low-
frequency variability by empirical reconstruc-
tion methods have been found in previous
studies, based either on short data sets (17)
or climate simulations with fixed external
forcing (23). In a study based entirely on an
instrumental data set (17), the spectrum of
the difference between the reconstructed and
observed global mean annual temperature is,
albeit consistent with a white noise assump-
tion, slightly red. In a further analysis of an
instrumental data set and data from a long
control simulation with the Geophysical Fluid
Dynamics Laboratory climate model (with con-
stant external forcing) and a relatively short
simulation of 143 years driven by varying ex-
ternal forcing (23), the spectra of the temper-
ature differences from the analysis of control
simulation are red (although again statistically
compatible with white noise assumption). In
this externally forced simulation, it was found
that the temperature reconstructions are biased
if the external forcing leads to nonstationary
behavior in the verification period. In a long
control simulation (1000 years) with the
model ECHO-G (24), the spectrum of the re-
constructed annual global temperature under-
estimates the spectrum of the simulated global
temperature at very low frequencies.
10
-2
10
-1
0.5
A B
Fig. 2. (A) The NH annual temperature evolution over the past 1000 years.
The NH annual temperature simulated by the model ECHO-G and MBH98
reconstructions of this temperature from 105 model grid points mimicking
the multiproxy network of MBH98. Increasing amounts of noise have been
added to the grid-point temperatures to mimic the presence of other than
temperature signals in the proxies. The corresponding local correlation is
also indicated. The 2G uncertainty range (derived as in MBH98 from the
variance of the interannual residuals) for the different noise levels is
indicated. The reconstruction with D 0 0.5 is shown with its 2G uncertainty
range. (B) The spectra of the NH annual temperatures shown in (A).
R E P O R T S
Climate simulations of the past millennium
are burdened by model limitations and uncer-
tainties in the external forcing, and therefore
their output must be considered with care.
However, they provide a surrogate climate
realistic enough to conclude that the use of the
regression methods considered here, which
exploit short-term cross-correlations to recon-
struct past climates, suffer from marked losses
of centennial and multidecadal variations. This
conclusion probably applies to most other
regression-based methods. Other methods
that estimate past temperatures with physical,
as opposed to statistical, methods Ee.g., bore-
hole temperature profiles (16)^, or regression
methods that address retention of the infor-
mation of the low-frequency variability con-
tained in the proxy indicators (25), may be in
theory free from this specific caveat. Our re-
sults indicate that a detailed testing of these
reconstruction methods in simulated climates
should be an essential part in the reconstruc-
tion process and may help in the design of
better reconstruction methods.
1. M. E. Mann, R. S. Bradley, M. K. Hughes, Nature 392,
779 (1998).
2. M. E. Mann, R. S. Bradley, M. K. Hughes, Geophys.
Res. Lett. 26, 759 (1999).
3. P. D. Jones, K. R. Briffa, T. P. Barnett, S. F. B. Tett,
Holocene 8, 455 (1998).
4. J. Luterbacher, C. Schmutz, D. Gyalistras, E. Xoplaki,
H. Wanner, Geophys. Res. Lett. 26, 2745 (1999).
5. J. M. Jones, M. Widmann, J. Clim. 16, 3511 (2003).
6. P. D. Jones, T. J. Osborn, K. R. Briffa, Science 292, 662
(2001).
7. J. T. Houghton et al., Climate Change 2001: The
Scientific Basis (Cambridge Univ. Press, Cambridge,
2001).
8. P. D. Jones, M. E. Mann, Rev. Geophys. 42, RG2002,
10.1029/2003RG000143 (2004).
9. E. Zorita et al., Meteorol. Z. 13, 271 (2004).
10. D. Rind et al., J. Clim. 17, 906 (2004).
11. T. J. Crowley, Science 289, 270 (2000).
12. S. Gerber et al., Clim. Dyn. 20, 281 (2003).
13. H. Goosse et al., Geophys. Res. Lett. 31, L06203,
10.1029/2003GL019140 (2004).
14. E. Bauer, M. Claussen, V. Brovkin, A. Huenerbein,
Geophys. Res. Lett. 30, 1276, 10.1029/2002GL016639
(2003).
15. J. Esper, E. R. Cook, F. H. Schweingruber, Science 295,
2250 (2002).
16. S. H. Huang, H. N. Pollack, P. Y. Shen, Nature 403,
756 (2000).
17. M. E. Mann, S. Rutherford, Geophys. Res. Lett. 29,
139, 10.1029/2001GL014554 (2002).
18. Materials and Methods are available as supporting
19. For instance, the standard deviation of the MBH99
reconstructed NH temperature in the period of 1948
to 1980 is 0.9K, compared with 1.4K derived from
the National Centers for Environmental Protection
reanalysis or from the Jones et al. (26) NH gridded
instrumental data.
20. H. von Storch, J. Clim. 12, 3505 (1999).
21. Local noise could in many cases be expected to be
either white or red. In any case, red noise degrades
the reconstructions, as in (17).
22. J. Luterbacher, D. Dietrich, E. Xoplaki, M. Grosjean,
H. Wanner, Science 303, 1499 (2004).
23. S. Rutherford, M. E. Mann, T. L. Delworth, R. J. Stouffer,
J. Clim. 16, 462 (2003).
24. E. Zorita, F. Gonza lez-Rouco, S. Legutke, J. Clim. 16,
1378 (2003).
25. T. J. Osborn, K. R. Briffa, Dendrochronologia 18, 9 (2000).
26. P. D. Jones, M. New, D. E. Parker, S. Martin, I. G. Rigor,
Rev. Geophys. 37, 173 (1999).
27. This work was carried out within the projects German
Climate Research Program (DEKLIM) [German Federal
Ministry for Education and Research (BMBF)]; Simula-
tions, Observations, and Palaeoclimate data (SOAP)
(European Union, EVK2-CT-2002-00160); and REN-
2000-0786cli (Comisio n Interministerial de Ciencia y
Tecnologia). S.F.B.T. was supported by a UK Govern-
ment Meteorological Research contract and SOAP.
Computer time for HadCM3 simulations was funded
by Defra under contract PECD/7/12. Data distribution
was through SOAP. Three anonymous reviewers greatly
contributed to the improvement of the original
manuscript. We acknowledge fruitful suggestions by
T. Stocker.
www.sciencemag.org/cgi/content/full/1096109/DC1
SOM Text
Figs. S1 and S2
References
27 January 2004; accepted 23 August 2004
10.1126/science.1096109
Fi g. 3. Si mul ated
(black) and estimated
NH temperature (pur-
ple, blue, orange, yel-
low, and pink) showing
the effect of noise. Dif-
ferent setups for the
estimation were used:
the standard method
with 50% noise added
(pink); with additional
pixels in Africa and
Asia (locations, Fig. 1)
with 50% noise (blue)
and without noise (pur-
ple); with a different
fitting period, namely
1680 to 1720 plus
1900 to 1940 (instead
of the standard 1900
to 1980) with 50%
noise (yellow) and with-
out noise (orange).
Fig. 4. Simulated NH
temperature compared
with an estimate with
a simple local linear
regression on the pseu-
doproxies used in Fig.
2. The local tempera-
ture is estimated from
each pseudoproxy and
the result is simply
averaged over all pseu-
doproxy locations to
obtain the NH esti-
mate. Also shown is
the arithmetic mean
of the perfect pseudo-
proxies and of the
pseudoproxies contain-
ing 50% noise.
R E P O R T S
Mycobacterial Ku and Ligase
Proteins Constitute a Two-
Component NHEJ Repair Machine
Marina Della,
1
*. Phillip L. Palmbos,
2
. Hui-Min Tseng,
3
.
Louise M. Tonkin,
1,4
James M. Daley,
2
Leana M. Topper,
2
Robert S. Pitcher,
4
Alan E. Tomkinson,
5
Thomas E. Wilson,
2
Aidan J. Doherty
4
-
In mammalian cells, repair of DNA double-strand breaks (DSBs) by nonhomolo-
gous end-joining (NHEJ) is critical for genome stability. Although the end-bridging
and ligation steps of NHEJ have been reconstituted in vitro, little is known about
the end-processing reactions that occur before ligation. Recently, functionally
homologous end-bridging and ligation activities have been identified in prokarya.
Consistent with its homology to polymerases and nucleases, we demonstrate that
DNA ligase D from Mycobacterium tuberculosis (Mt-Lig) possesses a unique va-
riety of nucleotidyl transferase activities, including gap-filling polymerase, ter-
minal transferase, and primase, and is also a 3 to 5 exonuclease. These enzyme
activities allow the Mt-Ku and Mt-Lig proteins to join incompatible DSB ends in
vitro, as well as to reconstitute NHEJ in vivo in yeast. These results demonstrate
that prokaryotic Ku and ligase form a bona fide NHEJ system that encodes all the
recognition, processing, and ligation activities required for DSB repair.
Nonhomologous end-joining (NHEJ) is the
major pathway for repairing DNA double-
strand breaks (DSBs) in mammalian cells (1).
Key factors in eukaryotic NHEJ are the Ku70/
Ku80 heterodimer (Ku) and DNA ligase IV
(2), which have functional homologs in pro-
karyotes (3). Ku binds directly to the termini
of DSBs and has end-bridging activity (4, 5),
as has the yeast Mre11/Rad50/Xrs2 (MRX)
complex (6). These factors interact with ligase
IV/XRCC4 to achieve repair (68). Many
DSBs generated in vivo have damaged or non-
complementary termini that require processing
by nucleases and polymerases to generate
ligatable termini. The molecular mechanisms
of these reactions are poorly understood.
The Mycobacterium tuberculosis DNA re-
pair ligase, Mt-Lig (Rv0938 and LigD), which
is specifically stimulated by Mt-Ku homo-
dimer (3), contains domains that exhibit sig-
nificant homology with polymerases (912)
and possibly nucleases (1012), suggesting
that Mt-Lig might process and then join
incompatible DNA ends. Mt-Lig is most ho-
mologous to eukaryotic primase (912); how-
ever, a requirement for primase during NHEJ
is not obvious. Eukaryotic primases share sig-
nificant sequence homology with the Pol X
family of nucleotidyl transferases active in
NHEJ (1317). Purified recombinant Mt-Lig
(3, 18) was an efficient DNA-dependent DNA
polymerase in template-dependent primer ex-
tension assays (fig. S1). Mt-Lig similarly acted
as a DNA-dependent RNA polymerase (fig.
S1). Mutating predicted catalytic aspartate
residues abolished all polymerase activities,
confirming that they are intrinsic properties of
Mt-Lig (fig. S1). Terminal transferase (TdT),
a Pol X family member, has also been
implicated in V(D)J recombination and NHEJ
(14, 15). Mt-Lig used adenosine triphosphate
(ATP) and, to a lesser extent, dATP to extend
single-strand (ss) DNA, demonstrating termi-
nal transferase activity (fig. S2). Finally, Mt-
Lig synthesized 10- to 25nucleotide (nt)
oligomers on unprimed circular M13 ssDNA
using nucleotide triphosphates (NTPs) but not
dNTPs (fig. S3). Thus, Mt-Lig has DNA-
dependent RNA primase activity.
To examine microhomology-mediated
joints typical of NHEJ, we performed poly-
merization assays with oligonucleotides that
generate a nonligatable 1-nt gap upon align-
ment (Fig. 1A). Mt-Lig efficiently filled in the
gap with no detectable strand displacement
synthesis (Fig. 1A, left). Addition of a phos-
phate group to the 5 terminus of the 1-nt gap
resulted in gap filling and ligation (Fig. 1A,
right), indicating the concerted action of
Mt-Lig polymerase and ligase activities.
Mt-Lig is also predicted to contain a distinct
nuclease domain (10, 11). Indeed, Mt-Lig
progressively digested the 3 but not the 5
ss tails of partial duplexes (18) until reach-
ing the double-strand (ds) region (Fig. 1B).
Thus, Mt-Lig possesses 3 to 5 ssDNA exo-
nuclease activity. Using DNA substrates that
generate a 3-flap adjacent to a nick, Mt-Lig
removed the flap by exonucleolytic digestion,
generating a base-paired linear duplex (Fig.
1C). At higher concentrations, the nuclease
progressed through the microhomology re-
1
Cambridge Institute for Medical Research, University
of Cambridge, Department of Haematology, Hills Road,
Cambridge CB2 2XY, UK.
2
Department of Pathology,
University of Michigan Medical School, Ann Arbor,
Michigan 481090602, USA.
3
Molecular Medicine Grad-
uate Program, Institute of Biotechnology, The University
of Texas Health Science Center at San Antonio, San
Antonio, TX 782453207, USA.
4
Genome Damage and
Stability Centre, University of Sussex, Falmer, Brighton
BN1 9RQ, UK.
5
Department of Radiation Oncology and
Greenebaum Cancer Center, University of Maryland
School of Medicine, Baltimore, MD 212011559, USA.
*Present address: The Institute of Cancer Research,
Chester Beatty Laboratories, London SW3 6JB, UK.
.These authors contributed equally to this work.
-To whom correspondence should be addressed.
E-mail: AJD21@sussex.ac.uk
Fig. 1. Mt-Ligase is a
multidomain enzyme
with polymerase, nu-
clease, and ligase ac-
tivities. (A) Left: In
assays with DNA du-
plexes that form a non-
ligatable 1-nt gap,
Mt-Lig (0.125, 0.25,
0.5, or 1 6M) efficient-
ly filled in the gap.
Right: When a phos-
phate group was added
to the 5 terminus, Mt-
Lig (1 6M) ligated the
filled intermediate. (B)
DNA duplexes with 3-
overhangs were incu-
bated with increasing
amounts of Mt-Lig
(0.25, 0.5, and 1 6M).
The positions of the
51-nt oligomer sub-
strate and 43-nt olig-
omer degr adat i on
products are indicated.
(C) DNA substrates were incubated with increasing amounts of Mt-Lig (0.25, 0.5, and 1 6M). The
positions of the 51-nt oligomer substrate and 48- and 43-nt oligomer 3 degradation products are
indicated.
1-nt gap
Ligation
product
Fill-in product
Substrate
5*
1-nt gap
5* P
43-mer
48-mer
43-mer
48-mer
A
C
43-mer
48-mer
flap
nick
5*
Mt-Lig
-
Mt-Lig
-
1-nt flap 3-nt flap 5-nt flap
Mt-Lig
-
48
48
53
51
49
48
43 43
43
43-mer
51-mer
-
51
43
Mt-Lig
B
3'
5'
-
-
Mt-Lig
+
5'*
R E P O R T S
gion and into the duplex (Fig. 1C and fig.
S4). Nuclease activity required the presence
of magnesium or manganese. Mutation of a
conserved histidine residue abolished this
exonuclease activity, confirming that it is also
an intrinsic property of Mt-Lig (fig. S5).
We thus asked whether this single poly-
peptide could repair a DSB that requires 3
resection, gap filling, and ligation. In the pre-
sence of NTPs, Mt-Lig joined aligned DNA
duplexes possessing a 1-nt 3 flap adjacent to
a 3-nt gap (Fig. 2A and fig. S6). A similar,
albeit less efficient, reaction was observed in
the presence of dNTPs (fig. S6). Neither the
nuclease nor polymerase mutants were able to
repair this junction, confirming that both ac-
tivities were required to process the DSB (fig.
S7). Sequencing of ligated junctions in equiv-
alent assays with substrates with a 3-nt flap
adjacent to a 5-nt gap revealed that the micro-
homology sequence was retained and that the
mismatched flap was replaced by nucleotides
complementary to the template strand (18).
Mt-Ku stimulates joining of fully comple-
mentary ends by Mt-Lig (3). At incompatible
ends, Mt-Ku did not have a significant effect
on the removal of mismatched flaps, but it did
inhibit further digestion into the microhomol-
ogy region (Fig. 2B), suggesting that Mt-Ku
remains physically associated with this region
during repair. Using an in vitro polymerase
chain reaction (PCR)based plasmid repair
assay (19), we observed that Mt-Ku markedly
stimulated the joining of long linear DNA
molecules with different incompatible ends
(Fig. 2C). This occurred in the presence of
either dNTPs or NTPs (Fig. 2C). In contrast,
we observed no rejoining by T4 ligase in the
presence or absence of Mt-Ku (Fig. 2C). Join-
ing of partially complementary 5 (Hind III
Nhe I) and 3 (Pst IKpn I) overhangs ap-
peared to require microhomology-mediated
alignments that need gap filling and, in some
instances, 3 flap removal on one strand (Fig.
2D). Joining of a blunt end3 ss overhang
(Sma IAat II) appeared to require the addition
of one nucleotide by the terminal transferase
activity, followed by microhomology pairing
with the 3 overhang, flap resection, gap filling,
and ligation (Fig. 2D). In all cases, gap filling
accurately copied the template strand (18).
To examine the rejoining of chromosome
breaks in vivo, we exploited a variant of the
yeast-based Bsuicide deletion[ assay (20, 21),
which allows the simultaneous determination
of NHEJ and recombination frequencies (fig.
S8A). Consistent with prior observations
(20, 21), 75% of wild-type yeast cells re-
paired an I-SceI endonuclease-induced DSB
by recombination and 2% by NHEJ, with the
remainder dying (Fig. 3A). NHEJ occurred
predominantly by simple religation [resulting
in Ade
colonies (fig. S8)] and was decreased

99% by yku70 (Ku) deletion. Introducing
plasmids that express Mt-Ku and Mt-Lig
restored NHEJ to about half its level in wild-
type yeast (Fig. 3A and fig. S9A). Combina-
tions of Mt-Ku, Mt-Lig, and yku70 and dnl4
(yeast ligase IV) mutations demonstrated that
Mt NHEJ was truly reconstituted by a con-
certed species-specific Ku-ligase interaction
(Fig. 3B).
In Saccharomyces cerevisiae, NHEJ is also
dependent on the MRXcomplex (21, 22). MRX
may act as an end-bridging factor and/or
functionally interact with yeast Ku and Dnl4
(6, 23). Expression of the Mt NHEJ proteins in
yeast rad50 mutants substantially recovered
NHEJ (Fig. 3C), although to a lesser extent
than seen with yku70 or dnl4 mutants. Thus,
Mt NHEJ can occur in the absence of both
MRX and its bacterial ortholog SbcCD (23).
As with NHEJ mediated by yeast proteins
(20, 21, 24), Mt NHEJ reconstituted in yeast
occasionally resulted in imperfect repair,
evident as Ade
colonies in the absence of

the gene conversion donor. Sequencing of
these colonies revealed a variety of junctions
that occurred predominantly through mis-
pairing of the I-SceI 3 overhangs (fig. S9A).
To create a suicide deletion system that
selected specifically for such NHEJ events,
we substituted HO (endonuclease) for I-SceI
so that joints with a relative reading frame of
2 nucleotides yielded Ade
colonies (fig.
S8). Again consistent with previous results
(16, 25), 0.75% of all NHEJ events in wild-
type yeast were Ade
(Fig. 4A), and 950% of

these were HO(2) joints (fig. S9). With Mt
NHEJ reconstituted, the overall frequency of
NHEJ remained high (18), but the percentage
of Ade
events was substantially decreased

(Fig. 4A). Although some HO(2) processed
joints were formed, the HO(j1) joint now
predominated (Fig. 4B), providing a signa-
ture for Mt NHEJ (Fig. 4C and fig. S9). Mt
NHEJ proteins shifted the NHEJ profile even
in wild-type yeast (Fig. 4, A and C, and fig.
S9). Mt-Ku and Mt-Lig proteins can there-
1
2
3
pUC18
cut
repair
ligated
product
S 15 20 60 120
Fill-in (51mer)
Removed flap
product (48-mer)
Substrate
43-mer
49-mer
1-nt flap
3-nt gap
5*
+ + + +
_
Mt-Lig
C
GACGT
CCC
GGG
C
CATGG
CTGCA
G
CTG
G CATGG
Aat II Sma I
PstI Kpn I
A
C
CTGTACC
GACATGG
CTAGC
G TTCGA
Hind III Nhe I
A
No. of
clones
Non-homologous
Ends
Repaired
junction
Predicted
intermediates
6
C
GACG
CCC
GGG
C
GAC
CCC
GGG
CTGCCCC
GACGGGG
CTGCCC
GACGGG
10
2
CTGCACC
GACGTGG
8
TAGC
G TTCGA
A
CTAGC
G TTCGA
A
TTCGATCG
AAGCTAGC
TTCGAATCG
AAGCTTAGC
10
1
C
TGG
CTGCA
G
C
A
C
T
T
G
C
-
+ + + + + Mt-Lig
Mt-Ku
48-mer
51-mer
- -
43-mer
51-mer
3-nt flap
1-nt gap
5*
A B
C D
Fig. 2. Joining of DNA molecules with incompatible ends by Mt NHEJ. (A) Mt-Lig (1 6M) and the
indicated DNA duplex were incubated with MgCl
2
, NTPs, and ATP for increasing times (minutes). S,
DNA substrate alone. The positions of the 49-nt oligomer (substrate), 48-nt oligomer (removal of
flap), 51-nt oligomer (gap fill-in product), and 92-nt oligomer (ligated product) are indicated.
Asterisks denote labeled strands. (B) Effect of Mt-Ku on the nuclease activity of Mt-Lig. DNA
duplexes were incubated with 1 6M Mt-Lig and increasing amounts of Mt-Ku (0.8, 1.6, 3.2, and
6.4 6M) without ATP. (C) Plasmid DNA cut with two restriction enzymes was incubated with Mt-Lig
(4 pmol) and increasing amounts of Mt-Ku (0.05, 0.1, 0.5, and 1 pmol), with either dNTPs or NTPs,
followed by PCR to monitor the rate of repair (bottom). Slower migrating products correspond to
uncut plasmid, and the fastest migrating products (white arrows) correspond to successfully repaired
junctions that have lost 500 base pairs between the restriction sites. pUC18, plasmid DNA.
(D) Plasmids from experiments similar to (C) were transformed into bacteria and sequenced. The
starting ends, final products, and inferred alignment intermediates are shown.
R E P O R T S
fore catalyze processed NHEJ in chromo-
somes, but despite this ability, repair is high-
ly accurate at compatible DSB ends.
With the exception of 5 trimming, Mt-Ku
and Mt-Lig thus encode a self-sufficient NHEJ
repair machine. The accuracy of this machine
suggests that its different activities must be
highly regulated to correctly handle different
end configurations. Eukaryotic NHEJ involves
many more proteins (1, 2), but we suggest that
the mechanisms of end-processing will be fun-
damentally similar, although coordinated by
protein-protein interactions rather than linkage
in a single polypeptide. Indeed, the variety of
nucleotidyl transferase activities in Mt-Lig
suggests a reduced dependence on template
similar to that of mammalian Pol 6 and Pol 4
(2628), which likely accommodates the
limited base pairing inherent to NHEJ. As
with Pol 6, Mt-Lig also appears most likely to
incorporate NTPs in vivo, which has been
suggested to be beneficial given the relative
concentrations of NTPs and dNTPs outside of
S phase (17). Incorporated ribonucleotides
could be replaced by described mechanisms
(29). Finally, Mt proteins catalyze NHEJ in
yeast even though prokaryotes lack histones.
This might suggest that a specific histone
interaction is not required during NHEJ.
1. L. Krejci, L. Chen, S. Van Komen, P. Sung, A. E.
Tomkinson, Prog. Nucleic Acid Res. Mol. Biol. 74,
159 (2003).
2. S. E. Critchlow, S. P. Jackson, Trends Biochem. Sci. 23,
394 (1998).
3. G. R. Weller et al., Science 297, 1686 (2002).
4. L. G. DeFazio, R. M. Stansel, J. D. Griffith, G. Chu,
EMBO J. 21, 3192 (2002).
5. D. Ramsden, M. Gellert, EMBO J. 17, 609 (1998).
6. L. Chen, K. Trujillo, W. Ramos, P. Sung, A. E. Tomkinson,
Mol. Cell 8, 1105 (2001).
7. L. Chen, K. Trujillo, P. Sung, A. E. Tomkinson, J. Biol.
Chem. 275, 26196 (2000).
8. S. A. Nick McElhinny, C. M. Snowden, J. McCarville,
D. Ramsden, Mol. Cell. Biol. 20, 2996 (2000).
9. L. Aravind, E. V. Koonin, Genome Res. 11, 1365 (2001).
10. E. V. Koonin, Y. I. Wolf, A. S. Kondrashov, L. Aravind,
J. Mol. Microbiol. Biotechnol. 2, 509 (2000).
11. G. R Weller, A. J. Doherty, FEBS Lett. 505, 340 (2001).
12. T. E. Wilson, L. M. Topper, P. L. Palmbos, Trends
Biochem. Sci. 28, 62 (2003).
13. B. Arezi, R. Kuchta, Trends Biochem. Sci. 11, 572 (2000).
14. K. N. Mahajan, S. A. Nick McElhinney, B. S. Mitchell,
D. A. Ramsden, Mol. Cell. Biol. 22, 5194 (2002).
15. H.-M. Tseng, A. E. Tomkinson, J. Biol. Chem. 277,
45630 (2002).
16. T. E. Wilson, M. R. Lieber, J. Biol. Chem. 274, 23599
(1999).
17. S. A. Nick McElhinney, D. Ramsden, Mol. Cell. Biol.
23, 2309 (2003).
18. M. Della et al., unpublished data.
19. D. A. Ramsden, T. T. Paull, M. Gellert, Nature 388,
488 (1997).
20. E. Karathanasis, T. E. Wilson, Genetics 161, 1015 (2002).
21. T. E. Wilson, Genetics 162, 677 (2002).
22. S. J. Boulton, S. P. Jackson, EMBO J. 17, 1819 (1998).
23. J. C. Connelly, D. R. Leach, Trends Biochem. Sci. 27,
410 (2002).
24. T. E. Wilson, U. Grawunder, M. R. Lieber, Nature 388,
495 (1997).
25. J. K. Moore, J. E. Haber, Mol. Cell. Biol. 16, 2164 (1996).
26. K. Bebenek, M. Garcia-Diaz, L. Blanco, T. A. Kunkel,
J. Biol. Chem. 278, 34685 (2003).
27. M. Garcia-Diaz et al., Mol. Cell 13, 561 (2004).
28. S. Covo, L. Blanco, Z. Livneh, J. Biol. Chem. 279, 859
(2004).
29. B. Rydberg, J. Game, Proc. Natl. Acad. Sci. U.S.A. 99,
16654 (2002).
30. We thank S. Bell and E. Koonin for discussions, S. Bell
for assistance with assays, and T. Carr, P. Jeggo, and
K. Caldecott for comments on the manuscript. A.J.D.
is a Royal Society University Research Fellow and
L.T. is supported by a Biotechnology and Biological
Sciences Research Council (BBSRC) studentship. Sup-
ported by grants from the BBSRC, the Association for
International Cancer Research, the Royal Society, the
U.S. Public Health Service, and the Pew Charitable
Trusts. Molecular interaction data have been depos-
ited in the Biomolecular Interaction Network Data-
base with accession code 153967.
DC1
Figs. S1 to S9
3 May 2004; accepted 3 September 2004
w
t
y
k
u
7
0
w
t
y
k
u
7
0
0
25
50
75
100
0
1
2
3
vectors bacterial
Ku + Lig
Ku + Lig
Lig
Ku
_
yeast
bacteria
K
u
+
L
ig
g i
L
u
K
_
0
0.5
1.0
1.5
2.0
d
n
4
l
d
n
4
l
r
a
d
5
0
r
a
d
5
0
d
n
4
l
r

a
d
5
0
1
0.5
0
vectors
bacterial
Ku + Lig
%

g
e
n
e

c
o
n
v
e
r
s
i
o
n
%

p
r
e
c
i
s
e

N
H
E
J
%

p
r
e
c
i
s
e

N
H
E
J
%

p
r
e
c
i
s
e

N
H
E
J
A B
C
Fig. 3. Reconstitution of NHEJ in yeast by Mt proteins. (A) Frequencies of gene conversion and
simple religation NHEJ in wild-type and yku70 mutant yeast demonstrate reconstitution of NHEJ
by combined expression of Mt-Ku and Mt-Lig. wt, wild-type. (B) All combinations of yeast and
bacterial Ku and ligase genes were tested for NHEJ function. Labels indicate those proteins that
were present in the cell. (C) NHEJ catalyzed by Mt proteins is only partially dependent on an intact
MRX complex. No Ade
colonies were recovered from dnl4 rad50 yeast with vectors only.
Fig. 4. Mt NHEJ exhib-
its a different pattern
of imprecise NHEJ than
yeast NHEJ. (A) The
extent of 2 frame-
shifted NHEJ was de-
termined as a fraction
of all NHEJ events. Mt
NHEJ showed a mark-
edly lower 2 fre-
quency than yeast
NHEJ. No Ade
colo-
nies recovered from
yku70 yeast with vec-
tors only. (B) Inferred
NHEJ intermediates for
t h e HO( 2 ) a n d
HO(j1) events that
givea2readingframe.
(C) Genomic DNA was
prepared from bulk cul-
tures of the indicated yeast bearing the HO suicide deletion system after outgrowth in adenine-
selective medium. Flanking PCR was performed, and products were electrophoresed alongside
standards derived from colonies of known joint types.
w
t
w
t
y
k
u
7
0
0.75
1
0.5
0.25
0
vectors bacterial
Ku + Lig
H
O
(
+
2
)
H
O
(
-
1
)
vec. Ku + Lig
wt yku70
HO(+2)
HO(-1)
%

+
2

o
f

t
o
t
a
l

N
H
E
J
A B
C
R E P O R T S
Elemental and Redox Analysis of
Single Bacterial Cells by X-ray
Microbeam Analysis
Kenneth M. Kemner,
1
* Shelly D. Kelly,
1
Barry Lai,
1
Joerg Maser,
1
Edward J. OLoughlin,
1
Deirdre Sholto-Douglas,
1
Zhonghou Cai,
1
Mark A. Schneegurt,
2
Charles F. Kulpa Jr.,
3
Kenneth H. Nealson
4
High-energy x-ray fluorescence measurements were used to make elemental
maps and qualitative chemical analyses of individual Pseudomonas fluorescens
strain NCIMB 11764 cells. Marked differences between planktonic and
adhered cells were seen in the morphology, elemental composition, and
sensitivity to Cr(VI) of hydrated cells at spatial scales of 150 nm. This
technology can be applied to natural geomicrobiological systems.
Attachment of prokaryotic cells to surfaces
during biofilm formation not only leads to
major changes in metabolism, resistance, and
survivability (1), but the substrates metabo-
lized by such communities (such as iron or
manganese oxides) and their products (such as
uraninite or chromium oxides) often are
insoluble and associated with the attached
bacteria (24). Thus, quantifying and specify-
ing the locations of cellular components,
metabolic reactants, and metabolic products
of prokaryotes is technologically challenging.
There are no techniques currently avail-
able that use noninvasive, nondestructive
analytical techniques with the spatial resolu-
tion to examine living and hydrated samples
at the nanometer scale and that are capable
of both imaging and chemical determina-
tions. In particular, most analytical methods
with high spatial resolution (typically with
electron or proton microprobes) use high-
vacuum systems, almost certain to alter the
cellular material itself and the locations of
key associated elements.
To address this issue, we have used high-
energy x-ray fluorescence (XRF) methods
with a spatial resolution of 150 nm to
examine elemental compositions of single
hydrated bacterial cells (5). We determined
elemental compositions of single cells of P.
fluorescens NCIMB 11764, an aquatic
Gram-negative bacterium that exists natural-
ly both in planktonic form and in surface-
adhered biofilms. We saw changes in cell
morphology and elemental composition of
single cells when attached to a solid sub-
strate. Many studies have investigated the
interactions of metals and ions with bacteria
(68) and aquatic protists (9), but not with
hydrated bacteria at the single-cell level.
Zone plate fabrication technology for the
production of high-energy x-ray microprobes
(10) combines high spatial resolution with
high elemental sensitivity (better than 1
10
4
atoms within a spot size of 150 nm) (11)
and x-ray absorption spectroscopy to probe
chemical interactions at this scale (12). The
high-energy x-ray microprobe has better
elemental sensitivity than most charged-
particle probes and allows the investigation
of heterogeneous hydrated samples. The
x-ray microprobe also allows investigation
of structures tens of microns beneath solid-
phase surfaces.
High-energy XRF microscopy showed
planktonic cells were rod shaped and contained
high concentrations of virtually all elements
expected in living cells (Fig. 1, table S1); the
attached cells were rounded and had notable
excesses of Ca and P associated with them
(Fig. 2, table S1). With surface adherence, EP^
associated with the cells increased from
16,000 to 650,000 ppm, and ECa^ in-
creased from 4,000 to 550,000 ppm,
yielding 1:1 stoichiometry for these two
elements (table S1). Other elements showed
only minor changes in the attached cells.
The greater ECa^ and EP^ in the adhered cells
versus the planktonic cells remains un-
1
Environmental Research Division and Advanced Pho-
ton Source, Argonne National Laboratory, Argonne, IL
604394843, USA.
2
Department of Biological Sciences,
Wichita State University, Wichita, KS 67260, USA.
3
Department of Biological Sciences, University of Notre
Dame, Notre Dame, IN 46556, USA.
4
Department of
Earth Sciences, Mail Code 0740, University of Southern
California, Los Angeles, CA 900890740, USA.
E-mail: kemner@anl.gov
Fig. 1. False-color
micro-XRF maps of
qualitative spatial dis-
tributions and concen-
tration gradients of
elements in and around
planktonic P. fluores-
cens microbes har-
vested before (A) and
after (B) exposure to
potassium dichromate
[Cr(VI)] solution (1000
ppm) for 6 hours.
Fig. 2. False-color
micro-XRF maps of
qualitative spatial dis-
tributions and concen-
tration gradients of
elements in and around
surface-adhered P.
fluorescens microbes
harvested before (A)
and after (B) exposure
to potassium dichro-
mate [Cr(VI)] solution
(1000 ppm) for 6 hours.
Because of the ex-
tremely high concen-
trations of Ca and P in
the surface-adherent
microbes, the intensi-
ties of neighboring
XRF signals from S,
Cl, and K could not be
determined accurate-
ly. Therefore, spatial distributions for these elements are not illustrated.
R E P O R T S
explained, but it probably involves the extra-
cellular polysaccharide (EPS) matrix laid down
by the P. fluorescens cells and the precip-
itation of apatite (5). Strains of this species
produce various acidic EPSs (1316) that can
bind calcium ions preferentially from solution
(1719). SEM imaging of the cells from our
experiments indicated the formation of an
extracellular matrix.
Because attached cells are thought to have
enhanced resistance to stress, we challenged
the planktonic cells with Cr(VI) at 1000 and
25 ppm for 6 hours and the attached cells with
Cr(VI) at 1000 ppm for 6 hours. After this
treatment, planktonic P. fluorescens cells
were markedly altered; they stained as dead
cells in the live or dead stain reaction (5), lost
almost all of their cellular transition ele-
ments, and accumulated substantial amounts
of Cr (Fig. 1). In contrast, the attached cells
showed virtually no change in elemental
composition and no uptake of Cr into the
cells (Fig. 2), and living cells were abundant.
Finally, we used the XRF microscopy
system to measure x-ray absorption near edge
spectra (XANES) (12) to determine the redox
state of the Cr in this system. With micro-
XANES we investigated the chemical speci-
ation of Cr in the Cr-enriched areas seen in
Fig. 3. Comparison of these Cr K-edge spectra
to standards (Fig. 4) indicates reduced Cr
Ei.e., Cr(III)^, consistent with association of
Cr with a phosphoryl functional group.
The XRF approach also revealed that,
when attached to a surface, the cells changed
shape slightly, lost some transition elements
(Co, Cu, Ni, and Zn), gained others (Fe and
Mn), and became resistant to exposure to
high levels of Cr(VI), which shows the
utility of this technique for investigating a
cell_s metabolic state. Combining elemental
analysis with the ability to measure redox
state and local chemistry is advantageous for
dissecting the activity of metal-active bacte-
ria in geomicrobiological systems.
1. J. W. Costerton et al., Annu. Rev. Microbiol. 41, 435
(1987).
2. D. R. Lovley, Microbiol. Rev. 55, 259 (1991).
3. K. H. Nealson, D. Saffarini, Annu. Rev. Microbiol. 48,
311 (1994).
4. J. W. Costerton, Z. Lewandowski, D. E. Caldwell, D. R.
Korber, H. M. Lappin-Scott, Annu. Rev. Microbiol. 49,
711 (1995).
6. T. J. Beveridge, Annu. Rev. Microbiol. 43, 147 (1989).
7. G. Winkelmann, C. J. Carrano, Transition Metals in
Microbial Metabolism (Harwood Academic Publish-
ers, Amsterdam, 1997).
8. S. Langley, T. J. Beveridge, Can. J. Microbiol. 45, 616
(1999).
9. B. S. Twinning et al., Anal. Chem. 75, 3806 (2003).
10. W. Yun et al., Rev. Sci. Instrum. 70, 2238 (1999).
11. Z. Cai et al., in X-ray Microscopy: Proceedings of
the Sixth International Conference, W. Meyer-Ilse,
T. Warwick, D. Attwood, Eds. (American Institute of
Physics, New York, 2000), pp. 472477.
12. K. M. Kemner et al., J. Synchrotron Radiat. 6, 639
(1999).
13. G. Skjak-Braek, H. Grasdalen, B. Larsen, Carbohydr.
Res. 154, 239 (1986).
14. R. R. Read, J. W. Costerton, Can. J. Microbiol. 33,
1080 (1987).
15. W. F. Fett, S. F. Osman, M. F. Dunn, Appl. Environ.
Microbiol. 55, 579 (1989).
16. E. Conti, A. Flaibani, M. ORegan, I. W. Sutherland,
Microbiology 140, 1125 (1994).
17. G. T. Grant, E. R. Morris, D. A. Rees, P. J. C. Smith,
D. Thom, FEBS Lett. 32, 195 (1973).
18. J. E. Gregor, E. Fenton, G. Brokenshire, P. Van den
Brink, B. OSullivan, Water Res. 30, 1319 (1996).
19. I. Braccini, R. P. Grasso, S. Perez, Carbohydr. Res. 317,
119 (1999).
20. We thank M. Boyanov, K. Germino, P. Illinski, D. Legnini,
M. Mundo, S. T. Pratt, W. Rodrigues, and W. Yun for
earlier contributions to this project. This work is
supported by the Natural and Accelerated Bioremedia-
tion (NABIR) Research Program, Office of Biological and
Environmental Research, Office of Science, U.S. Depart-
ment of Energy (DOE). Additional support for K.M.K. was
provided by the DOE-Office of Science Early Career
Scientist and Engineer Award. Work at the Advanced
Photon Source is supported by the DOE Office of
Science, Office of Basic Energy Sciences.
DC1
SOM Text
Table S1
2 August 2004; accepted 26 August 2004
A Chromosome 21 Critical Region
Does Not Cause Specific Down
Syndrome Phenotypes
L. E. Olson,
1
* J. T. Richtsmeier,
2
J. Leszl,
2
R. H. Reeves
1
.
The Down syndrome critical region (DSCR) is a chromosome 21 segment
purported to contain genes responsible for many features of Down syndrome
(DS), including craniofacial dysmorphology. We used chromosome engineer-
ing to create mice that were trisomic or monosomic for only the mouse
chromosome segment orthologous to the DSCR and assessed dysmorpholo-
gies of the craniofacial skeleton that show direct parallels with DS in mice
with a larger segmental trisomy. The DSCR genes were not sufficient and were
largely not necessary to produce the facial phenotype. These results refute
specific predictions of the prevailing hypothesis of gene action in DS.
Trisomy 21 is among the most complex
genetic insults compatible with human sur-
vival past term. The genetic complexity and
individual variability of DS phenotypes pose
a considerable challenge to understanding
mechanisms by which development is dis-
Fig. 3. False-color micro-XRF maps of Ca (A),
identifying the locations of surface-adherent mi-
crobes exposed to Cr(VI) solution (1000 ppm),
and Cr (B), depicting the locations of elevated
concentrations of Cr within 1 to 3 6m of
surface-adherent P. fluorescens. Cell positions
[determined from Ca distribution in (A)] are
drawn in white.
Fig. 4. Cr K-edge micro-XANES spectra repre-
senting potassium dichromate [Cr(VI)] solution
added to growth medium; a chromium phos-
phate dihydrate [Cr(III)] standard; a chromium
acetate [Cr(III)] standard; and the elevated
microbe-associated Cr concentrations illustrat-
ed in Fig. 3B.
R E P O R T S
rupted in trisomy. The DSCR concept sug-
gesting that most phenotypes are the product
of a few dosage-sensitive genes has dominat-
ed the field of DS research for three decades,
but until now there has been no method for
definitive testing of this hypothesis.
DS occurs in 1 out of 700 live births. The
disorder is the leading cause of congenital
heart disease and the most frequent genetic
cause of mental retardation. Characteristic
dysmorphology of the face and underlying
skeleton, alterations in brain structure, and
early onset Alzheimer pathology occur in
all individuals with DS, though with vari-
able severity and onset. The occurrence of
Hirschsprung_s disease, testicular cancer,
and childhood onset leukemia is elevated,
and many additional clinical features occur
with varying frequency in DS (1).
Rare individuals have chromosome rear-
rangements resulting in triplication of a sub-
set of HSA21 genes (segmental trisomy 21)
(2, 3). Correspondence of the smallest region
of overlap among individuals with the same
DS features led to the concept that a critical
chromosomal region (DSCR) delimited a
dosage-sensitive gene or genes that are re-
sponsible for various features of DS. The
most precise description of the DSCR ex-
tends about 5.4 Mb on HSA21 from a prox-
imal boundary between markers D21S17 and
D21S55 to a distal boundary between MX1
and BCEI (4, 5). This region has been asso-
ciated with several major DS phenotypes
including facial features that result from
dysmorphology of the craniofacial skeleton
(6, 7) (fig. S1). The DSCR hypothesis
predicts that a gene or genes in this region
are sufficient to produce this phenotype.
Several recent developments enable di-
rect testing of the genetic basis for disrupted
development in trisomy. Annotation of the
finished sequence of HSA21 provides a
complete gene catalog (8). Manipulation of
the mouse germ line through embryonic stem
(ES) cells supports the creation of defined
segmental aneuploidies (9). Quantitative
assessments of phenotypes with direct paral-
lels between DS and trisomic mice provide
readouts for these genetic experiments and
suggest that these features arise from pertur-
bations of common evolutionarily derived
developmental pathways (10).
Whereas some differences between mouse
and human gene content have been reported
for HSA21 predicted genes (11), established
genes are nearly perfectly conserved in con-
tent and order between species (12) (table S1).
Ts65Dn mice contain a marker chromosome
resulting in segmental trisomy for orthologs
of about half of the genes on HSA21. These
mice have been used widely in studies rel-
evant to DS and display a variety of phe-
notypes that parallel or predict those in DS
(10).
1
Department of Physiology, Johns Hopkins University
School of Medicine, Baltimore, MD 21205, USA.
2
Department of Anthropology and Program in Ge-
netics, Pennsylvania State University, University Park,
PA 16802, USA.
*Present address: Department of Biology, University of
Redlands, Redlands, CA 92373, USA.
E-mail: rreeves@jhmi.edu
Fig. 2. Trisomic segments represented in
mouse models relative to HSA21. The DSCR is
indicated as an open box where it is present
and by a dashed line where it is deleted.
Approximate size in megabases of triplicated
information is based on genomic sequence;
gene number is from Gardiner et al. (11). Gene
content for each segment is given in table S1.
0
5
10
15
20
25
30
10 20 30 40 50 60
W
e
i
g
h
t

(
g
)
Ms1Rhr
Euploid
Ts1Rhr
Days after Birth
5' Neo
R
gene
3' Neo
R
gene
LoxP site
Genomic flank
Genomic flank
Hyg
R
gene
Puro
R
gene
LoxP site
Genomic flank
Genomic flank
Reconstituted
Neo
R
gene
Cbr region
Mx region
Cre
Recombinase
Duplication
Deletion
Eco RV Hind III
WT 25
WT 25
WT MxP 25
9 kb
7.5 kb
5 kb
4.1 kb
1.7 kb
Eco RV Hind III
5'neo puro DTA
Mx1
Mx2
Eco RV Eco RV Hind III Hind III
P P
MXF1
SLK2R1
5 neo
hyg
3neo 5 neo puro
Bgl II
P
Mx region Cbr region
Deleted Chr
Duplicated Chr
3neo
hyg
puro
Bgl II Bgl II
Bgl II
P
390 bp
350 bp
A
D
E
F
G
B
C
Fig. 1. Construction of a duplication or deletion of the MMU16 region orthologous to the DSCR.
(A) LoxP sites were targeted to asymmetrical positions on MMU16 at Cbr1 (13) and adjacent to
Mx2. Each targeting vector contained a LoxP site (triangle), a selectable antibiotic resistance gene
(hyg or puro), and half of the neomycin resistance gene (5 or 3 neo). (B) The Mx-Lox vector
produced a 1.7-kb PCR product from the targeted allele in line MxP25. Wild-type (WT) and
targeted alleles produced 9.0- and 7.5-kb restriction fragments with EcoRV and 5.0- and 4.1-kb
fragments with HindIII. Arrows identify PCR primers; P designates probes. (C) PCR primers used to
screen vector sequences for recombination after Cre-mediated translocation. (D) PCR products
from neo
r
ES lines produced the 350 and 390base pair (bp) fragments expected for deletion and
duplication, respectively. (E) Metaphase FISH with one bacterial artificial chromosome (BAC) that
maps to the DSCR (red) and a second BAC proximal to it (green) shows one chromosome with a
single green signal (arrow) and a second with green and red plus a yellow signal, indicating overlap.
(F) Interphase FISH shows a green signal by itself representing the deleted MMU16 and a green
signal with two adjacent red signals representing the duplication. (G) Body weights of Ts1Rhr mice
are significantly larger than controls. Standard errors (bars) are indicated.
R E P O R T S
To assess the contribution of DSCR
genes to DS facial features that have been
mapped to this region, we generated ES cells
carrying a 3.9-Mb reciprocal duplication or
deletion (Fig. 1A). This segment contains
mouse orthologs of the 33 conserved and
minimally conserved genes in the human
DSCR (Fig. 2). Cre-mediated recombination
between LoxP sites located on different
chromosomes was used to produce the
duplication and deletion (9) (Fig. 1A).
Consecutive targeting with homologous
recombination vectors containing LoxP
sequences established the boundaries for the
translocation at Cbr1 (13) and at a site
adjacent to the Mx2 gene (Fig. 1B). The
targeting vectors were designed so that Cre-
mediated recombination between LoxP sites
would make cells resistant to G418 (neo
r
).
The CAGGS-Cre expression vector was
electroporated into doubly targeted ES cells
and two neo
r
clones were recovered. The
presence of both recombined chromosomes
was confirmed by Southern blotting and
polymerase chain reaction (PCR) followed
by sequencing (Fig. 1, C and D). Metaphase
and interphase fluorescence in situ hybrid-
ization (FISH) demonstrated the predicted
association of signals from the duplicated
and deleted chromosomes (Fig. 1, E and F).
These karyotypically normal ES lines were
injected into blastocysts and one of the re-
sulting chimeras exhibited germline transmis-
sion when bred to C57BL/6J (B6) mice. Mice
with segmental monosomy due to the deleted
MMU16 (Ms1Rhr mice) and with segmental
trisomy due to the duplication (Ts1Rhr) were
born, and both were fertile.
Ms1Rhr mice were smaller than (B6
129)F
1
controls from birth to adulthood
(Fig. 1G). However, adult Ts1Rhr mice
were significantly larger than euploid mice,
in contrast to Ts65Dn mice, which were
smaller than their euploid littermates (14).
Femur length, a commonly used index of
body size, was shorter in Ts65Dn mice than
in controls, whereas femurs were longer in
Ts1Rhr mice than in controls. Short stature
in DS is a feature attributed to the DSCR
(4, 5). Thus, whereas trisomy for orthologs
of about half of the genes on HSA21
produces an effect similar to that in DS in
Ts65Dn mice, the opposite effect occurred
in Ts1Rhr mice, which were trisomic only
for the DSCR.
Anomalies of the skull contribute to the
distinctive face in DS. The mandible is small
(7), contributing to the characteristic open
mouth and protruding tongue, features that
have been attributed to trisomy for genes
within the DSCR (4, 5). The DS skull shows
a compact midface with reduced maxilla and
zygomatic bones, and brachycephaly. Ts65Dn
mice show specific abnormalities in the man-
dible and in bones of the face and neuro-
cranium that directly parallel these patterns
in DS (6).
These parameters were assessed quanti-
tatively and compared in four mouse models.
The Ts1Rhr skull was larger than that of
euploid littermates (Fig. 3). These differ-
ences were not localized to specific bones of
the face or neurocranium but instead dem-
onstrated an overall rostrocaudal elongation
of the skull. There was no significant dif-
ference between Ts1Rhr and euploid litter-
mates along the mediolateral axis local to the
face or neurocranium.
These differences contrast sharply with the
findings of previous studies of Ts65Dn mice
(6). Skulls of Ts65Dn were significantly
smaller than euploid littermates. Differences
were most pronounced along the rostrocaudal
axis local to the bones of the face corre-
sponding with reduced facial dimensions in
humans with DS. The reduced rostrocaudal
dimension coupled with increased cranial
vault width local to the parietal bones pro-
duced a brachycephalic skull corresponding
with the human condition (Fig. 3).
The mandibles of Ts1Rhr mice were
larger overall than mandibles of their euploid
littermates. Out of 21 linear distances, 5
were significantly larger in Ts1Rhr, and
these converged on the condyle, inferior
ramus, and incisive alveolar of the murine
mandible (Fig. 4). As was the case for face
and skull, this pattern in Ts1Rhr mandibles
was in marked contrast to that in Ts65Dn
mice, which were significantly smaller over-
all. Further, differences in Ts65Dn were
centered on the coronoid and angular pro-
cesses (6). The DS mandible is significantly
reduced in size as well (7). Direct compar-
isons between mouse and human mandibles
are available for a few dimensions; these
show a DS-like pattern in Ts65Dn but not in
Ts1Rhr mice (Fig. 4).
Morphological assessment of the neuro-
cranium, face, and mandible demonstrate that
trisomy of only critical region genes is not
sufficient to produce the anomalies seen in
Ts65Dn mice that parallel those assigned to
the DSCR. To determine whether critical
region genes are necessary to produce pheno-
types mapped to this region, we examined
Fig. 3. Linear measurements that contribute to dysmorphology in Ts65Dn, Ts1Cje, and Ms1Rhr/
Ts65Dn form a closely related set, whereas trisomy for the DSCR alone produces a distinct
dysmorphology in Ts1Rhr. Red lines indicate linear distances between landmarks that were
significantly smaller in trisomic mice relative to euploid mice; blue lines indicate distances that were
larger. Statistical significance was determined using Euclidean Distance Matrix Analysis confidence
intervals (17). This is not a simple scaling difference because the magnitude of the differences
varies from one measure to another and the set of differences contributing to dysmorphology in
Ts1Rhr is different from that in the other models. Landmarks are defined in fig. S3.
Ts65Dn
Ts1Rhr
Down syndrome
Fig. 4. The Ts1Rhr mandible is larger overall
and has a distinctly different shape than
Ts65Dn. Results of three-dimensional compar-
ative analysis of mandibular shape show those
distances between landmarks that are signifi-
cantly smaller in trisomic mice as red lines,
whereas blue lines indicate distances that are
larger. Purple lines drawn on the DS and Ts65Dn
mandibles are measures that are significantly
smaller than those for euploid in both (6, 7).
Landmarks are described in fig. S3.
R E P O R T S
skull morphology of the Ms1Rhr/Ts65Dn
mouse. Crossing the Ms1Rhr deletion onto
the Ts65Dn background produced mice that
were trisomic for about 70% of the genes
trisomic in Ts65Dn but had the normal two
copies of genes in the critical region (Fig. 2).
Similar to Ts65Dn mice and in contrast to
Ts1Rhr mice, the skull was significantly
smaller than that of euploid littermates in
these mice (Fig. 3). Ms1Rhr/Ts65Dn mice did
not show disproportionate changes along the
rostrocaudal and mediolateral axis (i.e., they
did not have brachycephaly). Overall, how-
ever, the pattern of dysmorphology in the
Ms1Rhr/Ts65Dn craniofacial skeleton was
similar to Ts65Dn, whereas the anomalies
seen in Ts1Rhr mice were distinctly different.
Thus, although genes in the critical region
may contribute to skull development, the
Ts65Dn pattern is produced substantially
without their contribution.
One alternative to the critical region
concept suggests that anomalies in DS occur
primarily because of disruption of genetic
homeostasis and, therefore, the number of
triplicated genes rather than their specific
functions explain DS phenotypes (15). Thus,
Ts1Rhr mice, with only 32% of the trisomic
genes in Ts65Dn, should have a mild form of
the Ts65Dn phenotype. This was not the case.
Furthermore, we compared the craniofacial
dysmorphology in Ms1Rhr/Ts65Dn mice,
which are trisomic for about 70% of the genes
triplicated in Ts65Dn, with that in Ts1Cje
segmental trisomic mice (16), which are
trisomic for 78% of Ts65Dn genes (Fig. 2).
Ts65Dn, Ts1Cje, and Ms1Rhr/Ts65Dn mice
showed expansion, no change, and reduction
in measures of the middle aspect of the cranial
vault, respectively, but the models shared
reductions of the face and rostrocaudal aspects
of the cranial vault (Fig. 3).
Overall, the Ts65Dn, Ts1Cje, and Ms1Rhr/
Ts65Dn face and skull are affected similar-
ly, whereas the pattern produced by three
copies of the DSCR alone in Ts1Rhr mice is
very different. Statistical analyses of cranial
shape in these precisely constructed genetic
models show that DSCR genes alone are not
sufficient and are largely unnecessary to
produce the facial dysmorphology attributed
to this region. This result points to interac-
tions among noncontiguous genes that could
not have been predicted from or observed in
human studies. The critical region concept
proposes that a small number of dosage-
sensitive genes have large effects on phe-
notypes when present in three copies. The
developmental instability hypothesis consid-
ers that nonspecific, small effects of many
genes perturb genetic homeostasis. We favor
a synthesis of these two hypotheses, which
acknowledges that a triplicated gene, the sol-
itary effect of which is inconspicuous, could
contribute to a trisomic phenotype in com-
bination with other genes based on the spe-
cificity of effects and interactions of these
genes.
1. W. I. Cohen, Down Syndrome Q. 4, 3 (September,
1999).
2. E. Niebuhr, Humangenetik 21, 99 (1974).
3. M. K. McCormick et al., Genomics 5, 325 (1989).
4. J. M. Delabar et al., Eur. J. Hum. Genet. 1, 114
(1993).
5. J. R. Korenberg et al., Proc. Natl. Acad. Sci. U.S.A. 91,
4997 (1994).
6. J. Richtsmeier, L. Baxter, R. Reeves, Dev. Dyn. 217,
137 (2000).
7. E. Kisling, Cranial Morphology in Downs Syndrome: A
Comparative Roentgencephalometric Study in Adult
Males (Munksgaard, Copenhagen, 1966).
8. M. Hattori et al., Nature 405, 311 (2000).
9. R. Ramirez-Solis, P. Liu, A. Bradley, Nature 378, 720
(1995).
10. R. Reeves, L. Baxter, J. Richtsmeier, Trends Genet. 17,
83 (2001).
11. K. Gardiner, A. Fortna, L. Bechtel, M. T. Davisson,
Gene 318, 137 (2003).
12. R. J. Mural et al., Science 296, 1661 (2002).
13. L. E. Olson et al., Cancer Res. 63, 6602 (2003).
14. M. T. Davisson, personal communication.
15. B. Shapiro, Am. J. Med. Genet. 14, 241 (1983).
16. H. Sago et al., Proc. Natl. Acad. Sci. U.S.A. 95, 6256
(1998).
17. S. R. Lele, J. T. Richtsmeier, An Invariant Approach to
Statistical Analysis of Shapes (Chapman & Hall/CRC,
Boca Raton, FL, 2001).
18. We thank J. D. Gearhart for insightful comments,
M. Cowan for ES cells and chimeras, and S. South for
FISH analysis. This work was supported by a Howard
Hughes Predoctoral Fellowship to L.E.O. and by a Public
Health Service awards HD38384 and HD24605 (R.H.R.).
DC1
Figs. S1 to S3
Table S1
References
9 April 2004; accepted 23 August 2004
In Silico Genetics: Identification of
a Functional Element Regulating
H2-E! Gene Expression
Guochun Liao,
1
* Jianmei Wang,
1
* Jingshu Guo,
1
John Allard,
1
Janet Cheng,
1
Anh Ng,
1
Steve Shafer,
2
Anne Puech,
3
John D. McPherson,
4
Dorothee Foernzler,
5
Gary Peltz,
1
. Jonathan Usuka
1
Computational tools can markedly accelerate the rate at which murine
genetic models can be analyzed. We developed a computational method for
mapping phenotypic traits that vary among inbred strains onto haplotypic
blocks. This method correctly predicted the genetic basis for strain-specific
differences in several biologically important traits. It was also used to identify
an allele-specific functional genomic element regulating H2-E! gene expres-
sion. This functional element, which contained the binding sites for YY1 and a
second transcription factor that is probably serum response factor, is located
within the first intron of the H2-E! gene. This computational method will
greatly improve our ability to identify the genetic basis for a variety of
phenotypic traits, ranging from qualitative trait information to quantitative
gene expression data, which vary among inbred mouse strains.
Commonly available inbred mouse strains
can be used to genetically model many traits
that vary in the human population, including
those associated with disease susceptibility.
We have previously shown that chromosom-
al regions regulating phenotypic traits could
be computationally predicted by correlative
analysis of phenotypic data obtained from in-
bred mouse strains and the extent of allele
sharing within genomic regions (1). However,
this computational method had several lim-
itations. It identified 60-Mb chromosomal
regions that contained hundreds of genes, and
the predictions were assessed by relative
(percentile ranking), rather than absolute (P-
value), statistical criteria.
Single-nucleotide polymorphism (SNP)
and allelic information in our mouse SNP
database (2) were used to produce a high-
resolution map of the haplotypic block
structure of the mouse genome covering 16
commonly used inbred mouse strains (3). This
1
Department of Genetics and Genomics, Roche Palo
Alto, 3431 Hillview Avenue, Palo Alto, CA 94304
1397, USA.
2
Department of Anesthesia, Stanford
University Medical Center, Stanford, CA 94305,
USA.
3
Centre National de Ge notypage, 2 rue Gaston
Cre mieux, CP 5721, 91057 Evry Cedex, France.
4
Department of Genetics and Genome Sequencing
Center, Washington University School of Medicine,
St. Louis, MO 63108, USA.
5
Roche Center for Medical
Genomics F. HoffmannLa Roche, RCMG, Bau 93/
4.26, CH-4070 Basel, Switzerland.
E-mail: gary.peltz@roche.com
R E P O R T S
haplotype map was used to develop a
computational method for genetic analysis of
phenotypic traits in mice (4). Haplotypic
blocks that best correlate with phenotypic trait
data obtained from inbred strains were iden-
tified. As a first example, we tested the ability
of the haplotype-based computational map-
ping method to predict the chromosomal
location of the K locus of the major histo-
compatibility complex (MHC) located on
murine chromosome 17 (33 Mb). The
known properties of the MHC K locus of 16
inbred strains were used as input phenotypic
data for this analysis (5) (Table 1). The
categorical K locus data (b, k, d, u, v) were
transformed into points in multidimensional
metric space, such that each K locus category
was equidistant from any other category, and
then computationally analyzed for correlation
with the haplotypic blocks (Fig. 1A). One
haplotypic block best correlated with the
phenotypic data. The predicted haplotypic
block was on chromosome 17 E32.778 to
32.805 Mb, NCBI (National Center for
Biotechnology Information) Build 30^ and
contained five genes (H2-Dma, Psmb9,
Tap1, Psmb8, and Tap2). These genes are
known to determine the MHC class I and
class II phenotypes (6, 7). Consistent with this
prediction, the K, A$, A", and E$ phenotypes
were identical among the 16 mouse strains
tested (Table 1). The same computational
prediction profile was obtained when any of
the other class II phenotypes were used as
input phenotypic data (computational profile
identical to that in Fig. 1A).
We next determined if the computational
method could correctly identify haplotypic
blocks associated with two other MHC
phenotypes, the MHC class III S locus and
the class Ib Qa2 locus (5). In both cases, the
computationally identified haplotypic block
with the highest correlation contained the
MHC gene that corresponded with the
phenotype analyzed (Fig. 1, B and C, and
Table 2). The correct identification of Qa2
locus by the computational method was
particularly encouraging, because there were
only two different categorical phenotypes (a
and b) at this locus (Table 1). This analysis
of four MHC phenotypes demonstrates that
the computational method can analyze input
phenotypic data and accurately discriminate
among colinear, neighboring haplotypic
blocks in the mouse genome.
We next tested the ability of the compu-
tational mapping method to identify genetic
loci regulating the aromatic hydrocarbon
(AH) response among inbred mouse strains.
Genetic variation in this metabolic response
pathway can affect the response to drugs and
environmental toxins and is also likely to
affect cancer susceptibility (8, 9). The aro-
matic hydrocarbon receptor (AHR) is the
ligand-binding component of the intracellular
protein complex that regulates this response.
After ligand binding, this receptor translocates
to the nucleus where it forms a heterodimeric
protein complex that stimulates the expression
of genes involved in drug and xenobiotic
metabolism. The AH response was previously
characterized by measuring the amount of
hepatic aromatic hydrocarbon hydroxylase
activity induced after intraperitoneal injection
of 3-methylcholanthrene in 42 inbred mouse
strains (10). Thirteen of these strains were
characterized in our mouse SNP database (2).
The AKR/J, 129/SvlmJ, LP/J, NZB/BinJ,
NZW/LacJ, and DBA/2J strains were AH
Table 1. Phenotypic data used for computational haplotype-based genetic analysis. The MHC class Ia, Ib,
II, and III phenotypes; hepatic AH response; and pulmonary H2-E" gene expression data are shown. n/a,
not available.
Strain
H2 Haplotypes
H2-E" gene
expression
AH
response
Class Ia Class II Class III Class Ib
K A$ A" E$ E" S Qa-2
129/SvJ b b Null b a 5 j
A/J k k k d a 42
A/HeJ k k k d a 23
AKR/J k k k k b 44 j
B10.D2-H2/oSnJ d d d d b 1055
BALB/cJ d d d d b 1063
BALB/cByJ d d d d b n/a n/a
C3H/HeJ k k k k b 36
C57/6J b b Null b a 8
DBA/2J d d d d b 1343 j
LG/J d d d d b n/a n/a
LP/J b b Null b a n/a j
MRL/MpJ k k k k b 38 n/a
NZB/BinJ d d d d b n/a j
NZW/LacJ u u u z b n/a j
SM/J v v v v a n/a
Fig. 1. Computational mapping of MHC phenotypic data onto haplotypic blocks. The correlation
for each haplotypic block is represented as a bar, and the haplotypic blocks are arranged from
proximal (centromeric) to distal (telomeric) for all 20 chromosomes. The variable width of each
chromosome indicates the number of haplotypic blocks on that chromosome. The height of each
bar is the inverse of the score measuring the correlation between the phenotypic data and the
haplotypic block. The insets show predictions for haplotypic blocks on chromosome 17 for each
phenotype. The relative location of the MHC genes on chromosome 17 is shown in the block
diagram at the top of the figure. (A) Computational prediction of the location of the MHC class I-K
locus. The MHC class I-K phenotypic data were analyzed for correlation with strain groupings
within haplotypic blocks. The predicted haplotypic block was located in the MHC region on
chromosome 17 at 33.78 to 32.81 Mb and contained both the MHC K and class II genes. (B)
Computational analysis of MHC class III S locus phenotypic data. The predicted haplotypic block is
located within the MHC region and contains the S-locus gene. (C) Computational analysis of the
MHC class Ib Qa2 locus phenotypic data. The three predicted haplotypic blocks are all located on
chromosome 17 within the region 33.32 to 33.82 Mb. They contain the Qa2 locus.
R E P O R T S
nonresponsive, whereas the A/J, A/HeJ,
C57BL/6J, BALB/cJ, B10.D2-H2/oSnJ, SM/J,
and C3H/HeJ strains were AH responsive. The
AH response phenotypes were treated as a
qualitative binary phenotype (either or j)
for the computational analysis. The pattern
of genetic variation within only two haplo-
type blocks exactly matched the pattern of
AH responsiveness (calculated score 0 0)
among the inbred strains (Fig. 2, A and B).
Within these two blocks, all six nonrespon-
sive strains shared one haplotype, whereas
the seven responder strains had one of two
other haplotypes. Both of these haplotypic
blocks were within the Ahr locus on mouse
chromosome 12 (29.7 Mb). None of the other
haplotypic blocks exactly matched the AH
response pattern among the inbred strains.
Furthermore, because the AH phenotype was
measured in liver tissue, identification of the
computationally predicted genes that were
expressed in liver would also help to identify
the causative genetic locus. Analysis of gene
expression data indicated that only three of
the nine haplotypic blocks whose strain
groupings were best correlated with the phe-
notypic response were within genes (Ahr and
Dnpep) expressed in liver tissue (Fig. 2B). In
summary, haplotypic blocks within the Ahr
locus were the only blocks having a perfect
correlation with AH response phenotype,
and the Ahr gene was expressed in liver tis-
sue. Therefore, Ahr was the computationally
predicted gene candidate selected for further
analysis.
Analysis of genetic variation within the Ahr
locus identified the precise molecular basis for
the differential AH response among inbred
strains. The 211 SNPs within the Ahr locus
(2) divided the inbred mouse strains into
three haplotypic groups (Fig. 2C). The AH-
responsive strains were in haplotypic groups
I and II, whereas the nonresponsive strains
were in group III. Sixteen SNPs were located
in exons, and 7 (labeled a to g in Fig. 2, C and
D) altered the amino acid sequence of the
encoded protein. One polymorphism con-
verted a stop codon in the group I strains
(B10 and C57BL/6) to an Arg in all other
strains, adding 43 C-terminal amino acids to
the AHR protein expressed by the group II
and III strains. Three amino acid changes
differentiated the group II from the group III
strains. Although these polymorphisms in-
duced amino acid changes and were previous-
ly thought to be of importance to this
phenotype (11), their strain distribution indi-
cated that they were not responsible for
differences in AH response phenotype. The
precision of these computational predictions is
demonstrated by the fact that only one SNP
distinguished the AH responsive and nonre-
sponsive strains. This polymorphism (labeled
b in Fig. 2, C and D) converted Ala
444
in the
AH-responsive (group I and II) strains to
Val
444
in the AH-unresponsive (group III)
strains. This SNP was located within a PAS
Associated C-terminal (PAC) motif that con-
tributes to the folding of the ligand binding
and dimerization region of this protein (12, 13).
Consistent with the prediction of this compu-
Pola1 4 78.28~78.29 X 0.295
Cyp2b19 5 18.30 7 0.295
Ren1 21 133.67 1 0.289
Esr2
Vdr
Dusp2
Dnpep
Ahr
Ahr
Symbol
10 79.79~79.81 12 0.297
10 98.22 15 0.278
4 128.79~129.05 2 0.261
14 76.00~76.13 1 0.203
12 29.69 12 0
57 29.65~29.69 12 0
#
SNPs
Position Chr. Haplotype Score
SMJ
+
C57
+
C3H
+
B/C
+
B10
+
A/J
+
A/H
+
NZW
-
NZB
-
LPJ
-
DBA
-
AKR
-
129
-
Log
e
(Expression)
*
*
*
*
*
*
344
344
1134
A
B
C
a b c d g f e
+ + + + + + + - - - - - -
aa exon C57 B10 A/HeJ A/J B/cJ C3H SM/J 129 AKR DBA LP/J NZB NZW
g term875Arg 11 A A G G G G G G G G G G G
f Ala827Thr 10 C C T T T T C C C C C
e Leu658Met 10 G G T T T T T T T T T T T
d Ser602Asp 10 C C T T T T T T T T T T T
c Leu540Pro 10 A A G G G G G G G G G
b Ala444Val 9 G G G G G G G A A A A A A
a Leu417Phe 9 G G A A A A G G G G
AH Response
Group I Group III Group II
D
Fig. 2. Genetic regu-
lation of aromatic hy-
drocarbon responsive-
ness. (A) Aromatic
hydrocarbon respon-
siveness of 13 different
mouse strains was
used as input pheno-
typic data for compu-
tational mapping onto
haplotypic blocks. Two haplotypic blocks had a perfect correlation with the AHR response
phenotype, and both were within the Ahr locus on chromosome 12 (indicated by the arrow). (B)
The AH response phenotype for the 13 strains and a list of haplotypic blocks with the best
correlation with the pattern of AH response among the inbred strains are shown. For each
predicted block, the calculated score for partitioning of the phenotypic data, chromosomal
location, and the number of SNPs within a block and its gene symbol are shown. The haplotypes
for each strain are represented by colored blocks that are presented in the same order as shown
for the strain phenotypes. Strains sharing the same haplotype have the same colored block. A
calculated partition score of 0 indicates an exact match between the distribution of phenotypic
responses and haplotypic groupings among the inbred strains. An asterisk (*) indicates that the
gene was not expressed in liver tissue. (C) Three haplotypes within the Ahr locus on chromosome
12 (29.7 Mb). Of 211 SNPs within the Ahr locus, the seven exonic SNPs (labeled as a to g) that
change an amino acid are shown. (D) The relative location of amino acid changes in the AHR
protein for the three haplotypic groups. Four of the amino acid changes (c, d, e, and g) distinguish
group I from the other two haplotypic groups, including an amino acid change (g) that converts a
stop codon in the group I protein to an Arg in group II and III. Three other amino acid changes (a, b,
and f) differentiate the group II and group III proteins. However, only a single SNP (b), which
converts an Ala (group I and II) to Val (group III) and is located within the PAC domain,
distinguishes the responder and nonresponder strains.
R E P O R T S
tational genetic method, an Ala
444
Y Val
substitution in an expressed recombinant AHR
protein shifted the in vitro AH response from
a responder to a nonresponder state (11, 14).
Gene expression profiles across inbred
mouse strains provide a useful intermediate
phenotype that can be analyzed to under-
stand how phenotypic traits are genetically
regulated. In the same manner as phenotypic
trait information, strain-specific gene expres-
sion data can be computationally mapped
onto haplotypic blocks to identify genetic
loci that potentially regulate differential gene
expression. For example, the level of H2-E"
gene expression in mouse lung, which is an
MHC class II gene, differed by more than
20-fold across 10 inbred mouse strains
examined using oligonucleotide microarrays
(Fig. 3). Logarithm-transformed H2-E" gene
expression data were computationally ana-
lyzed using our haplotype-based computa-
tional mapping method. The haplotypic
block within the first intron of the H2-E"
gene had the strongest correlation with H2-
E" gene expression. A Bonferroni-adjusted
P-value of 0.001 for this prediction (table
S1) supported the possibility that a cis-acting
functional element within the H2-E" gene
could regulate its basal expression.
To identify the cis-acting element, we
analyzed 23 SNPs within this 1.0-kb hap-
lotypic block (Fig. 4A). The nucleotide
sequence within this haplotypic block was
scanned to identify known recognition ele-
ments for transcription factors in the
TRANSFAC database (15). Then, combina-
tions of SNPs that were located within or
near transcription factor recognition ele-
ments and that distinguished haplotypes
within this block were selected. By this
method, three haplotype-determining SNPs
within this block, all located within a 30
base pair (bp) sequence at nucleotide posi-
tions 974, 975, and 999 downstream
from the H2-E" transcription start site, were
selected for analysis. We examined the effect
of these polymorphisms on reporter gene
expression in a B lymphoma cell line. Two
constructs with insertion of the H2-E"
promoter, first exon segments, and first
intron segments were prepared (Fig. 4B).
Construct II has the 10- to 1029-bp region of
H2-E" inserted, with the intronic region
between positions 475 and 925 deleted.
Genomic segments for construct II were
amplified from three different inbred strains;
each strain had a different haplotype result-
ing from the three polymorphisms discussed
above. The intronic region containing these
three polymorphic sites is not contained in
construct I. Though similar to construct II,
construct I contains only the 10- to 957-bp
region of H2-E", with the intronic region
between positions 475 and 925 deleted.
The BALB/cJ haplotype in the first intron
sequence resulted in a threefold enhance-
ment of reporter gene expression, relative to
plasmids containing the same regions ampli-
fied from the other two haplotypes (Fig. 4C).
In fact, the AKR- and C57/BL6-derived
intronic sequences did not enhance expres-
sion of construct II plasmids relative to the
H2-E" promoter and exonic sequence pres-
ent in construct I. These results demonstrate
that an allele-specific regulatory element is
located between nucleotides 958 and
1029 in the first intron of the H2-E" gene.
To further characterize the regulatory
element within this intronic region, we per-
formed electrophoretic mobilityshift assay
(EMSA) experiments using three 45-bp oligo-
nucleotides. Each oligonucleotide had a differ-
ent SNP allele combination that corresponded
to one of the three H2-E" haplotypes at po-
sitions 974, 975, and 999 (Fig. 4A). The
EMSA experiments demonstrated that DNA-
protein complexes were formed after incuba-
tion of two different (CAIT and CAIG)
oligonucleotides with nuclear extracts pre-
pared from lung tissue. In contrast, the
(TGIG) oligonucleotide did not form detect-
able protein complexes (Fig. 4D). The adja-
cent SNPs (CA/TG) were located within a
Yin Yang 1 (YY1) transcription factor bind-
ing site, whereas the third SNP (T/G) was
within a serum response factor (SRF) binding
site. The CA-to-TG conversion eliminated the
YY1 binding site, whereas the G-to-T change
converts an SRF binding site to a potential
recognition element for another Activator
Protein 1 (AP1) transcription factor. YY1
binding to the CA-allele oligonucleotide was
confirmed by Bsupershift[ experiments per-
formed by preincubation of the nuclear lysate
proteins with an antibody to YY1. The
mobility of the upper band was altered in the
presence of the antibody, indicating that it was
a complex of YY1 bound to the oligonucleo-
tide (Fig. 4E). YY1 has been shown to be re-
quired for association of SRF with its serum
response element (SRE) (16). Of note, only
mouse strains with the CAIG haplotype ex-
hibited high H2-E" gene expression in the
lung. Taken together, the EMSA, reporter
gene, and gene expression results indicate
that a 45-bp functional element within the
H2-E" gene locus regulates its basal expres-
sion level in an allele-specific manner. Within
this functional element, alleles that enable the
A
L
o
g
e

(
G
e
n
e

E
x
p
.
)
0
2
4
6
8
A/J AKRC3HA/H MRL B/C B10 DBA C57129
B

Fig. 3. Cis-genetic regulation of
H2-E" gene expression in mouse
lung. (A) The level of pulmonary
H2-E" gene expression is shown as
the natural logarithm of the aver-
age of three independent measure-
ments for each of 10 mouse strains
examined. (B) Computational map-
ping of H2-E" gene expression data
onto haplotypic blocks. The haplo-
typic block with the highest predic-
tion contained the H2-E" gene.
Table 2. Haplotype-based computational predictions generated for five phenotypic data sets. For each
phenotype, the location of the haplotypic block, which contains the gene determining the phenotype,
along with its computational score are shown. The score for each haplotypic block was calculated as a
function of the correlation between the phenotypic data and strain grouping within the haplotypic block.
A smaller score indicates a better match, and 0 indicates a perfect match.
Phenotype Predicted locus Score Chr. MB
H2 K H2-K 0 17 32.7832.81
H2 A "/$, E "/$ H2-A "/$, H2-E "/$ 0 17 32.9432.94
H2 S H2-S 0.16 17 33.2633.32
H2 Qa-2 H2-Qa2 0 17 33.3233.42
H2-E" gene expression H2-E" 0.11 17 32.5733.26
AH response Ahr 0 12 29.6629.69
R E P O R T S
binding of YY1 and a second transcrip-
tion factor, which is probably SRF, are re-
quired for high H2-E" gene expression in
the lung.
These examples demonstrate that a vari-
ety of genetically regulated phenotypic traits,
ranging from qualitative trait information to
quantitative gene expression data that vary
among inbred strains, can be computational-
ly analyzed by this method. The haplotypic
blocks exhibiting a strong correlation with
phenotypic data provide a valuable starting
point that enables pathway analysis or
mechanistic studies to be performed. As
shown in the analysis of the H2-E" gene
expression and AH response data, a very
precise hypothesis about the genetic basis for
the phenotypic difference among mouse
strains can be generated using this computa-
tional method. In these examples, the SNPs
responsible for the phenotypic differences
were predicted.
A number of factors affect the perfor-
mance of this method. As shown here, it
performs extremely well when the pheno-
typic data reflect the genetic variation within
a haplotypic block in our SNP database.
However, if haplotypic information for a
key genetic locus or phenotypic informa-
tion for a sufficient number of strains is
not available, its performance is substan-
tially reduced. When phenotypic data for
only a few inbred strains are available, this
method usually produces a large number of
top predictions with comparable scores, and
most are false-positives. We have empiri-
cally determined that phenotypic data from
eight or more strains are needed to produce
statistically significant predictions (table S1).
The statistical power and SNP coverage
issues will diminish as the numbers of in-
bred strains and SNPs in the database in-
crease. A haplotypic map with more than
300,000 SNPs covering between 40 and 50
commonly used inbred mouse strains should
be available within 2 years. At present, it is
unlikely that this computational method can
analyze traits controlled by multiple genetic
loci, each with a small effect size. How-
ever, coverage of additional inbred strains
should enable this computational method to
5 3
ACATGGACAC[C/T][A/G]TATTTAAGATCACGAGAAAGAAA[G/T]GACATCATCC
YY1 SRF
Mouse Strains 129, C57 B10, Balb/c, A/J, A/H, AKR,
Balb/cBy, DBA C3H, MRL
SNP Haplotype CA..T CA..G TG..G
H2E Exp. Low High Low
YY1 / SRF + / - + / + - / +
1kb
1 2 3 4
2 1
I
II
H2E
Strain AKR BALB AKR BALB 129
Construct Vector I I II II II
0
3
6
9
L
u
c
i
f
e
r
a
s
e

a
c
t
i
v
i
t
y

(
1
0
6
)
Extract - A B C - A B C - A B C
SNP Haplotype CA..T CA..G TG..G
Extract: AKR Balb/c C57
Ab-YY1: - + - + - +
1 2 3 1 2 1 2
a b c
A
D E
B
C
Fig. 4. (A) The relative location of the
haplotypic block containing a cis-acting
functional genomic element in the H2-
E" gene. The filled boxes are exons and
the exon number is indicated. The bold
line below indicates the position of the
computationally identified haplotypic
block. The sequence of the oligonucleo-
tides used in the gel-shift experiments
are shown, and alleles at the three
polymorphic sites among the inbred
mouse strains are indicated in brackets.
The YY1 and a possible SRF recognition
element are underlined. The conversion
of the CA to TG alleles eliminates the
YY1 binding site. The CA..G allelecontaining haplotype has the YY1 and SRF binding sites and also had high H2-E" gene expression. (B) H2-E" gene
structure and reporter gene constructs. Construct II contains the 10- to 1029-bp region of H2-E", with a deletion of the 475- to 925-bp region. The
shorter construct (I) contains the 10- to 925-bp region of H2-E", with deletion of the region between 475 and 925 bp. The yellow arrow indicates the
location of the polymorphic regulatory element containing YY1 and SRF binding sites. (C) The haplotype-specific effect of the H2-E" cis-acting
element on reporter gene expression. The level of luciferase reporter gene expression in the A20 B lymphoma cell line after transfection of the
indicated plasmids is shown. Luciferase assays were performed using two different construct I plasmids, which had genomic DNA obtained from either
the AKR or Balb/cJ strains. Three construct II plasmids were prepared, and each had genomic DNA obtained from AKR, Balb/c, or 129 mice. These three
strains represent each of the three different haplotypes identified within the first intron of the H2-E" gene. The luciferase expression level was
normalized for transfection efficiency by cotransfection with a $-galactosidase plasmid. (D) Allele-specific formation of protein-DNA complexes
requires a YY1 recognition element. Three different
32
P-labeled oligonucleotides, each with different alleles at three polymorphic sites, were incubated
with nuclear extracts from lung tissue obtained from AKR (lane A), Balb/c (lane B), or C57 (lane C) mice. The arrowheads indicate the protein-DNA
complexes. (E) YY1 is part of the allele-specific protein-DNA complex. EMSAs were performed using a
32
P-labeled 45-bp CA..G oligonucleotide,
containing a YY1 recognition element, and nuclear extracts of lung tissue were prepared as in (D). The extracts were preincubated with an antibody to
YY1. The filled arrow indicates the position of the YY1 DNA complex, and the open arrow indicates the position of this complex after incubation with
the antibody to YY1.
R E P O R T S
have substantial power to identify genet-
ic loci for a wider range of phenotypic
traits, including those with increased under-
lying genetic complexity, than is currently
possible.
1. A. Grupe et al., Science 292, 1915 (2001).
2. See http://mouseSNP.Roche.com.
3. J. Wang et al., in Computational Genetics and
Genomics: New Tools for Understanding Disease,
G. Peltz, Ed. (Humana, Totowa, NJ, 2004).
4. Materials and methods are provided as supporting
online material in Science Online. The haplotype maps
used in this study are available at http://mouseSNP.
Roche.com.
5. Jackson Laboratory, Jackson Lab. Notes 475 (1998).
6. Y. Yang et al., J. Biol. Chem. 267, 11669 (1992).
7. S. G. Cho, M. Attaya, J. J. Monaco, Nature 353, 573
(1991).
8. D. W. Nebert, A. Puga, V. Vasiliou, Ann. N. Y. Acad.
Sci. 685, 624 (1993).
9. A. B. Okey, D. S. Riddick, P. A. Harper, Toxicol. Lett.
70, 1 (1994).
10. D. W. Nebert, N. M. Jensen, H. Shinozuka, H. W.
Kunz, T. J. Gill III, Genetics 100, 79 (1982).
11. M. Ema et al., J. Biol. Chem. 269, 27337 (1994).
12. C. P. Ponting, L. Aravind, Curr. Biol. 7, R674
(1997).
13. K. M. Burbach, A. Poland, C. A. Bradfield, Proc. Natl.
Acad. Sci. U.S.A. 89, 8185 (1992).
14. A. Maier et al., Environ. Health Perspect. 106, 421
(1998).
15. V. Matys et al., Nucleic Acids Res. 31, 374 (2003).
16. S. Natesan, M. Gilman, Mol. Cell. Biol. 15, 5975 (1995).
17. J.C., A.N., and J.W. were supported by an NIH Ge-
nome Research Institute grant (1 R01 HG02322-01)
awarded to G.P. We also thank M. Ott (Roche Center
for Medical Genomics) for his help with DNA
sequencing.
DC1
Table S1
References
The PP2A-Associated Protein !4 Is
an Essential Inhibitor of Apoptosis
Mei Kong,
1,3
Casey J. Fox,
1,3
James Mu,
2
Laura Solt,
1
Anne Xu,
1,3
Ryan M. Cinalli,
1,3
Morris J. Birnbaum,
2
Tullia Lindsten,
1,5
Craig B. Thompson
1,3,4
*
Despite evidence that protein kinases are regulators of apoptosis, a specific role
for phosphatases in regulating cell survival has not been established. Here we
show that !4, a noncatalytic subunit of protein phosphatase 2A (PP2A), is
required to repress apoptosis in murine cells. !4 is a nonredundant regulator
of the dephosphorylation of the transcription factors c-Jun and p53. As a result of
!4 deletion, multiple proapoptotic genes were transcribed. Either inhibition
of new protein synthesis or Bcl-x
L
overexpression suppressed apoptosis ini-
tiated by !4 deletion. Thus, mammalian cell viability depends on repression of
transcription-initiated apoptosis mediated by a component of PP2A.
The "4 protein was initially identified as a
component of receptor signal transduction
complexes in mammalian B and T lympho-
cytes (1, 2) and was later determined to be
broadly expressed (1, 3). It interacts with the
catalytic subunit of protein phosphatase
PP2A (PP2Ac) as well as those of PP4 and
PP6 (4, 5). Binding of "4 to PP2Ac displaces
PP2Ac from a dimeric regulatory complex
composed of the core A subunit and any of
more than 12 variable B components (6).
Interaction of "4 with PP2Ac both enhances
PP2Ac catalytic activity and alters its sub-
strate specificity (4, 7). Its yeast homolog,
Tap42, is a PP2A regulatory subunit that
functions in TOR-dependent nutrient sensing
(8). In mammalian cells, the association of
"4 with PP2Ac is regulated by growth fac-
tor signals and modulators of the TOR path-
way such as rapamycin (4, 7) (Fig. 1A).
However, rapamycin potentiates apoptosis
in growth factordeprived cells (9), so it is
difficult to determine whether the decline in
association of "4 with PP2Ac contributes to
the cellular response to such treatments or
occurs as a consequence of the decrease in
cell viability.
To determine whether "4 contributes to
the regulation of cell survival, we generated
mice deficient in "4. Two constructs were
created that each deleted exon I and adja-
cent sequences of the !4 gene. These were
introduced into an embryonic stem (ES)
cell line, but a homologous recombinant
was not recovered in either case. Because
the !4 gene is located on the X chromo-
some, this result raised the possibility that
!4 was an essential gene in the male ES
cells. Hence, we prepared a construct con-
taining the !4 gene in which exons III to V
were flanked by loxP (Fig. 1B). After elec-
troporation, 4 of 192 clones showed ho-
mologous recombination. Introduction of
recombinant Cre into these clones failed to
yield ES cell clones carrying a deleted !4
gene (10). Next, we generated mice carry-
ing a germline-transmitted !4-floxed allele
(!4
fl
) integrated by homologous recombi-
nation. These mice were bred to Lck-Cre
transgenic mice to determine the effect of
!4 deletion on developing T cells, a
nonessential lineage. In !4
fl
/Lck-Cre male
mice, the thymi were depleted of develop-
ing T cells (Fig. 1C) and the residual cells
were enriched in immature thymocytes
(10). Although the Cre-deleted form of the
!4 allele was present in the residual
thymocytes, these cells died in the thymus,
as no T cells with a deleted !4 allele
appeared in the periphery (Fig. 1D). Thus,
"4 is required for either T cell development
or survival.
In female heterozygotes carrying one
wild-type and one !4
fl
allele, thymocyte
numbers were reduced by about 60% rela-
tive to wild-type mice, but peripheral T cell
numbers were normal (Fig. 1C) (10). Most
peripheral T cells in female heterozygotes
carried a Cre-deleted form of !4
fl
(Fig. 1D).
Because T cell precursors have undergone
random X-chromosome inactivation, this
finding suggests that the decreased number
of thymocytes resulted from the death of
cells in which !4 was deleted on the active X
chromosome.
To further analyze the consequences of
!4 deletion, we generated immortalized
mouse embryonic fibroblasts (MEFs) from
male !4
fl
embryos and compared them to
littermate !4
wt
MEFs (fig. S1) (11). Retro-
viruses encoding either green fluorescent
protein (GFP) expressed from an internal
ribosome entry site (IRES) alone (vector) or
both a Cre recombinase and an IRES-GFP
(Cre) were introduced into !4
fl
or !4
wt
MEFs, and GFP-positive cells were isolated.
Immunoblot analysis of lysates from GFP-
positive cells 48 hours and 72 hours after
Cre infection showed a decrease and ab-
sence of "4 protein, respectively, in !4
fl
MEFs but not in !4
wt
MEFs (Fig. 2A). Cell
death was observed beginning 48 hours
after Cre infection in the !4
fl
MEFs, and
nearly all cells were dead by 120 hours
after infection (Fig. 2B). In contrast, the
viability of !4
wt
cells infected with Cre or
!4
fl
infected with vector was not affected
(Fig. 2B). Reconstitution of "4 into !4
fl
1
Abramson Family Cancer Research Institute,
2
Howard Hughes Medical Institute, University of
Pennsylvania, Philadelphia, PA 19104, USA.
3
Depart-
ment of Cancer Biology,
4
Department of Medicine,
5
Department of Pathology and Laboratory Medicine,
University of Pennsylvania School of Medicine,
Philadelphia, PA 19104, USA.
E-mail: craig@mail.med.upenn.edu
R E P O R T S
cells by stable transfection rescued cell
death in response to Cre infection (10). The
dying !4-deleted cells displayed the typi-
cal features of apoptosis, including cleavage
of caspase-3 and cleavage of the caspase sub-
strate PARP Epoly(ADP-ribose) polymerase^
(Fig. 2C).
Apoptosis can be initiated either through
biochemical modulation of existing apoptotic
regulatory proteins or through transcrip-
tion and translationdependent changes in
apoptotic regulatory proteins (1214). To
distinguish between these two possibilities,
we investigated the effect of protein synthe-
sis inhibition on !4 deletioninduced cell
death. Addition of the protein synthesis
inhibitor cycloheximide (CHX) 48 hours
after Cre infection rescued !4
fl
cells from
apoptosis, despite a decrease of "4 protein
expression in the presence or absence of
CHX (Fig. 3A) (fig. S2).
The transcription factor c-Jun is a PP2A
substrate and has been implicated in
transcription-dependent apoptotic death in
response to diverse cellular stresses, in-
cluding ultraviolet irradiation, heat and
osmotic shock, and growth factor with-
drawal (15). Its activation involves phospho-
rylation followed by nuclear translocation. At
72 hours after Cre-mediated !4 deletion, c-Jun
phosphorylation on Ser
63
increased (Fig. 3B)
and accumulated in the nucleus (Fig. 3C).
However, no changes in the expression or
activation status of the Jun kinases were
detected, as measured by their phosphoryla-
tion status (Fig. 3B).
To examine the transcriptional changes
that occur after !4 deletion, we performed
RNA microarray analysis with RNA from
!4
fl
MEFs 48 hours after infection with Cre
or vector. Isolated RNA was hybridized to
Affymetrix mouse expression microarrays
containing more than 39,000 transcripts
and variants. Pairwise analysis of the hy-
bridization profiles revealed that among
the 20 genes whose expression was most
highly increased in the Cre-infected sam-
ple (relative to the vector-infected con-
trol), six were established p53-dependent
targets: p21, Noxa, MDM2, cyclin G, Stk11,
and SIP (table S1). Several genes implicated
in the intrinsic mitochondrial apoptotic path-
way were also induced, including those en-
coding mDAP-3, Siva, and endonuclease G
(10). The up-regulation of p53-dependent
transcripts was associated with the accumu-
lation of p53 that was phosphorylated on
Ser
18
as "4 expression declined (Fig. 3D). In
addition to p53 Ser
18
phosphorylation, "4-
depleted cells accumulated p21 and Noxa
proteins.
Because induction of p53 activity is a
potent inducer of apoptosis, we made !4
fl
cells deficient in p53 by stimulating the
ubiquitin-dependent proteolysis of p53 (16).
0
0 2
0 4
0 6
0 8
0 0 1
0 2 1
0 4 1
T
h
y
m
o
c
y
t
e
N
u
m
b
e
r

(
m
i
l
l
i
o
n
)
e r C - t w
e r C - l f / t w
e r C - l f
B
D C
A
R K B
I I I I I I V V I
kb 1
X R B B B
t w
o k
R
R
X R B
o e n
l f
s u m y h T s l l e c T e n i l m r e G
t w
w
t
-
C
r
e
f
l
-
C
r
e
w
t
/
f
l
-
C
r
e
w
t
-
C
r
e
f
l
-
C
r
e
w
t
/
f
l
-
C
r
e
w
t
-
C
r
e
f
l
-
C
r
e
w
t
/
f
l
-
C
r
e
l f
o k
c A 2 P P : P I
4
c A 2 P P
+ - : a p a R
e t a s y L
4
Fig. 1. Conditional deletion of the PP2A-associated protein "4 in thymocytes. (A) Association of
"4 with PP2Ac is inhibited by rapamycin. The lymphoid progenitor cell line FL5.12 was cultured
overnight in growth factordeficient medium in the presence () or absence (j) of 20 nM
rapamycin (Rapa) as indicated. Cell lysates were immunoprecipitated (IP) with antibody to PP2Ac
followed by immunoblotting with antibodies to "4 or PP2Ac. As a control, an equivalent amount
of lysate was immunoblotted with antibody to "4 to detect total "4. (B) Generation of an !4 allele
containing loxP sites. Upper panel: Genomic organization of the !4 gene locus. R, Eco RI; B, Bam HI;
K, Kpn I; X, Xba I. The bar indicates the probe used for the Southern blots. Lower panel: Targeting
construct and different Eco RI fragments generated from different genotypes (wt, !4 wild-type
genotype; fl, floxed !4 allele; ko, Cre-deleted !4 allele). (C) Thymocyte cell number was
determined in !4
wt
/Lck-Cre male (wt-Cre), !4
wt/fl
/Lck-Cre female (wt/fl-Cre), and !4
fl
/Lck-Cre
male (fl-Cre) mice. Values are means T SD of three mice. (D) Southern blots of DNA prepared from
tail (germline), thymus, and purified splenic T cells of different genotypes as described in (C).
A
B
C
Cre: - + - +
fl wt
cleaved-
PARP
fl-Vec
fl-Cre
GFP/DAPI
cleaved-
caspase-3
fl-Vec
fl-Cre
wt-Vec
wt-Cre
s
l
l
e
C

d
a
e
D

%
Hours after retroviral infection
0
20
40
60
80
100
24 48 72 96 120
4
Cre (h): 0 48 72 0 48 72
fl wt
actin
Fig. 2. !4 deletion induces cell death in MEFs. (A) Depletion of "4 protein in !4
fl
male MEFs after
Cre introduction. !4
wt
(wt) or !4
fl
(fl) MEFs were infected with MIGR1-GFP-Cre (Cre), and the
resulting GFP-positive cells were isolated and analyzed at the time points indicated. Immunoblot-
ting was performed with antibodies to "4 or actin. (B) The percentage of dead cells was determined
by the ratio of 4,6-diamidino-2-phenylindole (DAPI)positive cells to GFP-positive cells isolated
after infection of !4
fl
MEFs with MIGR1-GFP-Cre (fl-Cre) or MIGR1-GFP (fl-Vec) or after infection
of !4
wt
MEFs with MIGR1-GFP-Cre (wt-Cre) or MIGR1-GFP (wt-Vec). (C) After 48 hours of Cre
infection, cells were fixed and the presence of cleaved caspase-3 was determined with a specific
antibody (red). GFP expression of the cells in the same field was visualized with a fluorescein
isothiocyanate (FITC) filter (green) and overlapped with DAPI staining (blue). Lower panel: Cell
lysates were analyzed by immunoblotting with a PARP-specific antibody.
R E P O R T S
Expression of a papilloma virus E6 protein
repressed p53 expression (fig. S3) and
partially inhibited apoptosis in response to
"4 loss (Fig. 3E). Like the proapoptotic p53
gene targets induced in !4-deleted cells, the
other proapoptotic genes transcriptionally
induced in !4-deleted cells also regulate the
intrinsic apoptotic pathway. Overexpression
of Bcl-x
L
, an inhibitor of the intrinsic
apoptotic pathway, protected !4
fl
-deleted
cells from cell death; this result indicates
that the transcriptional initiation of apoptosis
repressed by "4 is mediated through the
intrinsic apoptotic pathway (Fig. 3F).
To determine whether the requirement for
"4 is restricted to developing and proliferat-
ing cells, we assessed the effect of !4
deletion on differentiated adipocytes. Ex-
pression of PPAR, (peroxisome proliferator-
activated receptor ,), a nuclear hormone
receptor that is critical for adipogenesis (17),
caused the !4
fl
MEFs to differentiate into
adipocytes, as confirmed by the intracellular
accumulation of lipid droplets (Fig. 4A) and
by lipid staining with Oil Red O (10). Cells
were then infected at high multiplicity with
either an adenovirus encoding Cre recombi-
nase or a control adenovirus. Seven days after
infection, fewer adipocytes were observed
among the Cre-infected cells, most of which
were dead or dying (Fig. 4A) (fig. S4). This
Cre-induced death resulted in caspase-3 acti-
vation (fig. S4). As in proliferating cells,
increased phosphorylation of p53 and c-Jun
was detected in response to deletion of !4
(Fig. 4A).
Next, we assessed the role of "4 in
maintaining the viability of differentiated
cells in vivo. Three adult !4
fl
mice and three
!4
wt
mice were injected with an adenovirus
encoding Cre through the tail vein, a
technique that selectively infects the liver
parenchyma (18). At day 5 after injection, all
!4
fl
mice showed signs of illness, with
ruffled fur, hunched posture, and rapid
breathing. Over the next 24 hours, their
condition deteriorated while the !4
wt
mice
remained healthy. All six mice were killed
and their livers removed for histological and
biochemical analysis (Fig. 4B). Immunoblot
analysis of liver lysates revealed that "4 was
absent in the !4
fl
mice and present in the
!4
wt
mice. The lysates from the "4-depleted
liver revealed increased phosphorylation of
p53 and c-Jun and induction of p21 expres-
sion. Liver sections from Cre-infected !4
fl
mice revealed multiple apoptotic cells, as
determined by terminal deoxynucleotidyl
transferasemediated deoxyuridine triphos-
phate nick end labeling (TUNEL) staining.
In contrast, TUNEL analysis revealed an
absence of apoptotic cells in Cre-infected
!4
wt
mice.
It is surprising that the deletion of a
single PP2A regulatory subunit could have
such a profound phenotype. Previous studies
of !4 deletion in lymphocytes had concluded
that !4 is not an essential gene because ma-
ture lymphocytes were observed (19, 20).
However, our data show that the periph-
eral T cells that arose in !4
fl
/Lck-Cre mice
failed to delete the !4
fl
allele, a possibility
not addressed in the prior work. The "4
homolog in yeast, Tap42, plays an essential
role in suppressing stress response genes in
yeast by either repressing PP2Ac/Sit4 ac-
tivity or altering their substrate specificity
(21). Similarly, "4 appears to suppress the
stress response factor c-Jun by maintaining
it in a more dephosphorylated state. How-
ever, because yeast lack both p53 and an
apoptotic response, "4 has also been evo-
lutionarily adapted to repress p53-dependent
transcription and apoptosis. It is likely that,
in addition to its effect on c-Jun and p53,
!4 deletion alters the phosphorylation status
of more specialized substrates such as Mid1,
a protein involved in midline pattern forma-
tion that is also a substrate of the "4/PP2Ac
complex (22). Moreover, PP2Ac may play
an additional role in apoptosis through in-
teraction with other regulatory subunits (23).
Nonetheless, the observation that !4 deletion
leads rapidly to apoptosis in all cell types
tested demonstrates that specific phosphatase
complexes play nonredundant and essential
roles in the regulation of transcription-
induced apoptosis. The data support the
hypothesis that in animal cells, apoptosis is a
default cell fate (24). In the absence of spe-
cific and regulated inhibition, cells initiate
new transcription and translation to actively
initiate their apoptotic demise.
Fig. 3. p53 and c-Jun
are phosphorylated af-
ter !4 deletion. (A) Cy-
cloheximide treatment
rescues !4-deleted cells
from death. !4
fl
MEFs
were infected with
MIGR1-GFP (Vec) or
MIGR1-GFP-Cre (Cre).
Duplicate cultures were
prepared, and after 48
hours of retroviral in-
fection, CHX was added
to one of the paired
samples. The percent-
age of dead cells was
determined by the ratio
of DAPI-positive cells to
GFP-positive cells. (B)
!4
fl
(fl) or !4
wt
(wt)
MEFs were infected
with MIGR1-GFP (j)
or MIGR1-GFP-Cre ().
GFP-positive cells were
sorted by flow cytom-
etry after 24 hours of
infection and collected
at 72 hours after retro-
viral infection. Immuno-
blotting was performed
with different antibodies
as indicated. (C) !4
fl
(fl) MEFs were infected
with MIGR1-GFP (Vec)
or MIGR1-GFP-Cre
(Cre). After 72 hours,
cells were fixed and c-
Jun phosphorylation
was visualized with an
antibody to phospho-
rylated c-Jun Ser
63
(red). GFP expression
of the cells in the
same field was visualized with a FITC filter (green) and overlapped with DAPI staining
(blue). (D) !4
fl
MEFs were infected with MIGR1-GFP-Cre (Cre), and GFP-positive cells were
isolated and analyzed at 48 and 72 hours after Cre introduction. Immunoblotting was
performed with antibodies as indicated (11). (E) !4
fl
MEFs were infected with adeno-LacZ
(LacZ) or adeno-E6 (E6) followed by either MIGR1-GFP (Vec) or MIGR1-GFP-Cre (Cre) and
cell death was quantitated over time. (F) !4
fl
MEFs were stably transfected with pBabe-Bcl-x
L
(Bcl-x
L
) or pBabe (pBabe) retrovirus vector. The selected clones were infected with either
MIGR1-GFP (Vec) or MIGR1-GFP-Cre (Cre) and cell death was quantitated over time
(means T SD).
D
p-p53
actin
p53
Cre (h): 0 48 72
Noxa
p21
f
l
-
V
e
c
f
l
-
C
r
e
GFP/DAPI p-c-Jun
C
A
0
20
40
60
80
100
48 72 96 120
%

D
e
a
d

C
e
l
l
s
%

D
e
a
d

C
e
l
l
s
%

D
e
a
d

C
e
l
l
s
Vec
Cre
Vec-CHX
Cre-CHX
B
JNK
fl wt
p-c-Jun
Cre: - + - +
c-Jun
p54
p46
p-JNK
p54
p46
LacZ-Vec
LacZ-Cre
E6-Vec
E6-Cre
Hours after retrov
iral infection
0
10
20
30
40
50
24 48 72
F
Bcl-X
L
-Cre
pBabe-Vec
pBabe-Cre
Bcl-X
L
-Vec
24 48 72
0
10
20
30
40
50
60
70
E
R E P O R T S
1. S. Inui et al., J. Immunol. 154, 2714 (1995).
2. E. Chuang et al., Immunity 13, 313 (2000).
3. A. D. Everett, D. L. Brautigan, Dev. Dyn. 224, 461 (2002).
4. K. Murata, J. Wu, D. L. Brautigan, Proc. Natl. Acad.
Sci. U.S.A. 94, 10624 (1997).
5. J. Chen, R. T. Peterson, S. L. Schreiber, Biochem.
Biophys. Res. Commun. 247, 827 (1998).
6. S. Zolnierowicz, Biochem. Pharmacol. 60, 1225 (2000).
7. S. Inui et al., Blood 92, 539 (1998).
8. C. J. Di Como, K. T. Arndt, Genes Dev. 10, 1904 (1996).
9. A. L. Edinger, C. B. Thompson, Mol. Biol. Cell 13,
2276 (2002).
10. T. Lindsten, M. Kong, J. Mu, unpublished data.
12. K. Polyak, Y. Xia, J. L. Zweier, K. W. Kinzler, B.
Vogelstein, Nature 389, 300 (1997).
13. K. Nakano, K. H. Vousden, Mol. Cell 7, 683 (2001).
14. L. R. Devireddy, J. G. Teodoro, F. A. Richard, M. R.
Green, Science 293, 829 (2001).
15. R. J. Davis, Cell 103, 239 (2000).
16. A. Storey et al., Nature 393, 229 (1998).
17. D. Shao, M. A. Lazar, J. Biol. Chem. 272, 21473
(1997).
18. A. Rohlmann, M. Gotthardt, T. E. Willnow, R. E. Hammer,
J. Herz, Nature Biotechnol. 14, 1562 (1996).
19. D. R. Hua et al., Eur. J. Immunol. 33, 1899 (2003).
20. S. Inui et al., Int. Immunol. 14, 177 (2002).
21. K. Duvel, A. Santhanam, S. Garrett, L. Schneper, J. R.
Broach, Mol. Cell 11, 1467 (2003).
22. J. Liu, T. D. Prickett, E. Elliott, G. Meroni, D. L. Brautigan,
23. A. M. Silverstein, C. A. Barrow, A. J. Davis, M. C. Mumby,
24. M. C. Raff, Nature 356, 397 (1992).
25. We thank M. A. Lazar and W. El-Deiry for providing
the PPAR, and E6 constructs, and B. Keith, S. Reiner,
and members of the Thompson laboratory for helpful
discussions. Supported by a postdoctoral fellowship
from the Lymphoma Research Foundation (M.K.) and
by grants from the National Cancer Institute.
DC1
Figs. S1 to S4
Table S1
References
A Network of Control Mediated
by Regulator of Calcium/
Calmodulin-Dependent Signaling
S. V. Rakhilin,
1
P. A. Olson,
2
A. Nishi,
1,3
N. N. Starkova,
1
A. A. Fienberg,
1,4
A. C. Nairn,
1,5
* D. J. Surmeier,
2
* P. Greengard
1
*
Calmodulin (CaM) is a major effector for the intracellular actions of Ca
2
in
nearly all cell types. We identified a CaM-binding protein, designated
regulator of calmodulin signaling (RCS). G proteincoupled receptor (GPCR)
dependent activation of protein kinase A (PKA) led to phosphorylation of RCS
at Ser
55
and increased its binding to CaM. Phospho-RCS acted as a com-
petitive inhibitor of CaM-dependent enzymes, including protein phosphatase
2B (PP2B, also called calcineurin). Increasing RCS phosphorylation blocked GPCR-
and PP2B-mediated suppression of L-type Ca
2
currents in striatal neurons.
Conversely, genetic deletion of RCS significantly increased this modulation.
Through a molecular mechanism that amplifies GPCR- and PKA-mediated sig-
naling and attenuates GPCR- and PP2B-mediated signaling, RCS synergistically
increases the phosphorylation of key proteins whose phosphorylation is reg-
ulated by PKA and PP2B.
The regulator of calmodulin signaling (RCS),
previously referred to as ARPP-21, is a small
(9600 dalton), acidic (pI 0 4.6) neuronal
phosphoprotein that is highly expressed in
regions of the mammalian brain innervated
by dopamine-releasing neurons (1, 2). RCS
contains no obvious conserved domains and
displays no similarity to any known protein
except for the N terminus of TARPP (thymus-
specific cyclic adenosine monophosphate
regulated phospho-protein), a splice variant
that is not expressed in brain (3). RCS is
phosphorylated by protein kinase A (PKA)
at Ser
55
in neostriatal slices (4, 5). However,
the function of RCS or of its phosphorylation
has remained unknown.
We used the LexA yeast two-hybrid sys-
tem to study protein-protein interactions and
identified CaM as a binding partner for RCS
(fig. S1A). The RCS-CaM protein interac-
tion was confirmed by co-immunoprecipita-
tion of endogenous CaM from bovine striatal
extract with antibody to RCS (Fig. 1A). In
the presence but not the absence of Ca
2
,
CaM from bovine brain extract bound im-
mobilized recombinant RCS (fig. S1B), and
RCS also bound to CaM-Sepharose (fig. S1C).
Fig. 4. !4 deletion induces
cell death in nonproliferat-
ing tissues. (A) Adipocytes
generated from !4
fl
MEFs
(fl) as described in (11)
were infected with adeno-
LacZ (Vec) or adeno-Cre
(Cre). Seven days after in-
fection, cells were stained
with propidium iodide (PI)
and photographed with a
tetramethyl rhodamine iso-
thiocyanate (TRITC) filter or
bright field (left panel) or
analyzed for alterations in
protein expression or phos-
phorylation by immuno-
blotting with the indicated
antibodies (right panel). (B)
!4
wt
(wt) or !4
fl
(fl) mice
were injected with adeno-
Cre (Cre). After 6 days, the
mice were killed, livers
were removed, and liver sec-
tions were analyzed by he-
matoxylin and eosin (H/E) or
TUNEL staining. Solid arrow-
heads indicate apoptotic
cells; open arrowheads in-
dicate macrophages with
ingested apoptotic cells (left
panel). Immunoblotting was
performed with liver lysates
from !4
wt
(wt) or !4
fl
(fl)
mice after infection with adeno-Cre (Cre) using antibodies as indicated (right panel).
p-c-Jun
A
fl-Vec
fl-Cre
bright field PI
4
actin
p-p53
p21
c-Jun
p53
fl
Cre: - +
fl-Cre
TUNEL H/E
B
wt fl
p-p53
p-c-Jun
actin
p53
p21
4
c-Jun
Cre
wt-Cre
1
Laboratory of Molecular and Cellular Neuroscience,
Rockefeller University, New York, NY 10021, USA.
2
Department of Physiology, Feinberg School of Med-
icine, Northwestern University, Chicago, IL 60611,
USA.
3
Department of Pharmacology, Kurume Univer-
sity School of Medicine, Kurume, Fukuoka 830-0011,
Japan.
4
Intra-Cellular Therapies Incorporated, Audubon
Biomedical Science and Technology Park, New York, NY
10032, USA.
5
Department of Psychiatry, Yale University
School of Medicine, New Haven, CT 06508, USA.
E-mail: angus.nairn@yale.edu; j-surmeier@northwestern.
edu; greengard@rockefeller.edu
R E P O R T S
Thus, the interaction between RCS and CaM
has high affinity and is Ca
2
-dependent.
To assess the role of phosphorylation in
modulating the interaction of RCS and CaM,
we phosphorylated glutathione S-transferase
(GST)RCS stoichiometrically at Ser
55
with
PKA. Phospho-GST-RCS bound to CaM with
higher affinity than did dephospho-GST-RCS
(Fig. 1B) (apparent K
d
0 826 T 60 nM). In
support of this result, mutation of Ser
55
of
RCS to alanine greatly decreased yeast growth,
whereas substitution of Ser
55
with aspartic
acid sustained yeast growth, dependent on
the RCS-CaM interaction (fig. S1D). Nota-
bly, phosphorylated but not dephosphorylated
RCS effectively inhibited the activities of
CaM-dependent kinase I (CaMKI) and CaM-
dependent protein phosphatase 2B (PP2B),
with median inhibitory concentration (IC
50
)
values of 1 and 1.2 6M, respectively (Fig. 1,
C and D). Considering the dissimilar struc-
tural properties of CaMKI and PP2B, the
inhibitory activity of phospho-RCS can be
attributed to sequestration of CaM, thereby
preventing CaM binding to its targets.
Although CaM is expressed in very high
concentrations (possibly 9100 6M) in most
cell types, growing evidence suggests that
the level of CaM is substantially lower than
that of all of its targets (6). Thus, the amount
of free CaM is limited, and phosphorylation
of RCS would be expected to influence CaM-
dependent signaling by regulating its avail-
ability. A recent study in nonneuronal cells has
indicated that under resting conditions (low
intracellular Ca
2
) the available free CaM
concentration is only 9 6M (7). Moreover,
after elevation of intracellular Ca
2
by using
methods similar to those in this study, the
available concentration of CaM is G200 nM.
The concentration of RCS in medium spiny
neurons is very high Eestimated to be in the
range of 10 to 20 6M (1, 8)^. Assuming free
CaM in neurons is also G10 6M, regulation
of RCS phosphorylation would therefore be
expected to influence the CaM available for
activation of PP2B and other CaM targets.
In medium spiny neurons, PP2B is a po-
tent regulator of L-type Ca
2
channels (9, 10).
Mobilization of intracellular Ca
2
stores by
activation of either M
1
muscarinic or D
2
dopaminergic receptors leads to activation
of PP2B and suppression of L-type Ca
2
channel currents. PP2B-dependent modula-
tion of the L-type Ca
2
current was exam-
ined in medium spiny neurons from wild-type
mice or RCS knock-out mice (fig. S2). Appli-
cation of the M
1
muscarinic receptor agonist
muscarine (2 6M) or the D
2
dopamine
receptor agonist R()-propylnorapomorphine
hydrochloride (NPA, 10 6M) reduced L-type
Ca
2
channel current by roughly 20% in wild-
type medium spiny neurons (Fig. 2). How-
ever, in neurons from RCS knock-out mice,
the ability of both M
1
and D
2
receptor ac-
tivation to modulate L-type current was en-
hanced roughly twofold.
Our biochemical results suggested that in-
creased RCS phosphorylation should suppress
the ability of G proteincoupled receptors
(GPCRs) to activate PP2B and thereby mod-
ulate L-type Ca
2
channels. In neurons from
RCS knock-out mice, dialysis with phospho-
RCS diminished the ability of M
1
receptor
stimulation to reduce L-type Ca
2
currents
(IC
50
0 13%, n 0 5, P G 0.01, Kruskal Wallis),
whereas dialysis with dephospho-RCS had no
effect on the modulation (8). Activation of
PKA to phosphorylate endogenous RCS in
wild-type neurons nearly eliminated the mus-
carinic receptor modulation of L-type Ca
2
channel currents (n 0 8 and median modu-
lation of 0%) (fig. S3). In contrast, PKA ac-
tivation in neurons from RCS knock-out mice
had no significant effect on the muscarinic re-
ceptor modulation (n 0 9 and median modu-
lation 0 32%; untreated control, n 0 8 and
median modulation 0 42%; P 9 0.05, Kruskal
Wallis) (compare fig. S3D with Fig. 2F).
PP2B also has an important role in de-
phosphorylation of Thr
34
of DARPP-32 (dopa-
mine and cyclic adenosine monophosphate
regulated phospho-protein, 32,000 daltons), a
key component of dopaminergic signaling in
medium spiny neurons (11). Treatment of
neostriatal slices with the D
1
receptor agonist
SKF 81297 increased phosphorylation of
Ser
55
of RCS by threefold, reaching a peak
within 1 min and continuing for at least 30 min
(Fig. 3A). Treatment with SKF81297 also in-
creased phosphorylation of Thr
34
of DARPP-
32 by sevenfold in slices obtained from
wild-type mice, an effect that was sustained
for up to 30 min (Fig. 3B). In slices from
RCS knock-out mice, SKF 81297stimulated
phosphorylation of DARPP-32 was similar
over the first 5 min. However, DARPP-32
phosphorylation was transient, and dephos-
phorylation occurred rapidly in slices from
RCS knock-out mice (Fig. 3B).
Application of D
2
receptor agonists ac-
tivates PP2B and leads to dephosphoryla-
tion of Thr
34
of DARPP-32 (11). Treatment
Fig. 1. Phosphorylation-dependent interaction of RCS and CaM. (A) Co-immunoprecipitation of CaM
using an antibody against RCS. Striatal extract (lane 1 shows CaM in 10 6g lysate) was incubated
without (lane 2) or with (lane 3) rabbit polyclonal antibody against RCS followed by protein A
Sepharose beads. Co-immunoprecipitated proteins were separated by SDSpolyacrylamide gel
electrophoresis and immunoblotted with the use of mouse antibody against CaM. (B) Binding of
CaM to phospho- and dephospho-RCS. Increasing concentrations of CaM were incubated with
either dephospho-GST-RCS or phospho-Ser
55
-GST-RCS (90% phosphorylated). Results shown are
means T SEM from three independent experiments. Phosphorylated RCS inhibits CaMKI (C) and
PP2B (D). Increasing concentrations of unphosphorylated RCS or RCS phosphorylated by PKA (90%
phosphorylated) were incubated with CaM, either CaMKI or PP2B, and the respective substrate.
Data represent means T SEM of three independent experiments.
R E P O R T S
Fig. 2. Knock-out of RCS increases modulation of L-type
calcium currents by GPCRs. [(A), (C), and (E)] Current traces
from dorsal striatal medium spiny neurons. (Left) The
complete trace throughout the voltage step protocol; (right)
currents during the repolarizing step. (A) Neuron from an
RCS knock-out mouse. Control current (black) and tail cur-
rents elicited in the presence of 1 6M S()-BAY K 8644
(BayK), either alone (blue) or with 2 6M ()-muscarine
chloride (red). (B) Time course of current isolated from cell
in (A). Muscarine inhibited L-type current by 55% 5 ms after
initiation of the repolarization. The black trace shows the
normalized time course from a typical wild-type (wt) dorsal
striatal medium spiny neuron in which median inhibition of
current by 2 6M muscarine was 20%. (C) Neuron from an
RCS knock-out mouse. Control current (black) and tail
currents elicited in the presence of 1 6M BayK, either alone
(blue) or with 10 6M R()-propylnorapomorphine hydro-
chloride (red). (D) Time course of current isolated from cell
in (C). NPA inhibited L-type current by 44%. The black trace
represents the normalized time course from a typical wild-
type neuron, in which median inhibition of current by 10 6M
NPA in wild-type neurons was 18%. (E) Neuron from an
RCS knock-out mouse dialyzed with 20 6M thio-phospho-
RCS. Control current (black) and tail currents elicited in the
presence of 1 6M BayK, either alone (blue) or with 2 6M
muscarine (red). (F) Box plot (green) illustrates differences
in percent inhibition by M
1
receptor activation of L-type
current in medium spiny neurons from wild-type mice (n 0
5 and median of 20%), RCS knock-out (KO) mice (n 0
8 and median of 41%, P G 0.02, compared to wild-type,
Kruskal Wallis test) and RCS knock-out mice dialyzed with
thio-phospho-RCS (n 0 5 and median of 13%, P G 0.01 com-
pared with RCS knock-out mice). Box plot (yellow) illustrates
differences in percent inhibition by D
2
receptor activation
of L-type current in medium spiny neurons from wild-type
mice (n 0 3 and median of 18%) and RCS knock-out mice
(n 0 6 and median of 34%, P G 0.05).
Fig. 3. PP2B-dependent dephosphorylation of DARPP-32 in striatal neurons
from wild-type and RCS knock-out mice. (A) Effect of SKF81297 (1 6M) on
RCSSer
55
phosphorylation in neostriatal slices. The levels of phospho-Ser
55
RCS were analyzed by immunoblotting, and results were normalized to
values obtained with untreated slices from wild-type mice. Data represent
means T SEM for five experiments. A.U., arbitrary units. (B) Effect of D1
receptor agonist on DARPP-32 Thr
34
phosphorylation in neostriatal slices.
Slices from wild-type mice (solid) or RCS knock-out mice (open) were in-
cubated with SKF81297 (1 6M) for the indicated times. Data were nor-
malized to values obtained with untreated slices from wild-type mice and
represent means T SEM for 13 experiments. **P G 0.01 compared with wild-
type mice; two-way analysis of variance (ANOVA) followed by Bonferroni
posthoc test. (C) Effect of D2 receptor agonist on dephosphorylation of
DARPP-32 Thr
34
in neostriatal slices. Slices from wild-type mice (solid) or
RCS knock-out mice (open) were incubated for a total of 10 min in the
absence or presence of quinpirole (1 nM to 1 6M). SKF81297 (1 6M)
was added after 5 min of incubation. Data were normalized to values
obtained from slices treated with SKF81297 alone. The SKF81297-
stimulated levels of phospho-Thr
34
DARPP-32 were similar in slices
from wild-type and RCS knock-out mice. Data represent means T SEM
for five to seven experiments. *P G 0.05 and **P G 0.01 compared with
wild-type mice; Students t test. (D) Effect of quinpirole on dephos-
phorylation of DARPP-32 Thr
34
, using slices prepared from wild-type
mice (solid), and RCS knock-out mice (open) was examined in the
presence of the PP2B inhibitor cyclosporin A (CyA, 10 6M for 70 min).
Data were normalized to values obtained from slices treated with CyA
plus SKF81297. Data represent means T SEM for four experiments.
(Insert) Neostriatal slices prepared from wild-type mice (black) and
RCS knock-out mice (white) were incubated with no drug (control), CyA
(10 6M for 70 min), or CyA (10 6M for 70 min) plus SKF81297 (1 6M for
last 5 min). Data were normalized to values obtained with untreated
slices from wild-type mice. Data represent means T SEM for four to five
experiments. The effects of CyA and CyA plus SKF on DARPP-32 Thr
34
phosphorylation were similar in slices from wild-type and RCS knock-out mice.
R E P O R T S
of neostriatal slices from wild-type mice with
the D
2
receptor agonist quinpirole resulted in
a dose-dependent decrease in phosphoryla-
tion of Thr
34
of DARPP-32 (Fig. 3C). The
effect of quinpirole was greater in slices
from RCS knock-out mice. Half-maximal
inhibition of phosphorylation required 1 6M
quinpirole in slices from wild-type mice but
10 nM in slices from RCS-knock-out mice.
The effect of quinpirole was abolished when
slices from either wild-type or RCS knock-
out mice were pretreated with cyclosporin A
(10 6M), a specific PP2B inhibitor (Fig. 3D).
The present study indicates that RCS
plays a central role in integration of key
neurotransmitter inputs into medium spiny
neurons, placing it in a potentially pivotal
position to regulate striatal function in health
and disease through binding to CaM and af-
fecting the activation of multiple CaM tar-
gets. For example, RCS levels are reduced in
a presymptomatic transgenic mouse model
of Huntington_s disease (12). CaM is known
to regulate transglutaminase-mediated cross-
linking of the huntingtin protein (13), a pro-
cess that would be increased in the absence
of the inhibitory actions of phospho-RCS.
Administration of methamphetamine or co-
caine results in an increased phosphorylation
of RCS (14), suggesting that RCS-mediated
suppression of CaM signaling may be an
important mechanism underlying the effects
of psychomotor stimulants. Alterations in RCS
regulation of CaM and PP2B also may be an
important contributing factor in the patho-
genesis of schizophrenia, where there are clear
dopaminergic determinants (15, 16). Lastly,
RCS is positioned to arbitrate the interaction
between cholinergic and dopaminergic sig-
naling. The balance between these two trans-
mitters is critical for normal striatal function.
In Parkinson_s disease, the loss of striatal do-
pamine and the concomitant elevation of cho-
linergic tone is thought to be responsible for
the emergence of bradykinesia, rigidity, and
tremor (1719). Our work demonstrates that
RCS plays a central role in determining the
physiological impact of acetylcholine on the
principal neurons of the striatum and shows
how this regulation should change with dopa-
mine depletion (2022). The identification of
RCS and the determination of its importance
in striatal physiology creates a therapeutic
entry point for diseases of the basal ganglia.
By binding CaM and inhibiting PP2B,
phospho-RCS has the ability to amplify sig-
naling mediated by D
1
dopamine receptors
and other PKA-mediated GPCRs and to at-
tenuate signaling mediated by competing D
2
dopamine receptors, M1 muscarinic recep-
tors, and other phospholipase Cactivating
GPCRs (Fig. 4A). The ability of phospho-
RCS to inhibit PP2B is analogous to that of
phospho-DARPP-32 to inhibit PP1 (Fig. 4B).
Working in concert, these two signal trans-
duction mechanisms serve to amplify the
cellular consequences of PKA activation in
medium spiny neurons.
1. C. C. Ouimet, H. C. Hemmings Jr., P. Greengard, J.
Neurosci. 9, 865 (1989).
2. S. Brene et al., J. Neurosci. 14, 985 (1994).
3. J. Kisielow, A. C. Nairn, K. Karjalainen, Eur. J.
Immunol. 31, 1141 (2001).
4. H. C. Hemmings Jr., J. A. Girault, K. R. Williams, M. B.
LoPresti, P. Greengard, J. Biol. Chem. 264, 7726 (1989).
5. J. A. Girault, S. I. Walaas, H. C. Hemmings Jr., P. Greengard,
Neuroscience 37, 317 (1990).
6. A. Persechini, P. M. Stemmer, Trends Cardiovasc. Med.
12, 32 (2002).
7. D. J. Black, Q.-K. Tran, A. Persechini, Cell Calcium 35,
415 (2004).
8. S. V. Rakhilin et al., unpublished data.
9. A. R. Howe, D. J. Surmeier, J. Neurosci. 15, 458 (1995).
10. S. Hernandez-Lopez et al., J. Neurosci. 20, 8987
(2000).
11. A. Nishi, G. L. Snyder, P. Greengard, J. Neurosci. 17,
8147 (1997).
12. J. A. Bibb et al., Proc. Natl. Acad. Sci. U.S.A. 97, 6809
(2000).
13. G. M. Zainelli, C. A. Ross, J. C. Troncoso, J. K. Fitzgerald,
N. A. Muma, J. Neurosci. 24, 1954 (2004).
14. G. L. Caporaso et al., Neuropharmacology 39, 1637
(2000).
15. D. J. Gerber et al., Proc. Natl. Acad. Sci. U.S.A. 100,
8993 (2003).
16. T. Miyakawa et al., Proc. Natl. Acad. Sci. U.S.A. 100,
8987 (2003).
17. A. Barbeau, Can. Med. Assoc. J. 87, 802 (1962).
18. R. C. Duvoisin, Arch. Neurol. 17, 124 (1967).
19. O. Hornykiewicz, S. J. Kish, Adv. Neurol. 45, 19 (1987).
20. V. Bernard, E. Normand, B. Bloch, J. Neurosci. 12, 3591
(1992).
21. S. M. Hersch, C. A. Gutekunst, H. D. Rees, C. J. Heilman,
A. I. Levey, J. Neurosci. 14, 3351 (1994).
22. D. J. Surmeier, W. J. Song, Z. Yan, J. Neurosci. 16, 6579
(1996).
23. P. Svenningsson et al., Annu. Rev. Pharmacol. Toxicol.
44, 269 (2004).
24. B. Z. Peterson et al., Neuron 22, 549 (1999).
25. J. Kim, S. Ghosh, D. A. Nunziato, G. S. Pitt, Neuron 41,
745 (2004).
26. We thank P. Ingrassia, A. Horiuchi, F. Liu, and E. Griggs
for technical support. Work described in this paper
was supported by U.S. Public Health Service grants
MH40899 and DA10044 (A.C.N. and P.G.) and
NS34696 and DA12958 (D.J.S.), the Picower Founda-
tion, the Peter Jay Sharp Foundation, and the Simons
Foundation (P.G.), and a Grant-in-Aid for Scientific
Research from the Japan Society for the Promotion
of Science (A.N.). Molecular interaction data have
been deposited in the Biomolecular Interaction
Network Database with accession codes 169429 to
169431.
DC1
SOM Text
Figs. S1 to S3
Fig. 4. Role of RCS in
neuronal signaling. (A)
Regulation by RCS of
phosphorylation and
dephosphorylation of
PKA/PP2B substrates.
Stimulation of dopa-
mine D
1
receptors acti-
vates PKA. Stimulation
of muscarinic M
1
and
dopamine D
2
receptors
mobilizes intracellular
Ca
2
stores, resulting
in stimulation of PP2B,
which dephosphoryl-
ates a large number
of substrates. Many
proteins, including, for
example, L-type chan-
nels (illustrated here)
and DARPP-32, are
substrates for both
PKA and PP2B (23).
RCS regulates the state
of phosphorylation of
this major class of sub-
strates. The phospho-
rylated form of RCS inhibits the activation by free Ca
2
/CaM of PP2B (red arrow) and other
Ca
2
/CaM effectors that are likely to include CaM kinases, CaM-dependent phosphodiesterase,
and adenylyl cyclase (not shown in the figure). [Effects dependent upon bound CaM, like Ca
2
-
dependent inactivation of Ca
2
channels (24, 25), are less likely to be affected by phosphorylation
of soluble RCS.] As shown in the figure, phosphorylation of RCS amplifies the effects of PKA by
turning off the dephosphorylation of these PKA substrates by PP2B. Conversely, dephosphorylation
of RCS by PP1 and PP2A (14), through activation of PP2B, promotes Ca
2
/CaM signaling with a
consequent dampening of PKA signal transduction pathways. Thus, the state of phosphorylation
of RCS may regulate the balance between adenylyl cyclasemediated versus PP2B-mediated
GPCR signaling pathways. (B) Parallel roles of DARPP-32 and RCS in regulation of signal trans-
duction. The efficacy of phosphorylation by PKA of numerous PKA/PP1 substrates is increased
by PKA phosphorylation of DARPP-32, which inhibits PP1 (23). In an analogous manner, the
efficacy of phosphorylation by PKA of numerous PKA/PP2B substrates is increased by PKA phos-
phorylation of RCS, which inhibits PP2B. Thus, parallel signal transduction mechanisms have
evolved, with the use of DARPP-32 and RCS, to amplify PKA phosphorylation of two major and
distinct classes of substrates.
R E P O R T S
Plant Cuticular Lipid Export
Requires an ABC Transporter
Jamie A. Pighin,
1
* Huanquan Zheng,
1
* Laura J. Balakshin,
1
Ian P. Goodman,
1
Tamara L. Western,
1
Reinhard Jetter,
1,2
Ljerka Kunst,
1
A. Lacey Samuels
1
.
A waxy protective cuticle coats all primary aerial plant tissues. Its synthesis
requires extensive export of lipids from epidermal cells to the plant surface.
Arabidopsis cer5 mutants had reduced stem cuticular wax loads and ac-
cumulated sheetlike inclusions in the cytoplasm of wax-secreting cells. These
inclusions represented abnormal deposits of cuticular wax and resembled
inclusions found in a human disorder caused by a defective peroxisomal aden-
osine triphosphate binding cassette (ABC) transporter. We found that the
CER5 gene encodes an ABC transporter localized in the plasma membrane of
epidermal cells and conclude that it is required for wax export to the cuticle.
All primary aerial organs of land plants are
covered with a waxy cuticle that is essential
for their protection and interaction with the
environment. The cuticle is composed of
very-long-chain fatty acids and their deriva-
tives, collectively termed cuticular wax,
embedded within and encasing the cutin
matrix (1). Cuticle synthesis requires exten-
sive transport of lipids out of the epidermal
cells to the plant surface. The mechanism of
export of the cuticular lipids is unknown.
To identify mutants defective in lipid
transport to the cuticle, we examined a
collection of Arabidopsis thaliana eceriferum
(or cer) lines for changes in wax-secreting
epidermal cells by transmission electron mi-
croscopy (TEM). Cer mutants have a glossy,
bright green stem phenotype because of a
reduction or altered composition of cuticular
wax (2). TEM study (3) of the stem epider-
mis of the cer5 mutant revealed an unusual
cellular phenotype. Similar to the wild type,
cer5 cells were entirely filled with a central
vacuole with the cytoplasm present in a thin
rim around the edge of the cell (Fig. 1A), but
they also contained large protrusions of cyto-
plasm into the vacuole (Fig. 1B). Within
these protrusions, loose bundles of linear in-
clusions, distinct from the endoplasmic retic-
ulum, Golgi, and cytoskeletal elements, were
found (Fig. 1C). These inclusions were found
only in the epidermis; they were never ob-
served in other cell types. Morphologically
similar trilamellar inclusions had been de-
scribed in the cells of patients with X-linked
adrenoleukodystrophy (ALD), a neurodegen-
erative disease caused by a defect in an ABC
transporter involved in transport of saturated
very-long-chain fatty acids into the peroxi-
some for $-oxidation (4).
The cer5 stem epidermis was further ex-
amined by cryoscanning electron microsco-
py (SEM) (Fig. 1D) (3). The stem surface was
sparsely covered with epicuticular wax crys-
tals, consistent with reports that the wax load
on cer5 stems is merely reduced, not elimi-
nated (5). The cer5 epidermal cells contained
large sheetlike structures, which corresponded
in size and arrangement with the rod-shaped
inclusion profiles seen in TEM sections (Fig.
1, E and F). Nile red staining and examination
by light microscopy demonstrated that these
inclusions were lipidic in nature (fig. S1).
Morphological similarities between the
cer5 inclusions and those found in ALD cells
raised the question of whether both structures
had similar composition. Because the cer5
inclusions could not be prepared selectively,
we inferred their composition from compar-
isons between isolated epidermal cells with
and without inclusions (3). The total fatty acid
profiles of cer5 and wild-type epidermal peels
did not differ significantly. Thus, it is unlikely
that the cer5 inclusions consist of fatty acids,
distinguishing them from the corresponding
structures in ALD cells.
When cuticular wax components were
quantified (3), wild-type plants showed a
wax load of 0.24 6g/mm
2
, whereas the mu-
tant had only 0.11 6g/mm
2
of wax (Fig. 2A).
The amounts of all wax components (e.g.,
alkanes, ketones, and primary and secondary
alcohols) on the cer5 surface were signifi-
cantly reduced (Fig. 2B). In contrast, the
amounts of total epidermal wax (surface plus
intracellular) extracted from isolated epider-
mal peels of wild type (0.31 6g/mm
2
) and
cer5 (0.28 6g/mm
2
), did not differ signifi-
cantly. Thus, wax biosynthesis was not
compromised in cer5, but wax components
were retained within epidermal cells.
To determine the molecular basis of the
cer5 defect, we isolated the CER5 gene by
using a combination of positional cloning and
insertional mutagenesis (3). Complementa-
tion of the cer5 mutant with the wild-type
CER5 gene (At1g51500) rescued the wax-
deficient phenotype (Fig. 2A). Thus, CER5 is
the At1g51500 gene encoding an ABC
transporter. During cloning, we identified an
additional allele of CER5 in the Salk transfer-
DNA (T-DNA) insertional mutation collec-
tion (Salk 036776) (Fig. 2A). We designated
the original mutant allele cer5-1 and the Salk
036776 allele cer5-2. Sequencing of the
At1g51500 gene in cer5-1 identified a point
1
Department of Botany, University of British Colum-
bia (UBC), 6270 University Boulevard, Vancouver, BC
V6T 1Z4, Canada.
2
Department of Chemistry, UBC,
2036 Main Mall, Vancouver, BC V6T 1Z1, Canada.
E-mail: lsamuels@interchange.ubc.ca
Fig. 1. Epidermal wax-
secreting cells of Ara-
bi dopsi s stems i n
transverse section. (A)
Wild-type cells. c indi-
cates cytoplasm; cw,
cell wall. Scale bar, 2
6m. (B) cer5 cells with
intrusions of cyto-
plasm in vacuoles (ar-
rowhead). Scale bar, 2
6m. (C) cer5 cyto-
plasm contains unusu-
al linear inclusions
( ar r owheads) . ER,
endoplasmic reticu-
lum; G, Golgi. Scale
bar, 200 nm. (D)
Cryo-SEM of cer5 epi-
dermis, covered with
cuticle. Scale bar, 5
6m. (E) cer5 epidermal
cell with inclusions (ar-
row). Scale bar, 2 6m.
(F) High-magnification
view showing sheetlike
nature of inclusions.
Scale bar, 2 6m.
R E P O R T S
mutation that, in the predicted CER5 protein,
would cause the replacement of a glycine
with an aspartate within the consensus ABC C
motif (6) (fig. S2). The T-DNA insertion in
the cer5-2 allele was located in an exon
encoding the region after the predicted fifth
transmembrane domain of the CER5 protein.
We examined CER5 transcript levels in the
cer5 mutants to determine the extent of gene
disruption in each line. Whereas the abun-
dance of the CER5 transcript in cer5-1 was
comparable to the wild type, no transcript
could be detected in cer5-2, indicating that it
is a transcriptional knockout (fig. S3).
Analysis of the predicted CER5 protein
sequence revealed the presence of the char-
acteristic ABC transporter domains near the
N terminus, including the Walker A and B
boxes and C motif for nucleotide binding
and six transmembrane domains (TMD) near
the C terminus (fig. S2). When compared
with prototype ABC transporters, which
have the TMD near the N terminus followed
by the ABC domain, the ABC-TMD orien-
tation found in CER5 would be considered a
reverse arrangement. Known ABC trans-
porters consist of two (ABC-TMD) units
(7) either within one polypeptide or as two
Bhalf-transporters[ making up homo- or
heterodimers. CER5 sequence predicts that
it would encode a half-transporter, so pre-
sumably CER5 would require dimerization
to function.
The CER5 sequence has been designated
WBC12, a member of the white-brown com-
plex subfamily, in an analysis of the 129
putative ABC transporters of the Arabidopsis
genome (8). This is the largest subfamily of
ABC transporters in Arabidopsis, and, al-
though some have been cloned (9), CER5 is
the only member of the subfamily that has
been characterized functionally. Two other
putative Arabidopsis ABC transporters have
high similarity to CER5, At3g21090 (WBC
15) and At1g51460 (WBC 13) (10). Further-
more, there is similarity to two human ABC
transporters from the WBC/ABCG sub-
family: breast cancer resistance protein and
a placental ABCG2, which are localized to
the plasma membrane (11) and believed to
function in lipid and xenobiotic export (12).
The simplest hypothesis is that CER5, like
other WBC subfamily members, acts as a
primary transporter of lipids. However, it can-
not be ruled out that it acts indirectly by reg-
ulating the activities of other transporters.
CER5 was expressed exclusively in the
epidermal cells, as shown by GUS activity
assays in plants transformed with the CER5
promoter::GUS construct (Fig. 3A) (3). CER5
transcript was found in all examined plant
organs, including stems, leaves, siliques, flow-
ers, and roots (fig. S4). This was unexpected,
because the cer5 phenotype is only apparent
in stems or detectable by gas chromatography
in stems and leaves (85% of wild-type wax
load) (5). It suggests that additional trans-
porters must be involved in delivering wax
components to the cuticle in other tissues.
To investigate the subcellular localization
of CER5 in Arabidopsis, we introduced a
GFP-CER5 (where GFP indicates green
fluorescent protein) construct driven by the
native CER5 promoter into cer5-1 plants (3).
The wild-type phenotype was restored in 42
of 43 transgenic plants expressing the GFP-
CER5 fusion protein, indicating that the pro-
tein was fully functional. The GFP-tagged
CER5 was localized in the plasma mem-
brane of epidermal cells (Fig. 3B). When the
cell wall was stained with propidium iodide,
the GFP-CER5 plasma membrane fluores-
cence was clearly separated by the red pro-
pidium iodide signal (Fig. 3B, inset).
We identified the plasma membrane-
localized ABC transporter, CER5, involved
in wax export to the plant cuticle. CER5
must be an important component of the ex-
port machinery in the Arabidopsis stem,
because disruption of this transporter results
in striking accumulations of wax inside the
epidermal cells. The absence of a detectable
Fig. 2. Wax analyses of
Arabidopsis stem surface
(cuticle) or epidermal peel
extracts (total epidermis).
(A) Cuticular wax loads of
WT ecotypes are signifi-
cantly different from the
corresponding mutants:
Landsberg erecta (Ler) vs.
cer5-1; Columbia-2 (Col) vs.
cer5-2 (t test, P 0 0.05). F
1
progeny of a cer5-1 and
cer5-2 cross had a reduced
wax load similar to both
parents. Cer5-1 plants,
complemented with the
At1g51500 gene, showed
significantly increased wax loads. Total epidermis wax loads, intracellular and cuticular
(right), are not significantly different in WT and cer5-1. (B) Cuticular wax analysis,
including chain lengths of aliphatic compounds, revealed reductions of all major wax components of cer5-1 (Error bars represent means T SE).
W
a
x

l
o
a
d
m
/
g
[
m
]
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40 Total
epidermis
Cuticle
)
r
e
L
(
T
W
1
-
5
r
e
c
1
-
5
r
e
c
x
r
e
c
2
-
5
F
1
+
1
-
5
r
e
c
E
C
5
R
T
1
)
l
o
C
(
T
W
2
-
5
r
e
c
)
r
e
L
(
T
W
1
-
5
r
e
c
2830 262830 24262830 272931 282930 29 4244464850
]
m
m
/
g
[
W
a
x

l
o
a
d
0.00
0.02
0.04
0.06
0.08
0.10
0.12
WT (Ler)
cer5-1
Fatty
acids
Aldehydes
prim.
Alcohols
Alkanes
sec.
Alcohols
Ketone
Alkyl
esters
Not
identified
A B
Fig. 3. Expression of CER5 in the plasma membrane of the stem epidermis. (A) CER5 promoter
directed epidermis-specific expression of GUS in stem. Scale bar, 100 6m. (B) GFP-CER5 fusion
protein was localized to the plasma membrane (pm) of epidermal cells. Scale bar, 10 6m. (Inset)
High magnification of GFP-CER5 expressing cells labeled with propidium iodide, which stains the
cell wall red between adjacent GFP-labeled plasma membranes. Scale bar, 1 6m.
R E P O R T S
phenotype in tissues other than the stem and
leaf and accumulation of residual surface wax
on the stem of cer5-2 knockout line suggest
that additional wax export mechanisms must
exist in plants. Chemical analysis of the mu-
tant wax demonstrated that CER5, like many
ABC transporters, has broad substrate speci-
ficity and is capable of transporting a variety
of wax substrates. We conclude that in plants,
as in other eukaryotes, proteins of the WBC/
ABCG subfamily are key components of
lipid transport systems.
1. L. Kunst, A. L. Samuels, Prog. Lipid Res. 42, 51 (2003).
2. M. Koornneef, C. J. Hanhart, F. Thiel, J. Hered. 80,
118 (1989).
3. Materials and methods are presented as supporting
4. H. Powell, R. Tindall, P. Schultz, Arch. Neurol. 32, 250
(1975).
5. A. M. Rashotte, M. A. Jenks, K. A. Feldmann,
Phytochemistry 57, 115 (2001).
6. ABC transporter motifs were predicted by PROSITE
as referenced in (13).
7. M. Jasinski, E. Ducos, E. Martinoia, M. Boutry, Plant
Physiol. 131, 1169 (2003).
8. R. Sa nchez-Ferna ndez, T. G. E. Davies, J. O. D.
Coleman, P. A. Rea, J. Biol. Chem. 276, 30231 (2001).
9. C. T. Otsu et al., J. Exp. Bot. 55, 1643 (2004).
10. The analyses of R. Sa nchez-Ferna ndez et al. (8) agree
with these relationships; however, they erroneously
duplicated WBC15/WBC22 in their 2001 work. This
was corrected in (14).
11. G. L. Scheffer et al., Cancer Res. 60, 2589 (2000).
12. J. W. Jonker et al., Proc. Natl. Acad. Sci. U.S.A. 99,
15649 (2002).
13. L. Falquet et al., Nucleic Acids Res. 30, 235 (2002).
14. P. A. Rea et al., in ABC Transporters from Bacteria to
Man, I. B. Holland, S. P. C. Cole, K. Kuchler, C. F.
Higgins, Eds. (Academic Press, London, 2003), pp.
335355.
15. Thanks to G. Haughn, M. Smith, T. Hooker, and
O. Rowland for their insightful comments. The
financial support of the Natural Sciences and
Engineering Research Council of Canada, Canadian
Foundation for Innovation, BC Knowledge Devel-
opment Foundation, and the UBC Blusson fund are
gratefully acknowledged. We thank the Salk In-
stitute for Genomic Analysis Laboratory for pro-
viding sequence-indexed Arabidopsis T-DNA insertion
mutants (project funded by NSF). The CER5 gene
has been submitted to Genbank, and the accession
no. is AY734542.
DC1
Figs. S1 to S4
Oscillations in NF-.B Signaling
Control the Dynamics of
Gene Expression
D. E. Nelson,
1
A. E. C. Ihekwaba,
2
M. Elliott,
1
J. R. Johnson,
1
C. A. Gibney,
1
B. E. Foreman,
1
G. Nelson,
1
V. See,
1
C. A. Horton,
1
D. G. Spiller,
1
S. W. Edwards,
1
H. P. McDowell,
4
J. F. Unitt,
5
E. Sullivan,
6
R. Grimley,
7
N. Benson,
7
D. Broomhead,
3
D. B. Kell,
2
M. R. H. White
1
*
Signaling by the transcription factor nuclear factor kappa B (NF-.B) involves its
release from inhibitor kappa B (I.B) in the cytosol, followed by translocation
into the nucleus. NF-.B regulation of I.B! transcription represents a delayed
negative feedback loop that drives oscillations in NF-.B translocation. Single-
cell time-lapse imaging and computational modeling of NF-.B (RelA)
localization showed asynchronous oscillations following cell stimulation that
decreased in frequency with increased I.B! transcription. Transcription of
target genes depended on oscillation persistence, involving cycles of RelA
phosphorylation and dephosphorylation. The functional consequences of NF-
.B signaling may thus depend on number, period, and amplitude of oscillations.
NF-0B is a family of dimeric transcription
factors (usually RelA/p65:p50) that regulates
cell division, apoptosis, and inflammation
(1). NF-0B dimers are sequestered in the
cytoplasm of unstimulated cells by binding
to I0B proteins. NF-0Bactivating stimuli
activate the inhibitor kappa B kinase (IKK)
signalosome that phosphorylates I0B Eat
Ser32 and Ser36 on I0B" (2)^ and NF-0B
Eat Ser536 in RelA (3, 4)^. Phosphorylated
I0B proteins are then ubiquitinated and
degraded by the proteasome, liberating NF-
0B dimers to translocate to the nucleus and
regulate target gene transcription.
I0B" is a transcriptional target for NF-0B
(5), creating a negative feedback loop (Fig.
1A) in which its delayed expression gives the
system similar characteristics to the circadian
clock (6) and to ultradian oscillators such
as p53 (7, 8) and the segmentation clock
(8, 9). I0B" contains both nuclear localization
and export sequences, enabling its nuclear-
cytoplasmic (N-C) shuttling. Newly synthe-
sized free I0B" binds to nuclear NF-0B,
leading to export of the complex to the
cytoplasm (10). This complex, but not free
I0B", is the target for I0B" phosphorylation
by IKK (11, 12).
Oscillations in the temporal response of
NF-0B activity have been observed by
electromobility shift assay (EMSA) only in
studies of I0B$ and ( knockout mouse em-
bryonic fibroblast cell populations and have
been simulated in a computational model
(13). In the absence of time-lapse single-cell
analysis, it has remained unclear whether
asynchronous single-cell oscillations occur
in single cells following NF-0B stimulation
(8, 14). Like calcium signaling (15), NF-0B
could be a complex dynamic oscillator using
period and/or amplitude to regulate transcrip-
tion of target genes.
We have used fluorescence imaging of
NF-0B (RelA) and I0B" fluorescent fusion
proteins (11, 16) to study oscillations in
RelA N-C localization (N-C oscillations) in
HeLa (human cervical carcinoma) cells and
SK-N-AS cells Ehuman S-type neuroblasto-
ma cells that have been associated with
deregulated NF-0B signaling (17)^. In SK-
N-AS cells expressing RelA fused at the C
terminus to the red fluorescent protein
DsRed (RelA-DsRed) and I0B" fused at
the C terminus to the enhanced green
fluorescent protein EGFP (I0B"-EGFP)
(Fig. 1B and Fig. 2A), 96% showed an
NF-0B nuclear translocation response to
tumor necrosis factor alpha (TNF") stimu-
lation and 72% showed long-term N-C
oscillations in RelA-DsRed localization. Os-
cillations with a typical period of 100 min
continued for 920 hours after continuous
TNF" stimulation, damping slowly. In trans-
fected cells expressing RelA-DsRed and
control EGFP (Fig. 2C), 97% responded
and 91% of cells showed N-C oscillations.
These oscillations appeared more synchro-
nous between cells in the first three cycles
1
Centre for Cell Imaging, School of Biological Sci-
ences, Bioscience Research Building, Crown Street,
Liverpool, L69 7ZB, UK.
2
Department of Chemistry,
3
Department of Mathematics, University of Man-
chester Institute of Science and Technology, P.O. Box
88, Sackville Street, Manchester, M60 1QD, UK.
4
Royal Liverpool Childrens National Health Service
Trust, Alder Hey Hospital, Eaton Road, Liverpool, L12
2AP, UK.
5
Molecular Biology Department,
6
Advanced
Science and Technology Laboratory, AstraZeneca
Research and Development Charnwood, Bakewell
Road, Loughborough, Leicestershire, LE11 5RH, UK.
7
Pfizer Central Research, Ramsgate Road, Sandwich,
Kent, CT13 9NJ, UK.
E-mail: mwhite@liv.ac.uk
R E P O R T S
compared with cells that also expressed
I0B"-EGFP, which suggests that the sys-
tem was sensitive to variation in I0B"
levels, thus contributing to the degree of
cell-to-cell asynchrony. When HeLa cells
were continually stimulated with TNF"
(Fig. 2D), 86% of the cells responded and
30% exhibited up to three detectable N-C
oscillations that were markedly damped.
However, when TNF" was added to SK-N-
AS cells (Fig. 2B) or HeLa cells (Fig. 2E) as
a 5-min pulse, a single peak of nuclear
occupancy was observed with no subsequent
cycles of RelA movement.
TNF" treatment induced endogenous
RelA localization patterns in cells, consist-
ent with increasingly asynchronous N-C
oscillations (fig. S3). Western blot analysis
(figs. S4 and S5) showed that SK-N-AS and
HeLa cells continually treated with TNF"
gave biphasic dynamics of total I0B",
phosphorylated I0B" (Ser32 phospho-
I0B"), and phosphorylated RelA (Ser536
phospho-RelA). In HeLa cells, phosphopro-
tein expression levels diminished more
rapidly than in the SK-N-AS cells (fig.
S5). A 5-min TNF" pulse directed transient
accumulation of Ser32 phospho-I0B" and
Ser536 phospho-RelA (fig. S4B). These
data support the hypothesis that loss of
IKK activity (due to TNF" removal) results
in loss of N-C oscillations and that de-
phosphorylation of RelA occurs rapidly
without persistent IKK activity. When SK-
N-AS cells were treated with an alternative
stimulus, the topoisomerase II inhibitor
etoposide (VP16), 37% of the cells
responded and 24% showed N-C oscilla-
tions. Etoposide-induced N-C oscillations
had lower amplitude than those induced by
TNF", peaking after 300 min and then di-
minishing (Fig. 2F). The I0B" and RelA
phosphoprotein expression levels after etopo-
side treatment (fig. S4C) corresponded to
the timing of N-C oscillations.
We investigated whether N-C oscillation
persistence influenced the dynamics of NF-
0Bregulated gene expression using real-
time imaging of firefly luciferase activity
(18) driven by a 0B (5 consensus site)
promoter. SK-N-AS cells exhibited stable
luminescence for more than 25 hours in the
continual presence of TNF" (Fig. 2G). HeLa
cells showed a transient peak 10 hours after
TNF" treatment that decayed by 20 hours
(Fig. 2I). In SK-N-AS (Fig. 2H) or HeLa
cells (Fig. 2J) treated with a 5-min TNF"
pulse, a more transient peak of luminescence
occurred after 5 hours, which decayed by 10
hours. Etoposide treatment of SK-N-AS
cells elicited a lower luminescence signal,
reaching a peak at 15 hours after treat-
ment (Fig. 2K). With each stimulus, the
kinetics of NF-0B oscillations and mainte-
nance of phosphoprotein levels appeared
closely related to the kinetics of gene
expression. Thus, persistent NF-0B oscilla-
tions appear to maintain NF-0Bdependent
gene expression.
Analysis of successive peaks of RelA
nuclear occupancy (figs. S13 and S14 and
Fig. 3, E and F) showed that N-C oscilla-
tion damping and successive peak timing
were highly reproducible, but because of
phase differences, this was not apparent at
the population level. However, the pattern
of peak timing and amplitudes was dif-
ferent between HeLa and SK-N-AS cells.
The expression of I0B"EGFP affected
the amplitude and peak timing of the N-C
oscillations (fig. S14). To study the role
of I0B" synthesis rate on N-C oscilla-
tions, the rate of NF-0Bregulated I0B"
transcription was modulated. I0B"-EGFP
expression was driven by the 0B (5
consensus site) promoter and expressed in
HeLa cells together with a fusion protein
between RelA and the modified red
fluorescent protein DsRed-Express (RelA-
DsRed-Express). Continual TNF" stimula-
tion elicited oscillations in I0B"-EGFP
expression out of phase with the RelA
N-C oscillations (Fig. 3, A and B). This
caused a statistically significant delay in
the timing of nuclear RelA peaks 1, 2, and
3 (Fig. 3F). The amplitude was also slightly
reduced for peaks 2 and 3 in the presence
of the 0B-I0B"-EGFP expression vector
(Fig. 3E).
To investigate parameters affecting the
oscillation dynamics, we used a computa-
tional model (13) that predicted NF-0B
oscillations with a similar period and damp-
ing as those observed here. From this
model, we noted that changes in just two
molecular species (variables), free IKK and
I0B", were intimately coupled to the
oscillation dynamics of nuclear NF-0B
(fig. S16). Transfection with the 0B-I0B"-
EGFP expression vector (Fig. 3, A, B, E,
and F) was equivalent to increasing the rate
of NF-0Bdependent I0B" transcription;
thus, we chose to study the effect of this
parameter in the model (reaction 28 in
table S1; Fig. 3, C and D; and fig. S17).
Fig. 1. Oscillations in NF-0B localization. (A) Schematic diagram illustrating the potential mechanism
for repeated oscillations in NF-0B (p65/RelA) N-C localization. (B) Time-lapse confocal images of SK-
N-AS cells expressing RelA-DsRed (red) and I0B"-EGFP (green) showing single-cell asynchronous N:C
oscillations in RelA-DsRed localization after stimulation with 10 ng/ml TNF". The arrow marks one
oscillating cell. Times, min; scale bar, 50 6m.
R E P O R T S
See (19) for analysis of some other related
parameters (figs. S18 and S19). As the
rate of this reaction was increased, there
was a delay in simulated peaks 2 and
onward (Fig. 3, C, D, and G). Thus, the
computational analysis showed the effects
of this reaction rate to be similar to those
seen in the experimental studies. One
discrepancy between the computational
model and the experimental data was the
unpredicted delay in experimentally ob-
served peak 1 caused by 0B-I0B"-EGFP
transfection (Fig. 3, F and G). It is unclear
how the two cell types studied differ with
respect to the values of the parameters
used in the model. Given that the oscil-
lations are naturally asynchronous between
cells and that this might be associated
with varying levels of I0B proteins (13)
or a lack of optimization of the pre-
equilibration step in the model, this may
explain why the timing of peak 1 was im-
perfectly predicted.
The amplitude of oscillations in I0B"-
EGFP when expressed under the control of
the 0B promoter was not directly related to
the amplitude of the preceding peak in
RelA nuclear localization. In many HeLa
cells, peak 2 or 3 in RelA localization was
small in amplitude (Fig. 3B) compared
with peak 1 (and would not have been ob-
served in asynchronous populations). Never-
theless, these oscillations led to easily
observable I0B"-EGFP responses. Thus,
persistence of NF-0B oscillations maintains
NF-0Bdependent transcription. However,
NF-0B translocation cannot be the only
factor regulating transcriptional activation
(a property of the whole system), and further
NF-0B activating and inactivating reactions,
including modifications of RelA by phos-
phorylation (3, 4, 20), acetylation (21), or
prolyl isomerization/targeted degradation
(22), have also been described. The cessa-
tion of NF-0Bdependent transcription in
the nucleus, independent of nuclear export
(11), might occur as a consequence of
RelA inactivation. Thus, NF-0B oscillations
could repeatedly deliver newly activated
NF-0B into the nucleus, maintaining a
high nuclear ratio of active:inactive NF-
0B. To investigate this hypothesis, we used
the CRM1-dependent nuclear export inhib-
itor leptomycin B (LMB) to trap RelA in
the nucleus of SK-N-AS cells (Fig. 4, A
and B). This resulted in transitory 0B-
dependent luciferase reporter gene expres-
sion (11) that peaked after 5 hours (Fig.
4C). Western blot analysis indicated a tran-
sient increase in Ser32 phospho-I0B" ex-
pression after 5 min, with no subsequent
recovery (Fig. 4E). Ser536 phospho-RelA
expression was maximal at 5 min after
stimulation and decayed to the threshold
of detection by 180 min (in contrast to
cells treated with constant TNF!, Fig.
4D). These data support the hypothesis
(23) that rapid dephosphorylation of NF-
0B in the nucleus Eby PP2A activity (24)^
may be a key factor in the switch-off of
NF-0Bdependent gene expression.
We propose that oscillations in NF-0B
localization coupled to cycles of RelA and
I0B" phosphorylation maintain NF-0B
dependent gene expression. Calcium spikes
at intervals as long as 30 min have been
shown to maintain NF-0B activity in T cells
(25). The decoding of this ECa
2
^ spike fre-
quency might be related to the observed
kinetics of oscillatory transcription factor
shuttling and regulation (26). Specific, non-
Fig. 2. Analysis of the
dynamics of NF-0B
localization and 0B-
dependent reporter
gene expression. (A to
F) Time course of N:C
localization of RelA-
DsRed in cells co-
expressing I0B"-EGFP
[(A), (B), (D), (E), and
(F)] or EGFP control
(C). N:C ratio in RelA-
DsRed fluorescence
was normalized to
highest peak intensity.
The peak N:C ratio was
expressed as the aver-
age value for each set
of four cells. Data from
each cell is represented
by a different colored
line. (G to K) Lumines-
cence imaging (RLU,
relative light units) of
the dynamics of 0B-
dependent luciferase
reporter activity repre-
sented as a different
colored line for each
of four different cells.
The black line repre-
sents the average of
the cells. [(A), (C), and
(G)] SK-N-AS cells
treated with continual
10ng/ml TNF". [(D) and
(I)] HeLa cells treated
with continual 10 ng/ml
TNF". [(B) and (H)] SK-
N-AS cells treated with a 5-min TNF" pulse. [(E) and (J)] HeLa cells treated with a 5-min TNF" pulse. [(F) and (K)] SK-N-AS cells treated with
20 6M of etoposide. The black bar above each graph is a representation of the duration of TNF" treatment. For images of data in [(B) to (F)], (G), and (I), see
figs. S6 to S11.
0
0.2
0.4
0.6
0.8
0 200 400 600
0
1
2
3
0 200 400
0
1
2
3
0 200 400
0
3
6
9
0 200 400 600
0
3
6
9
12
0 200 400 600
0
3
6
9
0 200 400 600
Time (min) after TNF Time (min) after TNF
Time (min) after TNF
Time (h) after TNF Time (h) after TNF Time (h) after TNF Time (h) after TNF Time (h) after etoposide
Time (min) after TNF Time (min) after etoposide
A
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
1
3
5
7
9
1
3
5
7
9
1
3
5
7
9
1
3
5
7
9
1
3
5
7
9
0 6 12 18 24 0 6 12 18 24 0 6 12 18 24 0 6 12 18 24 0 6 12 18 24
R
L
U
R
L
U
R
L
U
R
L
U
R
L
U
I G H
E
C
A
D
J
10ng/ml TNF
10ng/ml TNF
10ng/ml TNF 10ng/ml TNF 10ng/ml TNF 10ng/ml TNF
10ng/ml TNF
10ng/ml TNF
10ng/ml TNF
B
K
20M Etoposide
F
20M Etoposide
R E P O R T S
linear Bnetwork motifs[ can decode frequen-
cies rather than amplitudes (27). Therefore,
the signal-processing elements of the NF-0B
signaling pathway, and its interaction with
other dynamic signaling systems, may involve
the encoding and decoding of specific time-
varying signals. Such temporal encoding
could avoid undesirable cross talk between
cellular signaling pathways that share com-
mon components. Furthermore, oscillatory
phosphorylation of RelA at Ser536 appears
to be a consequence of its shuttling between
the cytoplasm and the nucleus. Oscillatory
modifications at other regulatory amino
acids in RelA (21, 28) might also occur as
a consequence of N-C oscillations, whereas
changes in N-C oscillation frequency and
persistence might explain differential regu-
lation of cell fate in response to different
stimuli. Thus, in common, and perhaps in
combination, with other oscillatory tran-
scription factor pathways such as p53 (7, 8),
NF-0B may constitute a complex analog-
to-digital coding system that regulates cell
fate.
1. S. Ghosh, M. J. May, E. B. Kopp, Annu. Rev. Immunol.
16, 225 (1998).
2. J. A. DiDonato, M. Hayakawa, D. M. Rothwarf, E. Zandi,
M. Karin, Nature 388, 548 (1997).
0
1
2
3
4
Peak1 Time0:Peak1 Peak1:2 Peak2:3 Time0:Peak1 Peak1:2 Peak2:3 Peak2 Peak3
200 300 400 100 0 500 600
N
u
c
l
e
a
r

N
F
-
B
700 0.1
0.08
0.06
0.04
0.02
0
600
500
400
300
200
100
0
0.1 1 10 100
k28
T1
T2
T3
T4
T5
T6
A1
A2
A3
A4
A5
A6
1x
2x
5x
10x
1.3
0.1
0.08
0.06
0.04
0.02
0
1.1
0.9
0.7
0.5
0.3
0.1
800 600 400 200 0
Time (min)
Time (min)
T
i
m
e

(
m
i
n
)
T
i
m
e

(
m
i
n
)
800 600 400 200 0
Time (min)
800 600 400 200 0
Time (min)
N
:
C

R
e
I
A

a
m
p
l
i
t
u
d
e
N
:
C

R
e
I
A

a
m
p
l
i
t
u
d
e
N
:
C

R
e
I
A

a
m
p
l
i
t
u
d
e
N
:
C

R
e
I
A

a
m
p
l
i
t
u
d
e
A
m
p
l
i
t
u
d
e
I
-
E
G
F
P

a
m
p
l
i
t
u
d
e
I
-
E
G
F
P

a
m
p
l
i
t
u
d
e
I
-
E
G
F
P

a
m
p
l
i
t
u
d
e
1
0.7
0.4
0.1
1.3
1
0.4
0.1
1.3
1
0.4
0.1
Nuc: Cyto RelA
Nuc IB
Cyto IB
0
1
2
3
4 2.6
2.1
1.6
1.1
0.6
0.1
Minutes
F E G
A
B
C D
0 35 72 200 387 507 687 800
0
50
100
150
200
250
300
350
T
i
m
e

(
m
i
n
)
0
50
100
150
200
250
300
350
pB-luc
pB-lB
pB-luc
pB-lB
Fig. 3. NF-0Bdirected oscillations in I0B" expression. Experimental and
computational analysis of factors affecting the amplitude and period of
oscillations. (A, B, E, and F) HeLa cells were transfected to express RelA-
DsRed-Express and I0B"-EGFP under the control of either the consensus
0B promoter or a control 0B promoter vector. Cells were stimulated with
continual 10 ng/ml TNF". (A) Confocal time course of one typical cell
showing oscillations in both RelA-DsRed-Express (red) localization and
I0B"-EGFP (green) expression. Scale bar, 50 6m. (B) Analysis of three
typical cells showing RelA-DsRed-Express N:C ratio and cytoplasmic and
nuclear I0B"-EGFP levels. (C) The simulated time-dependent nuclear
localization of NF-0B for successively increasing the NF-0Bregulated
I0B" transcription rate constant by two orders of magnitude on either
side of the standard rate constant (reaction 28 in the computational
model, table S1) is shown by 41 lines changing in regularly increasing log
intervals from blue to green to yellow to red (scanned after
equilibration). (D) The peak amplitudes (A1 to A6) and timings (T1 to
T6) of the first six simulated peaks for different rate constant values for
NF-0B regulated I0B" transcription [as determined from data in (C)]. (E)
Experimentally determined relative amplitude (N:C ratio) of successive
RelA-DsRed-Express oscillations in HeLa cells continually stimulated with
TNF". Peak 1 set to 100%; subsequent peaks show relative amplitude
TSEM). (F) Average timing between successive peaks (TSEM) of
successive N-C oscillations in RelA-DsRed-Express. (G) Simulated peak
timings for 1x, 2x, 5x, and 10x standard reaction rate constant for NF-
0Bregulated I0B" transcription (reaction 28 in computational model,
table S1). The parameter was changed before the equilibration period.
R E P O R T S
3. H. Sakurai, H. Chiba, H. Miyoshi, T. Sugita, W. Toriumi,
J. Biol. Chem. 274, 30353 (1999).
4. X. Jiang, N. Takahashi, N. Matsui, T. Tetsuka, T. Okamoto,
J. Biol. Chem. 278, 919 (2003).
5. S. C. Sun, P. A. Ganchi, D. W. Ballard, W. C. Greene,
Science 259, 1912 (1993).
6. M. W. Young, S. A. Kay, Nature Rev. Genet. 2, 702
(2001).
7. G. Lahav et al., Nature Genet. 36, 147 (2004).
8. N. A. Monk, Curr. Biol. 13, 1409 (2003).
9. O. Pourquie , Science 301, 328 (2003).
10. F. Arenzana-Seisdedos et al., Mol. Cell. Biol. 15,
2689 (1995).
11. G. Nelson et al., J. Cell Sci. 115, 1137 (2002).
12. E. Zandi, Y. Chen, M. Karin, Science 281, 1360
(1998).
13. A. Hoffmann, A. Levchenko, M. L. Scott, D. Baltimore,
Science 298, 1241 (2002).
14. A. Y. Ting, D. Endy, Science 298, 1189 (2002).
15. M. J. Berridge, Nature 386, 759 (1997).
16. G. Nelson et al., J. Cell Sci. 116, 2495 (2003).
17. X. Bian et al., J. Biol. Chem. 277, 42144 (2002).
18. D. W. McFerran et al., Endocrinology 142, 3255
(2001).
20. L. Vermeulen, G. De Wilde, S. Notebaert, W. Vanden
Berghe, G. Haegeman, Biochem. Pharmacol. 64, 963
(2002).
21. L. Chen, W. Fischle, E. Verdin, W. C. Greene, Science
293, 1653 (2001).
22. A. Ryo et al., Mol. Cell 12, 1413 (2003).
23. H. Sakurai et al., J. Biol. Chem. 278, 36916 (2003).
24. J. Yang, G. H. Fan, B. E. Wadzinski, H. Sakurai, A. Richmond,
J. Biol. Chem. 276, 47828 (2001).
25. R. E. Dolmetsch, R. S. Lewis, C. C. Goodnow, J. I. Healy,
Nature 386, 855 (1997).
26. R. S. Lewis, Biochem. Soc. Trans. 31, 925 (2003).
27. D. B. Kell, Curr. Opin. Microbiol. 7, 296 (2004).
28. L. F. Chen, W. C. Greene, J. Mol. Med. 81, 549
(2003).
29. This work was supported by the Merseyside Neuro-
blastoma Research Fund, Alder Hey Oncology Fund,
North West Cancer Research Fund, AstraZeneca,
Medical Research Council, Department of Trade and
Industry, Engineering and Physical Sciences Research
Council, Royal Society of Chemistry, Biotechnology
and Biological Sciences Research Council, and Pfizer
UK. Carl Zeiss, Hamamatsu Photonics, and Kinetic
Imaging provided technical support. We thank A.
Hoffmann and A. Levchenko for assistance with the
NF-0B model and M. Begon, A. Hall, A. Loudon, A.
Millar, H. Rees, J. Turnbull, and the late Ray Paton for
helpful discussions.
DC1
Figs. S1 to S19
Movies S1 to S4
References
P-IB
P-ReIA
ReIA
IB
P-IB
P-ReIA
ReIA
IB
0 5 10 15 20 30 60 90 120 150 180 210 240 270
0 5 10 15 20 30 60 90 120 150 180 210 240 270
A
B
2.2
2
1.8
1.6
1.4
1.2
1
R
L
U
0 200 400 600
Minutes
Time (min) after TNF+LMB
Time (h) after TNF
R
e
l
A

A
m
p
l
i
t
u
d
e
D
E
0 50 100 150 300 600
C
0
25
50
75
100
10 5 0 15 20 25
Fig. 4. Effect of nuclear export inhibition on the dynamics of RelA localization, 0B-dependent
reporter gene expression, and NF-0B phosphoprotein expression. SK-N-AS cells were treated with
continuous 10 ng/ml TNF" and 10 ng/ml LMB (unless stated). (A) Time-lapse confocal images of
RelA-EGFP localization. (B) Time course of RelA-EGFP localization expressed as N:C fluorescence
ratio (each colored line represents data from one of four single cells). (C) 0B-dependent luciferase
reporter gene expression (each colored line represents data from one of four single cells, and the
black line represents the average). (D) Western blot analysis of Ser32 phospho-I0B" (P-I0B"), total
I0B" (I0B"), Ser536 phospho-RelA (P-RelA), and total RelA (RelA) protein levels in SK-N-AS cells
stimulated with continual 10 ng/ml TNF" for the indicated times before analysis. (E) Western blot
analysis of SK-N-AS cells stimulated with continual 10 ng/ml TNF" and 18 nM LMB for the
indicated times before analysis.
R E P O R T S
709
T CELL KIT
The T Cell Activation/Expansion Kit
consists of Anti-Biotin MACSiBead
particles and biotinylated antibod-
ies against human CD2, CD3, and
CD28. Anti-Biotin MACSiBead par-
ticles loaded with biotinylated antibodies are used to mimic
antigen-presenting cells and activate resting T cells from periph-
eral blood mononuclear cells as well as purified T cells. T cell ex-
pansion is achieved by culturing and reactivation at day 14 of
culture. Activated T cells can be used for any downstream pro-
cessing such as cytokine analysis or immunoprecipitation, and
can be efficiently transfected. The kit is also suitable for the ex-
pansion of antigen-specific T cell lines and clones.
GENE EXPRESSIONASSAYS
The BD ProLabel Screening and
Detection Kits provide sensitive
chemi l umi nescence assays for
measuring the expression levels of
any gene of interest in total cell
lysates. These fast, quantitative assay kits were designed to pro-
vide a functional assay for protein knockdown RNA interference
studies. The assay provides an excellent dynamic range, high sen-
sitivity, and a low threshold of detection. Crude cell lysates can
be tested without any purification. These assays can be easily run
in a high-throughput screen.
NMR SAMPLE RAIL
The Sample Rail enables fully auto-
mated sample preparation and
delivery for 24-hour, hands-off, high-
throughput nuclear magnetic reso-
nance (NMR) operation. The system
interfaces with a liquid handling robot, which prepares samples in
NMR tubes and then transports them on a rail system to the mag-
net. The liquid handler is synchronized with the NMR experiments
for just-in-time sample preparation, suitable for samples requiring
analysis immediately after preparation. The Sample Rail increases
the utility of NMR in various high-throughput applications, such as
metabolic studies, structure activity relations, protein-ligand bind-
ing detection, and other drug discovery methods. The Sample Rail is
compatible with all Bruker BioSpin Avance NMR spectrometers.
RNA INTERFERENCE STUDIES
The QArrayMini microarrayer and
aQuire microarray scanner with
blue laser option combine to carry
out RNA interference optimization
experiments. RNA interference is a
powerful molecular biology tool to achieve targeted gene silenc-
ing in cells. Double-stranded small-interfering RNA (siRNA)
molecules can be introduced to silence genes. It is usually neces-
sary to test a variety of oligonucleotide sequences for their effica-
cy in silencing a given gene, that can be done in a parallel and
high-throughput manner by microarraying. The synthetic siRNA
duplexes or the plasmids which are eventually processed into
siRNA molecules in the cell can be arrayed with a matrix of lipid
transfection regents onto a microarray slide using the QArrayMini
or QArray2. The slide is then overcoated with a monolayer of ad-
herent cells that subsequently become transfected upon contact
with the arrayed spots. In the case of an RNAi screen in which the
cell line is expressing a green fluorescent protein (GFP) fusion
protein to the target gene, the efficacy of each siRNA molecule is
assessed by the formation of plaques (nonfluorescent regions) in
the lawn of fluorescent GFP-expressing cells. Images of GFP fu-
sion protein expressing cells can be acquired and analyzed using
the Genetix aQuire scanner with optional blue laser.
PCR PRODUCT PURIFICATION
The 96-well Montage PCR96 clean-
up plates are for small-volume poly-
merase chain reaction (PCR) product
purification. The plates incorporate
Millipores patent-pending size-
exclusion technology to process
PCR reaction volumes from 1 to
150 l . The mi crowel l pl ate
design is optimized to purify,
concentrate, and recover PCR
products in as little as 20 l. The
15-min, vacuum-driven protocol
removes more than 99% of
primers and deoxynucleotide
triphosphates (dNTPs) in one
step. The plates yield purified
PCR product suitable for sensi-
tive genotyping applications,
direct sequencing, and micro-
array spotting.
GENE SILENCING REAGENTS
Two new small-interfering RNA
(siRNA) delivery reagents, TransIT-
siQuest and TransIT-TKO, are de-
signed to achieve optimal delivery
and high-efficiency target knock-
down in a broad range of cell types, including primary cells. They
are serum-compatible, one-step transfection reagents.
ENGINEERED M-MLV RT
ArrayScript reverse transcriptase is a
modified enzyme that produces up
to twice as much complementary
RNA compared with wild-type
Moloney murine leukemia virus (M-
MLV) and other engineered M-MLV reverse transcriptases when low
amounts of total RNA are used (~100 ng). It is suitable for applica-
tions such as complementary DNA (cDNA) library construction,
RNA ligase mediated-rapid amplification of cDNA ends, and RNA
amplification in which high yields of full-length cDNA are critical.
Newly offered instrumentation, apparatus, and laboratory materials of interest to re-
searchers in all disciplines in academic, industrial, and government organizations are
featured in this space. Emphasis is given to purpose, chief characteristics, and availabil-
ity of products and materials. Endorsement by Science or AAAS of any products or
materials mentioned is not implied. Additional information may be obtained from the
manufacturer or supplier by visiting http://science.labvelocity.com on the Web, where
you can request that the information be sent to you by e-mail, fax, mail, or telephone.
NE W PRODUCTS
Miltenyi Biotec
For more information
+49 2204-8306-0
www.MiltenyiBiotec.com
http://science.labvelocity.com
Mirus
608-441-2824
www.mirusbio.com
Ambion
800-888-8804
www.ambion.com
Millipore
800-MILLIPORE
www.millipore.com/montage
BD Biosciences
650-424-8222
www.bdbiosciences.com/clontech
Bruker
978-667-9580
www.bruker-biospin.com
Genetix
+44 1425 624600
www.genetix.com
Published by AAAS

Science Magazine 5696 2004-10-22

Uploaded by

Copyright:

Available Formats

Science Magazine 5696 2004-10-22

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Science Magazine 5696 2004-10-22

Uploaded by

Copyright:

Available Formats

EDITORIAL

www.sciencemag.org SCIENCE VOL 306 22 OCTOBER 2004 575

cells (DNaseI HS/NHGRI) or from K562 cells

) was excited through the e

(s) transition with

and 400 to 225 femtoseconds for (D

; the lifetimes decrease

), which invokes a 50-

(s)^ can be transiently confined

(s) state to the

( p) states (4, 5). The electron-

(s) excitation are more controversial, how-

, with 25 e n e 50. Water cluster

) was excited through the e

(s) transition with

and 400 to 225 femtoseconds for (D

; the lifetimes decrease

), which invokes a 50-

(s)^ can be transiently confined

(s) state to the

( p) states (4, 5). The electron-

(s) excitation are more controversial, how-

, with 25 e n e 50. Water cluster

clusters becomes appar-

( p) cluster state. TRPE

and all other clusters

(p) state (requir-

are plotted together (Fig. 3). These

. Features highlighted in (A) are dis-

clusters studied here (Fig. 4). This trend

VDEs, obtained from anion photo-

(n , 50) with significantly lower

. Size-dependent decay life-

clusters in the intramolecular bending

in the bending region

ion pair configuration, which

Photon Energy (cm

, reproduced from (29);

colonies (fig. S8)] and was decreased

colonies in the absence of

(Fig. 4A), and 950% of

events was substantially decreased

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.